Squashed 'third_party/abseil/' changes from ddf8e52a2..384af0e91
384af0e91 Export of internal Abseil changes
e7ca23aca Export of internal Abseil changes
4611a601a Export of internal Abseil changes
8a9ef3c5d Export of internal Abseil changes
9f8b87b71 Add missing word 'library' in the 'status' description (#868)
e2b1bab19 Export of internal Abseil changes
1bae23e32 Export of internal Abseil changes
6df644c56 Include the status library into the main README. (#863)
68f1ad932 Export of internal Abseil changes
52acfe6fc Export of internal Abseil changes
1918ad2ae Export of internal Abseil changes
938fd0f4e Export of internal Abseil changes
fbdff6f3a Export of internal Abseil changes
acf3390ca Export of internal Abseil changes
592924480 Export of internal Abseil changes
e80c0b353 Export of internal Abseil changes
5d8fc9192 Export of internal Abseil changes
e19260fd7 Export of internal Abseil changes
4fd9a1ec5 Export of internal Abseil changes
4ae673067 Export of internal Abseil changes
1b465af3b Export of internal Abseil changes
6b03bf543 fix build dll (#797)
0bbebc85c Export of internal Abseil changes
0453e1653 Export of internal Abseil changes
a4798817e Export of internal Abseil changes
e96d49687 Export of internal Abseil changes
731852f10 Fix stacktrace on aarch64 architecture. Fixes #805 (#827)
2e5f2bcfd moved deleted functions to public for better compiler errors. (#828)
e9e9b9fc7 Export of internal Abseil changes
b8e890f95 Export of internal Abseil changes
c9894d1dc Export of internal Abseil changes
e9b9e38f6 Export of internal Abseil changes
962b06754 Export of internal Abseil changes
5bf048b84 Export of internal Abseil changes
1e3d25b26 Export of internal Abseil changes
eb317a701 Export of internal Abseil changes
4b915e709 Export of internal Abseil changes
8f1c34a77 Export of internal Abseil changes
60d00a582 Export of internal Abseil changes
f3f785ab5 Export of internal Abseil changes
4b2fbb4ad Export of internal Abseil changes
e493d6acb fix compile fails with asan and -Wredundant-decls (#801)
e63a5a610 Export of internal Abseil changes
c678d6c6b Export of internal Abseil changes
4b4f9aae7 Export of internal Abseil changes
887d0eee6 Export of internal Abseil changes
b978fc02f Export of internal Abseil changes
093cc2760 Export of internal Abseil changes
40fdb59d3 btree: fix sign-compare warnings (#800)
1fd58b69c Export of internal Abseil changes
d1de75bf5 Export of internal Abseil changes
cad3f30b4 Export of internal Abseil changes
9927a0989 Export of internal Abseil changes
7680a5f8e Added missing asserts for seq.index() < capacity_ and unified their usage based on has_element(). (#781)
d3614de61 Export of internal Abseil changes
20feb1cdb Export of internal Abseil changes
c1ae0a497 Export of internal Abseil changes
6af91b351 Export of internal Abseil changes
f2c9c663d Export of internal Abseil changes
3c8b5d758 Export of internal Abseil changes
7ba8cdb56 Export of internal Abseil changes
930fbec75 Export of internal Abseil changes
0e9921b75 Export of internal Abseil changes
a4cbb5f69 Export of internal Abseil changes
4d2ff381a Export of internal Abseil changes
c03c18e7f Export of internal Abseil changes
b321ad86c Export of internal Abseil changes
fbf0fdab6 Export of internal Abseil changes
dc969f34a Export of internal Abseil changes
d0c433455 Export of internal Abseil changes
c6b3f2cf5 Export of internal Abseil changes
1beb3191c Export of internal Abseil changes
1b7e751e5 Export of internal Abseil changes
ce4bc9277 Export of internal Abseil changes
f72cc3516 Export of internal Abseil changes
f66bc7492 Export of internal Abseil changes
1995c6a3c Export of internal Abseil changes
184cf2524 Export of internal Abseil changes
82302f1e0 Export of internal Abseil changes
dea76486c Export of internal Abseil changes
d39fe6cd6 Export of internal Abseil changes
2c8a5b0d8 Export of internal Abseil changes
41a6263fd Export of internal Abseil changes
3c2bed2e7 Export of internal Abseil changes
ea8a689cf fix build on P9 (#739)
672d9e0ae Export of internal Abseil changes
f624790b7 Export of internal Abseil changes
55c04eb92 Export of internal Abseil changes
302b250e1 Disable pthread for standalone wasm build support (#721)
61d8bc057 Merge branch 'master' of https://github.com/abseil/abseil-cpp into master
4b5b25a28 cmake: remove unneeded enable_testing() cmd (#736)
63f2c695e cmake: flag conformance_testing as TESTONLY (#737)
d5269a8b6 Export of internal Abseil changes
23f1f9cf6 Typo in comment (#733)
259301a52 Fix CMake path for INSTALL_INCLUDEDIR (#723)
dc464c1dc Allow overriding Abseil IDE folder (#724)
bf655de09 Export of internal Abseil changes
38db52adb Export of internal Abseil changes
81f34df83 Export of internal Abseil changes
b86fff162 Export of internal Abseil changes
10cb35e45 Export of internal Abseil changes
4ccc0fce0 Export of internal Abseil changes
4a851046a Export of internal Abseil changes
ccdbb5941 Export of internal Abseil changes
01f5f81f9 Export of internal Abseil changes
2c92bdc7c Export of internal Abseil changes
e7ebf9803 Export of internal Abseil changes
2eba343b5 Export of internal Abseil changes
a8b03d90e Export of internal Abseil changes
1d31b5c36 Export of internal Abseil changes
da3a87690 Export of internal Abseil changes
8faf20461 Exclude empty directories (#697)
2069dc796 Export of internal Abseil changes
4832bf6bf Added a BUILD file in root to expose license. (#695)
af8f994af Export of internal Abseil changes
33caf1097 Export of internal Abseil changes
cf1a02e2d Export of internal Abseil changes
768eb2ca2 Export of internal Abseil changes
3f347c462 Fix build on riscv32 (#675)
62cf6a704 Export of internal Abseil changes
d118d4bb1 Export of internal Abseil changes
f2bc9d11e Fix public target name of the random library (#684)
0fecf0e63 Export of internal Abseil changes
cbfd0f0fe Export of internal Abseil changes
c45d1c09d Export of internal Abseil changes
a35ef8a62 Export of internal Abseil changes
bd317cae3 Export of internal Abseil changes
b11574465 fix MSVC warning 4245: conversion signed => unsigned during initialization (#678)
d85783fd0 Export of internal Abseil changes
a1d668990 Export of internal Abseil changes
ca9856cab Export of internal Abseil changes
6e18c7115 Export of internal Abseil changes
3f48ce1c4 init (#673)
cde2e2410 Export of internal Abseil changes
68494aae9 Fix CMake Threads dependency issue
902909a43 Export of internal Abseil changes
cb52b05ea Export of internal Abseil changes
1a02b7a20 Use "-lrt" instead of the resolved find_library result when linking librt (#665)
df60c82df Export of internal Abseil changes
b35973e3e Export of internal Abseil changes
db5773a72 Export of internal Abseil changes
71079e42c Export of internal Abseil changes
2946ac0de Use base_internal::AtomicHook instead of std::atomic (#661)
567bee2f7 Fix ABSL_RANDOM_RANDEN_COPTS setting on FreeBSD (#664)
bf6166a63 Export of internal Abseil changes
111260963 Export of internal Abseil changes
73ea9a957 Export of internal Abseil changes
c01b9916e Add option to use an externally provided GoogleTest target (for usage of abseil as add_subdirectory target) (#647)
d43b7997c Export of internal Abseil changes
62f05b1f5 Export of internal Abseil changes
fba8a316c Export of internal Abseil changes
79e0dc115 Export of internal Abseil changes
132d791b4 bazel: Add missing load statements for cc_binary (#645)
518f17501 Export of internal Abseil changes
092ed9793 Export of internal Abseil changes
2d2a8aea2 Export of internal Abseil changes
7853a7586 Export of internal Abseil changes
c6954897f Export of internal Abseil changes
b92f35f65 Fix CompressedTuple move constructor on MSVC (#637)
a877af1f2 Export of internal Abseil changes
d936052d3 Export of internal Abseil changes
238b9a59c Skip the .exe suffix in the helpshort filter on Windows (#629)
417ea99cb UWP doesn't allow reading regkeys (#594)
40a0e58eb Export of internal Abseil changes
cf3a1998e Export of internal Abseil changes
b19ba9676 Export of internal Abseil changes
06f0e767d BuildBreak: UWP apps can't call GetModuleHandle (#596)
bcefbdcdf Export of internal Abseil changes
0033c9ea9 Fix build on FreeBSD/powerpc (#616)
0d5ce2797 Export of internal Abseil changes
b69c7d880 Export of internal Abseil changes
2a5633fc0 Merge "Export of internal Abseil changes"
f9b3d6e49 Add RISCV support to GetProgramCounter() (#621)
914ff4451 Export of internal Abseil changes
0232c87f2 Add missing ABSL_HAVE_VDSO_SUPPORT conditional (#622)
3c8141051 Export of internal Abseil changes
c44657f55 Export of internal Abseil changes
98eb410c9 Export of internal Abseil changes
bf78e9773 Export of internal Abseil changes
d95d15671 Export of internal Abseil changes
24713a703 Export of internal Abseil changes
72382c21f Export of internal Abseil changes
08a7e7bf9 Export of internal Abseil changes
36bcd9599 Fix pointer format specifier in documentation (#614)
0f86336b6 Export of internal Abseil changes
c512f118d Export of internal Abseil changes
37dd2562e Export of internal Abseil changes
444277026 fix: Add support for more ARM processors detection (#608)
159bf2bf6 Export of internal Abseil changes
a2e6adecc Use https links. (#586)
564001ae5 Export of internal Abseil changes
b3aaac8a3 Export of internal Abseil changes
63ee2f887 Export of internal Abseil changes
a048203a8 Export of internal Abseil changes
1de016636 Export of internal Abseil changes
ad904b6cd Export of internal Abseil changes
7bd1935dc cmake: Fix x86_64 check on Windows for random copts (#518)
292351391 Export of internal Abseil changes
bf86cfe16 Export of internal Abseil changes
12bc53e03 Export of internal Abseil changes
1e39f8626 Export of internal Abseil changes
77f87009a Export of internal Abseil changes
d659fe54b Export of internal Abseil changes
a4b757b5d Export of internal Abseil changes
0514227d2 Export of internal Abseil changes
7f4fe64af Export of internal Abseil changes
16d9fd58a Export of internal Abseil changes
bcaae6009 Export of internal Abseil changes
8ba96a824 Export of internal Abseil changes
2103fd9ac Export of internal Abseil changes
3df7b52a6 Export of internal Abseil changes
fa8c75182 Export of internal Abseil changes
85092b4b6 Fix Conan builds (#400)
e96ae2203 Export of internal Abseil changes
20de2db74 Export of internal Abseil changes
846e5dbed Export of internal Abseil changes
83880e3d8 Merge branch 'master' of https://github.com/abseil/abseil-cpp
8207907f4 Export of internal Abseil changes
39d68a422 docs: fix typo (#397)
078b89b3c Export of internal Abseil changes
19b021cb3 Export of internal Abseil changes
ecc0033b5 Always enable proper symbolize implementation on Windows (#257)
2796d500a Export of internal Abseil changes
e4c8d0eb8 Export of internal Abseil changes
a15364ce4 Export of internal Abseil changes
ab3552a18 Export of internal Abseil changes
e9f9000c7 Fix ABSL_WAITER_MODE detection for mingw (#342)
abea769b5 Fix ABSL_HAVE_ALARM check on mingw (#341)
25597bdfc Export of internal Abseil changes
aad33fefa Export of internal Abseil changes
8fe7214fe Export of internal Abseil changes
debac94cf Export of internal Abseil changes
882b3501a Fix spelling errors (#384)
502efe6d7 Export of internal Abseil changes
ccdd1d57b Export of internal Abseil changes
Change-Id: I59864a0053f6e03d88cc9d5e2a92757039a05484
git-subtree-dir: third_party/abseil
git-subtree-split: 384af0e9141283172e2bff3210dae79fb7130d9c
diff --git a/absl/strings/BUILD.bazel b/absl/strings/BUILD.bazel
index eae75d8..7001679 100644
--- a/absl/strings/BUILD.bazel
+++ b/absl/strings/BUILD.bazel
@@ -25,7 +25,7 @@
features = ["parse_headers"],
)
-licenses(["notice"]) # Apache 2.0
+licenses(["notice"])
cc_library(
name = "strings",
@@ -54,6 +54,7 @@
"ascii.h",
"charconv.h",
"escaping.h",
+ "internal/string_constant.h",
"match.h",
"numbers.h",
"str_cat.h",
@@ -68,7 +69,6 @@
deps = [
":internal",
"//absl/base",
- "//absl/base:bits",
"//absl/base:config",
"//absl/base:core_headers",
"//absl/base:endian",
@@ -76,6 +76,7 @@
"//absl/base:throw_delegate",
"//absl/memory",
"//absl/meta:type_traits",
+ "//absl/numeric:bits",
"//absl/numeric:int128",
],
)
@@ -83,19 +84,23 @@
cc_library(
name = "internal",
srcs = [
+ "internal/escaping.cc",
"internal/ostringstream.cc",
"internal/utf8.cc",
],
hdrs = [
"internal/char_map.h",
+ "internal/escaping.h",
"internal/ostringstream.h",
"internal/resize_uninitialized.h",
"internal/utf8.h",
],
copts = ABSL_DEFAULT_COPTS,
deps = [
+ "//absl/base:config",
"//absl/base:core_headers",
"//absl/base:endian",
+ "//absl/base:raw_logging_internal",
"//absl/meta:type_traits",
],
)
@@ -122,6 +127,7 @@
copts = ABSL_TEST_COPTS,
visibility = ["//visibility:private"],
deps = [
+ ":cord",
":strings",
"//absl/base:core_headers",
"//absl/container:fixed_array",
@@ -218,6 +224,19 @@
)
cc_test(
+ name = "string_constant_test",
+ size = "small",
+ srcs = ["internal/string_constant_test.cc"],
+ copts = ABSL_TEST_COPTS,
+ visibility = ["//visibility:private"],
+ deps = [
+ ":strings",
+ "//absl/meta:type_traits",
+ "@com_google_googletest//:gtest_main",
+ ],
+)
+
+cc_test(
name = "string_view_benchmark",
srcs = ["string_view_benchmark.cc"],
copts = ABSL_TEST_COPTS,
@@ -246,6 +265,88 @@
],
)
+cc_library(
+ name = "cord_internal",
+ srcs = [
+ "internal/cord_internal.cc",
+ ],
+ hdrs = [
+ "internal/cord_internal.h",
+ "internal/cord_rep_flat.h",
+ ],
+ copts = ABSL_DEFAULT_COPTS,
+ visibility = [
+ "//visibility:private",
+ ],
+ deps = [
+ ":strings",
+ "//absl/base:base_internal",
+ "//absl/base:core_headers",
+ "//absl/container:compressed_tuple",
+ "//absl/container:inlined_vector",
+ "//absl/meta:type_traits",
+ ],
+)
+
+cc_library(
+ name = "cord",
+ srcs = [
+ "cord.cc",
+ ],
+ hdrs = [
+ "cord.h",
+ ],
+ copts = ABSL_DEFAULT_COPTS,
+ deps = [
+ ":cord_internal",
+ ":internal",
+ ":str_format",
+ ":strings",
+ "//absl/base",
+ "//absl/base:core_headers",
+ "//absl/base:endian",
+ "//absl/base:raw_logging_internal",
+ "//absl/container:fixed_array",
+ "//absl/container:inlined_vector",
+ "//absl/functional:function_ref",
+ "//absl/meta:type_traits",
+ "//absl/types:optional",
+ ],
+)
+
+cc_library(
+ name = "cord_test_helpers",
+ testonly = 1,
+ hdrs = [
+ "cord_test_helpers.h",
+ ],
+ copts = ABSL_DEFAULT_COPTS,
+ deps = [
+ ":cord",
+ ],
+)
+
+cc_test(
+ name = "cord_test",
+ size = "medium",
+ srcs = ["cord_test.cc"],
+ copts = ABSL_TEST_COPTS,
+ visibility = ["//visibility:private"],
+ deps = [
+ ":cord",
+ ":cord_test_helpers",
+ ":str_format",
+ ":strings",
+ "//absl/base",
+ "//absl/base:config",
+ "//absl/base:core_headers",
+ "//absl/base:endian",
+ "//absl/base:raw_logging_internal",
+ "//absl/container:fixed_array",
+ "@com_google_googletest//:gtest_main",
+ ],
+)
+
cc_test(
name = "substitute_test",
size = "small",
@@ -293,6 +394,8 @@
":strings",
"//absl/base:core_headers",
"//absl/base:dynamic_annotations",
+ "//absl/container:flat_hash_map",
+ "//absl/container:node_hash_map",
"@com_google_googletest//:gtest_main",
],
)
@@ -411,10 +514,13 @@
copts = ABSL_TEST_COPTS,
visibility = ["//visibility:private"],
deps = [
+ ":internal",
":pow10_helper",
":strings",
- "//absl/base:core_headers",
+ "//absl/base:config",
"//absl/base:raw_logging_internal",
+ "//absl/random",
+ "//absl/random:distributions",
"@com_google_googletest//:gtest_main",
],
)
@@ -428,6 +534,8 @@
deps = [
":strings",
"//absl/base:raw_logging_internal",
+ "//absl/random",
+ "//absl/random:distributions",
"@com_github_google_benchmark//:benchmark_main",
],
)
@@ -486,6 +594,7 @@
copts = ABSL_TEST_COPTS,
deps = [
":strings",
+ "//absl/base:config",
"//absl/base:raw_logging_internal",
"@com_google_googletest//:gtest_main",
],
@@ -501,6 +610,7 @@
copts = ABSL_TEST_COPTS,
deps = [
":strings",
+ "//absl/base:config",
"@com_google_googletest//:gtest_main",
],
)
@@ -555,8 +665,11 @@
":strings",
"//absl/base:config",
"//absl/base:core_headers",
+ "//absl/functional:function_ref",
"//absl/meta:type_traits",
+ "//absl/numeric:bits",
"//absl/numeric:int128",
+ "//absl/types:optional",
"//absl/types:span",
],
)
@@ -567,6 +680,7 @@
copts = ABSL_TEST_COPTS,
visibility = ["//visibility:private"],
deps = [
+ ":cord",
":str_format",
":strings",
"//absl/base:core_headers",
@@ -584,6 +698,7 @@
deps = [
":str_format",
":str_format_internal",
+ ":strings",
"@com_google_googletest//:gtest_main",
],
)
@@ -630,7 +745,9 @@
visibility = ["//visibility:private"],
deps = [
":str_format_internal",
- "//absl/numeric:int128",
+ ":strings",
+ "//absl/base:raw_logging_internal",
+ "//absl/types:optional",
"@com_google_googletest//:gtest_main",
],
)
@@ -641,6 +758,7 @@
copts = ABSL_TEST_COPTS,
visibility = ["//visibility:private"],
deps = [
+ ":cord",
":str_format_internal",
"@com_google_googletest//:gtest_main",
],
@@ -664,6 +782,7 @@
srcs = ["internal/pow10_helper.cc"],
hdrs = ["internal/pow10_helper.h"],
visibility = ["//visibility:private"],
+ deps = ["//absl/base:config"],
)
cc_test(
diff --git a/absl/strings/CMakeLists.txt b/absl/strings/CMakeLists.txt
index ccff444..654d0fd 100644
--- a/absl/strings/CMakeLists.txt
+++ b/absl/strings/CMakeLists.txt
@@ -21,6 +21,7 @@
"ascii.h"
"charconv.h"
"escaping.h"
+ "internal/string_constant.h"
"match.h"
"numbers.h"
"str_cat.h"
@@ -72,6 +73,8 @@
strings_internal
HDRS
"internal/char_map.h"
+ "internal/escaping.cc"
+ "internal/escaping.h"
"internal/ostringstream.h"
"internal/resize_uninitialized.h"
"internal/utf8.h"
@@ -81,8 +84,10 @@
COPTS
${ABSL_DEFAULT_COPTS}
DEPS
+ absl::config
absl::core_headers
absl::endian
+ absl::raw_logging_internal
absl::type_traits
)
@@ -156,6 +161,19 @@
absl_cc_test(
NAME
+ string_constant_test
+ SRCS
+ "internal/string_constant_test.cc"
+ COPTS
+ ${ABSL_TEST_COPTS}
+ DEPS
+ absl::strings
+ absl::type_traits
+ gmock_main
+)
+
+absl_cc_test(
+ NAME
string_view_test
SRCS
"string_view_test.cc"
@@ -206,6 +224,8 @@
absl::base
absl::core_headers
absl::dynamic_annotations
+ absl::flat_hash_map
+ absl::node_hash_map
gmock_main
)
@@ -276,7 +296,11 @@
absl::strings
absl::core_headers
absl::pow10_helper
+ absl::config
absl::raw_logging_internal
+ absl::random_random
+ absl::random_distributions
+ absl::strings_internal
gmock_main
)
@@ -329,6 +353,7 @@
${ABSL_TEST_COPTS}
DEPS
absl::strings
+ absl::config
absl::raw_logging_internal
gmock_main
)
@@ -344,6 +369,7 @@
${ABSL_TEST_COPTS}
DEPS
absl::strings
+ absl::config
gmock_main
)
@@ -380,6 +406,7 @@
COPTS
${ABSL_DEFAULT_COPTS}
DEPS
+ absl::bits
absl::strings
absl::config
absl::core_headers
@@ -397,6 +424,7 @@
${ABSL_TEST_COPTS}
DEPS
absl::str_format
+ absl::cord
absl::strings
absl::core_headers
gmock_main
@@ -412,6 +440,7 @@
DEPS
absl::str_format
absl::str_format_internal
+ absl::strings
gmock_main
)
@@ -460,7 +489,9 @@
COPTS
${ABSL_TEST_COPTS}
DEPS
+ absl::strings
absl::str_format_internal
+ absl::raw_logging_internal
absl::int128
gmock_main
)
@@ -474,6 +505,7 @@
${ABSL_TEST_COPTS}
DEPS
absl::str_format_internal
+ absl::cord
gmock_main
)
@@ -499,6 +531,8 @@
"internal/pow10_helper.cc"
COPTS
${ABSL_TEST_COPTS}
+ DEPS
+ absl::config
TESTONLY
)
@@ -514,3 +548,64 @@
absl::str_format
gmock_main
)
+
+absl_cc_library(
+ NAME
+ cord
+ HDRS
+ "cord.h"
+ SRCS
+ "cord.cc"
+ "internal/cord_internal.cc"
+ "internal/cord_internal.h"
+ "internal/cord_rep_flat.h"
+ COPTS
+ ${ABSL_DEFAULT_COPTS}
+ DEPS
+ absl::base
+ absl::base_internal
+ absl::compressed_tuple
+ absl::core_headers
+ absl::endian
+ absl::fixed_array
+ absl::function_ref
+ absl::inlined_vector
+ absl::optional
+ absl::raw_logging_internal
+ absl::strings
+ absl::strings_internal
+ absl::type_traits
+ PUBLIC
+)
+
+absl_cc_library(
+ NAME
+ cord_test_helpers
+ HDRS
+ "cord_test_helpers.h"
+ COPTS
+ ${ABSL_TEST_COPTS}
+ DEPS
+ absl::cord
+ TESTONLY
+)
+
+absl_cc_test(
+ NAME
+ cord_test
+ SRCS
+ "cord_test.cc"
+ COPTS
+ ${ABSL_TEST_COPTS}
+ DEPS
+ absl::cord
+ absl::str_format
+ absl::strings
+ absl::base
+ absl::config
+ absl::core_headers
+ absl::endian
+ absl::raw_logging_internal
+ absl::fixed_array
+ gmock_main
+)
diff --git a/absl/strings/ascii.cc b/absl/strings/ascii.cc
index 3f7c581..93bb03e 100644
--- a/absl/strings/ascii.cc
+++ b/absl/strings/ascii.cc
@@ -15,6 +15,7 @@
#include "absl/strings/ascii.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace ascii_internal {
// # Table generated by this Python code (bit 0x02 is currently unused):
@@ -56,7 +57,7 @@
// of these bits is tightly coupled to this implementation, the individual bits
// are not named. Note that bitfields for all characters above ASCII 127 are
// zero-initialized.
-const unsigned char kPropertyBits[256] = {
+ABSL_DLL const unsigned char kPropertyBits[256] = {
0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, // 0x00
0x40, 0x68, 0x48, 0x48, 0x48, 0x48, 0x40, 0x40,
0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, // 0x10
@@ -78,7 +79,7 @@
// Array of characters for the ascii_tolower() function. For values 'A'
// through 'Z', return the lower-case character; otherwise, return the
// identity of the passed character.
-const char kToLower[256] = {
+ABSL_DLL const char kToLower[256] = {
'\x00', '\x01', '\x02', '\x03', '\x04', '\x05', '\x06', '\x07',
'\x08', '\x09', '\x0a', '\x0b', '\x0c', '\x0d', '\x0e', '\x0f',
'\x10', '\x11', '\x12', '\x13', '\x14', '\x15', '\x16', '\x17',
@@ -116,7 +117,7 @@
// Array of characters for the ascii_toupper() function. For values 'a'
// through 'z', return the upper-case character; otherwise, return the
// identity of the passed character.
-const char kToUpper[256] = {
+ABSL_DLL const char kToUpper[256] = {
'\x00', '\x01', '\x02', '\x03', '\x04', '\x05', '\x06', '\x07',
'\x08', '\x09', '\x0a', '\x0b', '\x0c', '\x0d', '\x0e', '\x0f',
'\x10', '\x11', '\x12', '\x13', '\x14', '\x15', '\x16', '\x17',
@@ -195,4 +196,5 @@
str->erase(output_it - &(*str)[0]);
}
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/ascii.h b/absl/strings/ascii.h
index f9e4fd1..b46bc71 100644
--- a/absl/strings/ascii.h
+++ b/absl/strings/ascii.h
@@ -56,19 +56,21 @@
#include <string>
#include "absl/base/attributes.h"
+#include "absl/base/config.h"
#include "absl/strings/string_view.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace ascii_internal {
// Declaration for an array of bitfields holding character information.
-extern const unsigned char kPropertyBits[256];
+ABSL_DLL extern const unsigned char kPropertyBits[256];
// Declaration for the array of characters to upper-case characters.
-extern const char kToUpper[256];
+ABSL_DLL extern const char kToUpper[256];
// Declaration for the array of characters to lower-case characters.
-extern const char kToLower[256];
+ABSL_DLL extern const char kToLower[256];
} // namespace ascii_internal
@@ -234,6 +236,7 @@
// Removes leading, trailing, and consecutive internal whitespace.
void RemoveExtraAsciiWhitespace(std::string*);
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_ASCII_H_
diff --git a/absl/strings/charconv.cc b/absl/strings/charconv.cc
index bc07e7a..b8674c2 100644
--- a/absl/strings/charconv.cc
+++ b/absl/strings/charconv.cc
@@ -20,7 +20,7 @@
#include <cstring>
#include "absl/base/casts.h"
-#include "absl/base/internal/bits.h"
+#include "absl/numeric/bits.h"
#include "absl/numeric/int128.h"
#include "absl/strings/internal/charconv_bigint.h"
#include "absl/strings/internal/charconv_parse.h"
@@ -57,6 +57,7 @@
// narrower mantissas.
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace {
template <typename FloatType>
@@ -241,11 +242,11 @@
// Returns the bit width of the given uint128. (Equivalently, returns 128
// minus the number of leading zero bits.)
-int BitWidth(uint128 value) {
+unsigned BitWidth(uint128 value) {
if (Uint128High64(value) == 0) {
- return 64 - base_internal::CountLeadingZeros64(Uint128Low64(value));
+ return static_cast<unsigned>(bit_width(Uint128Low64(value)));
}
- return 128 - base_internal::CountLeadingZeros64(Uint128High64(value));
+ return 128 - countl_zero(Uint128High64(value));
}
// Calculates how far to the right a mantissa needs to be shifted to create a
@@ -518,7 +519,7 @@
const strings_internal::ParsedFloat& parsed_hex) {
uint64_t mantissa = parsed_hex.mantissa;
int exponent = parsed_hex.exponent;
- int mantissa_width = 64 - base_internal::CountLeadingZeros64(mantissa);
+ auto mantissa_width = static_cast<unsigned>(bit_width(mantissa));
const int shift = NormalizedShiftSize<FloatType>(mantissa_width, exponent);
bool result_exact;
exponent += shift;
@@ -618,10 +619,10 @@
// Either we failed to parse a hex float after the "0x", or we read
// "0xinf" or "0xnan" which we don't want to match.
//
- // However, a std::string that begins with "0x" also begins with "0", which
+ // However, a string that begins with "0x" also begins with "0", which
// is normally a valid match for the number zero. So we want these
// strings to match zero unless fmt_flags is `scientific`. (This flag
- // means an exponent is required, which the std::string "0" does not have.)
+ // means an exponent is required, which the string "0" does not have.)
if (fmt_flags == chars_format::scientific) {
result.ec = std::errc::invalid_argument;
} else {
@@ -672,7 +673,6 @@
EncodeResult(calculated, negative, &result, &value);
return result;
}
- return result;
}
} // namespace
@@ -980,4 +980,5 @@
};
} // namespace
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/charconv.h b/absl/strings/charconv.h
index 3f5891b..e04be32 100644
--- a/absl/strings/charconv.h
+++ b/absl/strings/charconv.h
@@ -17,7 +17,10 @@
#include <system_error> // NOLINT(build/c++11)
+#include "absl/base/config.h"
+
namespace absl {
+ABSL_NAMESPACE_BEGIN
// Workalike compatibilty version of std::chars_format from C++17.
//
@@ -110,6 +113,7 @@
return lhs;
}
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_CHARCONV_H_
diff --git a/absl/strings/charconv_benchmark.cc b/absl/strings/charconv_benchmark.cc
index 644b2ab..e8c7371 100644
--- a/absl/strings/charconv_benchmark.cc
+++ b/absl/strings/charconv_benchmark.cc
@@ -132,7 +132,7 @@
std::string MakeHardCase(int length) {
// The number 1.1521...e-297 is exactly halfway between 12345 * 2**-1000 and
// the next larger representable number. The digits of this number are in
- // the std::string below.
+ // the string below.
const std::string digits =
"1."
"152113937042223790993097181572444900347587985074226836242307364987727724"
diff --git a/absl/strings/charconv_test.cc b/absl/strings/charconv_test.cc
index b58fad2..b83de5a 100644
--- a/absl/strings/charconv_test.cc
+++ b/absl/strings/charconv_test.cc
@@ -511,6 +511,13 @@
EXPECT_EQ(f, std::numeric_limits<float>::max());
}
+TEST(FromChars, RegressionTestsFromFuzzer) {
+ absl::string_view src = "0x21900000p00000000099";
+ float f;
+ auto result = absl::from_chars(src.data(), src.data() + src.size(), f);
+ EXPECT_EQ(result.ec, std::errc::result_out_of_range);
+}
+
TEST(FromChars, ReturnValuePtr) {
// Check that `ptr` points one past the number scanned, even if that number
// is not representable.
@@ -646,7 +653,9 @@
negative_from_chars_float);
EXPECT_TRUE(std::signbit(negative_from_chars_float));
EXPECT_FALSE(Identical(negative_from_chars_float, from_chars_float));
- from_chars_float = std::copysign(from_chars_float, -1.0);
+ // Use the (float, float) overload of std::copysign to prevent narrowing;
+ // see https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98251.
+ from_chars_float = std::copysign(from_chars_float, -1.0f);
EXPECT_TRUE(Identical(negative_from_chars_float, from_chars_float));
}
}
diff --git a/absl/strings/cord.cc b/absl/strings/cord.cc
new file mode 100644
index 0000000..7cadf75
--- /dev/null
+++ b/absl/strings/cord.cc
@@ -0,0 +1,1825 @@
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "absl/strings/cord.h"
+
+#include <algorithm>
+#include <atomic>
+#include <cstddef>
+#include <cstdio>
+#include <cstdlib>
+#include <iomanip>
+#include <iostream>
+#include <limits>
+#include <ostream>
+#include <sstream>
+#include <type_traits>
+#include <unordered_set>
+#include <vector>
+
+#include "absl/base/casts.h"
+#include "absl/base/internal/raw_logging.h"
+#include "absl/base/macros.h"
+#include "absl/base/port.h"
+#include "absl/container/fixed_array.h"
+#include "absl/container/inlined_vector.h"
+#include "absl/strings/escaping.h"
+#include "absl/strings/internal/cord_internal.h"
+#include "absl/strings/internal/cord_rep_flat.h"
+#include "absl/strings/internal/resize_uninitialized.h"
+#include "absl/strings/str_cat.h"
+#include "absl/strings/str_format.h"
+#include "absl/strings/str_join.h"
+#include "absl/strings/string_view.h"
+
+namespace absl {
+ABSL_NAMESPACE_BEGIN
+
+using ::absl::cord_internal::CordRep;
+using ::absl::cord_internal::CordRepConcat;
+using ::absl::cord_internal::CordRepExternal;
+using ::absl::cord_internal::CordRepFlat;
+using ::absl::cord_internal::CordRepSubstring;
+
+using ::absl::cord_internal::kMinFlatLength;
+using ::absl::cord_internal::kMaxFlatLength;
+
+using ::absl::cord_internal::CONCAT;
+using ::absl::cord_internal::EXTERNAL;
+using ::absl::cord_internal::FLAT;
+using ::absl::cord_internal::RING;
+using ::absl::cord_internal::SUBSTRING;
+
+using ::absl::cord_internal::kInlinedVectorSize;
+using ::absl::cord_internal::kMaxBytesToCopy;
+
+constexpr uint64_t Fibonacci(unsigned char n, uint64_t a = 0, uint64_t b = 1) {
+ return n == 0 ? a : Fibonacci(n - 1, b, a + b);
+}
+
+static_assert(Fibonacci(63) == 6557470319842,
+ "Fibonacci values computed incorrectly");
+
+// Minimum length required for a given depth tree -- a tree is considered
+// balanced if
+// length(t) >= min_length[depth(t)]
+// The root node depth is allowed to become twice as large to reduce rebalancing
+// for larger strings (see IsRootBalanced).
+static constexpr uint64_t min_length[] = {
+ Fibonacci(2), Fibonacci(3), Fibonacci(4), Fibonacci(5),
+ Fibonacci(6), Fibonacci(7), Fibonacci(8), Fibonacci(9),
+ Fibonacci(10), Fibonacci(11), Fibonacci(12), Fibonacci(13),
+ Fibonacci(14), Fibonacci(15), Fibonacci(16), Fibonacci(17),
+ Fibonacci(18), Fibonacci(19), Fibonacci(20), Fibonacci(21),
+ Fibonacci(22), Fibonacci(23), Fibonacci(24), Fibonacci(25),
+ Fibonacci(26), Fibonacci(27), Fibonacci(28), Fibonacci(29),
+ Fibonacci(30), Fibonacci(31), Fibonacci(32), Fibonacci(33),
+ Fibonacci(34), Fibonacci(35), Fibonacci(36), Fibonacci(37),
+ Fibonacci(38), Fibonacci(39), Fibonacci(40), Fibonacci(41),
+ Fibonacci(42), Fibonacci(43), Fibonacci(44), Fibonacci(45),
+ Fibonacci(46), Fibonacci(47),
+ 0xffffffffffffffffull, // Avoid overflow
+};
+
+static const int kMinLengthSize = ABSL_ARRAYSIZE(min_length);
+
+static inline bool IsRootBalanced(CordRep* node) {
+ if (node->tag != CONCAT) {
+ return true;
+ } else if (node->concat()->depth() <= 15) {
+ return true;
+ } else if (node->concat()->depth() > kMinLengthSize) {
+ return false;
+ } else {
+ // Allow depth to become twice as large as implied by fibonacci rule to
+ // reduce rebalancing for larger strings.
+ return (node->length >= min_length[node->concat()->depth() / 2]);
+ }
+}
+
+static CordRep* Rebalance(CordRep* node);
+static void DumpNode(CordRep* rep, bool include_data, std::ostream* os);
+static bool VerifyNode(CordRep* root, CordRep* start_node,
+ bool full_validation);
+
+static inline CordRep* VerifyTree(CordRep* node) {
+ // Verification is expensive, so only do it in debug mode.
+ // Even in debug mode we normally do only light validation.
+ // If you are debugging Cord itself, you should define the
+ // macro EXTRA_CORD_VALIDATION, e.g. by adding
+ // --copt=-DEXTRA_CORD_VALIDATION to the blaze line.
+#ifdef EXTRA_CORD_VALIDATION
+ assert(node == nullptr || VerifyNode(node, node, /*full_validation=*/true));
+#else // EXTRA_CORD_VALIDATION
+ assert(node == nullptr || VerifyNode(node, node, /*full_validation=*/false));
+#endif // EXTRA_CORD_VALIDATION
+ static_cast<void>(&VerifyNode);
+
+ return node;
+}
+
+// Return the depth of a node
+static int Depth(const CordRep* rep) {
+ if (rep->tag == CONCAT) {
+ return rep->concat()->depth();
+ } else {
+ return 0;
+ }
+}
+
+static void SetConcatChildren(CordRepConcat* concat, CordRep* left,
+ CordRep* right) {
+ concat->left = left;
+ concat->right = right;
+
+ concat->length = left->length + right->length;
+ concat->set_depth(1 + std::max(Depth(left), Depth(right)));
+}
+
+// Create a concatenation of the specified nodes.
+// Does not change the refcounts of "left" and "right".
+// The returned node has a refcount of 1.
+static CordRep* RawConcat(CordRep* left, CordRep* right) {
+ // Avoid making degenerate concat nodes (one child is empty)
+ if (left == nullptr) return right;
+ if (right == nullptr) return left;
+ if (left->length == 0) {
+ CordRep::Unref(left);
+ return right;
+ }
+ if (right->length == 0) {
+ CordRep::Unref(right);
+ return left;
+ }
+
+ CordRepConcat* rep = new CordRepConcat();
+ rep->tag = CONCAT;
+ SetConcatChildren(rep, left, right);
+
+ return rep;
+}
+
+static CordRep* Concat(CordRep* left, CordRep* right) {
+ CordRep* rep = RawConcat(left, right);
+ if (rep != nullptr && !IsRootBalanced(rep)) {
+ rep = Rebalance(rep);
+ }
+ return VerifyTree(rep);
+}
+
+// Make a balanced tree out of an array of leaf nodes.
+static CordRep* MakeBalancedTree(CordRep** reps, size_t n) {
+ // Make repeated passes over the array, merging adjacent pairs
+ // until we are left with just a single node.
+ while (n > 1) {
+ size_t dst = 0;
+ for (size_t src = 0; src < n; src += 2) {
+ if (src + 1 < n) {
+ reps[dst] = Concat(reps[src], reps[src + 1]);
+ } else {
+ reps[dst] = reps[src];
+ }
+ dst++;
+ }
+ n = dst;
+ }
+
+ return reps[0];
+}
+
+// Create a new tree out of the specified array.
+// The returned node has a refcount of 1.
+static CordRep* NewTree(const char* data,
+ size_t length,
+ size_t alloc_hint) {
+ if (length == 0) return nullptr;
+ absl::FixedArray<CordRep*> reps((length - 1) / kMaxFlatLength + 1);
+ size_t n = 0;
+ do {
+ const size_t len = std::min(length, kMaxFlatLength);
+ CordRep* rep = CordRepFlat::New(len + alloc_hint);
+ rep->length = len;
+ memcpy(rep->data, data, len);
+ reps[n++] = VerifyTree(rep);
+ data += len;
+ length -= len;
+ } while (length != 0);
+ return MakeBalancedTree(reps.data(), n);
+}
+
+namespace cord_internal {
+
+void InitializeCordRepExternal(absl::string_view data, CordRepExternal* rep) {
+ assert(!data.empty());
+ rep->length = data.size();
+ rep->tag = EXTERNAL;
+ rep->base = data.data();
+ VerifyTree(rep);
+}
+
+} // namespace cord_internal
+
+static CordRep* NewSubstring(CordRep* child, size_t offset, size_t length) {
+ // Never create empty substring nodes
+ if (length == 0) {
+ CordRep::Unref(child);
+ return nullptr;
+ } else {
+ CordRepSubstring* rep = new CordRepSubstring();
+ assert((offset + length) <= child->length);
+ rep->length = length;
+ rep->tag = SUBSTRING;
+ rep->start = offset;
+ rep->child = child;
+ return VerifyTree(rep);
+ }
+}
+
+// --------------------------------------------------------------------
+// Cord::InlineRep functions
+
+constexpr unsigned char Cord::InlineRep::kMaxInline;
+
+inline void Cord::InlineRep::set_data(const char* data, size_t n,
+ bool nullify_tail) {
+ static_assert(kMaxInline == 15, "set_data is hard-coded for a length of 15");
+
+ cord_internal::SmallMemmove(data_.as_chars, data, n, nullify_tail);
+ set_tagged_size(static_cast<char>(n));
+}
+
+inline char* Cord::InlineRep::set_data(size_t n) {
+ assert(n <= kMaxInline);
+ ResetToEmpty();
+ set_tagged_size(static_cast<char>(n));
+ return data_.as_chars;
+}
+
+inline CordRep* Cord::InlineRep::force_tree(size_t extra_hint) {
+ size_t len = tagged_size();
+ if (len > kMaxInline) {
+ return data_.as_tree.rep;
+ }
+
+ CordRep* result = CordRepFlat::New(len + extra_hint);
+ result->length = len;
+ static_assert(kMinFlatLength >= sizeof(data_.as_chars), "");
+ memcpy(result->data, data_.as_chars, sizeof(data_.as_chars));
+ set_tree(result);
+ return result;
+}
+
+inline void Cord::InlineRep::reduce_size(size_t n) {
+ size_t tag = tagged_size();
+ assert(tag <= kMaxInline);
+ assert(tag >= n);
+ tag -= n;
+ memset(data_.as_chars + tag, 0, n);
+ set_tagged_size(static_cast<char>(tag));
+}
+
+inline void Cord::InlineRep::remove_prefix(size_t n) {
+ cord_internal::SmallMemmove(data_.as_chars, data_.as_chars + n,
+ tagged_size() - n);
+ reduce_size(n);
+}
+
+void Cord::InlineRep::AppendTree(CordRep* tree) {
+ if (tree == nullptr) return;
+ size_t len = tagged_size();
+ if (len == 0) {
+ set_tree(tree);
+ } else {
+ set_tree(Concat(force_tree(0), tree));
+ }
+}
+
+void Cord::InlineRep::PrependTree(CordRep* tree) {
+ assert(tree != nullptr);
+ size_t len = tagged_size();
+ if (len == 0) {
+ set_tree(tree);
+ } else {
+ set_tree(Concat(tree, force_tree(0)));
+ }
+}
+
+// Searches for a non-full flat node at the rightmost leaf of the tree. If a
+// suitable leaf is found, the function will update the length field for all
+// nodes to account for the size increase. The append region address will be
+// written to region and the actual size increase will be written to size.
+static inline bool PrepareAppendRegion(CordRep* root, char** region,
+ size_t* size, size_t max_length) {
+ // Search down the right-hand path for a non-full FLAT node.
+ CordRep* dst = root;
+ while (dst->tag == CONCAT && dst->refcount.IsOne()) {
+ dst = dst->concat()->right;
+ }
+
+ if (dst->tag < FLAT || !dst->refcount.IsOne()) {
+ *region = nullptr;
+ *size = 0;
+ return false;
+ }
+
+ const size_t in_use = dst->length;
+ const size_t capacity = dst->flat()->Capacity();
+ if (in_use == capacity) {
+ *region = nullptr;
+ *size = 0;
+ return false;
+ }
+
+ size_t size_increase = std::min(capacity - in_use, max_length);
+
+ // We need to update the length fields for all nodes, including the leaf node.
+ for (CordRep* rep = root; rep != dst; rep = rep->concat()->right) {
+ rep->length += size_increase;
+ }
+ dst->length += size_increase;
+
+ *region = dst->data + in_use;
+ *size = size_increase;
+ return true;
+}
+
+void Cord::InlineRep::GetAppendRegion(char** region, size_t* size,
+ size_t max_length) {
+ if (max_length == 0) {
+ *region = nullptr;
+ *size = 0;
+ return;
+ }
+
+ // Try to fit in the inline buffer if possible.
+ size_t inline_length = tagged_size();
+ if (inline_length < kMaxInline && max_length <= kMaxInline - inline_length) {
+ *region = data_.as_chars + inline_length;
+ *size = max_length;
+ set_tagged_size(static_cast<char>(inline_length + max_length));
+ return;
+ }
+
+ CordRep* root = force_tree(max_length);
+
+ if (PrepareAppendRegion(root, region, size, max_length)) {
+ return;
+ }
+
+ // Allocate new node.
+ CordRepFlat* new_node =
+ CordRepFlat::New(std::max(static_cast<size_t>(root->length), max_length));
+ new_node->length = std::min(new_node->Capacity(), max_length);
+ *region = new_node->data;
+ *size = new_node->length;
+ replace_tree(Concat(root, new_node));
+}
+
+void Cord::InlineRep::GetAppendRegion(char** region, size_t* size) {
+ const size_t max_length = std::numeric_limits<size_t>::max();
+
+ // Try to fit in the inline buffer if possible.
+ size_t inline_length = tagged_size();
+ if (inline_length < kMaxInline) {
+ *region = data_.as_chars + inline_length;
+ *size = kMaxInline - inline_length;
+ set_tagged_size(kMaxInline);
+ return;
+ }
+
+ CordRep* root = force_tree(max_length);
+
+ if (PrepareAppendRegion(root, region, size, max_length)) {
+ return;
+ }
+
+ // Allocate new node.
+ CordRepFlat* new_node = CordRepFlat::New(root->length);
+ new_node->length = new_node->Capacity();
+ *region = new_node->data;
+ *size = new_node->length;
+ replace_tree(Concat(root, new_node));
+}
+
+// If the rep is a leaf, this will increment the value at total_mem_usage and
+// will return true.
+static bool RepMemoryUsageLeaf(const CordRep* rep, size_t* total_mem_usage) {
+ if (rep->tag >= FLAT) {
+ *total_mem_usage += rep->flat()->AllocatedSize();
+ return true;
+ }
+ if (rep->tag == EXTERNAL) {
+ *total_mem_usage += sizeof(CordRepConcat) + rep->length;
+ return true;
+ }
+ return false;
+}
+
+void Cord::InlineRep::AssignSlow(const Cord::InlineRep& src) {
+ ClearSlow();
+
+ data_ = src.data_;
+ if (is_tree()) {
+ CordRep::Ref(tree());
+ }
+}
+
+void Cord::InlineRep::ClearSlow() {
+ if (is_tree()) {
+ CordRep::Unref(tree());
+ }
+ ResetToEmpty();
+}
+
+// --------------------------------------------------------------------
+// Constructors and destructors
+
+Cord::Cord(const Cord& src) : contents_(src.contents_) {
+ if (CordRep* tree = contents_.tree()) {
+ CordRep::Ref(tree);
+ }
+}
+
+Cord::Cord(absl::string_view src) {
+ const size_t n = src.size();
+ if (n <= InlineRep::kMaxInline) {
+ contents_.set_data(src.data(), n, false);
+ } else {
+ contents_.set_tree(NewTree(src.data(), n, 0));
+ }
+}
+
+template <typename T, Cord::EnableIfString<T>>
+Cord::Cord(T&& src) {
+ if (
+ // String is short: copy data to avoid external block overhead.
+ src.size() <= kMaxBytesToCopy ||
+ // String is wasteful: copy data to avoid pinning too much unused memory.
+ src.size() < src.capacity() / 2
+ ) {
+ if (src.size() <= InlineRep::kMaxInline) {
+ contents_.set_data(src.data(), src.size(), false);
+ } else {
+ contents_.set_tree(NewTree(src.data(), src.size(), 0));
+ }
+ } else {
+ struct StringReleaser {
+ void operator()(absl::string_view /* data */) {}
+ std::string data;
+ };
+ const absl::string_view original_data = src;
+ auto* rep = static_cast<
+ ::absl::cord_internal::CordRepExternalImpl<StringReleaser>*>(
+ absl::cord_internal::NewExternalRep(
+ original_data, StringReleaser{std::forward<T>(src)}));
+ // Moving src may have invalidated its data pointer, so adjust it.
+ rep->base = rep->template get<0>().data.data();
+ contents_.set_tree(rep);
+ }
+}
+
+template Cord::Cord(std::string&& src);
+
+// The destruction code is separate so that the compiler can determine
+// that it does not need to call the destructor on a moved-from Cord.
+void Cord::DestroyCordSlow() {
+ if (CordRep* tree = contents_.tree()) {
+ CordRep::Unref(VerifyTree(tree));
+ }
+}
+
+// --------------------------------------------------------------------
+// Mutators
+
+void Cord::Clear() {
+ if (CordRep* tree = contents_.clear()) {
+ CordRep::Unref(tree);
+ }
+}
+
+Cord& Cord::operator=(absl::string_view src) {
+
+ const char* data = src.data();
+ size_t length = src.size();
+ CordRep* tree = contents_.tree();
+ if (length <= InlineRep::kMaxInline) {
+ // Embed into this->contents_
+ contents_.set_data(data, length, true);
+ if (tree) CordRep::Unref(tree);
+ return *this;
+ }
+ if (tree != nullptr && tree->tag >= FLAT &&
+ tree->flat()->Capacity() >= length &&
+ tree->refcount.IsOne()) {
+ // Copy in place if the existing FLAT node is reusable.
+ memmove(tree->data, data, length);
+ tree->length = length;
+ VerifyTree(tree);
+ return *this;
+ }
+ contents_.set_tree(NewTree(data, length, 0));
+ if (tree) CordRep::Unref(tree);
+ return *this;
+}
+
+template <typename T, Cord::EnableIfString<T>>
+Cord& Cord::operator=(T&& src) {
+ if (src.size() <= kMaxBytesToCopy) {
+ *this = absl::string_view(src);
+ } else {
+ *this = Cord(std::forward<T>(src));
+ }
+ return *this;
+}
+
+template Cord& Cord::operator=(std::string&& src);
+
+// TODO(sanjay): Move to Cord::InlineRep section of file. For now,
+// we keep it here to make diffs easier.
+void Cord::InlineRep::AppendArray(const char* src_data, size_t src_size) {
+ if (src_size == 0) return; // memcpy(_, nullptr, 0) is undefined.
+ // Try to fit in the inline buffer if possible.
+ size_t inline_length = tagged_size();
+ if (inline_length < kMaxInline && src_size <= kMaxInline - inline_length) {
+ // Append new data to embedded array
+ set_tagged_size(static_cast<char>(inline_length + src_size));
+ memcpy(data_.as_chars + inline_length, src_data, src_size);
+ return;
+ }
+
+ CordRep* root = tree();
+
+ size_t appended = 0;
+ if (root) {
+ char* region;
+ if (PrepareAppendRegion(root, ®ion, &appended, src_size)) {
+ memcpy(region, src_data, appended);
+ }
+ } else {
+ // It is possible that src_data == data_, but when we transition from an
+ // InlineRep to a tree we need to assign data_ = root via set_tree. To
+ // avoid corrupting the source data before we copy it, delay calling
+ // set_tree until after we've copied data.
+ // We are going from an inline size to beyond inline size. Make the new size
+ // either double the inlined size, or the added size + 10%.
+ const size_t size1 = inline_length * 2 + src_size;
+ const size_t size2 = inline_length + src_size / 10;
+ root = CordRepFlat::New(std::max<size_t>(size1, size2));
+ appended = std::min(
+ src_size, root->flat()->Capacity() - inline_length);
+ memcpy(root->data, data_.as_chars, inline_length);
+ memcpy(root->data + inline_length, src_data, appended);
+ root->length = inline_length + appended;
+ set_tree(root);
+ }
+
+ src_data += appended;
+ src_size -= appended;
+ if (src_size == 0) {
+ return;
+ }
+
+ // Use new block(s) for any remaining bytes that were not handled above.
+ // Alloc extra memory only if the right child of the root of the new tree is
+ // going to be a FLAT node, which will permit further inplace appends.
+ size_t length = src_size;
+ if (src_size < kMaxFlatLength) {
+ // The new length is either
+ // - old size + 10%
+ // - old_size + src_size
+ // This will cause a reasonable conservative step-up in size that is still
+ // large enough to avoid excessive amounts of small fragments being added.
+ length = std::max<size_t>(root->length / 10, src_size);
+ }
+ set_tree(Concat(root, NewTree(src_data, src_size, length - src_size)));
+}
+
+inline CordRep* Cord::TakeRep() const& {
+ return CordRep::Ref(contents_.tree());
+}
+
+inline CordRep* Cord::TakeRep() && {
+ CordRep* rep = contents_.tree();
+ contents_.clear();
+ return rep;
+}
+
+template <typename C>
+inline void Cord::AppendImpl(C&& src) {
+ if (empty()) {
+ // In case of an empty destination avoid allocating a new node, do not copy
+ // data.
+ *this = std::forward<C>(src);
+ return;
+ }
+
+ // For short cords, it is faster to copy data if there is room in dst.
+ const size_t src_size = src.contents_.size();
+ if (src_size <= kMaxBytesToCopy) {
+ CordRep* src_tree = src.contents_.tree();
+ if (src_tree == nullptr) {
+ // src has embedded data.
+ contents_.AppendArray(src.contents_.data(), src_size);
+ return;
+ }
+ if (src_tree->tag >= FLAT) {
+ // src tree just has one flat node.
+ contents_.AppendArray(src_tree->data, src_size);
+ return;
+ }
+ if (&src == this) {
+ // ChunkIterator below assumes that src is not modified during traversal.
+ Append(Cord(src));
+ return;
+ }
+ // TODO(mec): Should we only do this if "dst" has space?
+ for (absl::string_view chunk : src.Chunks()) {
+ Append(chunk);
+ }
+ return;
+ }
+
+ // Guaranteed to be a tree (kMaxBytesToCopy > kInlinedSize)
+ contents_.AppendTree(std::forward<C>(src).TakeRep());
+}
+
+void Cord::Append(const Cord& src) { AppendImpl(src); }
+
+void Cord::Append(Cord&& src) { AppendImpl(std::move(src)); }
+
+template <typename T, Cord::EnableIfString<T>>
+void Cord::Append(T&& src) {
+ if (src.size() <= kMaxBytesToCopy) {
+ Append(absl::string_view(src));
+ } else {
+ Append(Cord(std::forward<T>(src)));
+ }
+}
+
+template void Cord::Append(std::string&& src);
+
+void Cord::Prepend(const Cord& src) {
+ CordRep* src_tree = src.contents_.tree();
+ if (src_tree != nullptr) {
+ CordRep::Ref(src_tree);
+ contents_.PrependTree(src_tree);
+ return;
+ }
+
+ // `src` cord is inlined.
+ absl::string_view src_contents(src.contents_.data(), src.contents_.size());
+ return Prepend(src_contents);
+}
+
+void Cord::Prepend(absl::string_view src) {
+ if (src.empty()) return; // memcpy(_, nullptr, 0) is undefined.
+ size_t cur_size = contents_.size();
+ if (!contents_.is_tree() && cur_size + src.size() <= InlineRep::kMaxInline) {
+ // Use embedded storage.
+ char data[InlineRep::kMaxInline + 1] = {0};
+ data[InlineRep::kMaxInline] = cur_size + src.size(); // set size
+ memcpy(data, src.data(), src.size());
+ memcpy(data + src.size(), contents_.data(), cur_size);
+ memcpy(reinterpret_cast<void*>(&contents_), data,
+ InlineRep::kMaxInline + 1);
+ } else {
+ contents_.PrependTree(NewTree(src.data(), src.size(), 0));
+ }
+}
+
+template <typename T, Cord::EnableIfString<T>>
+inline void Cord::Prepend(T&& src) {
+ if (src.size() <= kMaxBytesToCopy) {
+ Prepend(absl::string_view(src));
+ } else {
+ Prepend(Cord(std::forward<T>(src)));
+ }
+}
+
+template void Cord::Prepend(std::string&& src);
+
+static CordRep* RemovePrefixFrom(CordRep* node, size_t n) {
+ if (n >= node->length) return nullptr;
+ if (n == 0) return CordRep::Ref(node);
+ absl::InlinedVector<CordRep*, kInlinedVectorSize> rhs_stack;
+
+ while (node->tag == CONCAT) {
+ assert(n <= node->length);
+ if (n < node->concat()->left->length) {
+ // Push right to stack, descend left.
+ rhs_stack.push_back(node->concat()->right);
+ node = node->concat()->left;
+ } else {
+ // Drop left, descend right.
+ n -= node->concat()->left->length;
+ node = node->concat()->right;
+ }
+ }
+ assert(n <= node->length);
+
+ if (n == 0) {
+ CordRep::Ref(node);
+ } else {
+ size_t start = n;
+ size_t len = node->length - n;
+ if (node->tag == SUBSTRING) {
+ // Consider in-place update of node, similar to in RemoveSuffixFrom().
+ start += node->substring()->start;
+ node = node->substring()->child;
+ }
+ node = NewSubstring(CordRep::Ref(node), start, len);
+ }
+ while (!rhs_stack.empty()) {
+ node = Concat(node, CordRep::Ref(rhs_stack.back()));
+ rhs_stack.pop_back();
+ }
+ return node;
+}
+
+// RemoveSuffixFrom() is very similar to RemovePrefixFrom(), with the
+// exception that removing a suffix has an optimization where a node may be
+// edited in place iff that node and all its ancestors have a refcount of 1.
+static CordRep* RemoveSuffixFrom(CordRep* node, size_t n) {
+ if (n >= node->length) return nullptr;
+ if (n == 0) return CordRep::Ref(node);
+ absl::InlinedVector<CordRep*, kInlinedVectorSize> lhs_stack;
+ bool inplace_ok = node->refcount.IsOne();
+
+ while (node->tag == CONCAT) {
+ assert(n <= node->length);
+ if (n < node->concat()->right->length) {
+ // Push left to stack, descend right.
+ lhs_stack.push_back(node->concat()->left);
+ node = node->concat()->right;
+ } else {
+ // Drop right, descend left.
+ n -= node->concat()->right->length;
+ node = node->concat()->left;
+ }
+ inplace_ok = inplace_ok && node->refcount.IsOne();
+ }
+ assert(n <= node->length);
+
+ if (n == 0) {
+ CordRep::Ref(node);
+ } else if (inplace_ok && node->tag != EXTERNAL) {
+ // Consider making a new buffer if the current node capacity is much
+ // larger than the new length.
+ CordRep::Ref(node);
+ node->length -= n;
+ } else {
+ size_t start = 0;
+ size_t len = node->length - n;
+ if (node->tag == SUBSTRING) {
+ start = node->substring()->start;
+ node = node->substring()->child;
+ }
+ node = NewSubstring(CordRep::Ref(node), start, len);
+ }
+ while (!lhs_stack.empty()) {
+ node = Concat(CordRep::Ref(lhs_stack.back()), node);
+ lhs_stack.pop_back();
+ }
+ return node;
+}
+
+void Cord::RemovePrefix(size_t n) {
+ ABSL_INTERNAL_CHECK(n <= size(),
+ absl::StrCat("Requested prefix size ", n,
+ " exceeds Cord's size ", size()));
+ CordRep* tree = contents_.tree();
+ if (tree == nullptr) {
+ contents_.remove_prefix(n);
+ } else {
+ CordRep* newrep = RemovePrefixFrom(tree, n);
+ CordRep::Unref(tree);
+ contents_.replace_tree(VerifyTree(newrep));
+ }
+}
+
+void Cord::RemoveSuffix(size_t n) {
+ ABSL_INTERNAL_CHECK(n <= size(),
+ absl::StrCat("Requested suffix size ", n,
+ " exceeds Cord's size ", size()));
+ CordRep* tree = contents_.tree();
+ if (tree == nullptr) {
+ contents_.reduce_size(n);
+ } else {
+ CordRep* newrep = RemoveSuffixFrom(tree, n);
+ CordRep::Unref(tree);
+ contents_.replace_tree(VerifyTree(newrep));
+ }
+}
+
+// Work item for NewSubRange().
+struct SubRange {
+ SubRange(CordRep* a_node, size_t a_pos, size_t a_n)
+ : node(a_node), pos(a_pos), n(a_n) {}
+ CordRep* node; // nullptr means concat last 2 results.
+ size_t pos;
+ size_t n;
+};
+
+static CordRep* NewSubRange(CordRep* node, size_t pos, size_t n) {
+ absl::InlinedVector<CordRep*, kInlinedVectorSize> results;
+ absl::InlinedVector<SubRange, kInlinedVectorSize> todo;
+ todo.push_back(SubRange(node, pos, n));
+ do {
+ const SubRange& sr = todo.back();
+ node = sr.node;
+ pos = sr.pos;
+ n = sr.n;
+ todo.pop_back();
+
+ if (node == nullptr) {
+ assert(results.size() >= 2);
+ CordRep* right = results.back();
+ results.pop_back();
+ CordRep* left = results.back();
+ results.pop_back();
+ results.push_back(Concat(left, right));
+ } else if (pos == 0 && n == node->length) {
+ results.push_back(CordRep::Ref(node));
+ } else if (node->tag != CONCAT) {
+ if (node->tag == SUBSTRING) {
+ pos += node->substring()->start;
+ node = node->substring()->child;
+ }
+ results.push_back(NewSubstring(CordRep::Ref(node), pos, n));
+ } else if (pos + n <= node->concat()->left->length) {
+ todo.push_back(SubRange(node->concat()->left, pos, n));
+ } else if (pos >= node->concat()->left->length) {
+ pos -= node->concat()->left->length;
+ todo.push_back(SubRange(node->concat()->right, pos, n));
+ } else {
+ size_t left_n = node->concat()->left->length - pos;
+ todo.push_back(SubRange(nullptr, 0, 0)); // Concat()
+ todo.push_back(SubRange(node->concat()->right, 0, n - left_n));
+ todo.push_back(SubRange(node->concat()->left, pos, left_n));
+ }
+ } while (!todo.empty());
+ assert(results.size() == 1);
+ return results[0];
+}
+
+Cord Cord::Subcord(size_t pos, size_t new_size) const {
+ Cord sub_cord;
+ size_t length = size();
+ if (pos > length) pos = length;
+ if (new_size > length - pos) new_size = length - pos;
+ CordRep* tree = contents_.tree();
+ if (tree == nullptr) {
+ // sub_cord is newly constructed, no need to re-zero-out the tail of
+ // contents_ memory.
+ sub_cord.contents_.set_data(contents_.data() + pos, new_size, false);
+ } else if (new_size == 0) {
+ // We want to return empty subcord, so nothing to do.
+ } else if (new_size <= InlineRep::kMaxInline) {
+ Cord::ChunkIterator it = chunk_begin();
+ it.AdvanceBytes(pos);
+ char* dest = sub_cord.contents_.data_.as_chars;
+ size_t remaining_size = new_size;
+ while (remaining_size > it->size()) {
+ cord_internal::SmallMemmove(dest, it->data(), it->size());
+ remaining_size -= it->size();
+ dest += it->size();
+ ++it;
+ }
+ cord_internal::SmallMemmove(dest, it->data(), remaining_size);
+ sub_cord.contents_.set_tagged_size(new_size);
+ } else {
+ sub_cord.contents_.set_tree(NewSubRange(tree, pos, new_size));
+ }
+ return sub_cord;
+}
+
+// --------------------------------------------------------------------
+// Balancing
+
+class CordForest {
+ public:
+ explicit CordForest(size_t length)
+ : root_length_(length), trees_(kMinLengthSize, nullptr) {}
+
+ void Build(CordRep* cord_root) {
+ std::vector<CordRep*> pending = {cord_root};
+
+ while (!pending.empty()) {
+ CordRep* node = pending.back();
+ pending.pop_back();
+ CheckNode(node);
+ if (ABSL_PREDICT_FALSE(node->tag != CONCAT)) {
+ AddNode(node);
+ continue;
+ }
+
+ CordRepConcat* concat_node = node->concat();
+ if (concat_node->depth() >= kMinLengthSize ||
+ concat_node->length < min_length[concat_node->depth()]) {
+ pending.push_back(concat_node->right);
+ pending.push_back(concat_node->left);
+
+ if (concat_node->refcount.IsOne()) {
+ concat_node->left = concat_freelist_;
+ concat_freelist_ = concat_node;
+ } else {
+ CordRep::Ref(concat_node->right);
+ CordRep::Ref(concat_node->left);
+ CordRep::Unref(concat_node);
+ }
+ } else {
+ AddNode(node);
+ }
+ }
+ }
+
+ CordRep* ConcatNodes() {
+ CordRep* sum = nullptr;
+ for (auto* node : trees_) {
+ if (node == nullptr) continue;
+
+ sum = PrependNode(node, sum);
+ root_length_ -= node->length;
+ if (root_length_ == 0) break;
+ }
+ ABSL_INTERNAL_CHECK(sum != nullptr, "Failed to locate sum node");
+ return VerifyTree(sum);
+ }
+
+ private:
+ CordRep* AppendNode(CordRep* node, CordRep* sum) {
+ return (sum == nullptr) ? node : MakeConcat(sum, node);
+ }
+
+ CordRep* PrependNode(CordRep* node, CordRep* sum) {
+ return (sum == nullptr) ? node : MakeConcat(node, sum);
+ }
+
+ void AddNode(CordRep* node) {
+ CordRep* sum = nullptr;
+
+ // Collect together everything with which we will merge with node
+ int i = 0;
+ for (; node->length > min_length[i + 1]; ++i) {
+ auto& tree_at_i = trees_[i];
+
+ if (tree_at_i == nullptr) continue;
+ sum = PrependNode(tree_at_i, sum);
+ tree_at_i = nullptr;
+ }
+
+ sum = AppendNode(node, sum);
+
+ // Insert sum into appropriate place in the forest
+ for (; sum->length >= min_length[i]; ++i) {
+ auto& tree_at_i = trees_[i];
+ if (tree_at_i == nullptr) continue;
+
+ sum = MakeConcat(tree_at_i, sum);
+ tree_at_i = nullptr;
+ }
+
+ // min_length[0] == 1, which means sum->length >= min_length[0]
+ assert(i > 0);
+ trees_[i - 1] = sum;
+ }
+
+ // Make concat node trying to resue existing CordRepConcat nodes we
+ // already collected in the concat_freelist_.
+ CordRep* MakeConcat(CordRep* left, CordRep* right) {
+ if (concat_freelist_ == nullptr) return RawConcat(left, right);
+
+ CordRepConcat* rep = concat_freelist_;
+ if (concat_freelist_->left == nullptr) {
+ concat_freelist_ = nullptr;
+ } else {
+ concat_freelist_ = concat_freelist_->left->concat();
+ }
+ SetConcatChildren(rep, left, right);
+
+ return rep;
+ }
+
+ static void CheckNode(CordRep* node) {
+ ABSL_INTERNAL_CHECK(node->length != 0u, "");
+ if (node->tag == CONCAT) {
+ ABSL_INTERNAL_CHECK(node->concat()->left != nullptr, "");
+ ABSL_INTERNAL_CHECK(node->concat()->right != nullptr, "");
+ ABSL_INTERNAL_CHECK(node->length == (node->concat()->left->length +
+ node->concat()->right->length),
+ "");
+ }
+ }
+
+ size_t root_length_;
+
+ // use an inlined vector instead of a flat array to get bounds checking
+ absl::InlinedVector<CordRep*, kInlinedVectorSize> trees_;
+
+ // List of concat nodes we can re-use for Cord balancing.
+ CordRepConcat* concat_freelist_ = nullptr;
+};
+
+static CordRep* Rebalance(CordRep* node) {
+ VerifyTree(node);
+ assert(node->tag == CONCAT);
+
+ if (node->length == 0) {
+ return nullptr;
+ }
+
+ CordForest forest(node->length);
+ forest.Build(node);
+ return forest.ConcatNodes();
+}
+
+// --------------------------------------------------------------------
+// Comparators
+
+namespace {
+
+int ClampResult(int memcmp_res) {
+ return static_cast<int>(memcmp_res > 0) - static_cast<int>(memcmp_res < 0);
+}
+
+int CompareChunks(absl::string_view* lhs, absl::string_view* rhs,
+ size_t* size_to_compare) {
+ size_t compared_size = std::min(lhs->size(), rhs->size());
+ assert(*size_to_compare >= compared_size);
+ *size_to_compare -= compared_size;
+
+ int memcmp_res = ::memcmp(lhs->data(), rhs->data(), compared_size);
+ if (memcmp_res != 0) return memcmp_res;
+
+ lhs->remove_prefix(compared_size);
+ rhs->remove_prefix(compared_size);
+
+ return 0;
+}
+
+// This overload set computes comparison results from memcmp result. This
+// interface is used inside GenericCompare below. Differet implementations
+// are specialized for int and bool. For int we clamp result to {-1, 0, 1}
+// set. For bool we just interested in "value == 0".
+template <typename ResultType>
+ResultType ComputeCompareResult(int memcmp_res) {
+ return ClampResult(memcmp_res);
+}
+template <>
+bool ComputeCompareResult<bool>(int memcmp_res) {
+ return memcmp_res == 0;
+}
+
+} // namespace
+
+// Helper routine. Locates the first flat chunk of the Cord without
+// initializing the iterator.
+inline absl::string_view Cord::InlineRep::FindFlatStartPiece() const {
+ size_t n = tagged_size();
+ if (n <= kMaxInline) {
+ return absl::string_view(data_.as_chars, n);
+ }
+
+ CordRep* node = tree();
+ if (node->tag >= FLAT) {
+ return absl::string_view(node->data, node->length);
+ }
+
+ if (node->tag == EXTERNAL) {
+ return absl::string_view(node->external()->base, node->length);
+ }
+
+ // Walk down the left branches until we hit a non-CONCAT node.
+ while (node->tag == CONCAT) {
+ node = node->concat()->left;
+ }
+
+ // Get the child node if we encounter a SUBSTRING.
+ size_t offset = 0;
+ size_t length = node->length;
+ assert(length != 0);
+
+ if (node->tag == SUBSTRING) {
+ offset = node->substring()->start;
+ node = node->substring()->child;
+ }
+
+ if (node->tag >= FLAT) {
+ return absl::string_view(node->data + offset, length);
+ }
+
+ assert((node->tag == EXTERNAL) && "Expect FLAT or EXTERNAL node here");
+
+ return absl::string_view(node->external()->base + offset, length);
+}
+
+inline int Cord::CompareSlowPath(absl::string_view rhs, size_t compared_size,
+ size_t size_to_compare) const {
+ auto advance = [](Cord::ChunkIterator* it, absl::string_view* chunk) {
+ if (!chunk->empty()) return true;
+ ++*it;
+ if (it->bytes_remaining_ == 0) return false;
+ *chunk = **it;
+ return true;
+ };
+
+ Cord::ChunkIterator lhs_it = chunk_begin();
+
+ // compared_size is inside first chunk.
+ absl::string_view lhs_chunk =
+ (lhs_it.bytes_remaining_ != 0) ? *lhs_it : absl::string_view();
+ assert(compared_size <= lhs_chunk.size());
+ assert(compared_size <= rhs.size());
+ lhs_chunk.remove_prefix(compared_size);
+ rhs.remove_prefix(compared_size);
+ size_to_compare -= compared_size; // skip already compared size.
+
+ while (advance(&lhs_it, &lhs_chunk) && !rhs.empty()) {
+ int comparison_result = CompareChunks(&lhs_chunk, &rhs, &size_to_compare);
+ if (comparison_result != 0) return comparison_result;
+ if (size_to_compare == 0) return 0;
+ }
+
+ return static_cast<int>(rhs.empty()) - static_cast<int>(lhs_chunk.empty());
+}
+
+inline int Cord::CompareSlowPath(const Cord& rhs, size_t compared_size,
+ size_t size_to_compare) const {
+ auto advance = [](Cord::ChunkIterator* it, absl::string_view* chunk) {
+ if (!chunk->empty()) return true;
+ ++*it;
+ if (it->bytes_remaining_ == 0) return false;
+ *chunk = **it;
+ return true;
+ };
+
+ Cord::ChunkIterator lhs_it = chunk_begin();
+ Cord::ChunkIterator rhs_it = rhs.chunk_begin();
+
+ // compared_size is inside both first chunks.
+ absl::string_view lhs_chunk =
+ (lhs_it.bytes_remaining_ != 0) ? *lhs_it : absl::string_view();
+ absl::string_view rhs_chunk =
+ (rhs_it.bytes_remaining_ != 0) ? *rhs_it : absl::string_view();
+ assert(compared_size <= lhs_chunk.size());
+ assert(compared_size <= rhs_chunk.size());
+ lhs_chunk.remove_prefix(compared_size);
+ rhs_chunk.remove_prefix(compared_size);
+ size_to_compare -= compared_size; // skip already compared size.
+
+ while (advance(&lhs_it, &lhs_chunk) && advance(&rhs_it, &rhs_chunk)) {
+ int memcmp_res = CompareChunks(&lhs_chunk, &rhs_chunk, &size_to_compare);
+ if (memcmp_res != 0) return memcmp_res;
+ if (size_to_compare == 0) return 0;
+ }
+
+ return static_cast<int>(rhs_chunk.empty()) -
+ static_cast<int>(lhs_chunk.empty());
+}
+
+inline absl::string_view Cord::GetFirstChunk(const Cord& c) {
+ return c.contents_.FindFlatStartPiece();
+}
+inline absl::string_view Cord::GetFirstChunk(absl::string_view sv) {
+ return sv;
+}
+
+// Compares up to 'size_to_compare' bytes of 'lhs' with 'rhs'. It is assumed
+// that 'size_to_compare' is greater that size of smallest of first chunks.
+template <typename ResultType, typename RHS>
+ResultType GenericCompare(const Cord& lhs, const RHS& rhs,
+ size_t size_to_compare) {
+ absl::string_view lhs_chunk = Cord::GetFirstChunk(lhs);
+ absl::string_view rhs_chunk = Cord::GetFirstChunk(rhs);
+
+ size_t compared_size = std::min(lhs_chunk.size(), rhs_chunk.size());
+ assert(size_to_compare >= compared_size);
+ int memcmp_res = ::memcmp(lhs_chunk.data(), rhs_chunk.data(), compared_size);
+ if (compared_size == size_to_compare || memcmp_res != 0) {
+ return ComputeCompareResult<ResultType>(memcmp_res);
+ }
+
+ return ComputeCompareResult<ResultType>(
+ lhs.CompareSlowPath(rhs, compared_size, size_to_compare));
+}
+
+bool Cord::EqualsImpl(absl::string_view rhs, size_t size_to_compare) const {
+ return GenericCompare<bool>(*this, rhs, size_to_compare);
+}
+
+bool Cord::EqualsImpl(const Cord& rhs, size_t size_to_compare) const {
+ return GenericCompare<bool>(*this, rhs, size_to_compare);
+}
+
+template <typename RHS>
+inline int SharedCompareImpl(const Cord& lhs, const RHS& rhs) {
+ size_t lhs_size = lhs.size();
+ size_t rhs_size = rhs.size();
+ if (lhs_size == rhs_size) {
+ return GenericCompare<int>(lhs, rhs, lhs_size);
+ }
+ if (lhs_size < rhs_size) {
+ auto data_comp_res = GenericCompare<int>(lhs, rhs, lhs_size);
+ return data_comp_res == 0 ? -1 : data_comp_res;
+ }
+
+ auto data_comp_res = GenericCompare<int>(lhs, rhs, rhs_size);
+ return data_comp_res == 0 ? +1 : data_comp_res;
+}
+
+int Cord::Compare(absl::string_view rhs) const {
+ return SharedCompareImpl(*this, rhs);
+}
+
+int Cord::CompareImpl(const Cord& rhs) const {
+ return SharedCompareImpl(*this, rhs);
+}
+
+bool Cord::EndsWith(absl::string_view rhs) const {
+ size_t my_size = size();
+ size_t rhs_size = rhs.size();
+
+ if (my_size < rhs_size) return false;
+
+ Cord tmp(*this);
+ tmp.RemovePrefix(my_size - rhs_size);
+ return tmp.EqualsImpl(rhs, rhs_size);
+}
+
+bool Cord::EndsWith(const Cord& rhs) const {
+ size_t my_size = size();
+ size_t rhs_size = rhs.size();
+
+ if (my_size < rhs_size) return false;
+
+ Cord tmp(*this);
+ tmp.RemovePrefix(my_size - rhs_size);
+ return tmp.EqualsImpl(rhs, rhs_size);
+}
+
+// --------------------------------------------------------------------
+// Misc.
+
+Cord::operator std::string() const {
+ std::string s;
+ absl::CopyCordToString(*this, &s);
+ return s;
+}
+
+void CopyCordToString(const Cord& src, std::string* dst) {
+ if (!src.contents_.is_tree()) {
+ src.contents_.CopyTo(dst);
+ } else {
+ absl::strings_internal::STLStringResizeUninitialized(dst, src.size());
+ src.CopyToArraySlowPath(&(*dst)[0]);
+ }
+}
+
+void Cord::CopyToArraySlowPath(char* dst) const {
+ assert(contents_.is_tree());
+ absl::string_view fragment;
+ if (GetFlatAux(contents_.tree(), &fragment)) {
+ memcpy(dst, fragment.data(), fragment.size());
+ return;
+ }
+ for (absl::string_view chunk : Chunks()) {
+ memcpy(dst, chunk.data(), chunk.size());
+ dst += chunk.size();
+ }
+}
+
+Cord::ChunkIterator& Cord::ChunkIterator::AdvanceStack() {
+ auto& stack_of_right_children = stack_of_right_children_;
+ if (stack_of_right_children.empty()) {
+ assert(!current_chunk_.empty()); // Called on invalid iterator.
+ // We have reached the end of the Cord.
+ return *this;
+ }
+
+ // Process the next node on the stack.
+ CordRep* node = stack_of_right_children.back();
+ stack_of_right_children.pop_back();
+
+ // Walk down the left branches until we hit a non-CONCAT node. Save the
+ // right children to the stack for subsequent traversal.
+ while (node->tag == CONCAT) {
+ stack_of_right_children.push_back(node->concat()->right);
+ node = node->concat()->left;
+ }
+
+ // Get the child node if we encounter a SUBSTRING.
+ size_t offset = 0;
+ size_t length = node->length;
+ if (node->tag == SUBSTRING) {
+ offset = node->substring()->start;
+ node = node->substring()->child;
+ }
+
+ assert(node->tag == EXTERNAL || node->tag >= FLAT);
+ assert(length != 0);
+ const char* data =
+ node->tag == EXTERNAL ? node->external()->base : node->data;
+ current_chunk_ = absl::string_view(data + offset, length);
+ current_leaf_ = node;
+ return *this;
+}
+
+Cord Cord::ChunkIterator::AdvanceAndReadBytes(size_t n) {
+ ABSL_HARDENING_ASSERT(bytes_remaining_ >= n &&
+ "Attempted to iterate past `end()`");
+ Cord subcord;
+
+ if (n <= InlineRep::kMaxInline) {
+ // Range to read fits in inline data. Flatten it.
+ char* data = subcord.contents_.set_data(n);
+ while (n > current_chunk_.size()) {
+ memcpy(data, current_chunk_.data(), current_chunk_.size());
+ data += current_chunk_.size();
+ n -= current_chunk_.size();
+ ++*this;
+ }
+ memcpy(data, current_chunk_.data(), n);
+ if (n < current_chunk_.size()) {
+ RemoveChunkPrefix(n);
+ } else if (n > 0) {
+ ++*this;
+ }
+ return subcord;
+ }
+ auto& stack_of_right_children = stack_of_right_children_;
+ if (n < current_chunk_.size()) {
+ // Range to read is a proper subrange of the current chunk.
+ assert(current_leaf_ != nullptr);
+ CordRep* subnode = CordRep::Ref(current_leaf_);
+ const char* data =
+ subnode->tag == EXTERNAL ? subnode->external()->base : subnode->data;
+ subnode = NewSubstring(subnode, current_chunk_.data() - data, n);
+ subcord.contents_.set_tree(VerifyTree(subnode));
+ RemoveChunkPrefix(n);
+ return subcord;
+ }
+
+ // Range to read begins with a proper subrange of the current chunk.
+ assert(!current_chunk_.empty());
+ assert(current_leaf_ != nullptr);
+ CordRep* subnode = CordRep::Ref(current_leaf_);
+ if (current_chunk_.size() < subnode->length) {
+ const char* data =
+ subnode->tag == EXTERNAL ? subnode->external()->base : subnode->data;
+ subnode = NewSubstring(subnode, current_chunk_.data() - data,
+ current_chunk_.size());
+ }
+ n -= current_chunk_.size();
+ bytes_remaining_ -= current_chunk_.size();
+
+ // Process the next node(s) on the stack, reading whole subtrees depending on
+ // their length and how many bytes we are advancing.
+ CordRep* node = nullptr;
+ while (!stack_of_right_children.empty()) {
+ node = stack_of_right_children.back();
+ stack_of_right_children.pop_back();
+ if (node->length > n) break;
+ // TODO(qrczak): This might unnecessarily recreate existing concat nodes.
+ // Avoiding that would need pretty complicated logic (instead of
+ // current_leaf, keep current_subtree_ which points to the highest node
+ // such that the current leaf can be found on the path of left children
+ // starting from current_subtree_; delay creating subnode while node is
+ // below current_subtree_; find the proper node along the path of left
+ // children starting from current_subtree_ if this loop exits while staying
+ // below current_subtree_; etc.; alternatively, push parents instead of
+ // right children on the stack).
+ subnode = Concat(subnode, CordRep::Ref(node));
+ n -= node->length;
+ bytes_remaining_ -= node->length;
+ node = nullptr;
+ }
+
+ if (node == nullptr) {
+ // We have reached the end of the Cord.
+ assert(bytes_remaining_ == 0);
+ subcord.contents_.set_tree(VerifyTree(subnode));
+ return subcord;
+ }
+
+ // Walk down the appropriate branches until we hit a non-CONCAT node. Save the
+ // right children to the stack for subsequent traversal.
+ while (node->tag == CONCAT) {
+ if (node->concat()->left->length > n) {
+ // Push right, descend left.
+ stack_of_right_children.push_back(node->concat()->right);
+ node = node->concat()->left;
+ } else {
+ // Read left, descend right.
+ subnode = Concat(subnode, CordRep::Ref(node->concat()->left));
+ n -= node->concat()->left->length;
+ bytes_remaining_ -= node->concat()->left->length;
+ node = node->concat()->right;
+ }
+ }
+
+ // Get the child node if we encounter a SUBSTRING.
+ size_t offset = 0;
+ size_t length = node->length;
+ if (node->tag == SUBSTRING) {
+ offset = node->substring()->start;
+ node = node->substring()->child;
+ }
+
+ // Range to read ends with a proper (possibly empty) subrange of the current
+ // chunk.
+ assert(node->tag == EXTERNAL || node->tag >= FLAT);
+ assert(length > n);
+ if (n > 0) {
+ subnode = Concat(subnode, NewSubstring(CordRep::Ref(node), offset, n));
+ }
+ const char* data =
+ node->tag == EXTERNAL ? node->external()->base : node->data;
+ current_chunk_ = absl::string_view(data + offset + n, length - n);
+ current_leaf_ = node;
+ bytes_remaining_ -= n;
+ subcord.contents_.set_tree(VerifyTree(subnode));
+ return subcord;
+}
+
+void Cord::ChunkIterator::AdvanceBytesSlowPath(size_t n) {
+ assert(bytes_remaining_ >= n && "Attempted to iterate past `end()`");
+ assert(n >= current_chunk_.size()); // This should only be called when
+ // iterating to a new node.
+
+ n -= current_chunk_.size();
+ bytes_remaining_ -= current_chunk_.size();
+
+ if (stack_of_right_children_.empty()) {
+ // We have reached the end of the Cord.
+ assert(bytes_remaining_ == 0);
+ return;
+ }
+
+ // Process the next node(s) on the stack, skipping whole subtrees depending on
+ // their length and how many bytes we are advancing.
+ CordRep* node = nullptr;
+ auto& stack_of_right_children = stack_of_right_children_;
+ while (!stack_of_right_children.empty()) {
+ node = stack_of_right_children.back();
+ stack_of_right_children.pop_back();
+ if (node->length > n) break;
+ n -= node->length;
+ bytes_remaining_ -= node->length;
+ node = nullptr;
+ }
+
+ if (node == nullptr) {
+ // We have reached the end of the Cord.
+ assert(bytes_remaining_ == 0);
+ return;
+ }
+
+ // Walk down the appropriate branches until we hit a non-CONCAT node. Save the
+ // right children to the stack for subsequent traversal.
+ while (node->tag == CONCAT) {
+ if (node->concat()->left->length > n) {
+ // Push right, descend left.
+ stack_of_right_children.push_back(node->concat()->right);
+ node = node->concat()->left;
+ } else {
+ // Skip left, descend right.
+ n -= node->concat()->left->length;
+ bytes_remaining_ -= node->concat()->left->length;
+ node = node->concat()->right;
+ }
+ }
+
+ // Get the child node if we encounter a SUBSTRING.
+ size_t offset = 0;
+ size_t length = node->length;
+ if (node->tag == SUBSTRING) {
+ offset = node->substring()->start;
+ node = node->substring()->child;
+ }
+
+ assert(node->tag == EXTERNAL || node->tag >= FLAT);
+ assert(length > n);
+ const char* data =
+ node->tag == EXTERNAL ? node->external()->base : node->data;
+ current_chunk_ = absl::string_view(data + offset + n, length - n);
+ current_leaf_ = node;
+ bytes_remaining_ -= n;
+}
+
+char Cord::operator[](size_t i) const {
+ ABSL_HARDENING_ASSERT(i < size());
+ size_t offset = i;
+ const CordRep* rep = contents_.tree();
+ if (rep == nullptr) {
+ return contents_.data()[i];
+ }
+ while (true) {
+ assert(rep != nullptr);
+ assert(offset < rep->length);
+ if (rep->tag >= FLAT) {
+ // Get the "i"th character directly from the flat array.
+ return rep->data[offset];
+ } else if (rep->tag == EXTERNAL) {
+ // Get the "i"th character from the external array.
+ return rep->external()->base[offset];
+ } else if (rep->tag == CONCAT) {
+ // Recursively branch to the side of the concatenation that the "i"th
+ // character is on.
+ size_t left_length = rep->concat()->left->length;
+ if (offset < left_length) {
+ rep = rep->concat()->left;
+ } else {
+ offset -= left_length;
+ rep = rep->concat()->right;
+ }
+ } else {
+ // This must be a substring a node, so bypass it to get to the child.
+ assert(rep->tag == SUBSTRING);
+ offset += rep->substring()->start;
+ rep = rep->substring()->child;
+ }
+ }
+}
+
+absl::string_view Cord::FlattenSlowPath() {
+ size_t total_size = size();
+ CordRep* new_rep;
+ char* new_buffer;
+
+ // Try to put the contents into a new flat rep. If they won't fit in the
+ // biggest possible flat node, use an external rep instead.
+ if (total_size <= kMaxFlatLength) {
+ new_rep = CordRepFlat::New(total_size);
+ new_rep->length = total_size;
+ new_buffer = new_rep->data;
+ CopyToArraySlowPath(new_buffer);
+ } else {
+ new_buffer = std::allocator<char>().allocate(total_size);
+ CopyToArraySlowPath(new_buffer);
+ new_rep = absl::cord_internal::NewExternalRep(
+ absl::string_view(new_buffer, total_size), [](absl::string_view s) {
+ std::allocator<char>().deallocate(const_cast<char*>(s.data()),
+ s.size());
+ });
+ }
+ if (CordRep* tree = contents_.tree()) {
+ CordRep::Unref(tree);
+ }
+ contents_.set_tree(new_rep);
+ return absl::string_view(new_buffer, total_size);
+}
+
+/* static */ bool Cord::GetFlatAux(CordRep* rep, absl::string_view* fragment) {
+ assert(rep != nullptr);
+ if (rep->tag >= FLAT) {
+ *fragment = absl::string_view(rep->data, rep->length);
+ return true;
+ } else if (rep->tag == EXTERNAL) {
+ *fragment = absl::string_view(rep->external()->base, rep->length);
+ return true;
+ } else if (rep->tag == SUBSTRING) {
+ CordRep* child = rep->substring()->child;
+ if (child->tag >= FLAT) {
+ *fragment =
+ absl::string_view(child->data + rep->substring()->start, rep->length);
+ return true;
+ } else if (child->tag == EXTERNAL) {
+ *fragment = absl::string_view(
+ child->external()->base + rep->substring()->start, rep->length);
+ return true;
+ }
+ }
+ return false;
+}
+
+/* static */ void Cord::ForEachChunkAux(
+ absl::cord_internal::CordRep* rep,
+ absl::FunctionRef<void(absl::string_view)> callback) {
+ assert(rep != nullptr);
+ int stack_pos = 0;
+ constexpr int stack_max = 128;
+ // Stack of right branches for tree traversal
+ absl::cord_internal::CordRep* stack[stack_max];
+ absl::cord_internal::CordRep* current_node = rep;
+ while (true) {
+ if (current_node->tag == CONCAT) {
+ if (stack_pos == stack_max) {
+ // There's no more room on our stack array to add another right branch,
+ // and the idea is to avoid allocations, so call this function
+ // recursively to navigate this subtree further. (This is not something
+ // we expect to happen in practice).
+ ForEachChunkAux(current_node, callback);
+
+ // Pop the next right branch and iterate.
+ current_node = stack[--stack_pos];
+ continue;
+ } else {
+ // Save the right branch for later traversal and continue down the left
+ // branch.
+ stack[stack_pos++] = current_node->concat()->right;
+ current_node = current_node->concat()->left;
+ continue;
+ }
+ }
+ // This is a leaf node, so invoke our callback.
+ absl::string_view chunk;
+ bool success = GetFlatAux(current_node, &chunk);
+ assert(success);
+ if (success) {
+ callback(chunk);
+ }
+ if (stack_pos == 0) {
+ // end of traversal
+ return;
+ }
+ current_node = stack[--stack_pos];
+ }
+}
+
+static void DumpNode(CordRep* rep, bool include_data, std::ostream* os) {
+ const int kIndentStep = 1;
+ int indent = 0;
+ absl::InlinedVector<CordRep*, kInlinedVectorSize> stack;
+ absl::InlinedVector<int, kInlinedVectorSize> indents;
+ for (;;) {
+ *os << std::setw(3) << rep->refcount.Get();
+ *os << " " << std::setw(7) << rep->length;
+ *os << " [";
+ if (include_data) *os << static_cast<void*>(rep);
+ *os << "]";
+ *os << " " << (IsRootBalanced(rep) ? 'b' : 'u');
+ *os << " " << std::setw(indent) << "";
+ if (rep->tag == CONCAT) {
+ *os << "CONCAT depth=" << Depth(rep) << "\n";
+ indent += kIndentStep;
+ indents.push_back(indent);
+ stack.push_back(rep->concat()->right);
+ rep = rep->concat()->left;
+ } else if (rep->tag == SUBSTRING) {
+ *os << "SUBSTRING @ " << rep->substring()->start << "\n";
+ indent += kIndentStep;
+ rep = rep->substring()->child;
+ } else { // Leaf
+ if (rep->tag == EXTERNAL) {
+ *os << "EXTERNAL [";
+ if (include_data)
+ *os << absl::CEscape(std::string(rep->external()->base, rep->length));
+ *os << "]\n";
+ } else {
+ *os << "FLAT cap=" << rep->flat()->Capacity()
+ << " [";
+ if (include_data)
+ *os << absl::CEscape(std::string(rep->data, rep->length));
+ *os << "]\n";
+ }
+ if (stack.empty()) break;
+ rep = stack.back();
+ stack.pop_back();
+ indent = indents.back();
+ indents.pop_back();
+ }
+ }
+ ABSL_INTERNAL_CHECK(indents.empty(), "");
+}
+
+static std::string ReportError(CordRep* root, CordRep* node) {
+ std::ostringstream buf;
+ buf << "Error at node " << node << " in:";
+ DumpNode(root, true, &buf);
+ return buf.str();
+}
+
+static bool VerifyNode(CordRep* root, CordRep* start_node,
+ bool full_validation) {
+ absl::InlinedVector<CordRep*, 2> worklist;
+ worklist.push_back(start_node);
+ do {
+ CordRep* node = worklist.back();
+ worklist.pop_back();
+
+ ABSL_INTERNAL_CHECK(node != nullptr, ReportError(root, node));
+ if (node != root) {
+ ABSL_INTERNAL_CHECK(node->length != 0, ReportError(root, node));
+ }
+
+ if (node->tag == CONCAT) {
+ ABSL_INTERNAL_CHECK(node->concat()->left != nullptr,
+ ReportError(root, node));
+ ABSL_INTERNAL_CHECK(node->concat()->right != nullptr,
+ ReportError(root, node));
+ ABSL_INTERNAL_CHECK((node->length == node->concat()->left->length +
+ node->concat()->right->length),
+ ReportError(root, node));
+ if (full_validation) {
+ worklist.push_back(node->concat()->right);
+ worklist.push_back(node->concat()->left);
+ }
+ } else if (node->tag >= FLAT) {
+ ABSL_INTERNAL_CHECK(
+ node->length <= node->flat()->Capacity(),
+ ReportError(root, node));
+ } else if (node->tag == EXTERNAL) {
+ ABSL_INTERNAL_CHECK(node->external()->base != nullptr,
+ ReportError(root, node));
+ } else if (node->tag == SUBSTRING) {
+ ABSL_INTERNAL_CHECK(
+ node->substring()->start < node->substring()->child->length,
+ ReportError(root, node));
+ ABSL_INTERNAL_CHECK(node->substring()->start + node->length <=
+ node->substring()->child->length,
+ ReportError(root, node));
+ }
+ } while (!worklist.empty());
+ return true;
+}
+
+// Traverses the tree and computes the total memory allocated.
+/* static */ size_t Cord::MemoryUsageAux(const CordRep* rep) {
+ size_t total_mem_usage = 0;
+
+ // Allow a quick exit for the common case that the root is a leaf.
+ if (RepMemoryUsageLeaf(rep, &total_mem_usage)) {
+ return total_mem_usage;
+ }
+
+ // Iterate over the tree. cur_node is never a leaf node and leaf nodes will
+ // never be appended to tree_stack. This reduces overhead from manipulating
+ // tree_stack.
+ absl::InlinedVector<const CordRep*, kInlinedVectorSize> tree_stack;
+ const CordRep* cur_node = rep;
+ while (true) {
+ const CordRep* next_node = nullptr;
+
+ if (cur_node->tag == CONCAT) {
+ total_mem_usage += sizeof(CordRepConcat);
+ const CordRep* left = cur_node->concat()->left;
+ if (!RepMemoryUsageLeaf(left, &total_mem_usage)) {
+ next_node = left;
+ }
+
+ const CordRep* right = cur_node->concat()->right;
+ if (!RepMemoryUsageLeaf(right, &total_mem_usage)) {
+ if (next_node) {
+ tree_stack.push_back(next_node);
+ }
+ next_node = right;
+ }
+ } else {
+ // Since cur_node is not a leaf or a concat node it must be a substring.
+ assert(cur_node->tag == SUBSTRING);
+ total_mem_usage += sizeof(CordRepSubstring);
+ next_node = cur_node->substring()->child;
+ if (RepMemoryUsageLeaf(next_node, &total_mem_usage)) {
+ next_node = nullptr;
+ }
+ }
+
+ if (!next_node) {
+ if (tree_stack.empty()) {
+ return total_mem_usage;
+ }
+ next_node = tree_stack.back();
+ tree_stack.pop_back();
+ }
+ cur_node = next_node;
+ }
+}
+
+std::ostream& operator<<(std::ostream& out, const Cord& cord) {
+ for (absl::string_view chunk : cord.Chunks()) {
+ out.write(chunk.data(), chunk.size());
+ }
+ return out;
+}
+
+namespace strings_internal {
+size_t CordTestAccess::FlatOverhead() { return cord_internal::kFlatOverhead; }
+size_t CordTestAccess::MaxFlatLength() { return cord_internal::kMaxFlatLength; }
+size_t CordTestAccess::FlatTagToLength(uint8_t tag) {
+ return cord_internal::TagToLength(tag);
+}
+uint8_t CordTestAccess::LengthToTag(size_t s) {
+ ABSL_INTERNAL_CHECK(s <= kMaxFlatLength, absl::StrCat("Invalid length ", s));
+ return cord_internal::AllocatedSizeToTag(s + cord_internal::kFlatOverhead);
+}
+size_t CordTestAccess::SizeofCordRepConcat() { return sizeof(CordRepConcat); }
+size_t CordTestAccess::SizeofCordRepExternal() {
+ return sizeof(CordRepExternal);
+}
+size_t CordTestAccess::SizeofCordRepSubstring() {
+ return sizeof(CordRepSubstring);
+}
+} // namespace strings_internal
+ABSL_NAMESPACE_END
+} // namespace absl
diff --git a/absl/strings/cord.h b/absl/strings/cord.h
new file mode 100644
index 0000000..7136f03
--- /dev/null
+++ b/absl/strings/cord.h
@@ -0,0 +1,1322 @@
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+// -----------------------------------------------------------------------------
+// File: cord.h
+// -----------------------------------------------------------------------------
+//
+// This file defines the `absl::Cord` data structure and operations on that data
+// structure. A Cord is a string-like sequence of characters optimized for
+// specific use cases. Unlike a `std::string`, which stores an array of
+// contiguous characters, Cord data is stored in a structure consisting of
+// separate, reference-counted "chunks." (Currently, this implementation is a
+// tree structure, though that implementation may change.)
+//
+// Because a Cord consists of these chunks, data can be added to or removed from
+// a Cord during its lifetime. Chunks may also be shared between Cords. Unlike a
+// `std::string`, a Cord can therefore accomodate data that changes over its
+// lifetime, though it's not quite "mutable"; it can change only in the
+// attachment, detachment, or rearrangement of chunks of its constituent data.
+//
+// A Cord provides some benefit over `std::string` under the following (albeit
+// narrow) circumstances:
+//
+// * Cord data is designed to grow and shrink over a Cord's lifetime. Cord
+// provides efficient insertions and deletions at the start and end of the
+// character sequences, avoiding copies in those cases. Static data should
+// generally be stored as strings.
+// * External memory consisting of string-like data can be directly added to
+// a Cord without requiring copies or allocations.
+// * Cord data may be shared and copied cheaply. Cord provides a copy-on-write
+// implementation and cheap sub-Cord operations. Copying a Cord is an O(1)
+// operation.
+//
+// As a consequence to the above, Cord data is generally large. Small data
+// should generally use strings, as construction of a Cord requires some
+// overhead. Small Cords (<= 15 bytes) are represented inline, but most small
+// Cords are expected to grow over their lifetimes.
+//
+// Note that because a Cord is made up of separate chunked data, random access
+// to character data within a Cord is slower than within a `std::string`.
+//
+// Thread Safety
+//
+// Cord has the same thread-safety properties as many other types like
+// std::string, std::vector<>, int, etc -- it is thread-compatible. In
+// particular, if threads do not call non-const methods, then it is safe to call
+// const methods without synchronization. Copying a Cord produces a new instance
+// that can be used concurrently with the original in arbitrary ways.
+
+#ifndef ABSL_STRINGS_CORD_H_
+#define ABSL_STRINGS_CORD_H_
+
+#include <algorithm>
+#include <cstddef>
+#include <cstdint>
+#include <cstring>
+#include <iosfwd>
+#include <iterator>
+#include <string>
+#include <type_traits>
+
+#include "absl/base/internal/endian.h"
+#include "absl/base/internal/per_thread_tls.h"
+#include "absl/base/macros.h"
+#include "absl/base/port.h"
+#include "absl/container/inlined_vector.h"
+#include "absl/functional/function_ref.h"
+#include "absl/meta/type_traits.h"
+#include "absl/strings/internal/cord_internal.h"
+#include "absl/strings/internal/resize_uninitialized.h"
+#include "absl/strings/internal/string_constant.h"
+#include "absl/strings/string_view.h"
+#include "absl/types/optional.h"
+
+namespace absl {
+ABSL_NAMESPACE_BEGIN
+class Cord;
+class CordTestPeer;
+template <typename Releaser>
+Cord MakeCordFromExternal(absl::string_view, Releaser&&);
+void CopyCordToString(const Cord& src, std::string* dst);
+
+// Cord
+//
+// A Cord is a sequence of characters, designed to be more efficient than a
+// `std::string` in certain circumstances: namely, large string data that needs
+// to change over its lifetime or shared, especially when such data is shared
+// across API boundaries.
+//
+// A Cord stores its character data in a structure that allows efficient prepend
+// and append operations. This makes a Cord useful for large string data sent
+// over in a wire format that may need to be prepended or appended at some point
+// during the data exchange (e.g. HTTP, protocol buffers). For example, a
+// Cord is useful for storing an HTTP request, and prepending an HTTP header to
+// such a request.
+//
+// Cords should not be used for storing general string data, however. They
+// require overhead to construct and are slower than strings for random access.
+//
+// The Cord API provides the following common API operations:
+//
+// * Create or assign Cords out of existing string data, memory, or other Cords
+// * Append and prepend data to an existing Cord
+// * Create new Sub-Cords from existing Cord data
+// * Swap Cord data and compare Cord equality
+// * Write out Cord data by constructing a `std::string`
+//
+// Additionally, the API provides iterator utilities to iterate through Cord
+// data via chunks or character bytes.
+//
+class Cord {
+ private:
+ template <typename T>
+ using EnableIfString =
+ absl::enable_if_t<std::is_same<T, std::string>::value, int>;
+
+ public:
+ // Cord::Cord() Constructors.
+
+ // Creates an empty Cord.
+ constexpr Cord() noexcept;
+
+ // Creates a Cord from an existing Cord. Cord is copyable and efficiently
+ // movable. The moved-from state is valid but unspecified.
+ Cord(const Cord& src);
+ Cord(Cord&& src) noexcept;
+ Cord& operator=(const Cord& x);
+ Cord& operator=(Cord&& x) noexcept;
+
+ // Creates a Cord from a `src` string. This constructor is marked explicit to
+ // prevent implicit Cord constructions from arguments convertible to an
+ // `absl::string_view`.
+ explicit Cord(absl::string_view src);
+ Cord& operator=(absl::string_view src);
+
+ // Creates a Cord from a `std::string&&` rvalue. These constructors are
+ // templated to avoid ambiguities for types that are convertible to both
+ // `absl::string_view` and `std::string`, such as `const char*`.
+ template <typename T, EnableIfString<T> = 0>
+ explicit Cord(T&& src);
+ template <typename T, EnableIfString<T> = 0>
+ Cord& operator=(T&& src);
+
+ // Cord::~Cord()
+ //
+ // Destructs the Cord.
+ ~Cord() {
+ if (contents_.is_tree()) DestroyCordSlow();
+ }
+
+ // MakeCordFromExternal()
+ //
+ // Creates a Cord that takes ownership of external string memory. The
+ // contents of `data` are not copied to the Cord; instead, the external
+ // memory is added to the Cord and reference-counted. This data may not be
+ // changed for the life of the Cord, though it may be prepended or appended
+ // to.
+ //
+ // `MakeCordFromExternal()` takes a callable "releaser" that is invoked when
+ // the reference count for `data` reaches zero. As noted above, this data must
+ // remain live until the releaser is invoked. The callable releaser also must:
+ //
+ // * be move constructible
+ // * support `void operator()(absl::string_view) const` or `void operator()`
+ //
+ // Example:
+ //
+ // Cord MakeCord(BlockPool* pool) {
+ // Block* block = pool->NewBlock();
+ // FillBlock(block);
+ // return absl::MakeCordFromExternal(
+ // block->ToStringView(),
+ // [pool, block](absl::string_view v) {
+ // pool->FreeBlock(block, v);
+ // });
+ // }
+ //
+ // WARNING: Because a Cord can be reference-counted, it's likely a bug if your
+ // releaser doesn't do anything. For example, consider the following:
+ //
+ // void Foo(const char* buffer, int len) {
+ // auto c = absl::MakeCordFromExternal(absl::string_view(buffer, len),
+ // [](absl::string_view) {});
+ //
+ // // BUG: If Bar() copies its cord for any reason, including keeping a
+ // // substring of it, the lifetime of buffer might be extended beyond
+ // // when Foo() returns.
+ // Bar(c);
+ // }
+ template <typename Releaser>
+ friend Cord MakeCordFromExternal(absl::string_view data, Releaser&& releaser);
+
+ // Cord::Clear()
+ //
+ // Releases the Cord data. Any nodes that share data with other Cords, if
+ // applicable, will have their reference counts reduced by 1.
+ void Clear();
+
+ // Cord::Append()
+ //
+ // Appends data to the Cord, which may come from another Cord or other string
+ // data.
+ void Append(const Cord& src);
+ void Append(Cord&& src);
+ void Append(absl::string_view src);
+ template <typename T, EnableIfString<T> = 0>
+ void Append(T&& src);
+
+ // Cord::Prepend()
+ //
+ // Prepends data to the Cord, which may come from another Cord or other string
+ // data.
+ void Prepend(const Cord& src);
+ void Prepend(absl::string_view src);
+ template <typename T, EnableIfString<T> = 0>
+ void Prepend(T&& src);
+
+ // Cord::RemovePrefix()
+ //
+ // Removes the first `n` bytes of a Cord.
+ void RemovePrefix(size_t n);
+ void RemoveSuffix(size_t n);
+
+ // Cord::Subcord()
+ //
+ // Returns a new Cord representing the subrange [pos, pos + new_size) of
+ // *this. If pos >= size(), the result is empty(). If
+ // (pos + new_size) >= size(), the result is the subrange [pos, size()).
+ Cord Subcord(size_t pos, size_t new_size) const;
+
+ // Cord::swap()
+ //
+ // Swaps the contents of the Cord with `other`.
+ void swap(Cord& other) noexcept;
+
+ // swap()
+ //
+ // Swaps the contents of two Cords.
+ friend void swap(Cord& x, Cord& y) noexcept {
+ x.swap(y);
+ }
+
+ // Cord::size()
+ //
+ // Returns the size of the Cord.
+ size_t size() const;
+
+ // Cord::empty()
+ //
+ // Determines whether the given Cord is empty, returning `true` is so.
+ bool empty() const;
+
+ // Cord::EstimatedMemoryUsage()
+ //
+ // Returns the *approximate* number of bytes held in full or in part by this
+ // Cord (which may not remain the same between invocations). Note that Cords
+ // that share memory could each be "charged" independently for the same shared
+ // memory.
+ size_t EstimatedMemoryUsage() const;
+
+ // Cord::Compare()
+ //
+ // Compares 'this' Cord with rhs. This function and its relatives treat Cords
+ // as sequences of unsigned bytes. The comparison is a straightforward
+ // lexicographic comparison. `Cord::Compare()` returns values as follows:
+ //
+ // -1 'this' Cord is smaller
+ // 0 two Cords are equal
+ // 1 'this' Cord is larger
+ int Compare(absl::string_view rhs) const;
+ int Compare(const Cord& rhs) const;
+
+ // Cord::StartsWith()
+ //
+ // Determines whether the Cord starts with the passed string data `rhs`.
+ bool StartsWith(const Cord& rhs) const;
+ bool StartsWith(absl::string_view rhs) const;
+
+ // Cord::EndsWidth()
+ //
+ // Determines whether the Cord ends with the passed string data `rhs`.
+ bool EndsWith(absl::string_view rhs) const;
+ bool EndsWith(const Cord& rhs) const;
+
+ // Cord::operator std::string()
+ //
+ // Converts a Cord into a `std::string()`. This operator is marked explicit to
+ // prevent unintended Cord usage in functions that take a string.
+ explicit operator std::string() const;
+
+ // CopyCordToString()
+ //
+ // Copies the contents of a `src` Cord into a `*dst` string.
+ //
+ // This function optimizes the case of reusing the destination string since it
+ // can reuse previously allocated capacity. However, this function does not
+ // guarantee that pointers previously returned by `dst->data()` remain valid
+ // even if `*dst` had enough capacity to hold `src`. If `*dst` is a new
+ // object, prefer to simply use the conversion operator to `std::string`.
+ friend void CopyCordToString(const Cord& src, std::string* dst);
+
+ class CharIterator;
+
+ //----------------------------------------------------------------------------
+ // Cord::ChunkIterator
+ //----------------------------------------------------------------------------
+ //
+ // A `Cord::ChunkIterator` allows iteration over the constituent chunks of its
+ // Cord. Such iteration allows you to perform non-const operatons on the data
+ // of a Cord without modifying it.
+ //
+ // Generally, you do not instantiate a `Cord::ChunkIterator` directly;
+ // instead, you create one implicitly through use of the `Cord::Chunks()`
+ // member function.
+ //
+ // The `Cord::ChunkIterator` has the following properties:
+ //
+ // * The iterator is invalidated after any non-const operation on the
+ // Cord object over which it iterates.
+ // * The `string_view` returned by dereferencing a valid, non-`end()`
+ // iterator is guaranteed to be non-empty.
+ // * Two `ChunkIterator` objects can be compared equal if and only if they
+ // remain valid and iterate over the same Cord.
+ // * The iterator in this case is a proxy iterator; the `string_view`
+ // returned by the iterator does not live inside the Cord, and its
+ // lifetime is limited to the lifetime of the iterator itself. To help
+ // prevent lifetime issues, `ChunkIterator::reference` is not a true
+ // reference type and is equivalent to `value_type`.
+ // * The iterator keeps state that can grow for Cords that contain many
+ // nodes and are imbalanced due to sharing. Prefer to pass this type by
+ // const reference instead of by value.
+ class ChunkIterator {
+ public:
+ using iterator_category = std::input_iterator_tag;
+ using value_type = absl::string_view;
+ using difference_type = ptrdiff_t;
+ using pointer = const value_type*;
+ using reference = value_type;
+
+ ChunkIterator() = default;
+
+ ChunkIterator& operator++();
+ ChunkIterator operator++(int);
+ bool operator==(const ChunkIterator& other) const;
+ bool operator!=(const ChunkIterator& other) const;
+ reference operator*() const;
+ pointer operator->() const;
+
+ friend class Cord;
+ friend class CharIterator;
+
+ private:
+ // Stack of right children of concat nodes that we have to visit.
+ // Keep this at the end of the structure to avoid cache-thrashing.
+ // TODO(jgm): Benchmark to see if there's a more optimal value than 47 for
+ // the inlined vector size (47 exists for backward compatibility).
+ using Stack = absl::InlinedVector<absl::cord_internal::CordRep*, 47>;
+
+ // Constructs a `begin()` iterator from `cord`.
+ explicit ChunkIterator(const Cord* cord);
+
+ // Removes `n` bytes from `current_chunk_`. Expects `n` to be smaller than
+ // `current_chunk_.size()`.
+ void RemoveChunkPrefix(size_t n);
+ Cord AdvanceAndReadBytes(size_t n);
+ void AdvanceBytes(size_t n);
+
+ // Stack specific operator++
+ ChunkIterator& AdvanceStack();
+
+ // Iterates `n` bytes, where `n` is expected to be greater than or equal to
+ // `current_chunk_.size()`.
+ void AdvanceBytesSlowPath(size_t n);
+
+ // A view into bytes of the current `CordRep`. It may only be a view to a
+ // suffix of bytes if this is being used by `CharIterator`.
+ absl::string_view current_chunk_;
+ // The current leaf, or `nullptr` if the iterator points to short data.
+ // If the current chunk is a substring node, current_leaf_ points to the
+ // underlying flat or external node.
+ absl::cord_internal::CordRep* current_leaf_ = nullptr;
+ // The number of bytes left in the `Cord` over which we are iterating.
+ size_t bytes_remaining_ = 0;
+ // See 'Stack' alias definition.
+ Stack stack_of_right_children_;
+ };
+
+ // Cord::ChunkIterator::chunk_begin()
+ //
+ // Returns an iterator to the first chunk of the `Cord`.
+ //
+ // Generally, prefer using `Cord::Chunks()` within a range-based for loop for
+ // iterating over the chunks of a Cord. This method may be useful for getting
+ // a `ChunkIterator` where range-based for-loops are not useful.
+ //
+ // Example:
+ //
+ // absl::Cord::ChunkIterator FindAsChunk(const absl::Cord& c,
+ // absl::string_view s) {
+ // return std::find(c.chunk_begin(), c.chunk_end(), s);
+ // }
+ ChunkIterator chunk_begin() const;
+
+ // Cord::ChunkItertator::chunk_end()
+ //
+ // Returns an iterator one increment past the last chunk of the `Cord`.
+ //
+ // Generally, prefer using `Cord::Chunks()` within a range-based for loop for
+ // iterating over the chunks of a Cord. This method may be useful for getting
+ // a `ChunkIterator` where range-based for-loops may not be available.
+ ChunkIterator chunk_end() const;
+
+ //----------------------------------------------------------------------------
+ // Cord::ChunkIterator::ChunkRange
+ //----------------------------------------------------------------------------
+ //
+ // `ChunkRange` is a helper class for iterating over the chunks of the `Cord`,
+ // producing an iterator which can be used within a range-based for loop.
+ // Construction of a `ChunkRange` will return an iterator pointing to the
+ // first chunk of the Cord. Generally, do not construct a `ChunkRange`
+ // directly; instead, prefer to use the `Cord::Chunks()` method.
+ //
+ // Implementation note: `ChunkRange` is simply a convenience wrapper over
+ // `Cord::chunk_begin()` and `Cord::chunk_end()`.
+ class ChunkRange {
+ public:
+ explicit ChunkRange(const Cord* cord) : cord_(cord) {}
+
+ ChunkIterator begin() const;
+ ChunkIterator end() const;
+
+ private:
+ const Cord* cord_;
+ };
+
+ // Cord::Chunks()
+ //
+ // Returns a `Cord::ChunkIterator::ChunkRange` for iterating over the chunks
+ // of a `Cord` with a range-based for-loop. For most iteration tasks on a
+ // Cord, use `Cord::Chunks()` to retrieve this iterator.
+ //
+ // Example:
+ //
+ // void ProcessChunks(const Cord& cord) {
+ // for (absl::string_view chunk : cord.Chunks()) { ... }
+ // }
+ //
+ // Note that the ordinary caveats of temporary lifetime extension apply:
+ //
+ // void Process() {
+ // for (absl::string_view chunk : CordFactory().Chunks()) {
+ // // The temporary Cord returned by CordFactory has been destroyed!
+ // }
+ // }
+ ChunkRange Chunks() const;
+
+ //----------------------------------------------------------------------------
+ // Cord::CharIterator
+ //----------------------------------------------------------------------------
+ //
+ // A `Cord::CharIterator` allows iteration over the constituent characters of
+ // a `Cord`.
+ //
+ // Generally, you do not instantiate a `Cord::CharIterator` directly; instead,
+ // you create one implicitly through use of the `Cord::Chars()` member
+ // function.
+ //
+ // A `Cord::CharIterator` has the following properties:
+ //
+ // * The iterator is invalidated after any non-const operation on the
+ // Cord object over which it iterates.
+ // * Two `CharIterator` objects can be compared equal if and only if they
+ // remain valid and iterate over the same Cord.
+ // * The iterator keeps state that can grow for Cords that contain many
+ // nodes and are imbalanced due to sharing. Prefer to pass this type by
+ // const reference instead of by value.
+ // * This type cannot act as a forward iterator because a `Cord` can reuse
+ // sections of memory. This fact violates the requirement for forward
+ // iterators to compare equal if dereferencing them returns the same
+ // object.
+ class CharIterator {
+ public:
+ using iterator_category = std::input_iterator_tag;
+ using value_type = char;
+ using difference_type = ptrdiff_t;
+ using pointer = const char*;
+ using reference = const char&;
+
+ CharIterator() = default;
+
+ CharIterator& operator++();
+ CharIterator operator++(int);
+ bool operator==(const CharIterator& other) const;
+ bool operator!=(const CharIterator& other) const;
+ reference operator*() const;
+ pointer operator->() const;
+
+ friend Cord;
+
+ private:
+ explicit CharIterator(const Cord* cord) : chunk_iterator_(cord) {}
+
+ ChunkIterator chunk_iterator_;
+ };
+
+ // Cord::CharIterator::AdvanceAndRead()
+ //
+ // Advances the `Cord::CharIterator` by `n_bytes` and returns the bytes
+ // advanced as a separate `Cord`. `n_bytes` must be less than or equal to the
+ // number of bytes within the Cord; otherwise, behavior is undefined. It is
+ // valid to pass `char_end()` and `0`.
+ static Cord AdvanceAndRead(CharIterator* it, size_t n_bytes);
+
+ // Cord::CharIterator::Advance()
+ //
+ // Advances the `Cord::CharIterator` by `n_bytes`. `n_bytes` must be less than
+ // or equal to the number of bytes remaining within the Cord; otherwise,
+ // behavior is undefined. It is valid to pass `char_end()` and `0`.
+ static void Advance(CharIterator* it, size_t n_bytes);
+
+ // Cord::CharIterator::ChunkRemaining()
+ //
+ // Returns the longest contiguous view starting at the iterator's position.
+ //
+ // `it` must be dereferenceable.
+ static absl::string_view ChunkRemaining(const CharIterator& it);
+
+ // Cord::CharIterator::char_begin()
+ //
+ // Returns an iterator to the first character of the `Cord`.
+ //
+ // Generally, prefer using `Cord::Chars()` within a range-based for loop for
+ // iterating over the chunks of a Cord. This method may be useful for getting
+ // a `CharIterator` where range-based for-loops may not be available.
+ CharIterator char_begin() const;
+
+ // Cord::CharIterator::char_end()
+ //
+ // Returns an iterator to one past the last character of the `Cord`.
+ //
+ // Generally, prefer using `Cord::Chars()` within a range-based for loop for
+ // iterating over the chunks of a Cord. This method may be useful for getting
+ // a `CharIterator` where range-based for-loops are not useful.
+ CharIterator char_end() const;
+
+ // Cord::CharIterator::CharRange
+ //
+ // `CharRange` is a helper class for iterating over the characters of a
+ // producing an iterator which can be used within a range-based for loop.
+ // Construction of a `CharRange` will return an iterator pointing to the first
+ // character of the Cord. Generally, do not construct a `CharRange` directly;
+ // instead, prefer to use the `Cord::Chars()` method show below.
+ //
+ // Implementation note: `CharRange` is simply a convenience wrapper over
+ // `Cord::char_begin()` and `Cord::char_end()`.
+ class CharRange {
+ public:
+ explicit CharRange(const Cord* cord) : cord_(cord) {}
+
+ CharIterator begin() const;
+ CharIterator end() const;
+
+ private:
+ const Cord* cord_;
+ };
+
+ // Cord::CharIterator::Chars()
+ //
+ // Returns a `Cord::CharIterator` for iterating over the characters of a
+ // `Cord` with a range-based for-loop. For most character-based iteration
+ // tasks on a Cord, use `Cord::Chars()` to retrieve this iterator.
+ //
+ // Example:
+ //
+ // void ProcessCord(const Cord& cord) {
+ // for (char c : cord.Chars()) { ... }
+ // }
+ //
+ // Note that the ordinary caveats of temporary lifetime extension apply:
+ //
+ // void Process() {
+ // for (char c : CordFactory().Chars()) {
+ // // The temporary Cord returned by CordFactory has been destroyed!
+ // }
+ // }
+ CharRange Chars() const;
+
+ // Cord::operator[]
+ //
+ // Gets the "i"th character of the Cord and returns it, provided that
+ // 0 <= i < Cord.size().
+ //
+ // NOTE: This routine is reasonably efficient. It is roughly
+ // logarithmic based on the number of chunks that make up the cord. Still,
+ // if you need to iterate over the contents of a cord, you should
+ // use a CharIterator/ChunkIterator rather than call operator[] or Get()
+ // repeatedly in a loop.
+ char operator[](size_t i) const;
+
+ // Cord::TryFlat()
+ //
+ // If this cord's representation is a single flat array, returns a
+ // string_view referencing that array. Otherwise returns nullopt.
+ absl::optional<absl::string_view> TryFlat() const;
+
+ // Cord::Flatten()
+ //
+ // Flattens the cord into a single array and returns a view of the data.
+ //
+ // If the cord was already flat, the contents are not modified.
+ absl::string_view Flatten();
+
+ // Supports absl::Cord as a sink object for absl::Format().
+ friend void AbslFormatFlush(absl::Cord* cord, absl::string_view part) {
+ cord->Append(part);
+ }
+
+ template <typename H>
+ friend H AbslHashValue(H hash_state, const absl::Cord& c) {
+ absl::optional<absl::string_view> maybe_flat = c.TryFlat();
+ if (maybe_flat.has_value()) {
+ return H::combine(std::move(hash_state), *maybe_flat);
+ }
+ return c.HashFragmented(std::move(hash_state));
+ }
+
+ // Create a Cord with the contents of StringConstant<T>::value.
+ // No allocations will be done and no data will be copied.
+ // This is an INTERNAL API and subject to change or removal. This API can only
+ // be used by spelling absl::strings_internal::MakeStringConstant, which is
+ // also an internal API.
+ template <typename T>
+ explicit constexpr Cord(strings_internal::StringConstant<T>);
+
+ private:
+ friend class CordTestPeer;
+ friend bool operator==(const Cord& lhs, const Cord& rhs);
+ friend bool operator==(const Cord& lhs, absl::string_view rhs);
+
+ // Calls the provided function once for each cord chunk, in order. Unlike
+ // Chunks(), this API will not allocate memory.
+ void ForEachChunk(absl::FunctionRef<void(absl::string_view)>) const;
+
+ // Allocates new contiguous storage for the contents of the cord. This is
+ // called by Flatten() when the cord was not already flat.
+ absl::string_view FlattenSlowPath();
+
+ // Actual cord contents are hidden inside the following simple
+ // class so that we can isolate the bulk of cord.cc from changes
+ // to the representation.
+ //
+ // InlineRep holds either a tree pointer, or an array of kMaxInline bytes.
+ class InlineRep {
+ public:
+ static constexpr unsigned char kMaxInline = cord_internal::kMaxInline;
+ static_assert(kMaxInline >= sizeof(absl::cord_internal::CordRep*), "");
+ static constexpr unsigned char kTreeFlag = cord_internal::kTreeFlag;
+ static constexpr unsigned char kProfiledFlag = cord_internal::kProfiledFlag;
+
+ constexpr InlineRep() : data_() {}
+ InlineRep(const InlineRep& src);
+ InlineRep(InlineRep&& src);
+ InlineRep& operator=(const InlineRep& src);
+ InlineRep& operator=(InlineRep&& src) noexcept;
+
+ explicit constexpr InlineRep(cord_internal::InlineData data);
+
+ void Swap(InlineRep* rhs);
+ bool empty() const;
+ size_t size() const;
+ const char* data() const; // Returns nullptr if holding pointer
+ void set_data(const char* data, size_t n,
+ bool nullify_tail); // Discards pointer, if any
+ char* set_data(size_t n); // Write data to the result
+ // Returns nullptr if holding bytes
+ absl::cord_internal::CordRep* tree() const;
+ // Discards old pointer, if any
+ void set_tree(absl::cord_internal::CordRep* rep);
+ // Replaces a tree with a new root. This is faster than set_tree, but it
+ // should only be used when it's clear that the old rep was a tree.
+ void replace_tree(absl::cord_internal::CordRep* rep);
+ // Returns non-null iff was holding a pointer
+ absl::cord_internal::CordRep* clear();
+ // Converts to pointer if necessary.
+ absl::cord_internal::CordRep* force_tree(size_t extra_hint);
+ void reduce_size(size_t n); // REQUIRES: holding data
+ void remove_prefix(size_t n); // REQUIRES: holding data
+ void AppendArray(const char* src_data, size_t src_size);
+ absl::string_view FindFlatStartPiece() const;
+ void AppendTree(absl::cord_internal::CordRep* tree);
+ void PrependTree(absl::cord_internal::CordRep* tree);
+ void GetAppendRegion(char** region, size_t* size, size_t max_length);
+ void GetAppendRegion(char** region, size_t* size);
+ bool IsSame(const InlineRep& other) const {
+ return memcmp(&data_, &other.data_, sizeof(data_)) == 0;
+ }
+ int BitwiseCompare(const InlineRep& other) const {
+ uint64_t x, y;
+ // Use memcpy to avoid aliasing issues.
+ memcpy(&x, &data_, sizeof(x));
+ memcpy(&y, &other.data_, sizeof(y));
+ if (x == y) {
+ memcpy(&x, reinterpret_cast<const char*>(&data_) + 8, sizeof(x));
+ memcpy(&y, reinterpret_cast<const char*>(&other.data_) + 8, sizeof(y));
+ if (x == y) return 0;
+ }
+ return absl::big_endian::FromHost64(x) < absl::big_endian::FromHost64(y)
+ ? -1
+ : 1;
+ }
+ void CopyTo(std::string* dst) const {
+ // memcpy is much faster when operating on a known size. On most supported
+ // platforms, the small string optimization is large enough that resizing
+ // to 15 bytes does not cause a memory allocation.
+ absl::strings_internal::STLStringResizeUninitialized(dst,
+ sizeof(data_) - 1);
+ memcpy(&(*dst)[0], &data_, sizeof(data_) - 1);
+ // erase is faster than resize because the logic for memory allocation is
+ // not needed.
+ dst->erase(tagged_size());
+ }
+
+ // Copies the inline contents into `dst`. Assumes the cord is not empty.
+ void CopyToArray(char* dst) const;
+
+ bool is_tree() const { return tagged_size() > kMaxInline; }
+
+ private:
+ friend class Cord;
+
+ void AssignSlow(const InlineRep& src);
+ // Unrefs the tree, stops profiling, and zeroes the contents
+ void ClearSlow();
+
+ void ResetToEmpty() { data_ = {}; }
+
+ // This uses reinterpret_cast instead of the union to avoid accessing the
+ // inactive union element. The tagged size is not a common prefix.
+ void set_tagged_size(char new_tag) {
+ reinterpret_cast<char*>(&data_)[kMaxInline] = new_tag;
+ }
+ char tagged_size() const {
+ return reinterpret_cast<const char*>(&data_)[kMaxInline];
+ }
+
+ cord_internal::InlineData data_;
+ };
+ InlineRep contents_;
+
+ // Helper for MemoryUsage().
+ static size_t MemoryUsageAux(const absl::cord_internal::CordRep* rep);
+
+ // Helper for GetFlat() and TryFlat().
+ static bool GetFlatAux(absl::cord_internal::CordRep* rep,
+ absl::string_view* fragment);
+
+ // Helper for ForEachChunk().
+ static void ForEachChunkAux(
+ absl::cord_internal::CordRep* rep,
+ absl::FunctionRef<void(absl::string_view)> callback);
+
+ // The destructor for non-empty Cords.
+ void DestroyCordSlow();
+
+ // Out-of-line implementation of slower parts of logic.
+ void CopyToArraySlowPath(char* dst) const;
+ int CompareSlowPath(absl::string_view rhs, size_t compared_size,
+ size_t size_to_compare) const;
+ int CompareSlowPath(const Cord& rhs, size_t compared_size,
+ size_t size_to_compare) const;
+ bool EqualsImpl(absl::string_view rhs, size_t size_to_compare) const;
+ bool EqualsImpl(const Cord& rhs, size_t size_to_compare) const;
+ int CompareImpl(const Cord& rhs) const;
+
+ template <typename ResultType, typename RHS>
+ friend ResultType GenericCompare(const Cord& lhs, const RHS& rhs,
+ size_t size_to_compare);
+ static absl::string_view GetFirstChunk(const Cord& c);
+ static absl::string_view GetFirstChunk(absl::string_view sv);
+
+ // Returns a new reference to contents_.tree(), or steals an existing
+ // reference if called on an rvalue.
+ absl::cord_internal::CordRep* TakeRep() const&;
+ absl::cord_internal::CordRep* TakeRep() &&;
+
+ // Helper for Append().
+ template <typename C>
+ void AppendImpl(C&& src);
+
+ // Helper for AbslHashValue().
+ template <typename H>
+ H HashFragmented(H hash_state) const {
+ typename H::AbslInternalPiecewiseCombiner combiner;
+ ForEachChunk([&combiner, &hash_state](absl::string_view chunk) {
+ hash_state = combiner.add_buffer(std::move(hash_state), chunk.data(),
+ chunk.size());
+ });
+ return H::combine(combiner.finalize(std::move(hash_state)), size());
+ }
+};
+
+ABSL_NAMESPACE_END
+} // namespace absl
+
+namespace absl {
+ABSL_NAMESPACE_BEGIN
+
+// allow a Cord to be logged
+extern std::ostream& operator<<(std::ostream& out, const Cord& cord);
+
+// ------------------------------------------------------------------
+// Internal details follow. Clients should ignore.
+
+namespace cord_internal {
+
+// Fast implementation of memmove for up to 15 bytes. This implementation is
+// safe for overlapping regions. If nullify_tail is true, the destination is
+// padded with '\0' up to 16 bytes.
+inline void SmallMemmove(char* dst, const char* src, size_t n,
+ bool nullify_tail = false) {
+ if (n >= 8) {
+ assert(n <= 16);
+ uint64_t buf1;
+ uint64_t buf2;
+ memcpy(&buf1, src, 8);
+ memcpy(&buf2, src + n - 8, 8);
+ if (nullify_tail) {
+ memset(dst + 8, 0, 8);
+ }
+ memcpy(dst, &buf1, 8);
+ memcpy(dst + n - 8, &buf2, 8);
+ } else if (n >= 4) {
+ uint32_t buf1;
+ uint32_t buf2;
+ memcpy(&buf1, src, 4);
+ memcpy(&buf2, src + n - 4, 4);
+ if (nullify_tail) {
+ memset(dst + 4, 0, 4);
+ memset(dst + 8, 0, 8);
+ }
+ memcpy(dst, &buf1, 4);
+ memcpy(dst + n - 4, &buf2, 4);
+ } else {
+ if (n != 0) {
+ dst[0] = src[0];
+ dst[n / 2] = src[n / 2];
+ dst[n - 1] = src[n - 1];
+ }
+ if (nullify_tail) {
+ memset(dst + 8, 0, 8);
+ memset(dst + n, 0, 8);
+ }
+ }
+}
+
+// Does non-template-specific `CordRepExternal` initialization.
+// Expects `data` to be non-empty.
+void InitializeCordRepExternal(absl::string_view data, CordRepExternal* rep);
+
+// Creates a new `CordRep` that owns `data` and `releaser` and returns a pointer
+// to it, or `nullptr` if `data` was empty.
+template <typename Releaser>
+// NOLINTNEXTLINE - suppress clang-tidy raw pointer return.
+CordRep* NewExternalRep(absl::string_view data, Releaser&& releaser) {
+ using ReleaserType = absl::decay_t<Releaser>;
+ if (data.empty()) {
+ // Never create empty external nodes.
+ InvokeReleaser(Rank0{}, ReleaserType(std::forward<Releaser>(releaser)),
+ data);
+ return nullptr;
+ }
+
+ CordRepExternal* rep = new CordRepExternalImpl<ReleaserType>(
+ std::forward<Releaser>(releaser), 0);
+ InitializeCordRepExternal(data, rep);
+ return rep;
+}
+
+// Overload for function reference types that dispatches using a function
+// pointer because there are no `alignof()` or `sizeof()` a function reference.
+// NOLINTNEXTLINE - suppress clang-tidy raw pointer return.
+inline CordRep* NewExternalRep(absl::string_view data,
+ void (&releaser)(absl::string_view)) {
+ return NewExternalRep(data, &releaser);
+}
+
+} // namespace cord_internal
+
+template <typename Releaser>
+Cord MakeCordFromExternal(absl::string_view data, Releaser&& releaser) {
+ Cord cord;
+ cord.contents_.set_tree(::absl::cord_internal::NewExternalRep(
+ data, std::forward<Releaser>(releaser)));
+ return cord;
+}
+
+constexpr Cord::InlineRep::InlineRep(cord_internal::InlineData data)
+ : data_(data) {}
+
+inline Cord::InlineRep::InlineRep(const Cord::InlineRep& src) {
+ data_ = src.data_;
+}
+
+inline Cord::InlineRep::InlineRep(Cord::InlineRep&& src) {
+ data_ = src.data_;
+ src.ResetToEmpty();
+}
+
+inline Cord::InlineRep& Cord::InlineRep::operator=(const Cord::InlineRep& src) {
+ if (this == &src) {
+ return *this;
+ }
+ if (!is_tree() && !src.is_tree()) {
+ data_ = src.data_;
+ return *this;
+ }
+ AssignSlow(src);
+ return *this;
+}
+
+inline Cord::InlineRep& Cord::InlineRep::operator=(
+ Cord::InlineRep&& src) noexcept {
+ if (is_tree()) {
+ ClearSlow();
+ }
+ data_ = src.data_;
+ src.ResetToEmpty();
+ return *this;
+}
+
+inline void Cord::InlineRep::Swap(Cord::InlineRep* rhs) {
+ if (rhs == this) {
+ return;
+ }
+
+ std::swap(data_, rhs->data_);
+}
+
+inline const char* Cord::InlineRep::data() const {
+ return is_tree() ? nullptr : data_.as_chars;
+}
+
+inline absl::cord_internal::CordRep* Cord::InlineRep::tree() const {
+ if (is_tree()) {
+ return data_.as_tree.rep;
+ } else {
+ return nullptr;
+ }
+}
+
+inline bool Cord::InlineRep::empty() const { return tagged_size() == 0; }
+
+inline size_t Cord::InlineRep::size() const {
+ const char tag = tagged_size();
+ if (tag <= kMaxInline) return tag;
+ return static_cast<size_t>(tree()->length);
+}
+
+inline void Cord::InlineRep::set_tree(absl::cord_internal::CordRep* rep) {
+ if (rep == nullptr) {
+ ResetToEmpty();
+ } else {
+ bool was_tree = is_tree();
+ data_.as_tree = {rep, {}, tagged_size()};
+ if (!was_tree) {
+ // If we were not a tree already, set the tag.
+ // Otherwise, leave it alone because it might have the profile bit on.
+ set_tagged_size(kTreeFlag);
+ }
+ }
+}
+
+inline void Cord::InlineRep::replace_tree(absl::cord_internal::CordRep* rep) {
+ ABSL_ASSERT(is_tree());
+ if (ABSL_PREDICT_FALSE(rep == nullptr)) {
+ set_tree(rep);
+ return;
+ }
+ data_.as_tree = {rep, {}, tagged_size()};
+}
+
+inline absl::cord_internal::CordRep* Cord::InlineRep::clear() {
+ absl::cord_internal::CordRep* result = tree();
+ ResetToEmpty();
+ return result;
+}
+
+inline void Cord::InlineRep::CopyToArray(char* dst) const {
+ assert(!is_tree());
+ size_t n = tagged_size();
+ assert(n != 0);
+ cord_internal::SmallMemmove(dst, data_.as_chars, n);
+}
+
+constexpr inline Cord::Cord() noexcept {}
+
+template <typename T>
+constexpr Cord::Cord(strings_internal::StringConstant<T>)
+ : contents_(strings_internal::StringConstant<T>::value.size() <=
+ cord_internal::kMaxInline
+ ? cord_internal::InlineData(
+ strings_internal::StringConstant<T>::value)
+ : cord_internal::InlineData(cord_internal::AsTree{
+ &cord_internal::ConstInitExternalStorage<
+ strings_internal::StringConstant<T>>::value,
+ {},
+ cord_internal::kTreeFlag})) {}
+
+inline Cord& Cord::operator=(const Cord& x) {
+ contents_ = x.contents_;
+ return *this;
+}
+
+inline Cord::Cord(Cord&& src) noexcept : contents_(std::move(src.contents_)) {}
+
+inline void Cord::swap(Cord& other) noexcept {
+ contents_.Swap(&other.contents_);
+}
+
+inline Cord& Cord::operator=(Cord&& x) noexcept {
+ contents_ = std::move(x.contents_);
+ return *this;
+}
+
+extern template Cord::Cord(std::string&& src);
+extern template Cord& Cord::operator=(std::string&& src);
+
+inline size_t Cord::size() const {
+ // Length is 1st field in str.rep_
+ return contents_.size();
+}
+
+inline bool Cord::empty() const { return contents_.empty(); }
+
+inline size_t Cord::EstimatedMemoryUsage() const {
+ size_t result = sizeof(Cord);
+ if (const absl::cord_internal::CordRep* rep = contents_.tree()) {
+ result += MemoryUsageAux(rep);
+ }
+ return result;
+}
+
+inline absl::optional<absl::string_view> Cord::TryFlat() const {
+ absl::cord_internal::CordRep* rep = contents_.tree();
+ if (rep == nullptr) {
+ return absl::string_view(contents_.data(), contents_.size());
+ }
+ absl::string_view fragment;
+ if (GetFlatAux(rep, &fragment)) {
+ return fragment;
+ }
+ return absl::nullopt;
+}
+
+inline absl::string_view Cord::Flatten() {
+ absl::cord_internal::CordRep* rep = contents_.tree();
+ if (rep == nullptr) {
+ return absl::string_view(contents_.data(), contents_.size());
+ } else {
+ absl::string_view already_flat_contents;
+ if (GetFlatAux(rep, &already_flat_contents)) {
+ return already_flat_contents;
+ }
+ }
+ return FlattenSlowPath();
+}
+
+inline void Cord::Append(absl::string_view src) {
+ contents_.AppendArray(src.data(), src.size());
+}
+
+extern template void Cord::Append(std::string&& src);
+extern template void Cord::Prepend(std::string&& src);
+
+inline int Cord::Compare(const Cord& rhs) const {
+ if (!contents_.is_tree() && !rhs.contents_.is_tree()) {
+ return contents_.BitwiseCompare(rhs.contents_);
+ }
+
+ return CompareImpl(rhs);
+}
+
+// Does 'this' cord start/end with rhs
+inline bool Cord::StartsWith(const Cord& rhs) const {
+ if (contents_.IsSame(rhs.contents_)) return true;
+ size_t rhs_size = rhs.size();
+ if (size() < rhs_size) return false;
+ return EqualsImpl(rhs, rhs_size);
+}
+
+inline bool Cord::StartsWith(absl::string_view rhs) const {
+ size_t rhs_size = rhs.size();
+ if (size() < rhs_size) return false;
+ return EqualsImpl(rhs, rhs_size);
+}
+
+inline Cord::ChunkIterator::ChunkIterator(const Cord* cord)
+ : bytes_remaining_(cord->size()) {
+ if (cord->empty()) return;
+ if (cord->contents_.is_tree()) {
+ stack_of_right_children_.push_back(cord->contents_.tree());
+ operator++();
+ } else {
+ current_chunk_ = absl::string_view(cord->contents_.data(), cord->size());
+ }
+}
+
+inline Cord::ChunkIterator& Cord::ChunkIterator::operator++() {
+ ABSL_HARDENING_ASSERT(bytes_remaining_ > 0 &&
+ "Attempted to iterate past `end()`");
+ assert(bytes_remaining_ >= current_chunk_.size());
+ bytes_remaining_ -= current_chunk_.size();
+ if (bytes_remaining_ > 0) {
+ return AdvanceStack();
+ } else {
+ current_chunk_ = {};
+ }
+ return *this;
+}
+
+inline Cord::ChunkIterator Cord::ChunkIterator::operator++(int) {
+ ChunkIterator tmp(*this);
+ operator++();
+ return tmp;
+}
+
+inline bool Cord::ChunkIterator::operator==(const ChunkIterator& other) const {
+ return bytes_remaining_ == other.bytes_remaining_;
+}
+
+inline bool Cord::ChunkIterator::operator!=(const ChunkIterator& other) const {
+ return !(*this == other);
+}
+
+inline Cord::ChunkIterator::reference Cord::ChunkIterator::operator*() const {
+ ABSL_HARDENING_ASSERT(bytes_remaining_ != 0);
+ return current_chunk_;
+}
+
+inline Cord::ChunkIterator::pointer Cord::ChunkIterator::operator->() const {
+ ABSL_HARDENING_ASSERT(bytes_remaining_ != 0);
+ return ¤t_chunk_;
+}
+
+inline void Cord::ChunkIterator::RemoveChunkPrefix(size_t n) {
+ assert(n < current_chunk_.size());
+ current_chunk_.remove_prefix(n);
+ bytes_remaining_ -= n;
+}
+
+inline void Cord::ChunkIterator::AdvanceBytes(size_t n) {
+ if (ABSL_PREDICT_TRUE(n < current_chunk_.size())) {
+ RemoveChunkPrefix(n);
+ } else if (n != 0) {
+ AdvanceBytesSlowPath(n);
+ }
+}
+
+inline Cord::ChunkIterator Cord::chunk_begin() const {
+ return ChunkIterator(this);
+}
+
+inline Cord::ChunkIterator Cord::chunk_end() const { return ChunkIterator(); }
+
+inline Cord::ChunkIterator Cord::ChunkRange::begin() const {
+ return cord_->chunk_begin();
+}
+
+inline Cord::ChunkIterator Cord::ChunkRange::end() const {
+ return cord_->chunk_end();
+}
+
+inline Cord::ChunkRange Cord::Chunks() const { return ChunkRange(this); }
+
+inline Cord::CharIterator& Cord::CharIterator::operator++() {
+ if (ABSL_PREDICT_TRUE(chunk_iterator_->size() > 1)) {
+ chunk_iterator_.RemoveChunkPrefix(1);
+ } else {
+ ++chunk_iterator_;
+ }
+ return *this;
+}
+
+inline Cord::CharIterator Cord::CharIterator::operator++(int) {
+ CharIterator tmp(*this);
+ operator++();
+ return tmp;
+}
+
+inline bool Cord::CharIterator::operator==(const CharIterator& other) const {
+ return chunk_iterator_ == other.chunk_iterator_;
+}
+
+inline bool Cord::CharIterator::operator!=(const CharIterator& other) const {
+ return !(*this == other);
+}
+
+inline Cord::CharIterator::reference Cord::CharIterator::operator*() const {
+ return *chunk_iterator_->data();
+}
+
+inline Cord::CharIterator::pointer Cord::CharIterator::operator->() const {
+ return chunk_iterator_->data();
+}
+
+inline Cord Cord::AdvanceAndRead(CharIterator* it, size_t n_bytes) {
+ assert(it != nullptr);
+ return it->chunk_iterator_.AdvanceAndReadBytes(n_bytes);
+}
+
+inline void Cord::Advance(CharIterator* it, size_t n_bytes) {
+ assert(it != nullptr);
+ it->chunk_iterator_.AdvanceBytes(n_bytes);
+}
+
+inline absl::string_view Cord::ChunkRemaining(const CharIterator& it) {
+ return *it.chunk_iterator_;
+}
+
+inline Cord::CharIterator Cord::char_begin() const {
+ return CharIterator(this);
+}
+
+inline Cord::CharIterator Cord::char_end() const { return CharIterator(); }
+
+inline Cord::CharIterator Cord::CharRange::begin() const {
+ return cord_->char_begin();
+}
+
+inline Cord::CharIterator Cord::CharRange::end() const {
+ return cord_->char_end();
+}
+
+inline Cord::CharRange Cord::Chars() const { return CharRange(this); }
+
+inline void Cord::ForEachChunk(
+ absl::FunctionRef<void(absl::string_view)> callback) const {
+ absl::cord_internal::CordRep* rep = contents_.tree();
+ if (rep == nullptr) {
+ callback(absl::string_view(contents_.data(), contents_.size()));
+ } else {
+ return ForEachChunkAux(rep, callback);
+ }
+}
+
+// Nonmember Cord-to-Cord relational operarators.
+inline bool operator==(const Cord& lhs, const Cord& rhs) {
+ if (lhs.contents_.IsSame(rhs.contents_)) return true;
+ size_t rhs_size = rhs.size();
+ if (lhs.size() != rhs_size) return false;
+ return lhs.EqualsImpl(rhs, rhs_size);
+}
+
+inline bool operator!=(const Cord& x, const Cord& y) { return !(x == y); }
+inline bool operator<(const Cord& x, const Cord& y) {
+ return x.Compare(y) < 0;
+}
+inline bool operator>(const Cord& x, const Cord& y) {
+ return x.Compare(y) > 0;
+}
+inline bool operator<=(const Cord& x, const Cord& y) {
+ return x.Compare(y) <= 0;
+}
+inline bool operator>=(const Cord& x, const Cord& y) {
+ return x.Compare(y) >= 0;
+}
+
+// Nonmember Cord-to-absl::string_view relational operators.
+//
+// Due to implicit conversions, these also enable comparisons of Cord with
+// with std::string, ::string, and const char*.
+inline bool operator==(const Cord& lhs, absl::string_view rhs) {
+ size_t lhs_size = lhs.size();
+ size_t rhs_size = rhs.size();
+ if (lhs_size != rhs_size) return false;
+ return lhs.EqualsImpl(rhs, rhs_size);
+}
+
+inline bool operator==(absl::string_view x, const Cord& y) { return y == x; }
+inline bool operator!=(const Cord& x, absl::string_view y) { return !(x == y); }
+inline bool operator!=(absl::string_view x, const Cord& y) { return !(x == y); }
+inline bool operator<(const Cord& x, absl::string_view y) {
+ return x.Compare(y) < 0;
+}
+inline bool operator<(absl::string_view x, const Cord& y) {
+ return y.Compare(x) > 0;
+}
+inline bool operator>(const Cord& x, absl::string_view y) { return y < x; }
+inline bool operator>(absl::string_view x, const Cord& y) { return y < x; }
+inline bool operator<=(const Cord& x, absl::string_view y) { return !(y < x); }
+inline bool operator<=(absl::string_view x, const Cord& y) { return !(y < x); }
+inline bool operator>=(const Cord& x, absl::string_view y) { return !(x < y); }
+inline bool operator>=(absl::string_view x, const Cord& y) { return !(x < y); }
+
+// Some internals exposed to test code.
+namespace strings_internal {
+class CordTestAccess {
+ public:
+ static size_t FlatOverhead();
+ static size_t MaxFlatLength();
+ static size_t SizeofCordRepConcat();
+ static size_t SizeofCordRepExternal();
+ static size_t SizeofCordRepSubstring();
+ static size_t FlatTagToLength(uint8_t tag);
+ static uint8_t LengthToTag(size_t s);
+};
+} // namespace strings_internal
+ABSL_NAMESPACE_END
+} // namespace absl
+
+#endif // ABSL_STRINGS_CORD_H_
diff --git a/absl/strings/cord_test.cc b/absl/strings/cord_test.cc
new file mode 100644
index 0000000..7942bfc
--- /dev/null
+++ b/absl/strings/cord_test.cc
@@ -0,0 +1,1711 @@
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "absl/strings/cord.h"
+
+#include <algorithm>
+#include <climits>
+#include <cstdio>
+#include <iterator>
+#include <map>
+#include <numeric>
+#include <random>
+#include <sstream>
+#include <type_traits>
+#include <utility>
+#include <vector>
+
+#include "gmock/gmock.h"
+#include "gtest/gtest.h"
+#include "absl/base/casts.h"
+#include "absl/base/config.h"
+#include "absl/base/internal/endian.h"
+#include "absl/base/internal/raw_logging.h"
+#include "absl/base/macros.h"
+#include "absl/container/fixed_array.h"
+#include "absl/strings/cord_test_helpers.h"
+#include "absl/strings/str_cat.h"
+#include "absl/strings/str_format.h"
+#include "absl/strings/string_view.h"
+
+typedef std::mt19937_64 RandomEngine;
+
+static std::string RandomLowercaseString(RandomEngine* rng);
+static std::string RandomLowercaseString(RandomEngine* rng, size_t length);
+
+static int GetUniformRandomUpTo(RandomEngine* rng, int upper_bound) {
+ if (upper_bound > 0) {
+ std::uniform_int_distribution<int> uniform(0, upper_bound - 1);
+ return uniform(*rng);
+ } else {
+ return 0;
+ }
+}
+
+static size_t GetUniformRandomUpTo(RandomEngine* rng, size_t upper_bound) {
+ if (upper_bound > 0) {
+ std::uniform_int_distribution<size_t> uniform(0, upper_bound - 1);
+ return uniform(*rng);
+ } else {
+ return 0;
+ }
+}
+
+static int32_t GenerateSkewedRandom(RandomEngine* rng, int max_log) {
+ const uint32_t base = (*rng)() % (max_log + 1);
+ const uint32_t mask = ((base < 32) ? (1u << base) : 0u) - 1u;
+ return (*rng)() & mask;
+}
+
+static std::string RandomLowercaseString(RandomEngine* rng) {
+ int length;
+ std::bernoulli_distribution one_in_1k(0.001);
+ std::bernoulli_distribution one_in_10k(0.0001);
+ // With low probability, make a large fragment
+ if (one_in_10k(*rng)) {
+ length = GetUniformRandomUpTo(rng, 1048576);
+ } else if (one_in_1k(*rng)) {
+ length = GetUniformRandomUpTo(rng, 10000);
+ } else {
+ length = GenerateSkewedRandom(rng, 10);
+ }
+ return RandomLowercaseString(rng, length);
+}
+
+static std::string RandomLowercaseString(RandomEngine* rng, size_t length) {
+ std::string result(length, '\0');
+ std::uniform_int_distribution<int> chars('a', 'z');
+ std::generate(result.begin(), result.end(),
+ [&]() { return static_cast<char>(chars(*rng)); });
+ return result;
+}
+
+static void DoNothing(absl::string_view /* data */, void* /* arg */) {}
+
+static void DeleteExternalString(absl::string_view data, void* arg) {
+ std::string* s = reinterpret_cast<std::string*>(arg);
+ EXPECT_EQ(data, *s);
+ delete s;
+}
+
+// Add "s" to *dst via `MakeCordFromExternal`
+static void AddExternalMemory(absl::string_view s, absl::Cord* dst) {
+ std::string* str = new std::string(s.data(), s.size());
+ dst->Append(absl::MakeCordFromExternal(*str, [str](absl::string_view data) {
+ DeleteExternalString(data, str);
+ }));
+}
+
+static void DumpGrowth() {
+ absl::Cord str;
+ for (int i = 0; i < 1000; i++) {
+ char c = 'a' + i % 26;
+ str.Append(absl::string_view(&c, 1));
+ }
+}
+
+// Make a Cord with some number of fragments. Return the size (in bytes)
+// of the smallest fragment.
+static size_t AppendWithFragments(const std::string& s, RandomEngine* rng,
+ absl::Cord* cord) {
+ size_t j = 0;
+ const size_t max_size = s.size() / 5; // Make approx. 10 fragments
+ size_t min_size = max_size; // size of smallest fragment
+ while (j < s.size()) {
+ size_t N = 1 + GetUniformRandomUpTo(rng, max_size);
+ if (N > (s.size() - j)) {
+ N = s.size() - j;
+ }
+ if (N < min_size) {
+ min_size = N;
+ }
+
+ std::bernoulli_distribution coin_flip(0.5);
+ if (coin_flip(*rng)) {
+ // Grow by adding an external-memory.
+ AddExternalMemory(absl::string_view(s.data() + j, N), cord);
+ } else {
+ cord->Append(absl::string_view(s.data() + j, N));
+ }
+ j += N;
+ }
+ return min_size;
+}
+
+// Add an external memory that contains the specified std::string to cord
+static void AddNewStringBlock(const std::string& str, absl::Cord* dst) {
+ char* data = new char[str.size()];
+ memcpy(data, str.data(), str.size());
+ dst->Append(absl::MakeCordFromExternal(
+ absl::string_view(data, str.size()),
+ [](absl::string_view s) { delete[] s.data(); }));
+}
+
+// Make a Cord out of many different types of nodes.
+static absl::Cord MakeComposite() {
+ absl::Cord cord;
+ cord.Append("the");
+ AddExternalMemory(" quick brown", &cord);
+ AddExternalMemory(" fox jumped", &cord);
+
+ absl::Cord full(" over");
+ AddExternalMemory(" the lazy", &full);
+ AddNewStringBlock(" dog slept the whole day away", &full);
+ absl::Cord substring = full.Subcord(0, 18);
+
+ // Make substring long enough to defeat the copying fast path in Append.
+ substring.Append(std::string(1000, '.'));
+ cord.Append(substring);
+ cord = cord.Subcord(0, cord.size() - 998); // Remove most of extra junk
+
+ return cord;
+}
+
+namespace absl {
+ABSL_NAMESPACE_BEGIN
+
+class CordTestPeer {
+ public:
+ static void ForEachChunk(
+ const Cord& c, absl::FunctionRef<void(absl::string_view)> callback) {
+ c.ForEachChunk(callback);
+ }
+
+ static bool IsTree(const Cord& c) { return c.contents_.is_tree(); }
+};
+
+ABSL_NAMESPACE_END
+} // namespace absl
+
+TEST(Cord, AllFlatSizes) {
+ using absl::strings_internal::CordTestAccess;
+
+ for (size_t s = 0; s < CordTestAccess::MaxFlatLength(); s++) {
+ // Make a string of length s.
+ std::string src;
+ while (src.size() < s) {
+ src.push_back('a' + (src.size() % 26));
+ }
+
+ absl::Cord dst(src);
+ EXPECT_EQ(std::string(dst), src) << s;
+ }
+}
+
+// We create a Cord at least 128GB in size using the fact that Cords can
+// internally reference-count; thus the Cord is enormous without actually
+// consuming very much memory.
+TEST(GigabyteCord, FromExternal) {
+ const size_t one_gig = 1024U * 1024U * 1024U;
+ size_t max_size = 2 * one_gig;
+ if (sizeof(max_size) > 4) max_size = 128 * one_gig;
+
+ size_t length = 128 * 1024;
+ char* data = new char[length];
+ absl::Cord from = absl::MakeCordFromExternal(
+ absl::string_view(data, length),
+ [](absl::string_view sv) { delete[] sv.data(); });
+
+ // This loop may seem odd due to its combination of exponential doubling of
+ // size and incremental size increases. We do it incrementally to be sure the
+ // Cord will need rebalancing and will exercise code that, in the past, has
+ // caused crashes in production. We grow exponentially so that the code will
+ // execute in a reasonable amount of time.
+ absl::Cord c;
+ ABSL_RAW_LOG(INFO, "Made a Cord with %zu bytes!", c.size());
+ c.Append(from);
+ while (c.size() < max_size) {
+ c.Append(c);
+ c.Append(from);
+ c.Append(from);
+ c.Append(from);
+ c.Append(from);
+ }
+
+ for (int i = 0; i < 1024; ++i) {
+ c.Append(from);
+ }
+ ABSL_RAW_LOG(INFO, "Made a Cord with %zu bytes!", c.size());
+ // Note: on a 32-bit build, this comes out to 2,818,048,000 bytes.
+ // Note: on a 64-bit build, this comes out to 171,932,385,280 bytes.
+}
+
+static absl::Cord MakeExternalCord(int size) {
+ char* buffer = new char[size];
+ memset(buffer, 'x', size);
+ absl::Cord cord;
+ cord.Append(absl::MakeCordFromExternal(
+ absl::string_view(buffer, size),
+ [](absl::string_view s) { delete[] s.data(); }));
+ return cord;
+}
+
+// Extern to fool clang that this is not constant. Needed to suppress
+// a warning of unsafe code we want to test.
+extern bool my_unique_true_boolean;
+bool my_unique_true_boolean = true;
+
+TEST(Cord, Assignment) {
+ absl::Cord x(absl::string_view("hi there"));
+ absl::Cord y(x);
+ ASSERT_EQ(std::string(x), "hi there");
+ ASSERT_EQ(std::string(y), "hi there");
+ ASSERT_TRUE(x == y);
+ ASSERT_TRUE(x <= y);
+ ASSERT_TRUE(y <= x);
+
+ x = absl::string_view("foo");
+ ASSERT_EQ(std::string(x), "foo");
+ ASSERT_EQ(std::string(y), "hi there");
+ ASSERT_TRUE(x < y);
+ ASSERT_TRUE(y > x);
+ ASSERT_TRUE(x != y);
+ ASSERT_TRUE(x <= y);
+ ASSERT_TRUE(y >= x);
+
+ x = "foo";
+ ASSERT_EQ(x, "foo");
+
+ // Test that going from inline rep to tree we don't leak memory.
+ std::vector<std::pair<absl::string_view, absl::string_view>>
+ test_string_pairs = {{"hi there", "foo"},
+ {"loooooong coooooord", "short cord"},
+ {"short cord", "loooooong coooooord"},
+ {"loooooong coooooord1", "loooooong coooooord2"}};
+ for (std::pair<absl::string_view, absl::string_view> test_strings :
+ test_string_pairs) {
+ absl::Cord tmp(test_strings.first);
+ absl::Cord z(std::move(tmp));
+ ASSERT_EQ(std::string(z), test_strings.first);
+ tmp = test_strings.second;
+ z = std::move(tmp);
+ ASSERT_EQ(std::string(z), test_strings.second);
+ }
+ {
+ // Test that self-move assignment doesn't crash/leak.
+ // Do not write such code!
+ absl::Cord my_small_cord("foo");
+ absl::Cord my_big_cord("loooooong coooooord");
+ // Bypass clang's warning on self move-assignment.
+ absl::Cord* my_small_alias =
+ my_unique_true_boolean ? &my_small_cord : &my_big_cord;
+ absl::Cord* my_big_alias =
+ !my_unique_true_boolean ? &my_small_cord : &my_big_cord;
+
+ *my_small_alias = std::move(my_small_cord);
+ *my_big_alias = std::move(my_big_cord);
+ // my_small_cord and my_big_cord are in an unspecified but valid
+ // state, and will be correctly destroyed here.
+ }
+}
+
+TEST(Cord, StartsEndsWith) {
+ absl::Cord x(absl::string_view("abcde"));
+ absl::Cord empty("");
+
+ ASSERT_TRUE(x.StartsWith(absl::Cord("abcde")));
+ ASSERT_TRUE(x.StartsWith(absl::Cord("abc")));
+ ASSERT_TRUE(x.StartsWith(absl::Cord("")));
+ ASSERT_TRUE(empty.StartsWith(absl::Cord("")));
+ ASSERT_TRUE(x.EndsWith(absl::Cord("abcde")));
+ ASSERT_TRUE(x.EndsWith(absl::Cord("cde")));
+ ASSERT_TRUE(x.EndsWith(absl::Cord("")));
+ ASSERT_TRUE(empty.EndsWith(absl::Cord("")));
+
+ ASSERT_TRUE(!x.StartsWith(absl::Cord("xyz")));
+ ASSERT_TRUE(!empty.StartsWith(absl::Cord("xyz")));
+ ASSERT_TRUE(!x.EndsWith(absl::Cord("xyz")));
+ ASSERT_TRUE(!empty.EndsWith(absl::Cord("xyz")));
+
+ ASSERT_TRUE(x.StartsWith("abcde"));
+ ASSERT_TRUE(x.StartsWith("abc"));
+ ASSERT_TRUE(x.StartsWith(""));
+ ASSERT_TRUE(empty.StartsWith(""));
+ ASSERT_TRUE(x.EndsWith("abcde"));
+ ASSERT_TRUE(x.EndsWith("cde"));
+ ASSERT_TRUE(x.EndsWith(""));
+ ASSERT_TRUE(empty.EndsWith(""));
+
+ ASSERT_TRUE(!x.StartsWith("xyz"));
+ ASSERT_TRUE(!empty.StartsWith("xyz"));
+ ASSERT_TRUE(!x.EndsWith("xyz"));
+ ASSERT_TRUE(!empty.EndsWith("xyz"));
+}
+
+TEST(Cord, Subcord) {
+ RandomEngine rng(testing::GTEST_FLAG(random_seed));
+ const std::string s = RandomLowercaseString(&rng, 1024);
+
+ absl::Cord a;
+ AppendWithFragments(s, &rng, &a);
+ ASSERT_EQ(s.size(), a.size());
+
+ // Check subcords of a, from a variety of interesting points.
+ std::set<size_t> positions;
+ for (int i = 0; i <= 32; ++i) {
+ positions.insert(i);
+ positions.insert(i * 32 - 1);
+ positions.insert(i * 32);
+ positions.insert(i * 32 + 1);
+ positions.insert(a.size() - i);
+ }
+ positions.insert(237);
+ positions.insert(732);
+ for (size_t pos : positions) {
+ if (pos > a.size()) continue;
+ for (size_t end_pos : positions) {
+ if (end_pos < pos || end_pos > a.size()) continue;
+ absl::Cord sa = a.Subcord(pos, end_pos - pos);
+ EXPECT_EQ(absl::string_view(s).substr(pos, end_pos - pos),
+ std::string(sa))
+ << a;
+ }
+ }
+
+ // Do the same thing for an inline cord.
+ const std::string sh = "short";
+ absl::Cord c(sh);
+ for (size_t pos = 0; pos <= sh.size(); ++pos) {
+ for (size_t n = 0; n <= sh.size() - pos; ++n) {
+ absl::Cord sc = c.Subcord(pos, n);
+ EXPECT_EQ(sh.substr(pos, n), std::string(sc)) << c;
+ }
+ }
+
+ // Check subcords of subcords.
+ absl::Cord sa = a.Subcord(0, a.size());
+ std::string ss = s.substr(0, s.size());
+ while (sa.size() > 1) {
+ sa = sa.Subcord(1, sa.size() - 2);
+ ss = ss.substr(1, ss.size() - 2);
+ EXPECT_EQ(ss, std::string(sa)) << a;
+ if (HasFailure()) break; // halt cascade
+ }
+
+ // It is OK to ask for too much.
+ sa = a.Subcord(0, a.size() + 1);
+ EXPECT_EQ(s, std::string(sa));
+
+ // It is OK to ask for something beyond the end.
+ sa = a.Subcord(a.size() + 1, 0);
+ EXPECT_TRUE(sa.empty());
+ sa = a.Subcord(a.size() + 1, 1);
+ EXPECT_TRUE(sa.empty());
+}
+
+TEST(Cord, Swap) {
+ absl::string_view a("Dexter");
+ absl::string_view b("Mandark");
+ absl::Cord x(a);
+ absl::Cord y(b);
+ swap(x, y);
+ ASSERT_EQ(x, absl::Cord(b));
+ ASSERT_EQ(y, absl::Cord(a));
+ x.swap(y);
+ ASSERT_EQ(x, absl::Cord(a));
+ ASSERT_EQ(y, absl::Cord(b));
+}
+
+static void VerifyCopyToString(const absl::Cord& cord) {
+ std::string initially_empty;
+ absl::CopyCordToString(cord, &initially_empty);
+ EXPECT_EQ(initially_empty, cord);
+
+ constexpr size_t kInitialLength = 1024;
+ std::string has_initial_contents(kInitialLength, 'x');
+ const char* address_before_copy = has_initial_contents.data();
+ absl::CopyCordToString(cord, &has_initial_contents);
+ EXPECT_EQ(has_initial_contents, cord);
+
+ if (cord.size() <= kInitialLength) {
+ EXPECT_EQ(has_initial_contents.data(), address_before_copy)
+ << "CopyCordToString allocated new string storage; "
+ "has_initial_contents = \""
+ << has_initial_contents << "\"";
+ }
+}
+
+TEST(Cord, CopyToString) {
+ VerifyCopyToString(absl::Cord());
+ VerifyCopyToString(absl::Cord("small cord"));
+ VerifyCopyToString(
+ absl::MakeFragmentedCord({"fragmented ", "cord ", "to ", "test ",
+ "copying ", "to ", "a ", "string."}));
+}
+
+TEST(TryFlat, Empty) {
+ absl::Cord c;
+ EXPECT_EQ(c.TryFlat(), "");
+}
+
+TEST(TryFlat, Flat) {
+ absl::Cord c("hello");
+ EXPECT_EQ(c.TryFlat(), "hello");
+}
+
+TEST(TryFlat, SubstrInlined) {
+ absl::Cord c("hello");
+ c.RemovePrefix(1);
+ EXPECT_EQ(c.TryFlat(), "ello");
+}
+
+TEST(TryFlat, SubstrFlat) {
+ absl::Cord c("longer than 15 bytes");
+ c.RemovePrefix(1);
+ EXPECT_EQ(c.TryFlat(), "onger than 15 bytes");
+}
+
+TEST(TryFlat, Concat) {
+ absl::Cord c = absl::MakeFragmentedCord({"hel", "lo"});
+ EXPECT_EQ(c.TryFlat(), absl::nullopt);
+}
+
+TEST(TryFlat, External) {
+ absl::Cord c = absl::MakeCordFromExternal("hell", [](absl::string_view) {});
+ EXPECT_EQ(c.TryFlat(), "hell");
+}
+
+TEST(TryFlat, SubstrExternal) {
+ absl::Cord c = absl::MakeCordFromExternal("hell", [](absl::string_view) {});
+ c.RemovePrefix(1);
+ EXPECT_EQ(c.TryFlat(), "ell");
+}
+
+TEST(TryFlat, SubstrConcat) {
+ absl::Cord c = absl::MakeFragmentedCord({"hello", " world"});
+ c.RemovePrefix(1);
+ EXPECT_EQ(c.TryFlat(), absl::nullopt);
+}
+
+static bool IsFlat(const absl::Cord& c) {
+ return c.chunk_begin() == c.chunk_end() || ++c.chunk_begin() == c.chunk_end();
+}
+
+static void VerifyFlatten(absl::Cord c) {
+ std::string old_contents(c);
+ absl::string_view old_flat;
+ bool already_flat_and_non_empty = IsFlat(c) && !c.empty();
+ if (already_flat_and_non_empty) {
+ old_flat = *c.chunk_begin();
+ }
+ absl::string_view new_flat = c.Flatten();
+
+ // Verify that the contents of the flattened Cord are correct.
+ EXPECT_EQ(new_flat, old_contents);
+ EXPECT_EQ(std::string(c), old_contents);
+
+ // If the Cord contained data and was already flat, verify that the data
+ // wasn't copied.
+ if (already_flat_and_non_empty) {
+ EXPECT_EQ(old_flat.data(), new_flat.data())
+ << "Allocated new memory even though the Cord was already flat.";
+ }
+
+ // Verify that the flattened Cord is in fact flat.
+ EXPECT_TRUE(IsFlat(c));
+}
+
+TEST(Cord, Flatten) {
+ VerifyFlatten(absl::Cord());
+ VerifyFlatten(absl::Cord("small cord"));
+ VerifyFlatten(absl::Cord("larger than small buffer optimization"));
+ VerifyFlatten(absl::MakeFragmentedCord({"small ", "fragmented ", "cord"}));
+
+ // Test with a cord that is longer than the largest flat buffer
+ RandomEngine rng(testing::GTEST_FLAG(random_seed));
+ VerifyFlatten(absl::Cord(RandomLowercaseString(&rng, 8192)));
+}
+
+// Test data
+namespace {
+class TestData {
+ private:
+ std::vector<std::string> data_;
+
+ // Return a std::string of the specified length.
+ static std::string MakeString(int length) {
+ std::string result;
+ char buf[30];
+ snprintf(buf, sizeof(buf), "(%d)", length);
+ while (result.size() < length) {
+ result += buf;
+ }
+ result.resize(length);
+ return result;
+ }
+
+ public:
+ TestData() {
+ // short strings increasing in length by one
+ for (int i = 0; i < 30; i++) {
+ data_.push_back(MakeString(i));
+ }
+
+ // strings around half kMaxFlatLength
+ static const int kMaxFlatLength = 4096 - 9;
+ static const int kHalf = kMaxFlatLength / 2;
+
+ for (int i = -10; i <= +10; i++) {
+ data_.push_back(MakeString(kHalf + i));
+ }
+
+ for (int i = -10; i <= +10; i++) {
+ data_.push_back(MakeString(kMaxFlatLength + i));
+ }
+ }
+
+ size_t size() const { return data_.size(); }
+ const std::string& data(size_t i) const { return data_[i]; }
+};
+} // namespace
+
+TEST(Cord, MultipleLengths) {
+ TestData d;
+ for (size_t i = 0; i < d.size(); i++) {
+ std::string a = d.data(i);
+
+ { // Construct from Cord
+ absl::Cord tmp(a);
+ absl::Cord x(tmp);
+ EXPECT_EQ(a, std::string(x)) << "'" << a << "'";
+ }
+
+ { // Construct from absl::string_view
+ absl::Cord x(a);
+ EXPECT_EQ(a, std::string(x)) << "'" << a << "'";
+ }
+
+ { // Append cord to self
+ absl::Cord self(a);
+ self.Append(self);
+ EXPECT_EQ(a + a, std::string(self)) << "'" << a << "' + '" << a << "'";
+ }
+
+ { // Prepend cord to self
+ absl::Cord self(a);
+ self.Prepend(self);
+ EXPECT_EQ(a + a, std::string(self)) << "'" << a << "' + '" << a << "'";
+ }
+
+ // Try to append/prepend others
+ for (size_t j = 0; j < d.size(); j++) {
+ std::string b = d.data(j);
+
+ { // CopyFrom Cord
+ absl::Cord x(a);
+ absl::Cord y(b);
+ x = y;
+ EXPECT_EQ(b, std::string(x)) << "'" << a << "' + '" << b << "'";
+ }
+
+ { // CopyFrom absl::string_view
+ absl::Cord x(a);
+ x = b;
+ EXPECT_EQ(b, std::string(x)) << "'" << a << "' + '" << b << "'";
+ }
+
+ { // Cord::Append(Cord)
+ absl::Cord x(a);
+ absl::Cord y(b);
+ x.Append(y);
+ EXPECT_EQ(a + b, std::string(x)) << "'" << a << "' + '" << b << "'";
+ }
+
+ { // Cord::Append(absl::string_view)
+ absl::Cord x(a);
+ x.Append(b);
+ EXPECT_EQ(a + b, std::string(x)) << "'" << a << "' + '" << b << "'";
+ }
+
+ { // Cord::Prepend(Cord)
+ absl::Cord x(a);
+ absl::Cord y(b);
+ x.Prepend(y);
+ EXPECT_EQ(b + a, std::string(x)) << "'" << b << "' + '" << a << "'";
+ }
+
+ { // Cord::Prepend(absl::string_view)
+ absl::Cord x(a);
+ x.Prepend(b);
+ EXPECT_EQ(b + a, std::string(x)) << "'" << b << "' + '" << a << "'";
+ }
+ }
+ }
+}
+
+namespace {
+
+TEST(Cord, RemoveSuffixWithExternalOrSubstring) {
+ absl::Cord cord = absl::MakeCordFromExternal(
+ "foo bar baz", [](absl::string_view s) { DoNothing(s, nullptr); });
+
+ EXPECT_EQ("foo bar baz", std::string(cord));
+
+ // This RemoveSuffix() will wrap the EXTERNAL node in a SUBSTRING node.
+ cord.RemoveSuffix(4);
+ EXPECT_EQ("foo bar", std::string(cord));
+
+ // This RemoveSuffix() will adjust the SUBSTRING node in-place.
+ cord.RemoveSuffix(4);
+ EXPECT_EQ("foo", std::string(cord));
+}
+
+TEST(Cord, RemoveSuffixMakesZeroLengthNode) {
+ absl::Cord c;
+ c.Append(absl::Cord(std::string(100, 'x')));
+ absl::Cord other_ref = c; // Prevent inplace appends
+ c.Append(absl::Cord(std::string(200, 'y')));
+ c.RemoveSuffix(200);
+ EXPECT_EQ(std::string(100, 'x'), std::string(c));
+}
+
+} // namespace
+
+// CordSpliceTest contributed by hendrie.
+namespace {
+
+// Create a cord with an external memory block filled with 'z'
+absl::Cord CordWithZedBlock(size_t size) {
+ char* data = new char[size];
+ if (size > 0) {
+ memset(data, 'z', size);
+ }
+ absl::Cord cord = absl::MakeCordFromExternal(
+ absl::string_view(data, size),
+ [](absl::string_view s) { delete[] s.data(); });
+ return cord;
+}
+
+// Establish that ZedBlock does what we think it does.
+TEST(CordSpliceTest, ZedBlock) {
+ absl::Cord blob = CordWithZedBlock(10);
+ EXPECT_EQ(10, blob.size());
+ std::string s;
+ absl::CopyCordToString(blob, &s);
+ EXPECT_EQ("zzzzzzzzzz", s);
+}
+
+TEST(CordSpliceTest, ZedBlock0) {
+ absl::Cord blob = CordWithZedBlock(0);
+ EXPECT_EQ(0, blob.size());
+ std::string s;
+ absl::CopyCordToString(blob, &s);
+ EXPECT_EQ("", s);
+}
+
+TEST(CordSpliceTest, ZedBlockSuffix1) {
+ absl::Cord blob = CordWithZedBlock(10);
+ EXPECT_EQ(10, blob.size());
+ absl::Cord suffix(blob);
+ suffix.RemovePrefix(9);
+ EXPECT_EQ(1, suffix.size());
+ std::string s;
+ absl::CopyCordToString(suffix, &s);
+ EXPECT_EQ("z", s);
+}
+
+// Remove all of a prefix block
+TEST(CordSpliceTest, ZedBlockSuffix0) {
+ absl::Cord blob = CordWithZedBlock(10);
+ EXPECT_EQ(10, blob.size());
+ absl::Cord suffix(blob);
+ suffix.RemovePrefix(10);
+ EXPECT_EQ(0, suffix.size());
+ std::string s;
+ absl::CopyCordToString(suffix, &s);
+ EXPECT_EQ("", s);
+}
+
+absl::Cord BigCord(size_t len, char v) {
+ std::string s(len, v);
+ return absl::Cord(s);
+}
+
+// Splice block into cord.
+absl::Cord SpliceCord(const absl::Cord& blob, int64_t offset,
+ const absl::Cord& block) {
+ ABSL_RAW_CHECK(offset >= 0, "");
+ ABSL_RAW_CHECK(offset + block.size() <= blob.size(), "");
+ absl::Cord result(blob);
+ result.RemoveSuffix(blob.size() - offset);
+ result.Append(block);
+ absl::Cord suffix(blob);
+ suffix.RemovePrefix(offset + block.size());
+ result.Append(suffix);
+ ABSL_RAW_CHECK(blob.size() == result.size(), "");
+ return result;
+}
+
+// Taking an empty suffix of a block breaks appending.
+TEST(CordSpliceTest, RemoveEntireBlock1) {
+ absl::Cord zero = CordWithZedBlock(10);
+ absl::Cord suffix(zero);
+ suffix.RemovePrefix(10);
+ absl::Cord result;
+ result.Append(suffix);
+}
+
+TEST(CordSpliceTest, RemoveEntireBlock2) {
+ absl::Cord zero = CordWithZedBlock(10);
+ absl::Cord prefix(zero);
+ prefix.RemoveSuffix(10);
+ absl::Cord suffix(zero);
+ suffix.RemovePrefix(10);
+ absl::Cord result(prefix);
+ result.Append(suffix);
+}
+
+TEST(CordSpliceTest, RemoveEntireBlock3) {
+ absl::Cord blob = CordWithZedBlock(10);
+ absl::Cord block = BigCord(10, 'b');
+ blob = SpliceCord(blob, 0, block);
+}
+
+struct CordCompareTestCase {
+ template <typename LHS, typename RHS>
+ CordCompareTestCase(const LHS& lhs, const RHS& rhs)
+ : lhs_cord(lhs), rhs_cord(rhs) {}
+
+ absl::Cord lhs_cord;
+ absl::Cord rhs_cord;
+};
+
+const auto sign = [](int x) { return x == 0 ? 0 : (x > 0 ? 1 : -1); };
+
+void VerifyComparison(const CordCompareTestCase& test_case) {
+ std::string lhs_string(test_case.lhs_cord);
+ std::string rhs_string(test_case.rhs_cord);
+ int expected = sign(lhs_string.compare(rhs_string));
+ EXPECT_EQ(expected, test_case.lhs_cord.Compare(test_case.rhs_cord))
+ << "LHS=" << lhs_string << "; RHS=" << rhs_string;
+ EXPECT_EQ(expected, test_case.lhs_cord.Compare(rhs_string))
+ << "LHS=" << lhs_string << "; RHS=" << rhs_string;
+ EXPECT_EQ(-expected, test_case.rhs_cord.Compare(test_case.lhs_cord))
+ << "LHS=" << rhs_string << "; RHS=" << lhs_string;
+ EXPECT_EQ(-expected, test_case.rhs_cord.Compare(lhs_string))
+ << "LHS=" << rhs_string << "; RHS=" << lhs_string;
+}
+
+TEST(Cord, Compare) {
+ absl::Cord subcord("aaaaaBBBBBcccccDDDDD");
+ subcord = subcord.Subcord(3, 10);
+
+ absl::Cord tmp("aaaaaaaaaaaaaaaa");
+ tmp.Append("BBBBBBBBBBBBBBBB");
+ absl::Cord concat = absl::Cord("cccccccccccccccc");
+ concat.Append("DDDDDDDDDDDDDDDD");
+ concat.Prepend(tmp);
+
+ absl::Cord concat2("aaaaaaaaaaaaa");
+ concat2.Append("aaaBBBBBBBBBBBBBBBBccccc");
+ concat2.Append("cccccccccccDDDDDDDDDDDDDD");
+ concat2.Append("DD");
+
+ std::vector<CordCompareTestCase> test_cases = {{
+ // Inline cords
+ {"abcdef", "abcdef"},
+ {"abcdef", "abcdee"},
+ {"abcdef", "abcdeg"},
+ {"bbcdef", "abcdef"},
+ {"bbcdef", "abcdeg"},
+ {"abcdefa", "abcdef"},
+ {"abcdef", "abcdefa"},
+
+ // Small flat cords
+ {"aaaaaBBBBBcccccDDDDD", "aaaaaBBBBBcccccDDDDD"},
+ {"aaaaaBBBBBcccccDDDDD", "aaaaaBBBBBxccccDDDDD"},
+ {"aaaaaBBBBBcxcccDDDDD", "aaaaaBBBBBcccccDDDDD"},
+ {"aaaaaBBBBBxccccDDDDD", "aaaaaBBBBBcccccDDDDX"},
+ {"aaaaaBBBBBcccccDDDDDa", "aaaaaBBBBBcccccDDDDD"},
+ {"aaaaaBBBBBcccccDDDDD", "aaaaaBBBBBcccccDDDDDa"},
+
+ // Subcords
+ {subcord, subcord},
+ {subcord, "aaBBBBBccc"},
+ {subcord, "aaBBBBBccd"},
+ {subcord, "aaBBBBBccb"},
+ {subcord, "aaBBBBBxcb"},
+ {subcord, "aaBBBBBccca"},
+ {subcord, "aaBBBBBcc"},
+
+ // Concats
+ {concat, concat},
+ {concat,
+ "aaaaaaaaaaaaaaaaBBBBBBBBBBBBBBBBccccccccccccccccDDDDDDDDDDDDDDDD"},
+ {concat,
+ "aaaaaaaaaaaaaaaaBBBBBBBBBBBBBBBBcccccccccccccccxDDDDDDDDDDDDDDDD"},
+ {concat,
+ "aaaaaaaaaaaaaaaaBBBBBBBBBBBBBBBBacccccccccccccccDDDDDDDDDDDDDDDD"},
+ {concat,
+ "aaaaaaaaaaaaaaaaBBBBBBBBBBBBBBBBccccccccccccccccDDDDDDDDDDDDDDD"},
+ {concat,
+ "aaaaaaaaaaaaaaaaBBBBBBBBBBBBBBBBccccccccccccccccDDDDDDDDDDDDDDDDe"},
+
+ {concat, concat2},
+ }};
+
+ for (const auto& tc : test_cases) {
+ VerifyComparison(tc);
+ }
+}
+
+TEST(Cord, CompareAfterAssign) {
+ absl::Cord a("aaaaaa1111111");
+ absl::Cord b("aaaaaa2222222");
+ a = "cccccc";
+ b = "cccccc";
+ EXPECT_EQ(a, b);
+ EXPECT_FALSE(a < b);
+
+ a = "aaaa";
+ b = "bbbbb";
+ a = "";
+ b = "";
+ EXPECT_EQ(a, b);
+ EXPECT_FALSE(a < b);
+}
+
+// Test CompareTo() and ComparePrefix() against string and substring
+// comparison methods from basic_string.
+static void TestCompare(const absl::Cord& c, const absl::Cord& d,
+ RandomEngine* rng) {
+ typedef std::basic_string<uint8_t> ustring;
+ ustring cs(reinterpret_cast<const uint8_t*>(std::string(c).data()), c.size());
+ ustring ds(reinterpret_cast<const uint8_t*>(std::string(d).data()), d.size());
+ // ustring comparison is ideal because we expect Cord comparisons to be
+ // based on unsigned byte comparisons regardless of whether char is signed.
+ int expected = sign(cs.compare(ds));
+ EXPECT_EQ(expected, sign(c.Compare(d))) << c << ", " << d;
+}
+
+TEST(Compare, ComparisonIsUnsigned) {
+ RandomEngine rng(testing::GTEST_FLAG(random_seed));
+ std::uniform_int_distribution<uint32_t> uniform_uint8(0, 255);
+ char x = static_cast<char>(uniform_uint8(rng));
+ TestCompare(
+ absl::Cord(std::string(GetUniformRandomUpTo(&rng, 100), x)),
+ absl::Cord(std::string(GetUniformRandomUpTo(&rng, 100), x ^ 0x80)), &rng);
+}
+
+TEST(Compare, RandomComparisons) {
+ const int kIters = 5000;
+ RandomEngine rng(testing::GTEST_FLAG(random_seed));
+
+ int n = GetUniformRandomUpTo(&rng, 5000);
+ absl::Cord a[] = {MakeExternalCord(n),
+ absl::Cord("ant"),
+ absl::Cord("elephant"),
+ absl::Cord("giraffe"),
+ absl::Cord(std::string(GetUniformRandomUpTo(&rng, 100),
+ GetUniformRandomUpTo(&rng, 100))),
+ absl::Cord(""),
+ absl::Cord("x"),
+ absl::Cord("A"),
+ absl::Cord("B"),
+ absl::Cord("C")};
+ for (int i = 0; i < kIters; i++) {
+ absl::Cord c, d;
+ for (int j = 0; j < (i % 7) + 1; j++) {
+ c.Append(a[GetUniformRandomUpTo(&rng, ABSL_ARRAYSIZE(a))]);
+ d.Append(a[GetUniformRandomUpTo(&rng, ABSL_ARRAYSIZE(a))]);
+ }
+ std::bernoulli_distribution coin_flip(0.5);
+ TestCompare(coin_flip(rng) ? c : absl::Cord(std::string(c)),
+ coin_flip(rng) ? d : absl::Cord(std::string(d)), &rng);
+ }
+}
+
+template <typename T1, typename T2>
+void CompareOperators() {
+ const T1 a("a");
+ const T2 b("b");
+
+ EXPECT_TRUE(a == a);
+ // For pointer type (i.e. `const char*`), operator== compares the address
+ // instead of the string, so `a == const char*("a")` isn't necessarily true.
+ EXPECT_TRUE(std::is_pointer<T1>::value || a == T1("a"));
+ EXPECT_TRUE(std::is_pointer<T2>::value || a == T2("a"));
+ EXPECT_FALSE(a == b);
+
+ EXPECT_TRUE(a != b);
+ EXPECT_FALSE(a != a);
+
+ EXPECT_TRUE(a < b);
+ EXPECT_FALSE(b < a);
+
+ EXPECT_TRUE(b > a);
+ EXPECT_FALSE(a > b);
+
+ EXPECT_TRUE(a >= a);
+ EXPECT_TRUE(b >= a);
+ EXPECT_FALSE(a >= b);
+
+ EXPECT_TRUE(a <= a);
+ EXPECT_TRUE(a <= b);
+ EXPECT_FALSE(b <= a);
+}
+
+TEST(ComparisonOperators, Cord_Cord) {
+ CompareOperators<absl::Cord, absl::Cord>();
+}
+
+TEST(ComparisonOperators, Cord_StringPiece) {
+ CompareOperators<absl::Cord, absl::string_view>();
+}
+
+TEST(ComparisonOperators, StringPiece_Cord) {
+ CompareOperators<absl::string_view, absl::Cord>();
+}
+
+TEST(ComparisonOperators, Cord_string) {
+ CompareOperators<absl::Cord, std::string>();
+}
+
+TEST(ComparisonOperators, string_Cord) {
+ CompareOperators<std::string, absl::Cord>();
+}
+
+TEST(ComparisonOperators, stdstring_Cord) {
+ CompareOperators<std::string, absl::Cord>();
+}
+
+TEST(ComparisonOperators, Cord_stdstring) {
+ CompareOperators<absl::Cord, std::string>();
+}
+
+TEST(ComparisonOperators, charstar_Cord) {
+ CompareOperators<const char*, absl::Cord>();
+}
+
+TEST(ComparisonOperators, Cord_charstar) {
+ CompareOperators<absl::Cord, const char*>();
+}
+
+TEST(ConstructFromExternal, ReleaserInvoked) {
+ // Empty external memory means the releaser should be called immediately.
+ {
+ bool invoked = false;
+ auto releaser = [&invoked](absl::string_view) { invoked = true; };
+ {
+ auto c = absl::MakeCordFromExternal("", releaser);
+ EXPECT_TRUE(invoked);
+ }
+ }
+
+ // If the size of the data is small enough, a future constructor
+ // implementation may copy the bytes and immediately invoke the releaser
+ // instead of creating an external node. We make a large dummy std::string to
+ // make this test independent of such an optimization.
+ std::string large_dummy(2048, 'c');
+ {
+ bool invoked = false;
+ auto releaser = [&invoked](absl::string_view) { invoked = true; };
+ {
+ auto c = absl::MakeCordFromExternal(large_dummy, releaser);
+ EXPECT_FALSE(invoked);
+ }
+ EXPECT_TRUE(invoked);
+ }
+
+ {
+ bool invoked = false;
+ auto releaser = [&invoked](absl::string_view) { invoked = true; };
+ {
+ absl::Cord copy;
+ {
+ auto c = absl::MakeCordFromExternal(large_dummy, releaser);
+ copy = c;
+ EXPECT_FALSE(invoked);
+ }
+ EXPECT_FALSE(invoked);
+ }
+ EXPECT_TRUE(invoked);
+ }
+}
+
+TEST(ConstructFromExternal, CompareContents) {
+ RandomEngine rng(testing::GTEST_FLAG(random_seed));
+
+ for (int length = 1; length <= 2048; length *= 2) {
+ std::string data = RandomLowercaseString(&rng, length);
+ auto* external = new std::string(data);
+ auto cord =
+ absl::MakeCordFromExternal(*external, [external](absl::string_view sv) {
+ EXPECT_EQ(external->data(), sv.data());
+ EXPECT_EQ(external->size(), sv.size());
+ delete external;
+ });
+ EXPECT_EQ(data, cord);
+ }
+}
+
+TEST(ConstructFromExternal, LargeReleaser) {
+ RandomEngine rng(testing::GTEST_FLAG(random_seed));
+ constexpr size_t kLength = 256;
+ std::string data = RandomLowercaseString(&rng, kLength);
+ std::array<char, kLength> data_array;
+ for (size_t i = 0; i < kLength; ++i) data_array[i] = data[i];
+ bool invoked = false;
+ auto releaser = [data_array, &invoked](absl::string_view data) {
+ EXPECT_EQ(data, absl::string_view(data_array.data(), data_array.size()));
+ invoked = true;
+ };
+ (void)absl::MakeCordFromExternal(data, releaser);
+ EXPECT_TRUE(invoked);
+}
+
+TEST(ConstructFromExternal, FunctionPointerReleaser) {
+ static absl::string_view data("hello world");
+ static bool invoked;
+ auto* releaser =
+ static_cast<void (*)(absl::string_view)>([](absl::string_view sv) {
+ EXPECT_EQ(data, sv);
+ invoked = true;
+ });
+ invoked = false;
+ (void)absl::MakeCordFromExternal(data, releaser);
+ EXPECT_TRUE(invoked);
+
+ invoked = false;
+ (void)absl::MakeCordFromExternal(data, *releaser);
+ EXPECT_TRUE(invoked);
+}
+
+TEST(ConstructFromExternal, MoveOnlyReleaser) {
+ struct Releaser {
+ explicit Releaser(bool* invoked) : invoked(invoked) {}
+ Releaser(Releaser&& other) noexcept : invoked(other.invoked) {}
+ void operator()(absl::string_view) const { *invoked = true; }
+
+ bool* invoked;
+ };
+
+ bool invoked = false;
+ (void)absl::MakeCordFromExternal("dummy", Releaser(&invoked));
+ EXPECT_TRUE(invoked);
+}
+
+TEST(ConstructFromExternal, NoArgLambda) {
+ bool invoked = false;
+ (void)absl::MakeCordFromExternal("dummy", [&invoked]() { invoked = true; });
+ EXPECT_TRUE(invoked);
+}
+
+TEST(ConstructFromExternal, StringViewArgLambda) {
+ bool invoked = false;
+ (void)absl::MakeCordFromExternal(
+ "dummy", [&invoked](absl::string_view) { invoked = true; });
+ EXPECT_TRUE(invoked);
+}
+
+TEST(ConstructFromExternal, NonTrivialReleaserDestructor) {
+ struct Releaser {
+ explicit Releaser(bool* destroyed) : destroyed(destroyed) {}
+ ~Releaser() { *destroyed = true; }
+ void operator()(absl::string_view) const {}
+
+ bool* destroyed;
+ };
+
+ bool destroyed = false;
+ Releaser releaser(&destroyed);
+ (void)absl::MakeCordFromExternal("dummy", releaser);
+ EXPECT_TRUE(destroyed);
+}
+
+TEST(ConstructFromExternal, ReferenceQualifierOverloads) {
+ struct Releaser {
+ void operator()(absl::string_view) & { *lvalue_invoked = true; }
+ void operator()(absl::string_view) && { *rvalue_invoked = true; }
+
+ bool* lvalue_invoked;
+ bool* rvalue_invoked;
+ };
+
+ bool lvalue_invoked = false;
+ bool rvalue_invoked = false;
+ Releaser releaser = {&lvalue_invoked, &rvalue_invoked};
+ (void)absl::MakeCordFromExternal("", releaser);
+ EXPECT_FALSE(lvalue_invoked);
+ EXPECT_TRUE(rvalue_invoked);
+ rvalue_invoked = false;
+
+ (void)absl::MakeCordFromExternal("dummy", releaser);
+ EXPECT_FALSE(lvalue_invoked);
+ EXPECT_TRUE(rvalue_invoked);
+ rvalue_invoked = false;
+
+ // NOLINTNEXTLINE: suppress clang-tidy std::move on trivially copyable type.
+ (void)absl::MakeCordFromExternal("dummy", std::move(releaser));
+ EXPECT_FALSE(lvalue_invoked);
+ EXPECT_TRUE(rvalue_invoked);
+}
+
+TEST(ExternalMemory, BasicUsage) {
+ static const char* strings[] = {"", "hello", "there"};
+ for (const char* str : strings) {
+ absl::Cord dst("(prefix)");
+ AddExternalMemory(str, &dst);
+ dst.Append("(suffix)");
+ EXPECT_EQ((std::string("(prefix)") + str + std::string("(suffix)")),
+ std::string(dst));
+ }
+}
+
+TEST(ExternalMemory, RemovePrefixSuffix) {
+ // Exhaustively try all sub-strings.
+ absl::Cord cord = MakeComposite();
+ std::string s = std::string(cord);
+ for (int offset = 0; offset <= s.size(); offset++) {
+ for (int length = 0; length <= s.size() - offset; length++) {
+ absl::Cord result(cord);
+ result.RemovePrefix(offset);
+ result.RemoveSuffix(result.size() - length);
+ EXPECT_EQ(s.substr(offset, length), std::string(result))
+ << offset << " " << length;
+ }
+ }
+}
+
+TEST(ExternalMemory, Get) {
+ absl::Cord cord("hello");
+ AddExternalMemory(" world!", &cord);
+ AddExternalMemory(" how are ", &cord);
+ cord.Append(" you?");
+ std::string s = std::string(cord);
+ for (int i = 0; i < s.size(); i++) {
+ EXPECT_EQ(s[i], cord[i]);
+ }
+}
+
+// CordMemoryUsage tests verify the correctness of the EstimatedMemoryUsage()
+// These tests take into account that the reported memory usage is approximate
+// and non-deterministic. For all tests, We verify that the reported memory
+// usage is larger than `size()`, and less than `size() * 1.5` as a cord should
+// never reserve more 'extra' capacity than half of its size as it grows.
+// Additionally we have some whiteboxed expectations based on our knowledge of
+// the layout and size of empty and inlined cords, and flat nodes.
+
+TEST(CordMemoryUsage, Empty) {
+ EXPECT_EQ(sizeof(absl::Cord), absl::Cord().EstimatedMemoryUsage());
+}
+
+TEST(CordMemoryUsage, Embedded) {
+ absl::Cord a("hello");
+ EXPECT_EQ(a.EstimatedMemoryUsage(), sizeof(absl::Cord));
+}
+
+TEST(CordMemoryUsage, EmbeddedAppend) {
+ absl::Cord a("a");
+ absl::Cord b("bcd");
+ EXPECT_EQ(b.EstimatedMemoryUsage(), sizeof(absl::Cord));
+ a.Append(b);
+ EXPECT_EQ(a.EstimatedMemoryUsage(), sizeof(absl::Cord));
+}
+
+TEST(CordMemoryUsage, ExternalMemory) {
+ static const int kLength = 1000;
+ absl::Cord cord;
+ AddExternalMemory(std::string(kLength, 'x'), &cord);
+ EXPECT_GT(cord.EstimatedMemoryUsage(), kLength);
+ EXPECT_LE(cord.EstimatedMemoryUsage(), kLength * 1.5);
+}
+
+TEST(CordMemoryUsage, Flat) {
+ static const int kLength = 125;
+ absl::Cord a(std::string(kLength, 'a'));
+ EXPECT_GT(a.EstimatedMemoryUsage(), kLength);
+ EXPECT_LE(a.EstimatedMemoryUsage(), kLength * 1.5);
+}
+
+TEST(CordMemoryUsage, AppendFlat) {
+ using absl::strings_internal::CordTestAccess;
+ absl::Cord a(std::string(CordTestAccess::MaxFlatLength(), 'a'));
+ size_t length = a.EstimatedMemoryUsage();
+ a.Append(std::string(CordTestAccess::MaxFlatLength(), 'b'));
+ size_t delta = a.EstimatedMemoryUsage() - length;
+ EXPECT_GT(delta, CordTestAccess::MaxFlatLength());
+ EXPECT_LE(delta, CordTestAccess::MaxFlatLength() * 1.5);
+}
+
+// Regtest for a change that had to be rolled back because it expanded out
+// of the InlineRep too soon, which was observable through MemoryUsage().
+TEST(CordMemoryUsage, InlineRep) {
+ constexpr size_t kMaxInline = 15; // Cord::InlineRep::N
+ const std::string small_string(kMaxInline, 'x');
+ absl::Cord c1(small_string);
+
+ absl::Cord c2;
+ c2.Append(small_string);
+ EXPECT_EQ(c1, c2);
+ EXPECT_EQ(c1.EstimatedMemoryUsage(), c2.EstimatedMemoryUsage());
+}
+
+} // namespace
+
+// Regtest for 7510292 (fix a bug introduced by 7465150)
+TEST(Cord, Concat_Append) {
+ // Create a rep of type CONCAT
+ absl::Cord s1("foobarbarbarbarbar");
+ s1.Append("abcdefgabcdefgabcdefgabcdefgabcdefgabcdefgabcdefg");
+ size_t size = s1.size();
+
+ // Create a copy of s1 and append to it.
+ absl::Cord s2 = s1;
+ s2.Append("x");
+
+ // 7465150 modifies s1 when it shouldn't.
+ EXPECT_EQ(s1.size(), size);
+ EXPECT_EQ(s2.size(), size + 1);
+}
+
+TEST(MakeFragmentedCord, MakeFragmentedCordFromInitializerList) {
+ absl::Cord fragmented =
+ absl::MakeFragmentedCord({"A ", "fragmented ", "Cord"});
+
+ EXPECT_EQ("A fragmented Cord", fragmented);
+
+ auto chunk_it = fragmented.chunk_begin();
+
+ ASSERT_TRUE(chunk_it != fragmented.chunk_end());
+ EXPECT_EQ("A ", *chunk_it);
+
+ ASSERT_TRUE(++chunk_it != fragmented.chunk_end());
+ EXPECT_EQ("fragmented ", *chunk_it);
+
+ ASSERT_TRUE(++chunk_it != fragmented.chunk_end());
+ EXPECT_EQ("Cord", *chunk_it);
+
+ ASSERT_TRUE(++chunk_it == fragmented.chunk_end());
+}
+
+TEST(MakeFragmentedCord, MakeFragmentedCordFromVector) {
+ std::vector<absl::string_view> chunks = {"A ", "fragmented ", "Cord"};
+ absl::Cord fragmented = absl::MakeFragmentedCord(chunks);
+
+ EXPECT_EQ("A fragmented Cord", fragmented);
+
+ auto chunk_it = fragmented.chunk_begin();
+
+ ASSERT_TRUE(chunk_it != fragmented.chunk_end());
+ EXPECT_EQ("A ", *chunk_it);
+
+ ASSERT_TRUE(++chunk_it != fragmented.chunk_end());
+ EXPECT_EQ("fragmented ", *chunk_it);
+
+ ASSERT_TRUE(++chunk_it != fragmented.chunk_end());
+ EXPECT_EQ("Cord", *chunk_it);
+
+ ASSERT_TRUE(++chunk_it == fragmented.chunk_end());
+}
+
+TEST(CordChunkIterator, Traits) {
+ static_assert(std::is_copy_constructible<absl::Cord::ChunkIterator>::value,
+ "");
+ static_assert(std::is_copy_assignable<absl::Cord::ChunkIterator>::value, "");
+
+ // Move semantics to satisfy swappable via std::swap
+ static_assert(std::is_move_constructible<absl::Cord::ChunkIterator>::value,
+ "");
+ static_assert(std::is_move_assignable<absl::Cord::ChunkIterator>::value, "");
+
+ static_assert(
+ std::is_same<
+ std::iterator_traits<absl::Cord::ChunkIterator>::iterator_category,
+ std::input_iterator_tag>::value,
+ "");
+ static_assert(
+ std::is_same<std::iterator_traits<absl::Cord::ChunkIterator>::value_type,
+ absl::string_view>::value,
+ "");
+ static_assert(
+ std::is_same<
+ std::iterator_traits<absl::Cord::ChunkIterator>::difference_type,
+ ptrdiff_t>::value,
+ "");
+ static_assert(
+ std::is_same<std::iterator_traits<absl::Cord::ChunkIterator>::pointer,
+ const absl::string_view*>::value,
+ "");
+ static_assert(
+ std::is_same<std::iterator_traits<absl::Cord::ChunkIterator>::reference,
+ absl::string_view>::value,
+ "");
+}
+
+static void VerifyChunkIterator(const absl::Cord& cord,
+ size_t expected_chunks) {
+ EXPECT_EQ(cord.chunk_begin() == cord.chunk_end(), cord.empty()) << cord;
+ EXPECT_EQ(cord.chunk_begin() != cord.chunk_end(), !cord.empty());
+
+ absl::Cord::ChunkRange range = cord.Chunks();
+ EXPECT_EQ(range.begin() == range.end(), cord.empty());
+ EXPECT_EQ(range.begin() != range.end(), !cord.empty());
+
+ std::string content(cord);
+ size_t pos = 0;
+ auto pre_iter = cord.chunk_begin(), post_iter = cord.chunk_begin();
+ size_t n_chunks = 0;
+ while (pre_iter != cord.chunk_end() && post_iter != cord.chunk_end()) {
+ EXPECT_FALSE(pre_iter == cord.chunk_end()); // NOLINT: explicitly test ==
+ EXPECT_FALSE(post_iter == cord.chunk_end()); // NOLINT
+
+ EXPECT_EQ(pre_iter, post_iter);
+ EXPECT_EQ(*pre_iter, *post_iter);
+
+ EXPECT_EQ(pre_iter->data(), (*pre_iter).data());
+ EXPECT_EQ(pre_iter->size(), (*pre_iter).size());
+
+ absl::string_view chunk = *pre_iter;
+ EXPECT_FALSE(chunk.empty());
+ EXPECT_LE(pos + chunk.size(), content.size());
+ EXPECT_EQ(absl::string_view(content.c_str() + pos, chunk.size()), chunk);
+
+ int n_equal_iterators = 0;
+ for (absl::Cord::ChunkIterator it = range.begin(); it != range.end();
+ ++it) {
+ n_equal_iterators += static_cast<int>(it == pre_iter);
+ }
+ EXPECT_EQ(n_equal_iterators, 1);
+
+ ++pre_iter;
+ EXPECT_EQ(*post_iter++, chunk);
+
+ pos += chunk.size();
+ ++n_chunks;
+ }
+ EXPECT_EQ(expected_chunks, n_chunks);
+ EXPECT_EQ(pos, content.size());
+ EXPECT_TRUE(pre_iter == cord.chunk_end()); // NOLINT: explicitly test ==
+ EXPECT_TRUE(post_iter == cord.chunk_end()); // NOLINT
+}
+
+TEST(CordChunkIterator, Operations) {
+ absl::Cord empty_cord;
+ VerifyChunkIterator(empty_cord, 0);
+
+ absl::Cord small_buffer_cord("small cord");
+ VerifyChunkIterator(small_buffer_cord, 1);
+
+ absl::Cord flat_node_cord("larger than small buffer optimization");
+ VerifyChunkIterator(flat_node_cord, 1);
+
+ VerifyChunkIterator(
+ absl::MakeFragmentedCord({"a ", "small ", "fragmented ", "cord ", "for ",
+ "testing ", "chunk ", "iterations."}),
+ 8);
+
+ absl::Cord reused_nodes_cord(std::string(40, 'c'));
+ reused_nodes_cord.Prepend(absl::Cord(std::string(40, 'b')));
+ reused_nodes_cord.Prepend(absl::Cord(std::string(40, 'a')));
+ size_t expected_chunks = 3;
+ for (int i = 0; i < 8; ++i) {
+ reused_nodes_cord.Prepend(reused_nodes_cord);
+ expected_chunks *= 2;
+ VerifyChunkIterator(reused_nodes_cord, expected_chunks);
+ }
+
+ RandomEngine rng(testing::GTEST_FLAG(random_seed));
+ absl::Cord flat_cord(RandomLowercaseString(&rng, 256));
+ absl::Cord subcords;
+ for (int i = 0; i < 128; ++i) subcords.Prepend(flat_cord.Subcord(i, 128));
+ VerifyChunkIterator(subcords, 128);
+}
+
+TEST(CordCharIterator, Traits) {
+ static_assert(std::is_copy_constructible<absl::Cord::CharIterator>::value,
+ "");
+ static_assert(std::is_copy_assignable<absl::Cord::CharIterator>::value, "");
+
+ // Move semantics to satisfy swappable via std::swap
+ static_assert(std::is_move_constructible<absl::Cord::CharIterator>::value,
+ "");
+ static_assert(std::is_move_assignable<absl::Cord::CharIterator>::value, "");
+
+ static_assert(
+ std::is_same<
+ std::iterator_traits<absl::Cord::CharIterator>::iterator_category,
+ std::input_iterator_tag>::value,
+ "");
+ static_assert(
+ std::is_same<std::iterator_traits<absl::Cord::CharIterator>::value_type,
+ char>::value,
+ "");
+ static_assert(
+ std::is_same<
+ std::iterator_traits<absl::Cord::CharIterator>::difference_type,
+ ptrdiff_t>::value,
+ "");
+ static_assert(
+ std::is_same<std::iterator_traits<absl::Cord::CharIterator>::pointer,
+ const char*>::value,
+ "");
+ static_assert(
+ std::is_same<std::iterator_traits<absl::Cord::CharIterator>::reference,
+ const char&>::value,
+ "");
+}
+
+static void VerifyCharIterator(const absl::Cord& cord) {
+ EXPECT_EQ(cord.char_begin() == cord.char_end(), cord.empty());
+ EXPECT_EQ(cord.char_begin() != cord.char_end(), !cord.empty());
+
+ absl::Cord::CharRange range = cord.Chars();
+ EXPECT_EQ(range.begin() == range.end(), cord.empty());
+ EXPECT_EQ(range.begin() != range.end(), !cord.empty());
+
+ size_t i = 0;
+ absl::Cord::CharIterator pre_iter = cord.char_begin();
+ absl::Cord::CharIterator post_iter = cord.char_begin();
+ std::string content(cord);
+ while (pre_iter != cord.char_end() && post_iter != cord.char_end()) {
+ EXPECT_FALSE(pre_iter == cord.char_end()); // NOLINT: explicitly test ==
+ EXPECT_FALSE(post_iter == cord.char_end()); // NOLINT
+
+ EXPECT_LT(i, cord.size());
+ EXPECT_EQ(content[i], *pre_iter);
+
+ EXPECT_EQ(pre_iter, post_iter);
+ EXPECT_EQ(*pre_iter, *post_iter);
+ EXPECT_EQ(&*pre_iter, &*post_iter);
+
+ EXPECT_EQ(&*pre_iter, pre_iter.operator->());
+
+ const char* character_address = &*pre_iter;
+ absl::Cord::CharIterator copy = pre_iter;
+ ++copy;
+ EXPECT_EQ(character_address, &*pre_iter);
+
+ int n_equal_iterators = 0;
+ for (absl::Cord::CharIterator it = range.begin(); it != range.end(); ++it) {
+ n_equal_iterators += static_cast<int>(it == pre_iter);
+ }
+ EXPECT_EQ(n_equal_iterators, 1);
+
+ absl::Cord::CharIterator advance_iter = range.begin();
+ absl::Cord::Advance(&advance_iter, i);
+ EXPECT_EQ(pre_iter, advance_iter);
+
+ advance_iter = range.begin();
+ EXPECT_EQ(absl::Cord::AdvanceAndRead(&advance_iter, i), cord.Subcord(0, i));
+ EXPECT_EQ(pre_iter, advance_iter);
+
+ advance_iter = pre_iter;
+ absl::Cord::Advance(&advance_iter, cord.size() - i);
+ EXPECT_EQ(range.end(), advance_iter);
+
+ advance_iter = pre_iter;
+ EXPECT_EQ(absl::Cord::AdvanceAndRead(&advance_iter, cord.size() - i),
+ cord.Subcord(i, cord.size() - i));
+ EXPECT_EQ(range.end(), advance_iter);
+
+ ++i;
+ ++pre_iter;
+ post_iter++;
+ }
+ EXPECT_EQ(i, cord.size());
+ EXPECT_TRUE(pre_iter == cord.char_end()); // NOLINT: explicitly test ==
+ EXPECT_TRUE(post_iter == cord.char_end()); // NOLINT
+
+ absl::Cord::CharIterator zero_advanced_end = cord.char_end();
+ absl::Cord::Advance(&zero_advanced_end, 0);
+ EXPECT_EQ(zero_advanced_end, cord.char_end());
+
+ absl::Cord::CharIterator it = cord.char_begin();
+ for (absl::string_view chunk : cord.Chunks()) {
+ while (!chunk.empty()) {
+ EXPECT_EQ(absl::Cord::ChunkRemaining(it), chunk);
+ chunk.remove_prefix(1);
+ ++it;
+ }
+ }
+}
+
+TEST(CordCharIterator, Operations) {
+ absl::Cord empty_cord;
+ VerifyCharIterator(empty_cord);
+
+ absl::Cord small_buffer_cord("small cord");
+ VerifyCharIterator(small_buffer_cord);
+
+ absl::Cord flat_node_cord("larger than small buffer optimization");
+ VerifyCharIterator(flat_node_cord);
+
+ VerifyCharIterator(
+ absl::MakeFragmentedCord({"a ", "small ", "fragmented ", "cord ", "for ",
+ "testing ", "character ", "iteration."}));
+
+ absl::Cord reused_nodes_cord("ghi");
+ reused_nodes_cord.Prepend(absl::Cord("def"));
+ reused_nodes_cord.Prepend(absl::Cord("abc"));
+ for (int i = 0; i < 4; ++i) {
+ reused_nodes_cord.Prepend(reused_nodes_cord);
+ VerifyCharIterator(reused_nodes_cord);
+ }
+
+ RandomEngine rng(testing::GTEST_FLAG(random_seed));
+ absl::Cord flat_cord(RandomLowercaseString(&rng, 256));
+ absl::Cord subcords;
+ for (int i = 0; i < 4; ++i) subcords.Prepend(flat_cord.Subcord(16 * i, 128));
+ VerifyCharIterator(subcords);
+}
+
+TEST(Cord, StreamingOutput) {
+ absl::Cord c =
+ absl::MakeFragmentedCord({"A ", "small ", "fragmented ", "Cord", "."});
+ std::stringstream output;
+ output << c;
+ EXPECT_EQ("A small fragmented Cord.", output.str());
+}
+
+TEST(Cord, ForEachChunk) {
+ for (int num_elements : {1, 10, 200}) {
+ SCOPED_TRACE(num_elements);
+ std::vector<std::string> cord_chunks;
+ for (int i = 0; i < num_elements; ++i) {
+ cord_chunks.push_back(absl::StrCat("[", i, "]"));
+ }
+ absl::Cord c = absl::MakeFragmentedCord(cord_chunks);
+
+ std::vector<std::string> iterated_chunks;
+ absl::CordTestPeer::ForEachChunk(c,
+ [&iterated_chunks](absl::string_view sv) {
+ iterated_chunks.emplace_back(sv);
+ });
+ EXPECT_EQ(iterated_chunks, cord_chunks);
+ }
+}
+
+TEST(Cord, SmallBufferAssignFromOwnData) {
+ constexpr size_t kMaxInline = 15;
+ std::string contents = "small buff cord";
+ EXPECT_EQ(contents.size(), kMaxInline);
+ for (size_t pos = 0; pos < contents.size(); ++pos) {
+ for (size_t count = contents.size() - pos; count > 0; --count) {
+ absl::Cord c(contents);
+ absl::string_view flat = c.Flatten();
+ c = flat.substr(pos, count);
+ EXPECT_EQ(c, contents.substr(pos, count))
+ << "pos = " << pos << "; count = " << count;
+ }
+ }
+}
+
+TEST(Cord, Format) {
+ absl::Cord c;
+ absl::Format(&c, "There were %04d little %s.", 3, "pigs");
+ EXPECT_EQ(c, "There were 0003 little pigs.");
+ absl::Format(&c, "And %-3llx bad wolf!", 1);
+ EXPECT_EQ(c, "There were 0003 little pigs.And 1 bad wolf!");
+}
+
+TEST(CordDeathTest, Hardening) {
+ absl::Cord cord("hello");
+ // These statement should abort the program in all builds modes.
+ EXPECT_DEATH_IF_SUPPORTED(cord.RemovePrefix(6), "");
+ EXPECT_DEATH_IF_SUPPORTED(cord.RemoveSuffix(6), "");
+
+ bool test_hardening = false;
+ ABSL_HARDENING_ASSERT([&]() {
+ // This only runs when ABSL_HARDENING_ASSERT is active.
+ test_hardening = true;
+ return true;
+ }());
+ if (!test_hardening) return;
+
+ EXPECT_DEATH_IF_SUPPORTED(cord[5], "");
+ EXPECT_DEATH_IF_SUPPORTED(*cord.chunk_end(), "");
+ EXPECT_DEATH_IF_SUPPORTED(static_cast<void>(cord.chunk_end()->empty()), "");
+ EXPECT_DEATH_IF_SUPPORTED(++cord.chunk_end(), "");
+}
+
+class AfterExitCordTester {
+ public:
+ bool Set(absl::Cord* cord, absl::string_view expected) {
+ cord_ = cord;
+ expected_ = expected;
+ return true;
+ }
+
+ ~AfterExitCordTester() {
+ EXPECT_EQ(*cord_, expected_);
+ }
+ private:
+ absl::Cord* cord_;
+ absl::string_view expected_;
+};
+
+template <typename Str>
+void TestConstinitConstructor(Str) {
+ const auto expected = Str::value;
+ // Defined before `cord` to be destroyed after it.
+ static AfterExitCordTester exit_tester; // NOLINT
+ ABSL_CONST_INIT static absl::Cord cord(Str{}); // NOLINT
+ static bool init_exit_tester = exit_tester.Set(&cord, expected);
+ (void)init_exit_tester;
+
+ EXPECT_EQ(cord, expected);
+ // Copy the object and test the copy, and the original.
+ {
+ absl::Cord copy = cord;
+ EXPECT_EQ(copy, expected);
+ }
+ // The original still works
+ EXPECT_EQ(cord, expected);
+
+ // Try making adding more structure to the tree.
+ {
+ absl::Cord copy = cord;
+ std::string expected_copy(expected);
+ for (int i = 0; i < 10; ++i) {
+ copy.Append(cord);
+ absl::StrAppend(&expected_copy, expected);
+ EXPECT_EQ(copy, expected_copy);
+ }
+ }
+
+ // Make sure we are using the right branch during constant evaluation.
+ EXPECT_EQ(absl::CordTestPeer::IsTree(cord), cord.size() >= 16);
+
+ for (int i = 0; i < 10; ++i) {
+ // Make a few more Cords from the same global rep.
+ // This tests what happens when the refcount for it gets below 1.
+ EXPECT_EQ(expected, absl::Cord(Str{}));
+ }
+}
+
+constexpr int SimpleStrlen(const char* p) {
+ return *p ? 1 + SimpleStrlen(p + 1) : 0;
+}
+
+struct ShortView {
+ constexpr absl::string_view operator()() const {
+ return absl::string_view("SSO string", SimpleStrlen("SSO string"));
+ }
+};
+
+struct LongView {
+ constexpr absl::string_view operator()() const {
+ return absl::string_view("String that does not fit SSO.",
+ SimpleStrlen("String that does not fit SSO."));
+ }
+};
+
+
+TEST(Cord, ConstinitConstructor) {
+ TestConstinitConstructor(
+ absl::strings_internal::MakeStringConstant(ShortView{}));
+ TestConstinitConstructor(
+ absl::strings_internal::MakeStringConstant(LongView{}));
+}
diff --git a/absl/strings/cord_test_helpers.h b/absl/strings/cord_test_helpers.h
new file mode 100644
index 0000000..f1036e3
--- /dev/null
+++ b/absl/strings/cord_test_helpers.h
@@ -0,0 +1,60 @@
+//
+// Copyright 2018 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+
+#ifndef ABSL_STRINGS_CORD_TEST_HELPERS_H_
+#define ABSL_STRINGS_CORD_TEST_HELPERS_H_
+
+#include "absl/strings/cord.h"
+
+namespace absl {
+ABSL_NAMESPACE_BEGIN
+
+// Creates a multi-segment Cord from an iterable container of strings. The
+// resulting Cord is guaranteed to have one segment for every string in the
+// container. This allows code to be unit tested with multi-segment Cord
+// inputs.
+//
+// Example:
+//
+// absl::Cord c = absl::MakeFragmentedCord({"A ", "fragmented ", "Cord"});
+// EXPECT_FALSE(c.GetFlat(&unused));
+//
+// The mechanism by which this Cord is created is an implementation detail. Any
+// implementation that produces a multi-segment Cord may produce a flat Cord in
+// the future as new optimizations are added to the Cord class.
+// MakeFragmentedCord will, however, always be updated to return a multi-segment
+// Cord.
+template <typename Container>
+Cord MakeFragmentedCord(const Container& c) {
+ Cord result;
+ for (const auto& s : c) {
+ auto* external = new std::string(s);
+ Cord tmp = absl::MakeCordFromExternal(
+ *external, [external](absl::string_view) { delete external; });
+ tmp.Prepend(result);
+ result = tmp;
+ }
+ return result;
+}
+
+inline Cord MakeFragmentedCord(std::initializer_list<absl::string_view> list) {
+ return MakeFragmentedCord<std::initializer_list<absl::string_view>>(list);
+}
+
+ABSL_NAMESPACE_END
+} // namespace absl
+
+#endif // ABSL_STRINGS_CORD_TEST_HELPERS_H_
diff --git a/absl/strings/escaping.cc b/absl/strings/escaping.cc
index 0d336e3..18b20b8 100644
--- a/absl/strings/escaping.cc
+++ b/absl/strings/escaping.cc
@@ -26,6 +26,7 @@
#include "absl/base/internal/raw_logging.h"
#include "absl/base/internal/unaligned_access.h"
#include "absl/strings/internal/char_map.h"
+#include "absl/strings/internal/escaping.h"
#include "absl/strings/internal/resize_uninitialized.h"
#include "absl/strings/internal/utf8.h"
#include "absl/strings/str_cat.h"
@@ -33,29 +34,9 @@
#include "absl/strings/string_view.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace {
-// Digit conversion.
-constexpr char kHexChar[] = "0123456789abcdef";
-
-constexpr char kHexTable[513] =
- "000102030405060708090a0b0c0d0e0f"
- "101112131415161718191a1b1c1d1e1f"
- "202122232425262728292a2b2c2d2e2f"
- "303132333435363738393a3b3c3d3e3f"
- "404142434445464748494a4b4c4d4e4f"
- "505152535455565758595a5b5c5d5e5f"
- "606162636465666768696a6b6c6d6e6f"
- "707172737475767778797a7b7c7d7e7f"
- "808182838485868788898a8b8c8d8e8f"
- "909192939495969798999a9b9c9d9e9f"
- "a0a1a2a3a4a5a6a7a8a9aaabacadaeaf"
- "b0b1b2b3b4b5b6b7b8b9babbbcbdbebf"
- "c0c1c2c3c4c5c6c7c8c9cacbcccdcecf"
- "d0d1d2d3d4d5d6d7d8d9dadbdcdddedf"
- "e0e1e2e3e4e5e6e7e8e9eaebecedeeef"
- "f0f1f2f3f4f5f6f7f8f9fafbfcfdfeff";
-
// These are used for the leave_nulls_escaped argument to CUnescapeInternal().
constexpr bool kUnescapeNulls = false;
@@ -156,7 +137,7 @@
// Copy the escape sequence for the null character
const ptrdiff_t octal_size = p + 1 - octal_start;
*d++ = '\\';
- memcpy(d, octal_start, octal_size);
+ memmove(d, octal_start, octal_size);
d += octal_size;
break;
}
@@ -189,7 +170,7 @@
// Copy the escape sequence for the null character
const ptrdiff_t hex_size = p + 1 - hex_start;
*d++ = '\\';
- memcpy(d, hex_start, hex_size);
+ memmove(d, hex_start, hex_size);
d += hex_size;
break;
}
@@ -222,7 +203,7 @@
if ((rune == 0) && leave_nulls_escaped) {
// Copy the escape sequence for the null character
*d++ = '\\';
- memcpy(d, hex_start, 5); // u0000
+ memmove(d, hex_start, 5); // u0000
d += 5;
break;
}
@@ -270,7 +251,7 @@
if ((rune == 0) && leave_nulls_escaped) {
// Copy the escape sequence for the null character
*d++ = '\\';
- memcpy(d, hex_start, 9); // U00000000
+ memmove(d, hex_start, 9); // U00000000
d += 9;
break;
}
@@ -348,14 +329,14 @@
(last_hex_escape && absl::ascii_isxdigit(c)))) {
if (use_hex) {
dest.append("\\" "x");
- dest.push_back(kHexChar[c / 16]);
- dest.push_back(kHexChar[c % 16]);
+ dest.push_back(numbers_internal::kHexChar[c / 16]);
+ dest.push_back(numbers_internal::kHexChar[c % 16]);
is_hex_escape = true;
} else {
dest.append("\\");
- dest.push_back(kHexChar[c / 64]);
- dest.push_back(kHexChar[(c % 64) / 8]);
- dest.push_back(kHexChar[c % 8]);
+ dest.push_back(numbers_internal::kHexChar[c / 64]);
+ dest.push_back(numbers_internal::kHexChar[(c % 64) / 8]);
+ dest.push_back(numbers_internal::kHexChar[c % 8]);
}
} else {
dest.push_back(c);
@@ -469,7 +450,7 @@
// The GET_INPUT macro gets the next input character, skipping
// over any whitespace, and stopping when we reach the end of the
- // std::string or when we read any non-data character. The arguments are
+ // string or when we read any non-data character. The arguments are
// an arbitrary identifier (used as a label for goto) and the number
// of data bytes that must remain in the input to avoid aborting the
// loop.
@@ -492,18 +473,18 @@
if (dest) {
// This loop consumes 4 input bytes and produces 3 output bytes
// per iteration. We can't know at the start that there is enough
- // data left in the std::string for a full iteration, so the loop may
+ // data left in the string for a full iteration, so the loop may
// break out in the middle; if so 'state' will be set to the
// number of input bytes read.
while (szsrc >= 4) {
// We'll start by optimistically assuming that the next four
- // bytes of the std::string (src[0..3]) are four good data bytes
+ // bytes of the string (src[0..3]) are four good data bytes
// (that is, no nulls, whitespace, padding chars, or illegal
// chars). We need to test src[0..2] for nulls individually
// before constructing temp to preserve the property that we
- // never read past a null in the std::string (no matter how long
- // szsrc claims the std::string is).
+ // never read past a null in the string (no matter how long
+ // szsrc claims the string is).
if (!src[0] || !src[1] || !src[2] ||
((temp = ((unsigned(unbase64[src[0]]) << 18) |
@@ -528,7 +509,7 @@
temp = (temp << 6) | decode;
} else {
// We really did have four good data bytes, so advance four
- // characters in the std::string.
+ // characters in the string.
szsrc -= 4;
src += 4;
@@ -663,7 +644,7 @@
state);
}
- // The remainder of the std::string should be all whitespace, mixed with
+ // The remainder of the string should be all whitespace, mixed with
// exactly 0 equals signs, or exactly 'expected_equals' equals
// signs. (Always accepting 0 equals signs is an Abseil extension
// not covered in the RFC, as is accepting dot as the pad character.)
@@ -784,180 +765,13 @@
};
/* clang-format on */
-size_t CalculateBase64EscapedLenInternal(size_t input_len, bool do_padding) {
- // Base64 encodes three bytes of input at a time. If the input is not
- // divisible by three, we pad as appropriate.
- //
- // (from https://tools.ietf.org/html/rfc3548)
- // Special processing is performed if fewer than 24 bits are available
- // at the end of the data being encoded. A full encoding quantum is
- // always completed at the end of a quantity. When fewer than 24 input
- // bits are available in an input group, zero bits are added (on the
- // right) to form an integral number of 6-bit groups. Padding at the
- // end of the data is performed using the '=' character. Since all base
- // 64 input is an integral number of octets, only the following cases
- // can arise:
-
- // Base64 encodes each three bytes of input into four bytes of output.
- size_t len = (input_len / 3) * 4;
-
- if (input_len % 3 == 0) {
- // (from https://tools.ietf.org/html/rfc3548)
- // (1) the final quantum of encoding input is an integral multiple of 24
- // bits; here, the final unit of encoded output will be an integral
- // multiple of 4 characters with no "=" padding,
- } else if (input_len % 3 == 1) {
- // (from https://tools.ietf.org/html/rfc3548)
- // (2) the final quantum of encoding input is exactly 8 bits; here, the
- // final unit of encoded output will be two characters followed by two
- // "=" padding characters, or
- len += 2;
- if (do_padding) {
- len += 2;
- }
- } else { // (input_len % 3 == 2)
- // (from https://tools.ietf.org/html/rfc3548)
- // (3) the final quantum of encoding input is exactly 16 bits; here, the
- // final unit of encoded output will be three characters followed by one
- // "=" padding character.
- len += 3;
- if (do_padding) {
- len += 1;
- }
- }
-
- assert(len >= input_len); // make sure we didn't overflow
- return len;
-}
-
-size_t Base64EscapeInternal(const unsigned char* src, size_t szsrc, char* dest,
- size_t szdest, const char* base64,
- bool do_padding) {
- static const char kPad64 = '=';
-
- if (szsrc * 4 > szdest * 3) return 0;
-
- char* cur_dest = dest;
- const unsigned char* cur_src = src;
-
- char* const limit_dest = dest + szdest;
- const unsigned char* const limit_src = src + szsrc;
-
- // Three bytes of data encodes to four characters of cyphertext.
- // So we can pump through three-byte chunks atomically.
- if (szsrc >= 3) { // "limit_src - 3" is UB if szsrc < 3.
- while (cur_src < limit_src - 3) { // While we have >= 32 bits.
- uint32_t in = absl::big_endian::Load32(cur_src) >> 8;
-
- cur_dest[0] = base64[in >> 18];
- in &= 0x3FFFF;
- cur_dest[1] = base64[in >> 12];
- in &= 0xFFF;
- cur_dest[2] = base64[in >> 6];
- in &= 0x3F;
- cur_dest[3] = base64[in];
-
- cur_dest += 4;
- cur_src += 3;
- }
- }
- // To save time, we didn't update szdest or szsrc in the loop. So do it now.
- szdest = limit_dest - cur_dest;
- szsrc = limit_src - cur_src;
-
- /* now deal with the tail (<=3 bytes) */
- switch (szsrc) {
- case 0:
- // Nothing left; nothing more to do.
- break;
- case 1: {
- // One byte left: this encodes to two characters, and (optionally)
- // two pad characters to round out the four-character cypherblock.
- if (szdest < 2) return 0;
- uint32_t in = cur_src[0];
- cur_dest[0] = base64[in >> 2];
- in &= 0x3;
- cur_dest[1] = base64[in << 4];
- cur_dest += 2;
- szdest -= 2;
- if (do_padding) {
- if (szdest < 2) return 0;
- cur_dest[0] = kPad64;
- cur_dest[1] = kPad64;
- cur_dest += 2;
- szdest -= 2;
- }
- break;
- }
- case 2: {
- // Two bytes left: this encodes to three characters, and (optionally)
- // one pad character to round out the four-character cypherblock.
- if (szdest < 3) return 0;
- uint32_t in = absl::big_endian::Load16(cur_src);
- cur_dest[0] = base64[in >> 10];
- in &= 0x3FF;
- cur_dest[1] = base64[in >> 4];
- in &= 0x00F;
- cur_dest[2] = base64[in << 2];
- cur_dest += 3;
- szdest -= 3;
- if (do_padding) {
- if (szdest < 1) return 0;
- cur_dest[0] = kPad64;
- cur_dest += 1;
- szdest -= 1;
- }
- break;
- }
- case 3: {
- // Three bytes left: same as in the big loop above. We can't do this in
- // the loop because the loop above always reads 4 bytes, and the fourth
- // byte is past the end of the input.
- if (szdest < 4) return 0;
- uint32_t in = (cur_src[0] << 16) + absl::big_endian::Load16(cur_src + 1);
- cur_dest[0] = base64[in >> 18];
- in &= 0x3FFFF;
- cur_dest[1] = base64[in >> 12];
- in &= 0xFFF;
- cur_dest[2] = base64[in >> 6];
- in &= 0x3F;
- cur_dest[3] = base64[in];
- cur_dest += 4;
- szdest -= 4;
- break;
- }
- default:
- // Should not be reached: blocks of 4 bytes are handled
- // in the while loop before this switch statement.
- ABSL_RAW_LOG(FATAL, "Logic problem? szsrc = %zu", szsrc);
- break;
- }
- return (cur_dest - dest);
-}
-
-constexpr char kBase64Chars[] =
- "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/";
-
constexpr char kWebSafeBase64Chars[] =
"ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789-_";
template <typename String>
-void Base64EscapeInternal(const unsigned char* src, size_t szsrc, String* dest,
- bool do_padding, const char* base64_chars) {
- const size_t calc_escaped_size =
- CalculateBase64EscapedLenInternal(szsrc, do_padding);
- strings_internal::STLStringResizeUninitialized(dest, calc_escaped_size);
-
- const size_t escaped_len = Base64EscapeInternal(
- src, szsrc, &(*dest)[0], dest->size(), base64_chars, do_padding);
- assert(calc_escaped_size == escaped_len);
- dest->erase(escaped_len);
-}
-
-template <typename String>
bool Base64UnescapeInternal(const char* src, size_t slen, String* dest,
const signed char* unbase64) {
- // Determine the size of the output std::string. Base64 encodes every 3 bytes into
+ // Determine the size of the output string. Base64 encodes every 3 bytes into
// 4 characters. any leftover chars are added directly for good measure.
// This is documented in the base64 RFC: http://tools.ietf.org/html/rfc3548
const size_t dest_len = 3 * (slen / 4) + (slen % 4);
@@ -965,7 +779,7 @@
strings_internal::STLStringResizeUninitialized(dest, dest_len);
// We are getting the destination buffer by getting the beginning of the
- // std::string and converting it into a char *.
+ // string and converting it into a char *.
size_t len;
const bool ok =
Base64UnescapeInternal(src, slen, &(*dest)[0], dest_len, unbase64, &len);
@@ -982,7 +796,7 @@
}
/* clang-format off */
-constexpr char kHexValue[256] = {
+constexpr char kHexValueLenient[256] = {
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
@@ -998,8 +812,9 @@
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
- 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
+ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
};
+
/* clang-format on */
// This is a templated function so that T can be either a char*
@@ -1008,8 +823,8 @@
template <typename T>
void HexStringToBytesInternal(const char* from, T to, ptrdiff_t num) {
for (int i = 0; i < num; i++) {
- to[i] = (kHexValue[from[i * 2] & 0xFF] << 4) +
- (kHexValue[from[i * 2 + 1] & 0xFF]);
+ to[i] = (kHexValueLenient[from[i * 2] & 0xFF] << 4) +
+ (kHexValueLenient[from[i * 2 + 1] & 0xFF]);
}
}
@@ -1019,7 +834,7 @@
void BytesToHexStringInternal(const unsigned char* src, T dest, ptrdiff_t num) {
auto dest_ptr = &dest[0];
for (auto src_ptr = src; src_ptr != (src + num); ++src_ptr, dest_ptr += 2) {
- const char* hex_p = &kHexTable[*src_ptr * 2];
+ const char* hex_p = &numbers_internal::kHexTable[*src_ptr * 2];
std::copy(hex_p, hex_p + 2, dest_ptr);
}
}
@@ -1087,26 +902,30 @@
}
void Base64Escape(absl::string_view src, std::string* dest) {
- Base64EscapeInternal(reinterpret_cast<const unsigned char*>(src.data()),
- src.size(), dest, true, kBase64Chars);
+ strings_internal::Base64EscapeInternal(
+ reinterpret_cast<const unsigned char*>(src.data()), src.size(), dest,
+ true, strings_internal::kBase64Chars);
}
void WebSafeBase64Escape(absl::string_view src, std::string* dest) {
- Base64EscapeInternal(reinterpret_cast<const unsigned char*>(src.data()),
- src.size(), dest, false, kWebSafeBase64Chars);
+ strings_internal::Base64EscapeInternal(
+ reinterpret_cast<const unsigned char*>(src.data()), src.size(), dest,
+ false, kWebSafeBase64Chars);
}
std::string Base64Escape(absl::string_view src) {
std::string dest;
- Base64EscapeInternal(reinterpret_cast<const unsigned char*>(src.data()),
- src.size(), &dest, true, kBase64Chars);
+ strings_internal::Base64EscapeInternal(
+ reinterpret_cast<const unsigned char*>(src.data()), src.size(), &dest,
+ true, strings_internal::kBase64Chars);
return dest;
}
std::string WebSafeBase64Escape(absl::string_view src) {
std::string dest;
- Base64EscapeInternal(reinterpret_cast<const unsigned char*>(src.data()),
- src.size(), &dest, false, kWebSafeBase64Chars);
+ strings_internal::Base64EscapeInternal(
+ reinterpret_cast<const unsigned char*>(src.data()), src.size(), &dest,
+ false, kWebSafeBase64Chars);
return dest;
}
@@ -1126,4 +945,5 @@
return result;
}
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/escaping.h b/absl/strings/escaping.h
index 198b934..f5ca26c 100644
--- a/absl/strings/escaping.h
+++ b/absl/strings/escaping.h
@@ -33,6 +33,7 @@
#include "absl/strings/string_view.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
// CUnescape()
//
@@ -157,6 +158,7 @@
// `2*from.size()`.
std::string BytesToHexString(absl::string_view from);
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_ESCAPING_H_
diff --git a/absl/strings/escaping_test.cc b/absl/strings/escaping_test.cc
index 1967975..45671a0 100644
--- a/absl/strings/escaping_test.cc
+++ b/absl/strings/escaping_test.cc
@@ -300,7 +300,7 @@
absl::string_view plaintext;
absl::string_view cyphertext;
} const base64_tests[] = {
- // Empty std::string.
+ // Empty string.
{{"", 0}, {"", 0}},
{{nullptr, 0},
{"", 0}}, // if length is zero, plaintext ptr must be ignored!
@@ -586,7 +586,7 @@
EXPECT_EQ(encoded, websafe);
EXPECT_EQ(absl::WebSafeBase64Escape(tc.plaintext), websafe);
- // Let's try the std::string version of the decoder
+ // Let's try the string version of the decoder
decoded = "this junk should be ignored";
EXPECT_TRUE(absl::WebSafeBase64Unescape(websafe, &decoded));
EXPECT_EQ(decoded, tc.plaintext);
@@ -625,7 +625,7 @@
std::string escaped;
absl::Base64Escape(huge, &escaped);
- // Generates the std::string that should match a base64 encoded "xxx..." std::string.
+ // Generates the string that should match a base64 encoded "xxx..." string.
// "xxx" in base64 is "eHh4".
std::string expected_encoding;
expected_encoding.reserve(kSize / 3 * 4);
diff --git a/absl/strings/internal/char_map.h b/absl/strings/internal/char_map.h
index b9108b8..61484de 100644
--- a/absl/strings/internal/char_map.h
+++ b/absl/strings/internal/char_map.h
@@ -28,6 +28,7 @@
#include "absl/base/port.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace strings_internal {
class Charmap {
@@ -71,7 +72,7 @@
CharMaskForWord(x, 2), CharMaskForWord(x, 3));
}
- // Containing all the chars in the C-std::string 's'.
+ // Containing all the chars in the C-string 's'.
// Note that this is expensively recursive because of the C++11 constexpr
// formulation. Use only in constexpr initializers.
static constexpr Charmap FromString(const char* s) {
@@ -149,6 +150,7 @@
constexpr Charmap PunctCharmap() { return GraphCharmap() & ~AlnumCharmap(); }
} // namespace strings_internal
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_INTERNAL_CHAR_MAP_H_
diff --git a/absl/strings/internal/charconv_bigint.cc b/absl/strings/internal/charconv_bigint.cc
index 95d471d..ebf8c07 100644
--- a/absl/strings/internal/charconv_bigint.cc
+++ b/absl/strings/internal/charconv_bigint.cc
@@ -19,6 +19,7 @@
#include <string>
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace strings_internal {
namespace {
@@ -157,12 +158,12 @@
int LargePowerOfFiveSize(int i) { return 2 * i; }
} // namespace
-const uint32_t kFiveToNth[14] = {
+ABSL_DLL const uint32_t kFiveToNth[14] = {
1, 5, 25, 125, 625, 3125, 15625,
78125, 390625, 1953125, 9765625, 48828125, 244140625, 1220703125,
};
-const uint32_t kTenToNth[10] = {
+ABSL_DLL const uint32_t kTenToNth[10] = {
1, 10, 100, 1000, 10000, 100000, 1000000, 10000000, 100000000, 1000000000,
};
@@ -207,7 +208,7 @@
++dropped_digits;
}
if (begin < end && *std::prev(end) == '.') {
- // If the std::string ends in '.', either before or after dropping zeroes, then
+ // If the string ends in '.', either before or after dropping zeroes, then
// drop the decimal point and look for more digits to drop.
dropped_digits = 0;
--end;
@@ -354,4 +355,5 @@
template class BigUnsigned<84>;
} // namespace strings_internal
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/internal/charconv_bigint.h b/absl/strings/internal/charconv_bigint.h
index 7da9a7e..8f70297 100644
--- a/absl/strings/internal/charconv_bigint.h
+++ b/absl/strings/internal/charconv_bigint.h
@@ -20,11 +20,13 @@
#include <iostream>
#include <string>
+#include "absl/base/config.h"
#include "absl/strings/ascii.h"
#include "absl/strings/internal/charconv_parse.h"
#include "absl/strings/string_view.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace strings_internal {
// The largest power that 5 that can be raised to, and still fit in a uint32_t.
@@ -32,8 +34,9 @@
// The largest power that 10 that can be raised to, and still fit in a uint32_t.
constexpr int kMaxSmallPowerOfTen = 9;
-extern const uint32_t kFiveToNth[kMaxSmallPowerOfFive + 1];
-extern const uint32_t kTenToNth[kMaxSmallPowerOfTen + 1];
+ABSL_DLL extern const uint32_t
+ kFiveToNth[kMaxSmallPowerOfFive + 1];
+ABSL_DLL extern const uint32_t kTenToNth[kMaxSmallPowerOfTen + 1];
// Large, fixed-width unsigned integer.
//
@@ -63,7 +66,7 @@
static_cast<uint32_t>(v >> 32)} {}
// Constructs a BigUnsigned from the given string_view containing a decimal
- // value. If the input std::string is not a decimal integer, constructs a 0
+ // value. If the input string is not a decimal integer, constructs a 0
// instead.
explicit BigUnsigned(absl::string_view sv) : size_(0), words_{} {
// Check for valid input, returning a 0 otherwise. This is reasonable
@@ -207,7 +210,7 @@
return words_[index];
}
- // Returns this integer as a decimal std::string. This is not used in the decimal-
+ // Returns this integer as a decimal string. This is not used in the decimal-
// to-binary conversion; it is intended to aid in testing.
std::string ToString() const;
@@ -414,6 +417,7 @@
extern template class BigUnsigned<84>;
} // namespace strings_internal
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_INTERNAL_CHARCONV_BIGINT_H_
diff --git a/absl/strings/internal/charconv_bigint_test.cc b/absl/strings/internal/charconv_bigint_test.cc
index 745714a..a8b9945 100644
--- a/absl/strings/internal/charconv_bigint_test.cc
+++ b/absl/strings/internal/charconv_bigint_test.cc
@@ -19,6 +19,7 @@
#include "gtest/gtest.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace strings_internal {
TEST(BigUnsigned, ShiftLeft) {
@@ -68,6 +69,61 @@
// And we should have fully rotated all bits off by now:
EXPECT_EQ(a, BigUnsigned<84>(0u));
}
+ {
+ // Bit shifting large and small numbers by large and small offsets.
+ // Intended to exercise bounds-checking corner on ShiftLeft() (directly
+ // and under asan).
+
+ // 2**(32*84)-1
+ const BigUnsigned<84> all_bits_one(
+ "1474444211396924248063325089479706787923460402125687709454567433186613"
+ "6228083464060749874845919674257665016359189106695900028098437021384227"
+ "3285029708032466536084583113729486015826557532750465299832071590813090"
+ "2011853039837649252477307070509704043541368002938784757296893793903797"
+ "8180292336310543540677175225040919704702800559606097685920595947397024"
+ "8303316808753252115729411497720357971050627997031988036134171378490368"
+ "6008000778741115399296162550786288457245180872759047016734959330367829"
+ "5235612397427686310674725251378116268607113017720538636924549612987647"
+ "5767411074510311386444547332882472126067840027882117834454260409440463"
+ "9345147252664893456053258463203120637089916304618696601333953616715125"
+ "2115882482473279040772264257431663818610405673876655957323083702713344"
+ "4201105427930770976052393421467136557055");
+ const BigUnsigned<84> zero(0u);
+ const BigUnsigned<84> one(1u);
+ // in bounds shifts
+ for (int i = 1; i < 84*32; ++i) {
+ // shifting all_bits_one to the left should result in a smaller number,
+ // since the high bits rotate off and the low bits are replaced with
+ // zeroes.
+ BigUnsigned<84> big_shifted = all_bits_one;
+ big_shifted.ShiftLeft(i);
+ EXPECT_GT(all_bits_one, big_shifted);
+ // Shifting 1 to the left should instead result in a larger number.
+ BigUnsigned<84> small_shifted = one;
+ small_shifted.ShiftLeft(i);
+ EXPECT_LT(one, small_shifted);
+ }
+ // Shifting by zero or a negative number has no effect
+ for (int no_op_shift : {0, -1, -84 * 32, std::numeric_limits<int>::min()}) {
+ BigUnsigned<84> big_shifted = all_bits_one;
+ big_shifted.ShiftLeft(no_op_shift);
+ EXPECT_EQ(all_bits_one, big_shifted);
+ BigUnsigned<84> small_shifted = one;
+ big_shifted.ShiftLeft(no_op_shift);
+ EXPECT_EQ(one, small_shifted);
+ }
+ // Shifting by an amount greater than the number of bits should result in
+ // zero.
+ for (int out_of_bounds_shift :
+ {84 * 32, 84 * 32 + 1, std::numeric_limits<int>::max()}) {
+ BigUnsigned<84> big_shifted = all_bits_one;
+ big_shifted.ShiftLeft(out_of_bounds_shift);
+ EXPECT_EQ(zero, big_shifted);
+ BigUnsigned<84> small_shifted = one;
+ small_shifted.ShiftLeft(out_of_bounds_shift);
+ EXPECT_EQ(zero, small_shifted);
+ }
+ }
}
TEST(BigUnsigned, MultiplyByUint32) {
@@ -200,4 +256,5 @@
} // namespace strings_internal
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/internal/charconv_parse.cc b/absl/strings/internal/charconv_parse.cc
index f3c7232..8b11868 100644
--- a/absl/strings/internal/charconv_parse.cc
+++ b/absl/strings/internal/charconv_parse.cc
@@ -22,6 +22,7 @@
#include "absl/strings/internal/memutil.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace {
// ParseFloat<10> will read the first 19 significant digits of the mantissa.
@@ -245,14 +246,20 @@
// ConsumeDigits does not protect against overflow on *out; max_digits must
// be chosen with respect to type T to avoid the possibility of overflow.
template <int base, typename T>
-std::size_t ConsumeDigits(const char* begin, const char* end, int max_digits,
- T* out, bool* dropped_nonzero_digit) {
+int ConsumeDigits(const char* begin, const char* end, int max_digits, T* out,
+ bool* dropped_nonzero_digit) {
if (base == 10) {
assert(max_digits <= std::numeric_limits<T>::digits10);
} else if (base == 16) {
assert(max_digits * 4 <= std::numeric_limits<T>::digits);
}
const char* const original_begin = begin;
+
+ // Skip leading zeros, but only if *out is zero.
+ // They don't cause an overflow so we don't have to count them for
+ // `max_digits`.
+ while (!*out && end != begin && *begin == '0') ++begin;
+
T accumulator = *out;
const char* significant_digits_end =
(end - begin > max_digits) ? begin + max_digits : end;
@@ -275,7 +282,7 @@
*dropped_nonzero_digit = true;
}
*out = accumulator;
- return begin - original_begin;
+ return static_cast<int>(begin - original_begin);
}
// Returns true if `v` is one of the chars allowed inside parentheses following
@@ -295,7 +302,7 @@
switch (*begin) {
case 'i':
case 'I': {
- // An infinity std::string consists of the characters "inf" or "infinity",
+ // An infinity string consists of the characters "inf" or "infinity",
// case insensitive.
if (strings_internal::memcasecmp(begin + 1, "nf", 2) != 0) {
return false;
@@ -319,7 +326,7 @@
}
out->type = strings_internal::FloatType::kNan;
out->end = begin + 3;
- // NaN is allowed to be followed by a parenthesized std::string, consisting of
+ // NaN is allowed to be followed by a parenthesized string, consisting of
// only the characters [a-zA-Z0-9_]. Match that if it's present.
begin += 3;
if (begin < end && *begin == '(') {
@@ -365,7 +372,7 @@
int exponent_adjustment = 0;
bool mantissa_is_inexact = false;
- std::size_t pre_decimal_digits = ConsumeDigits<base>(
+ int pre_decimal_digits = ConsumeDigits<base>(
begin, end, MantissaDigitsMax<base>(), &mantissa, &mantissa_is_inexact);
begin += pre_decimal_digits;
int digits_left;
@@ -391,14 +398,14 @@
while (begin < end && *begin == '0') {
++begin;
}
- std::size_t zeros_skipped = begin - begin_zeros;
+ int zeros_skipped = static_cast<int>(begin - begin_zeros);
if (zeros_skipped >= DigitLimit<base>()) {
// refuse to parse pathological inputs
return result;
}
exponent_adjustment -= static_cast<int>(zeros_skipped);
}
- std::size_t post_decimal_digits = ConsumeDigits<base>(
+ int post_decimal_digits = ConsumeDigits<base>(
begin, end, digits_left, &mantissa, &mantissa_is_inexact);
begin += post_decimal_digits;
@@ -493,4 +500,5 @@
chars_format format_flags);
} // namespace strings_internal
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/internal/charconv_parse.h b/absl/strings/internal/charconv_parse.h
index 44d06b2..505998b 100644
--- a/absl/strings/internal/charconv_parse.h
+++ b/absl/strings/internal/charconv_parse.h
@@ -17,9 +17,11 @@
#include <cstdint>
+#include "absl/base/config.h"
#include "absl/strings/charconv.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace strings_internal {
// Enum indicating whether a parsed float is a number or special value.
@@ -92,5 +94,6 @@
absl::chars_format format_flags);
} // namespace strings_internal
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_INTERNAL_CHARCONV_PARSE_H_
diff --git a/absl/strings/internal/charconv_parse_test.cc b/absl/strings/internal/charconv_parse_test.cc
index 9511c98..bc2d111 100644
--- a/absl/strings/internal/charconv_parse_test.cc
+++ b/absl/strings/internal/charconv_parse_test.cc
@@ -63,7 +63,7 @@
}
const std::string::size_type expected_characters_matched = s.find('$');
ABSL_RAW_CHECK(expected_characters_matched != std::string::npos,
- "Input std::string must contain $");
+ "Input string must contain $");
s.replace(expected_characters_matched, 1, "");
ParsedFloat parsed =
diff --git a/absl/strings/internal/cord_internal.cc b/absl/strings/internal/cord_internal.cc
new file mode 100644
index 0000000..59f2e4d
--- /dev/null
+++ b/absl/strings/internal/cord_internal.cc
@@ -0,0 +1,79 @@
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+#include "absl/strings/internal/cord_internal.h"
+
+#include <atomic>
+#include <cassert>
+#include <memory>
+
+#include "absl/container/inlined_vector.h"
+#include "absl/strings/internal/cord_rep_flat.h"
+
+namespace absl {
+ABSL_NAMESPACE_BEGIN
+namespace cord_internal {
+
+ABSL_CONST_INIT std::atomic<bool> cord_ring_buffer_enabled(
+ kCordEnableRingBufferDefault);
+ABSL_CONST_INIT std::atomic<bool> shallow_subcords_enabled(
+ kCordShallowSubcordsDefault);
+
+void CordRep::Destroy(CordRep* rep) {
+ assert(rep != nullptr);
+
+ absl::InlinedVector<CordRep*, Constants::kInlinedVectorSize> pending;
+ while (true) {
+ assert(!rep->refcount.IsImmortal());
+ if (rep->tag == CONCAT) {
+ CordRepConcat* rep_concat = rep->concat();
+ CordRep* right = rep_concat->right;
+ if (!right->refcount.Decrement()) {
+ pending.push_back(right);
+ }
+ CordRep* left = rep_concat->left;
+ delete rep_concat;
+ rep = nullptr;
+ if (!left->refcount.Decrement()) {
+ rep = left;
+ continue;
+ }
+ } else if (rep->tag == EXTERNAL) {
+ CordRepExternal::Delete(rep);
+ rep = nullptr;
+ } else if (rep->tag == SUBSTRING) {
+ CordRepSubstring* rep_substring = rep->substring();
+ CordRep* child = rep_substring->child;
+ delete rep_substring;
+ rep = nullptr;
+ if (!child->refcount.Decrement()) {
+ rep = child;
+ continue;
+ }
+ } else {
+ CordRepFlat::Delete(rep);
+ rep = nullptr;
+ }
+
+ if (!pending.empty()) {
+ rep = pending.back();
+ pending.pop_back();
+ } else {
+ break;
+ }
+ }
+}
+
+} // namespace cord_internal
+ABSL_NAMESPACE_END
+} // namespace absl
diff --git a/absl/strings/internal/cord_internal.h b/absl/strings/internal/cord_internal.h
new file mode 100644
index 0000000..57e7046
--- /dev/null
+++ b/absl/strings/internal/cord_internal.h
@@ -0,0 +1,408 @@
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#ifndef ABSL_STRINGS_INTERNAL_CORD_INTERNAL_H_
+#define ABSL_STRINGS_INTERNAL_CORD_INTERNAL_H_
+
+#include <atomic>
+#include <cassert>
+#include <cstddef>
+#include <cstdint>
+#include <type_traits>
+
+#include "absl/base/internal/invoke.h"
+#include "absl/base/optimization.h"
+#include "absl/container/internal/compressed_tuple.h"
+#include "absl/meta/type_traits.h"
+#include "absl/strings/string_view.h"
+
+namespace absl {
+ABSL_NAMESPACE_BEGIN
+namespace cord_internal {
+
+// Default feature enable states for cord ring buffers
+enum CordFeatureDefaults {
+ kCordEnableRingBufferDefault = false,
+ kCordShallowSubcordsDefault = false
+};
+
+extern std::atomic<bool> cord_ring_buffer_enabled;
+extern std::atomic<bool> shallow_subcords_enabled;
+
+inline void enable_cord_ring_buffer(bool enable) {
+ cord_ring_buffer_enabled.store(enable, std::memory_order_relaxed);
+}
+
+inline void enable_shallow_subcords(bool enable) {
+ shallow_subcords_enabled.store(enable, std::memory_order_relaxed);
+}
+
+enum Constants {
+ // The inlined size to use with absl::InlinedVector.
+ //
+ // Note: The InlinedVectors in this file (and in cord.h) do not need to use
+ // the same value for their inlined size. The fact that they do is historical.
+ // It may be desirable for each to use a different inlined size optimized for
+ // that InlinedVector's usage.
+ //
+ // TODO(jgm): Benchmark to see if there's a more optimal value than 47 for
+ // the inlined vector size (47 exists for backward compatibility).
+ kInlinedVectorSize = 47,
+
+ // Prefer copying blocks of at most this size, otherwise reference count.
+ kMaxBytesToCopy = 511
+};
+
+// Wraps std::atomic for reference counting.
+class Refcount {
+ public:
+ constexpr Refcount() : count_{kRefIncrement} {}
+ struct Immortal {};
+ explicit constexpr Refcount(Immortal) : count_(kImmortalTag) {}
+
+ // Increments the reference count. Imposes no memory ordering.
+ inline void Increment() {
+ count_.fetch_add(kRefIncrement, std::memory_order_relaxed);
+ }
+
+ // Asserts that the current refcount is greater than 0. If the refcount is
+ // greater than 1, decrements the reference count.
+ //
+ // Returns false if there are no references outstanding; true otherwise.
+ // Inserts barriers to ensure that state written before this method returns
+ // false will be visible to a thread that just observed this method returning
+ // false.
+ inline bool Decrement() {
+ int32_t refcount = count_.load(std::memory_order_acquire);
+ assert(refcount > 0 || refcount & kImmortalTag);
+ return refcount != kRefIncrement &&
+ count_.fetch_sub(kRefIncrement, std::memory_order_acq_rel) !=
+ kRefIncrement;
+ }
+
+ // Same as Decrement but expect that refcount is greater than 1.
+ inline bool DecrementExpectHighRefcount() {
+ int32_t refcount =
+ count_.fetch_sub(kRefIncrement, std::memory_order_acq_rel);
+ assert(refcount > 0 || refcount & kImmortalTag);
+ return refcount != kRefIncrement;
+ }
+
+ // Returns the current reference count using acquire semantics.
+ inline int32_t Get() const {
+ return count_.load(std::memory_order_acquire) >> kImmortalShift;
+ }
+
+ // Returns whether the atomic integer is 1.
+ // If the reference count is used in the conventional way, a
+ // reference count of 1 implies that the current thread owns the
+ // reference and no other thread shares it.
+ // This call performs the test for a reference count of one, and
+ // performs the memory barrier needed for the owning thread
+ // to act on the object, knowing that it has exclusive access to the
+ // object.
+ inline bool IsOne() {
+ return count_.load(std::memory_order_acquire) == kRefIncrement;
+ }
+
+ bool IsImmortal() const {
+ return (count_.load(std::memory_order_relaxed) & kImmortalTag) != 0;
+ }
+
+ private:
+ // We reserve the bottom bit to tag a reference count as immortal.
+ // By making it `1` we ensure that we never reach `0` when adding/subtracting
+ // `2`, thus it never looks as if it should be destroyed.
+ // These are used for the StringConstant constructor where we do not increase
+ // the refcount at construction time (due to constinit requirements) but we
+ // will still decrease it at destruction time to avoid branching on Unref.
+ enum {
+ kImmortalShift = 1,
+ kRefIncrement = 1 << kImmortalShift,
+ kImmortalTag = kRefIncrement - 1
+ };
+
+ std::atomic<int32_t> count_;
+};
+
+// The overhead of a vtable is too much for Cord, so we roll our own subclasses
+// using only a single byte to differentiate classes from each other - the "tag"
+// byte. Define the subclasses first so we can provide downcasting helper
+// functions in the base class.
+
+struct CordRepConcat;
+struct CordRepExternal;
+struct CordRepFlat;
+struct CordRepSubstring;
+
+// Various representations that we allow
+enum CordRepKind {
+ CONCAT = 0,
+ EXTERNAL = 1,
+ SUBSTRING = 2,
+ RING = 3,
+
+ // We have different tags for different sized flat arrays,
+ // starting with FLAT, and limited to MAX_FLAT_TAG. The 224 value is based on
+ // the current 'size to tag' encoding of 8 / 32 bytes. If a new tag is needed
+ // in the future, then 'FLAT' and 'MAX_FLAT_TAG' should be adjusted as well
+ // as the Tag <---> Size logic so that FLAT stil represents the minimum flat
+ // allocation size. (32 bytes as of now).
+ FLAT = 4,
+ MAX_FLAT_TAG = 224,
+};
+
+struct CordRep {
+ CordRep() = default;
+ constexpr CordRep(Refcount::Immortal immortal, size_t l)
+ : length(l), refcount(immortal), tag(EXTERNAL), data{} {}
+
+ // The following three fields have to be less than 32 bytes since
+ // that is the smallest supported flat node size.
+ size_t length;
+ Refcount refcount;
+ // If tag < FLAT, it represents CordRepKind and indicates the type of node.
+ // Otherwise, the node type is CordRepFlat and the tag is the encoded size.
+ uint8_t tag;
+ char data[1]; // Starting point for flat array: MUST BE LAST FIELD of CordRep
+
+ inline CordRepConcat* concat();
+ inline const CordRepConcat* concat() const;
+ inline CordRepSubstring* substring();
+ inline const CordRepSubstring* substring() const;
+ inline CordRepExternal* external();
+ inline const CordRepExternal* external() const;
+ inline CordRepFlat* flat();
+ inline const CordRepFlat* flat() const;
+
+ // --------------------------------------------------------------------
+ // Memory management
+
+ // This internal routine is called from the cold path of Unref below. Keeping
+ // it in a separate routine allows good inlining of Unref into many profitable
+ // call sites. However, the call to this function can be highly disruptive to
+ // the register pressure in those callers. To minimize the cost to callers, we
+ // use a special LLVM calling convention that preserves most registers. This
+ // allows the call to this routine in cold paths to not disrupt the caller's
+ // register pressure. This calling convention is not available on all
+ // platforms; we intentionally allow LLVM to ignore the attribute rather than
+ // attempting to hardcode the list of supported platforms.
+#if defined(__clang__) && !defined(__i386__)
+#pragma clang diagnostic push
+#pragma clang diagnostic ignored "-Wattributes"
+ __attribute__((preserve_most))
+#pragma clang diagnostic pop
+#endif
+ static void Destroy(CordRep* rep);
+
+ // Increments the reference count of `rep`.
+ // Requires `rep` to be a non-null pointer value.
+ static inline CordRep* Ref(CordRep* rep);
+
+ // Decrements the reference count of `rep`. Destroys rep if count reaches
+ // zero. Requires `rep` to be a non-null pointer value.
+ static inline void Unref(CordRep* rep);
+};
+
+struct CordRepConcat : public CordRep {
+ CordRep* left;
+ CordRep* right;
+
+ uint8_t depth() const { return static_cast<uint8_t>(data[0]); }
+ void set_depth(uint8_t depth) { data[0] = static_cast<char>(depth); }
+};
+
+struct CordRepSubstring : public CordRep {
+ size_t start; // Starting offset of substring in child
+ CordRep* child;
+};
+
+// Type for function pointer that will invoke the releaser function and also
+// delete the `CordRepExternalImpl` corresponding to the passed in
+// `CordRepExternal`.
+using ExternalReleaserInvoker = void (*)(CordRepExternal*);
+
+// External CordReps are allocated together with a type erased releaser. The
+// releaser is stored in the memory directly following the CordRepExternal.
+struct CordRepExternal : public CordRep {
+ CordRepExternal() = default;
+ explicit constexpr CordRepExternal(absl::string_view str)
+ : CordRep(Refcount::Immortal{}, str.size()),
+ base(str.data()),
+ releaser_invoker(nullptr) {}
+
+ const char* base;
+ // Pointer to function that knows how to call and destroy the releaser.
+ ExternalReleaserInvoker releaser_invoker;
+
+ // Deletes (releases) the external rep.
+ // Requires rep != nullptr and rep->tag == EXTERNAL
+ static void Delete(CordRep* rep);
+};
+
+struct Rank1 {};
+struct Rank0 : Rank1 {};
+
+template <typename Releaser, typename = ::absl::base_internal::invoke_result_t<
+ Releaser, absl::string_view>>
+void InvokeReleaser(Rank0, Releaser&& releaser, absl::string_view data) {
+ ::absl::base_internal::invoke(std::forward<Releaser>(releaser), data);
+}
+
+template <typename Releaser,
+ typename = ::absl::base_internal::invoke_result_t<Releaser>>
+void InvokeReleaser(Rank1, Releaser&& releaser, absl::string_view) {
+ ::absl::base_internal::invoke(std::forward<Releaser>(releaser));
+}
+
+// We use CompressedTuple so that we can benefit from EBCO.
+template <typename Releaser>
+struct CordRepExternalImpl
+ : public CordRepExternal,
+ public ::absl::container_internal::CompressedTuple<Releaser> {
+ // The extra int arg is so that we can avoid interfering with copy/move
+ // constructors while still benefitting from perfect forwarding.
+ template <typename T>
+ CordRepExternalImpl(T&& releaser, int)
+ : CordRepExternalImpl::CompressedTuple(std::forward<T>(releaser)) {
+ this->releaser_invoker = &Release;
+ }
+
+ ~CordRepExternalImpl() {
+ InvokeReleaser(Rank0{}, std::move(this->template get<0>()),
+ absl::string_view(base, length));
+ }
+
+ static void Release(CordRepExternal* rep) {
+ delete static_cast<CordRepExternalImpl*>(rep);
+ }
+};
+
+inline void CordRepExternal::Delete(CordRep* rep) {
+ assert(rep != nullptr && rep->tag == EXTERNAL);
+ auto* rep_external = static_cast<CordRepExternal*>(rep);
+ assert(rep_external->releaser_invoker != nullptr);
+ rep_external->releaser_invoker(rep_external);
+}
+
+template <typename Str>
+struct ConstInitExternalStorage {
+ ABSL_CONST_INIT static CordRepExternal value;
+};
+
+template <typename Str>
+CordRepExternal ConstInitExternalStorage<Str>::value(Str::value);
+
+enum {
+ kMaxInline = 15,
+ // Tag byte & kMaxInline means we are storing a pointer.
+ kTreeFlag = 1 << 4,
+ // Tag byte & kProfiledFlag means we are profiling the Cord.
+ kProfiledFlag = 1 << 5
+};
+
+// If the data has length <= kMaxInline, we store it in `as_chars`, and
+// store the size in `tagged_size`.
+// Else we store it in a tree and store a pointer to that tree in
+// `as_tree.rep` and store a tag in `tagged_size`.
+struct AsTree {
+ absl::cord_internal::CordRep* rep;
+ char padding[kMaxInline + 1 - sizeof(absl::cord_internal::CordRep*) - 1];
+ char tagged_size;
+};
+
+constexpr char GetOrNull(absl::string_view data, size_t pos) {
+ return pos < data.size() ? data[pos] : '\0';
+}
+
+union InlineData {
+ constexpr InlineData() : as_chars{} {}
+ explicit constexpr InlineData(AsTree tree) : as_tree(tree) {}
+ explicit constexpr InlineData(absl::string_view chars)
+ : as_chars{GetOrNull(chars, 0), GetOrNull(chars, 1),
+ GetOrNull(chars, 2), GetOrNull(chars, 3),
+ GetOrNull(chars, 4), GetOrNull(chars, 5),
+ GetOrNull(chars, 6), GetOrNull(chars, 7),
+ GetOrNull(chars, 8), GetOrNull(chars, 9),
+ GetOrNull(chars, 10), GetOrNull(chars, 11),
+ GetOrNull(chars, 12), GetOrNull(chars, 13),
+ GetOrNull(chars, 14), static_cast<char>(chars.size())} {}
+
+ AsTree as_tree;
+ char as_chars[kMaxInline + 1];
+};
+static_assert(sizeof(InlineData) == kMaxInline + 1, "");
+static_assert(sizeof(AsTree) == sizeof(InlineData), "");
+static_assert(offsetof(AsTree, tagged_size) == kMaxInline, "");
+
+inline CordRepConcat* CordRep::concat() {
+ assert(tag == CONCAT);
+ return static_cast<CordRepConcat*>(this);
+}
+
+inline const CordRepConcat* CordRep::concat() const {
+ assert(tag == CONCAT);
+ return static_cast<const CordRepConcat*>(this);
+}
+
+inline CordRepSubstring* CordRep::substring() {
+ assert(tag == SUBSTRING);
+ return static_cast<CordRepSubstring*>(this);
+}
+
+inline const CordRepSubstring* CordRep::substring() const {
+ assert(tag == SUBSTRING);
+ return static_cast<const CordRepSubstring*>(this);
+}
+
+inline CordRepExternal* CordRep::external() {
+ assert(tag == EXTERNAL);
+ return static_cast<CordRepExternal*>(this);
+}
+
+inline const CordRepExternal* CordRep::external() const {
+ assert(tag == EXTERNAL);
+ return static_cast<const CordRepExternal*>(this);
+}
+
+inline CordRepFlat* CordRep::flat() {
+ assert(tag >= FLAT && tag <= MAX_FLAT_TAG);
+ return reinterpret_cast<CordRepFlat*>(this);
+}
+
+inline const CordRepFlat* CordRep::flat() const {
+ assert(tag >= FLAT && tag <= MAX_FLAT_TAG);
+ return reinterpret_cast<const CordRepFlat*>(this);
+}
+
+inline CordRep* CordRep::Ref(CordRep* rep) {
+ assert(rep != nullptr);
+ rep->refcount.Increment();
+ return rep;
+}
+
+inline void CordRep::Unref(CordRep* rep) {
+ assert(rep != nullptr);
+ // Expect refcount to be 0. Avoiding the cost of an atomic decrement should
+ // typically outweigh the cost of an extra branch checking for ref == 1.
+ if (ABSL_PREDICT_FALSE(!rep->refcount.DecrementExpectHighRefcount())) {
+ Destroy(rep);
+ }
+}
+
+} // namespace cord_internal
+
+ABSL_NAMESPACE_END
+} // namespace absl
+#endif // ABSL_STRINGS_INTERNAL_CORD_INTERNAL_H_
diff --git a/absl/strings/internal/cord_rep_flat.h b/absl/strings/internal/cord_rep_flat.h
new file mode 100644
index 0000000..80391a5
--- /dev/null
+++ b/absl/strings/internal/cord_rep_flat.h
@@ -0,0 +1,129 @@
+// Copyright 2020 The Abseil Authors
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#ifndef ABSL_STRINGS_INTERNAL_CORD_REP_FLAT_H_
+#define ABSL_STRINGS_INTERNAL_CORD_REP_FLAT_H_
+
+#include <cassert>
+#include <cstddef>
+#include <cstdint>
+#include <memory>
+
+#include "absl/strings/internal/cord_internal.h"
+
+namespace absl {
+ABSL_NAMESPACE_BEGIN
+namespace cord_internal {
+
+// Note: all constants below are never ODR used and internal to cord, we define
+// these as static constexpr to avoid 'in struct' definition and usage clutter.
+
+// Largest and smallest flat node lengths we are willing to allocate
+// Flat allocation size is stored in tag, which currently can encode sizes up
+// to 4K, encoded as multiple of either 8 or 32 bytes.
+// If we allow for larger sizes, we need to change this to 8/64, 16/128, etc.
+// kMinFlatSize is bounded by tag needing to be at least FLAT * 8 bytes, and
+// ideally a 'nice' size aligning with allocation and cacheline sizes like 32.
+// kMaxFlatSize is bounded by the size resulting in a computed tag no greater
+// than MAX_FLAT_TAG. MAX_FLAT_TAG provides for additional 'high' tag values.
+static constexpr size_t kFlatOverhead = offsetof(CordRep, data);
+static constexpr size_t kMinFlatSize = 32;
+static constexpr size_t kMaxFlatSize = 4096;
+static constexpr size_t kMaxFlatLength = kMaxFlatSize - kFlatOverhead;
+static constexpr size_t kMinFlatLength = kMinFlatSize - kFlatOverhead;
+
+constexpr size_t AllocatedSizeToTagUnchecked(size_t size) {
+ return (size <= 1024) ? size / 8 : 128 + size / 32 - 1024 / 32;
+}
+
+static_assert(kMinFlatSize / 8 >= FLAT, "");
+static_assert(AllocatedSizeToTagUnchecked(kMaxFlatSize) <= MAX_FLAT_TAG, "");
+
+// Helper functions for rounded div, and rounding to exact sizes.
+constexpr size_t DivUp(size_t n, size_t m) { return (n + m - 1) / m; }
+constexpr size_t RoundUp(size_t n, size_t m) { return DivUp(n, m) * m; }
+
+// Returns the size to the nearest equal or larger value that can be
+// expressed exactly as a tag value.
+inline size_t RoundUpForTag(size_t size) {
+ return RoundUp(size, (size <= 1024) ? 8 : 32);
+}
+
+// Converts the allocated size to a tag, rounding down if the size
+// does not exactly match a 'tag expressible' size value. The result is
+// undefined if the size exceeds the maximum size that can be encoded in
+// a tag, i.e., if size is larger than TagToAllocatedSize(<max tag>).
+inline uint8_t AllocatedSizeToTag(size_t size) {
+ const size_t tag = AllocatedSizeToTagUnchecked(size);
+ assert(tag <= MAX_FLAT_TAG);
+ return tag;
+}
+
+// Converts the provided tag to the corresponding allocated size
+constexpr size_t TagToAllocatedSize(uint8_t tag) {
+ return (tag <= 128) ? (tag * 8) : (1024 + (tag - 128) * 32);
+}
+
+// Converts the provided tag to the corresponding available data length
+constexpr size_t TagToLength(uint8_t tag) {
+ return TagToAllocatedSize(tag) - kFlatOverhead;
+}
+
+// Enforce that kMaxFlatSize maps to a well-known exact tag value.
+static_assert(TagToAllocatedSize(224) == kMaxFlatSize, "Bad tag logic");
+
+struct CordRepFlat : public CordRep {
+ // Creates a new flat node.
+ static CordRepFlat* New(size_t len) {
+ if (len <= kMinFlatLength) {
+ len = kMinFlatLength;
+ } else if (len > kMaxFlatLength) {
+ len = kMaxFlatLength;
+ }
+
+ // Round size up so it matches a size we can exactly express in a tag.
+ const size_t size = RoundUpForTag(len + kFlatOverhead);
+ void* const raw_rep = ::operator new(size);
+ CordRepFlat* rep = new (raw_rep) CordRepFlat();
+ rep->tag = AllocatedSizeToTag(size);
+ return rep;
+ }
+
+ // Deletes a CordRepFlat instance created previously through a call to New().
+ // Flat CordReps are allocated and constructed with raw ::operator new and
+ // placement new, and must be destructed and deallocated accordingly.
+ static void Delete(CordRep*rep) {
+ assert(rep->tag >= FLAT);
+#if defined(__cpp_sized_deallocation)
+ size_t size = TagToAllocatedSize(rep->tag);
+ rep->~CordRep();
+ ::operator delete(rep, size);
+#else
+ rep->~CordRep();
+ ::operator delete(rep);
+#endif
+ }
+
+ // Returns the maximum capacity (payload size) of this instance.
+ size_t Capacity() const { return TagToLength(tag); }
+
+ // Returns the allocated size (payload + overhead) of this instance.
+ size_t AllocatedSize() const { return TagToAllocatedSize(tag); }
+};
+
+} // namespace cord_internal
+ABSL_NAMESPACE_END
+} // namespace absl
+
+#endif // ABSL_STRINGS_INTERNAL_CORD_REP_FLAT_H_
diff --git a/absl/strings/internal/escaping.cc b/absl/strings/internal/escaping.cc
new file mode 100644
index 0000000..c527128
--- /dev/null
+++ b/absl/strings/internal/escaping.cc
@@ -0,0 +1,180 @@
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "absl/strings/internal/escaping.h"
+
+#include "absl/base/internal/endian.h"
+#include "absl/base/internal/raw_logging.h"
+
+namespace absl {
+ABSL_NAMESPACE_BEGIN
+namespace strings_internal {
+
+const char kBase64Chars[] =
+ "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/";
+
+size_t CalculateBase64EscapedLenInternal(size_t input_len, bool do_padding) {
+ // Base64 encodes three bytes of input at a time. If the input is not
+ // divisible by three, we pad as appropriate.
+ //
+ // (from https://tools.ietf.org/html/rfc3548)
+ // Special processing is performed if fewer than 24 bits are available
+ // at the end of the data being encoded. A full encoding quantum is
+ // always completed at the end of a quantity. When fewer than 24 input
+ // bits are available in an input group, zero bits are added (on the
+ // right) to form an integral number of 6-bit groups. Padding at the
+ // end of the data is performed using the '=' character. Since all base
+ // 64 input is an integral number of octets, only the following cases
+ // can arise:
+
+ // Base64 encodes each three bytes of input into four bytes of output.
+ size_t len = (input_len / 3) * 4;
+
+ if (input_len % 3 == 0) {
+ // (from https://tools.ietf.org/html/rfc3548)
+ // (1) the final quantum of encoding input is an integral multiple of 24
+ // bits; here, the final unit of encoded output will be an integral
+ // multiple of 4 characters with no "=" padding,
+ } else if (input_len % 3 == 1) {
+ // (from https://tools.ietf.org/html/rfc3548)
+ // (2) the final quantum of encoding input is exactly 8 bits; here, the
+ // final unit of encoded output will be two characters followed by two
+ // "=" padding characters, or
+ len += 2;
+ if (do_padding) {
+ len += 2;
+ }
+ } else { // (input_len % 3 == 2)
+ // (from https://tools.ietf.org/html/rfc3548)
+ // (3) the final quantum of encoding input is exactly 16 bits; here, the
+ // final unit of encoded output will be three characters followed by one
+ // "=" padding character.
+ len += 3;
+ if (do_padding) {
+ len += 1;
+ }
+ }
+
+ assert(len >= input_len); // make sure we didn't overflow
+ return len;
+}
+
+size_t Base64EscapeInternal(const unsigned char* src, size_t szsrc, char* dest,
+ size_t szdest, const char* base64,
+ bool do_padding) {
+ static const char kPad64 = '=';
+
+ if (szsrc * 4 > szdest * 3) return 0;
+
+ char* cur_dest = dest;
+ const unsigned char* cur_src = src;
+
+ char* const limit_dest = dest + szdest;
+ const unsigned char* const limit_src = src + szsrc;
+
+ // Three bytes of data encodes to four characters of cyphertext.
+ // So we can pump through three-byte chunks atomically.
+ if (szsrc >= 3) { // "limit_src - 3" is UB if szsrc < 3.
+ while (cur_src < limit_src - 3) { // While we have >= 32 bits.
+ uint32_t in = absl::big_endian::Load32(cur_src) >> 8;
+
+ cur_dest[0] = base64[in >> 18];
+ in &= 0x3FFFF;
+ cur_dest[1] = base64[in >> 12];
+ in &= 0xFFF;
+ cur_dest[2] = base64[in >> 6];
+ in &= 0x3F;
+ cur_dest[3] = base64[in];
+
+ cur_dest += 4;
+ cur_src += 3;
+ }
+ }
+ // To save time, we didn't update szdest or szsrc in the loop. So do it now.
+ szdest = limit_dest - cur_dest;
+ szsrc = limit_src - cur_src;
+
+ /* now deal with the tail (<=3 bytes) */
+ switch (szsrc) {
+ case 0:
+ // Nothing left; nothing more to do.
+ break;
+ case 1: {
+ // One byte left: this encodes to two characters, and (optionally)
+ // two pad characters to round out the four-character cypherblock.
+ if (szdest < 2) return 0;
+ uint32_t in = cur_src[0];
+ cur_dest[0] = base64[in >> 2];
+ in &= 0x3;
+ cur_dest[1] = base64[in << 4];
+ cur_dest += 2;
+ szdest -= 2;
+ if (do_padding) {
+ if (szdest < 2) return 0;
+ cur_dest[0] = kPad64;
+ cur_dest[1] = kPad64;
+ cur_dest += 2;
+ szdest -= 2;
+ }
+ break;
+ }
+ case 2: {
+ // Two bytes left: this encodes to three characters, and (optionally)
+ // one pad character to round out the four-character cypherblock.
+ if (szdest < 3) return 0;
+ uint32_t in = absl::big_endian::Load16(cur_src);
+ cur_dest[0] = base64[in >> 10];
+ in &= 0x3FF;
+ cur_dest[1] = base64[in >> 4];
+ in &= 0x00F;
+ cur_dest[2] = base64[in << 2];
+ cur_dest += 3;
+ szdest -= 3;
+ if (do_padding) {
+ if (szdest < 1) return 0;
+ cur_dest[0] = kPad64;
+ cur_dest += 1;
+ szdest -= 1;
+ }
+ break;
+ }
+ case 3: {
+ // Three bytes left: same as in the big loop above. We can't do this in
+ // the loop because the loop above always reads 4 bytes, and the fourth
+ // byte is past the end of the input.
+ if (szdest < 4) return 0;
+ uint32_t in = (cur_src[0] << 16) + absl::big_endian::Load16(cur_src + 1);
+ cur_dest[0] = base64[in >> 18];
+ in &= 0x3FFFF;
+ cur_dest[1] = base64[in >> 12];
+ in &= 0xFFF;
+ cur_dest[2] = base64[in >> 6];
+ in &= 0x3F;
+ cur_dest[3] = base64[in];
+ cur_dest += 4;
+ szdest -= 4;
+ break;
+ }
+ default:
+ // Should not be reached: blocks of 4 bytes are handled
+ // in the while loop before this switch statement.
+ ABSL_RAW_LOG(FATAL, "Logic problem? szsrc = %zu", szsrc);
+ break;
+ }
+ return (cur_dest - dest);
+}
+
+} // namespace strings_internal
+ABSL_NAMESPACE_END
+} // namespace absl
diff --git a/absl/strings/internal/escaping.h b/absl/strings/internal/escaping.h
new file mode 100644
index 0000000..6a9ce60
--- /dev/null
+++ b/absl/strings/internal/escaping.h
@@ -0,0 +1,58 @@
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#ifndef ABSL_STRINGS_INTERNAL_ESCAPING_H_
+#define ABSL_STRINGS_INTERNAL_ESCAPING_H_
+
+#include <cassert>
+
+#include "absl/strings/internal/resize_uninitialized.h"
+
+namespace absl {
+ABSL_NAMESPACE_BEGIN
+namespace strings_internal {
+
+ABSL_CONST_INIT extern const char kBase64Chars[];
+
+// Calculates how long a string will be when it is base64 encoded given its
+// length and whether or not the result should be padded.
+size_t CalculateBase64EscapedLenInternal(size_t input_len, bool do_padding);
+
+// Base64-encodes `src` using the alphabet provided in `base64` and writes the
+// result to `dest`. If `do_padding` is true, `dest` is padded with '=' chars
+// until its length is a multiple of 3. Returns the length of `dest`.
+size_t Base64EscapeInternal(const unsigned char* src, size_t szsrc, char* dest,
+ size_t szdest, const char* base64, bool do_padding);
+
+// Base64-encodes `src` using the alphabet provided in `base64` and writes the
+// result to `dest`. If `do_padding` is true, `dest` is padded with '=' chars
+// until its length is a multiple of 3.
+template <typename String>
+void Base64EscapeInternal(const unsigned char* src, size_t szsrc, String* dest,
+ bool do_padding, const char* base64_chars) {
+ const size_t calc_escaped_size =
+ CalculateBase64EscapedLenInternal(szsrc, do_padding);
+ STLStringResizeUninitialized(dest, calc_escaped_size);
+
+ const size_t escaped_len = Base64EscapeInternal(
+ src, szsrc, &(*dest)[0], dest->size(), base64_chars, do_padding);
+ assert(calc_escaped_size == escaped_len);
+ dest->erase(escaped_len);
+}
+
+} // namespace strings_internal
+ABSL_NAMESPACE_END
+} // namespace absl
+
+#endif // ABSL_STRINGS_INTERNAL_ESCAPING_H_
diff --git a/absl/strings/internal/escaping_test_common.h b/absl/strings/internal/escaping_test_common.h
index bd80303..7b18017 100644
--- a/absl/strings/internal/escaping_test_common.h
+++ b/absl/strings/internal/escaping_test_common.h
@@ -22,6 +22,7 @@
#include "absl/strings/string_view.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace strings_internal {
struct base64_testcase {
@@ -126,6 +127,7 @@
}
} // namespace strings_internal
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_INTERNAL_ESCAPING_TEST_COMMON_H_
diff --git a/absl/strings/internal/memutil.cc b/absl/strings/internal/memutil.cc
index 77aa63c..2519c68 100644
--- a/absl/strings/internal/memutil.cc
+++ b/absl/strings/internal/memutil.cc
@@ -17,6 +17,7 @@
#include <cstdlib>
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace strings_internal {
int memcasecmp(const char* s1, const char* s2, size_t len) {
@@ -107,4 +108,5 @@
}
} // namespace strings_internal
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/internal/memutil.h b/absl/strings/internal/memutil.h
index 7c071a8..9ad0535 100644
--- a/absl/strings/internal/memutil.h
+++ b/absl/strings/internal/memutil.h
@@ -69,6 +69,7 @@
#include "absl/strings/ascii.h" // for absl::ascii_tolower
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace strings_internal {
inline char* memcat(char* dest, size_t destlen, const char* src,
@@ -141,6 +142,7 @@
size_t neelen);
} // namespace strings_internal
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_INTERNAL_MEMUTIL_H_
diff --git a/absl/strings/internal/numbers_test_common.h b/absl/strings/internal/numbers_test_common.h
index 2824720..eaa88a8 100644
--- a/absl/strings/internal/numbers_test_common.h
+++ b/absl/strings/internal/numbers_test_common.h
@@ -23,7 +23,10 @@
#include <limits>
#include <string>
+#include "absl/base/config.h"
+
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace strings_internal {
template <typename IntType>
@@ -42,8 +45,9 @@
while (value != 0) {
const IntType next_value = value / base;
// Can't use std::abs here because of problems when IntType is unsigned.
- int remainder = value > next_value * base ? value - next_value * base
- : next_value * base - value;
+ int remainder =
+ static_cast<int>(value > next_value * base ? value - next_value * base
+ : next_value * base - value);
char c = remainder < 10 ? '0' + remainder : 'A' + remainder - 10;
destination->insert(0, 1, c);
value = next_value;
@@ -166,7 +170,7 @@
{"0x1234", true, 16, 0x1234},
- // Base-10 std::string version.
+ // Base-10 string version.
{"1234", true, 0, 1234},
{nullptr, false, 0, 0},
}};
@@ -174,6 +178,7 @@
}
} // namespace strings_internal
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_INTERNAL_NUMBERS_TEST_COMMON_H_
diff --git a/absl/strings/internal/ostringstream.cc b/absl/strings/internal/ostringstream.cc
index d0f0f84..05324c7 100644
--- a/absl/strings/internal/ostringstream.cc
+++ b/absl/strings/internal/ostringstream.cc
@@ -15,6 +15,7 @@
#include "absl/strings/internal/ostringstream.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace strings_internal {
OStringStream::Buf::int_type OStringStream::overflow(int c) {
@@ -31,4 +32,5 @@
}
} // namespace strings_internal
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/internal/ostringstream.h b/absl/strings/internal/ostringstream.h
index 2079201..d25d604 100644
--- a/absl/strings/internal/ostringstream.h
+++ b/absl/strings/internal/ostringstream.h
@@ -23,6 +23,7 @@
#include "absl/base/port.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace strings_internal {
// The same as std::ostringstream but appends to a user-specified std::string,
@@ -82,6 +83,7 @@
};
} // namespace strings_internal
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_INTERNAL_OSTRINGSTREAM_H_
diff --git a/absl/strings/internal/pow10_helper.cc b/absl/strings/internal/pow10_helper.cc
index 03ed8d0..42e96c3 100644
--- a/absl/strings/internal/pow10_helper.cc
+++ b/absl/strings/internal/pow10_helper.cc
@@ -17,6 +17,7 @@
#include <cmath>
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace strings_internal {
namespace {
@@ -117,4 +118,5 @@
}
} // namespace strings_internal
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/internal/pow10_helper.h b/absl/strings/internal/pow10_helper.h
index 9d1aa71..c37c2c3 100644
--- a/absl/strings/internal/pow10_helper.h
+++ b/absl/strings/internal/pow10_helper.h
@@ -22,7 +22,10 @@
#include <vector>
+#include "absl/base/config.h"
+
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace strings_internal {
// Computes the precise value of 10^exp. (I.e. the nearest representable
@@ -31,6 +34,7 @@
double Pow10(int exp);
} // namespace strings_internal
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_INTERNAL_POW10_HELPER_H_
diff --git a/absl/strings/internal/pow10_helper_test.cc b/absl/strings/internal/pow10_helper_test.cc
index a4a68b5..a4ff76d 100644
--- a/absl/strings/internal/pow10_helper_test.cc
+++ b/absl/strings/internal/pow10_helper_test.cc
@@ -20,6 +20,7 @@
#include "absl/strings/str_format.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace strings_internal {
namespace {
@@ -117,4 +118,5 @@
} // namespace
} // namespace strings_internal
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/internal/resize_uninitialized.h b/absl/strings/internal/resize_uninitialized.h
index 469962b..e42628e 100644
--- a/absl/strings/internal/resize_uninitialized.h
+++ b/absl/strings/internal/resize_uninitialized.h
@@ -25,6 +25,7 @@
#include "absl/meta/type_traits.h" // for void_t
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace strings_internal {
// Is a subclass of true_type or false_type, depending on whether or not
@@ -35,8 +36,7 @@
static void Resize(string_type* s, size_t new_size) { s->resize(new_size); }
};
-// __resize_default_init is provided by libc++ >= 8.0 and by Google's internal
-// ::string implementation.
+// __resize_default_init is provided by libc++ >= 8.0
template <typename string_type>
struct ResizeUninitializedTraits<
string_type, absl::void_t<decltype(std::declval<string_type&>()
@@ -67,6 +67,7 @@
}
} // namespace strings_internal
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_INTERNAL_RESIZE_UNINITIALIZED_H_
diff --git a/absl/strings/internal/resize_uninitialized_test.cc b/absl/strings/internal/resize_uninitialized_test.cc
index c5be0b1..0f8b3c2 100644
--- a/absl/strings/internal/resize_uninitialized_test.cc
+++ b/absl/strings/internal/resize_uninitialized_test.cc
@@ -20,13 +20,27 @@
int resize_call_count = 0;
+// A mock string class whose only purpose is to track how many times its
+// resize() method has been called.
struct resizable_string {
+ size_t size() const { return 0; }
+ char& operator[](size_t) {
+ static char c = '\0';
+ return c;
+ }
void resize(size_t) { resize_call_count += 1; }
};
int resize_default_init_call_count = 0;
+// A mock string class whose only purpose is to track how many times its
+// resize() and __resize_default_init() methods have been called.
struct resize_default_init_string {
+ size_t size() const { return 0; }
+ char& operator[](size_t) {
+ static char c = '\0';
+ return c;
+ }
void resize(size_t) { resize_call_count += 1; }
void __resize_default_init(size_t) { resize_default_init_call_count += 1; }
};
diff --git a/absl/strings/internal/stl_type_traits.h b/absl/strings/internal/stl_type_traits.h
index 202ab37..6035ca4 100644
--- a/absl/strings/internal/stl_type_traits.h
+++ b/absl/strings/internal/stl_type_traits.h
@@ -40,6 +40,7 @@
#include "absl/meta/type_traits.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace strings_internal {
template <typename C, template <typename...> class T>
@@ -242,5 +243,6 @@
IsConvertibleToSTLContainer<C>> {};
} // namespace strings_internal
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_INTERNAL_STL_TYPE_TRAITS_H_
diff --git a/absl/strings/internal/str_format/arg.cc b/absl/strings/internal/str_format/arg.cc
index b40be8f..e28a29b 100644
--- a/absl/strings/internal/str_format/arg.cc
+++ b/absl/strings/internal/str_format/arg.cc
@@ -1,3 +1,17 @@
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
//
// POSIX spec:
// http://pubs.opengroup.org/onlinepubs/009695399/functions/fprintf.html
@@ -12,13 +26,13 @@
#include "absl/base/port.h"
#include "absl/strings/internal/str_format/float_conversion.h"
+#include "absl/strings/numbers.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace str_format_internal {
namespace {
-const char kDigit[2][32] = { "0123456789abcdef", "0123456789ABCDEF" };
-
// Reduce *capacity by s.size(), clipped to a 0 minimum.
void ReducePadding(string_view s, size_t *capacity) {
*capacity = Excess(s.size(), *capacity);
@@ -32,6 +46,10 @@
template <typename T>
struct MakeUnsigned : std::make_unsigned<T> {};
template <>
+struct MakeUnsigned<absl::int128> {
+ using type = absl::uint128;
+};
+template <>
struct MakeUnsigned<absl::uint128> {
using type = absl::uint128;
};
@@ -39,128 +57,183 @@
template <typename T>
struct IsSigned : std::is_signed<T> {};
template <>
+struct IsSigned<absl::int128> : std::true_type {};
+template <>
struct IsSigned<absl::uint128> : std::false_type {};
-class ConvertedIntInfo {
+// Integral digit printer.
+// Call one of the PrintAs* routines after construction once.
+// Use with_neg_and_zero/without_neg_or_zero/is_negative to access the results.
+class IntDigits {
public:
+ // Print the unsigned integer as octal.
+ // Supports unsigned integral types and uint128.
template <typename T>
- ConvertedIntInfo(T v, ConversionChar conv) {
- using Unsigned = typename MakeUnsigned<T>::type;
- auto u = static_cast<Unsigned>(v);
- if (IsNeg(v)) {
- is_neg_ = true;
- u = Unsigned{} - u;
- } else {
- is_neg_ = false;
- }
- UnsignedToStringRight(u, conv);
+ void PrintAsOct(T v) {
+ static_assert(!IsSigned<T>::value, "");
+ char *p = storage_ + sizeof(storage_);
+ do {
+ *--p = static_cast<char>('0' + (static_cast<size_t>(v) & 7));
+ v >>= 3;
+ } while (v);
+ start_ = p;
+ size_ = storage_ + sizeof(storage_) - p;
}
- string_view digits() const {
- return {end() - size_, static_cast<size_t>(size_)};
+ // Print the signed or unsigned integer as decimal.
+ // Supports all integral types.
+ template <typename T>
+ void PrintAsDec(T v) {
+ static_assert(std::is_integral<T>::value, "");
+ start_ = storage_;
+ size_ = numbers_internal::FastIntToBuffer(v, storage_) - storage_;
}
- bool is_neg() const { return is_neg_; }
+
+ void PrintAsDec(int128 v) {
+ auto u = static_cast<uint128>(v);
+ bool add_neg = false;
+ if (v < 0) {
+ add_neg = true;
+ u = uint128{} - u;
+ }
+ PrintAsDec(u, add_neg);
+ }
+
+ void PrintAsDec(uint128 v, bool add_neg = false) {
+ // This function can be sped up if needed. We can call FastIntToBuffer
+ // twice, or fix FastIntToBuffer to support uint128.
+ char *p = storage_ + sizeof(storage_);
+ do {
+ p -= 2;
+ numbers_internal::PutTwoDigits(static_cast<size_t>(v % 100), p);
+ v /= 100;
+ } while (v);
+ if (p[0] == '0') {
+ // We printed one too many hexits.
+ ++p;
+ }
+ if (add_neg) {
+ *--p = '-';
+ }
+ size_ = storage_ + sizeof(storage_) - p;
+ start_ = p;
+ }
+
+ // Print the unsigned integer as hex using lowercase.
+ // Supports unsigned integral types and uint128.
+ template <typename T>
+ void PrintAsHexLower(T v) {
+ static_assert(!IsSigned<T>::value, "");
+ char *p = storage_ + sizeof(storage_);
+
+ do {
+ p -= 2;
+ constexpr const char* table = numbers_internal::kHexTable;
+ std::memcpy(p, table + 2 * (static_cast<size_t>(v) & 0xFF), 2);
+ if (sizeof(T) == 1) break;
+ v >>= 8;
+ } while (v);
+ if (p[0] == '0') {
+ // We printed one too many digits.
+ ++p;
+ }
+ start_ = p;
+ size_ = storage_ + sizeof(storage_) - p;
+ }
+
+ // Print the unsigned integer as hex using uppercase.
+ // Supports unsigned integral types and uint128.
+ template <typename T>
+ void PrintAsHexUpper(T v) {
+ static_assert(!IsSigned<T>::value, "");
+ char *p = storage_ + sizeof(storage_);
+
+ // kHexTable is only lowercase, so do it manually for uppercase.
+ do {
+ *--p = "0123456789ABCDEF"[static_cast<size_t>(v) & 15];
+ v >>= 4;
+ } while (v);
+ start_ = p;
+ size_ = storage_ + sizeof(storage_) - p;
+ }
+
+ // The printed value including the '-' sign if available.
+ // For inputs of value `0`, this will return "0"
+ string_view with_neg_and_zero() const { return {start_, size_}; }
+
+ // The printed value not including the '-' sign.
+ // For inputs of value `0`, this will return "".
+ string_view without_neg_or_zero() const {
+ static_assert('-' < '0', "The check below verifies both.");
+ size_t advance = start_[0] <= '0' ? 1 : 0;
+ return {start_ + advance, size_ - advance};
+ }
+
+ bool is_negative() const { return start_[0] == '-'; }
private:
- template <typename T, bool IsSigned>
- struct IsNegImpl {
- static bool Eval(T v) { return v < 0; }
- };
- template <typename T>
- struct IsNegImpl<T, false> {
- static bool Eval(T) {
- return false;
- }
- };
-
- template <typename T>
- bool IsNeg(T v) {
- return IsNegImpl<T, IsSigned<T>::value>::Eval(v);
- }
-
- template <typename T>
- void UnsignedToStringRight(T u, ConversionChar conv) {
- char *p = end();
- switch (conv.radix()) {
- default:
- case 10:
- for (; u; u /= 10)
- *--p = static_cast<char>('0' + static_cast<size_t>(u % 10));
- break;
- case 8:
- for (; u; u /= 8)
- *--p = static_cast<char>('0' + static_cast<size_t>(u % 8));
- break;
- case 16: {
- const char *digits = kDigit[conv.upper() ? 1 : 0];
- for (; u; u /= 16) *--p = digits[static_cast<size_t>(u % 16)];
- break;
- }
- }
- size_ = static_cast<int>(end() - p);
- }
-
- const char *end() const { return storage_ + sizeof(storage_); }
- char *end() { return storage_ + sizeof(storage_); }
-
- bool is_neg_;
- int size_;
- // Max size: 128 bit value as octal -> 43 digits
- char storage_[128 / 3 + 1];
+ const char *start_;
+ size_t size_;
+ // Max size: 128 bit value as octal -> 43 digits, plus sign char
+ char storage_[128 / 3 + 1 + 1];
};
// Note: 'o' conversions do not have a base indicator, it's just that
// the '#' flag is specified to modify the precision for 'o' conversions.
-string_view BaseIndicator(const ConvertedIntInfo &info,
- const ConversionSpec conv) {
- bool alt = conv.flags().alt;
- int radix = conv.conv().radix();
- if (conv.conv().id() == ConversionChar::p)
- alt = true; // always show 0x for %p.
+string_view BaseIndicator(const IntDigits &as_digits,
+ const FormatConversionSpecImpl conv) {
+ // always show 0x for %p.
+ bool alt = conv.has_alt_flag() ||
+ conv.conversion_char() == FormatConversionCharInternal::p;
+ bool hex = (conv.conversion_char() == FormatConversionCharInternal::x ||
+ conv.conversion_char() == FormatConversionCharInternal::X ||
+ conv.conversion_char() == FormatConversionCharInternal::p);
// From the POSIX description of '#' flag:
// "For x or X conversion specifiers, a non-zero result shall have
// 0x (or 0X) prefixed to it."
- if (alt && radix == 16 && !info.digits().empty()) {
- if (conv.conv().upper()) return "0X";
- return "0x";
+ if (alt && hex && !as_digits.without_neg_or_zero().empty()) {
+ return conv.conversion_char() == FormatConversionCharInternal::X ? "0X"
+ : "0x";
}
return {};
}
-string_view SignColumn(bool neg, const ConversionSpec conv) {
- if (conv.conv().is_signed()) {
+string_view SignColumn(bool neg, const FormatConversionSpecImpl conv) {
+ if (conv.conversion_char() == FormatConversionCharInternal::d ||
+ conv.conversion_char() == FormatConversionCharInternal::i) {
if (neg) return "-";
- if (conv.flags().show_pos) return "+";
- if (conv.flags().sign_col) return " ";
+ if (conv.has_show_pos_flag()) return "+";
+ if (conv.has_sign_col_flag()) return " ";
}
return {};
}
-bool ConvertCharImpl(unsigned char v, const ConversionSpec conv,
+bool ConvertCharImpl(unsigned char v, const FormatConversionSpecImpl conv,
FormatSinkImpl *sink) {
size_t fill = 0;
if (conv.width() >= 0) fill = conv.width();
ReducePadding(1, &fill);
- if (!conv.flags().left) sink->Append(fill, ' ');
+ if (!conv.has_left_flag()) sink->Append(fill, ' ');
sink->Append(1, v);
- if (conv.flags().left) sink->Append(fill, ' ');
+ if (conv.has_left_flag()) sink->Append(fill, ' ');
return true;
}
-bool ConvertIntImplInner(const ConvertedIntInfo &info,
- const ConversionSpec conv, FormatSinkImpl *sink) {
+bool ConvertIntImplInnerSlow(const IntDigits &as_digits,
+ const FormatConversionSpecImpl conv,
+ FormatSinkImpl *sink) {
// Print as a sequence of Substrings:
// [left_spaces][sign][base_indicator][zeroes][formatted][right_spaces]
size_t fill = 0;
if (conv.width() >= 0) fill = conv.width();
- string_view formatted = info.digits();
+ string_view formatted = as_digits.without_neg_or_zero();
ReducePadding(formatted, &fill);
- string_view sign = SignColumn(info.is_neg(), conv);
+ string_view sign = SignColumn(as_digits.is_negative(), conv);
ReducePadding(sign, &fill);
- string_view base_indicator = BaseIndicator(info, conv);
+ string_view base_indicator = BaseIndicator(as_digits, conv);
ReducePadding(base_indicator, &fill);
int precision = conv.precision();
@@ -168,7 +241,8 @@
if (!precision_specified)
precision = 1;
- if (conv.flags().alt && conv.conv().id() == ConversionChar::o) {
+ if (conv.has_alt_flag() &&
+ conv.conversion_char() == FormatConversionCharInternal::o) {
// From POSIX description of the '#' (alt) flag:
// "For o conversion, it increases the precision (if necessary) to
// force the first digit of the result to be zero."
@@ -181,13 +255,13 @@
size_t num_zeroes = Excess(formatted.size(), precision);
ReducePadding(num_zeroes, &fill);
- size_t num_left_spaces = !conv.flags().left ? fill : 0;
- size_t num_right_spaces = conv.flags().left ? fill : 0;
+ size_t num_left_spaces = !conv.has_left_flag() ? fill : 0;
+ size_t num_right_spaces = conv.has_left_flag() ? fill : 0;
// From POSIX description of the '0' (zero) flag:
// "For d, i, o, u, x, and X conversion specifiers, if a precision
// is specified, the '0' flag is ignored."
- if (!precision_specified && conv.flags().zero) {
+ if (!precision_specified && conv.has_zero_flag()) {
num_zeroes += num_left_spaces;
num_left_spaces = 0;
}
@@ -202,72 +276,97 @@
}
template <typename T>
-bool ConvertIntImplInner(T v, const ConversionSpec conv, FormatSinkImpl *sink) {
- ConvertedIntInfo info(v, conv.conv());
- if (conv.flags().basic && conv.conv().id() != ConversionChar::p) {
- if (info.is_neg()) sink->Append(1, '-');
- if (info.digits().empty()) {
- sink->Append(1, '0');
- } else {
- sink->Append(info.digits());
- }
+bool ConvertIntArg(T v, const FormatConversionSpecImpl conv,
+ FormatSinkImpl *sink) {
+ using U = typename MakeUnsigned<T>::type;
+ IntDigits as_digits;
+
+ // This odd casting is due to a bug in -Wswitch behavior in gcc49 which causes
+ // it to complain about a switch/case type mismatch, even though both are
+ // FormatConverionChar. Likely this is because at this point
+ // FormatConversionChar is declared, but not defined.
+ switch (static_cast<uint8_t>(conv.conversion_char())) {
+ case static_cast<uint8_t>(FormatConversionCharInternal::c):
+ return ConvertCharImpl(static_cast<unsigned char>(v), conv, sink);
+
+ case static_cast<uint8_t>(FormatConversionCharInternal::o):
+ as_digits.PrintAsOct(static_cast<U>(v));
+ break;
+
+ case static_cast<uint8_t>(FormatConversionCharInternal::x):
+ as_digits.PrintAsHexLower(static_cast<U>(v));
+ break;
+ case static_cast<uint8_t>(FormatConversionCharInternal::X):
+ as_digits.PrintAsHexUpper(static_cast<U>(v));
+ break;
+
+ case static_cast<uint8_t>(FormatConversionCharInternal::u):
+ as_digits.PrintAsDec(static_cast<U>(v));
+ break;
+
+ case static_cast<uint8_t>(FormatConversionCharInternal::d):
+ case static_cast<uint8_t>(FormatConversionCharInternal::i):
+ as_digits.PrintAsDec(v);
+ break;
+
+ case static_cast<uint8_t>(FormatConversionCharInternal::a):
+ case static_cast<uint8_t>(FormatConversionCharInternal::e):
+ case static_cast<uint8_t>(FormatConversionCharInternal::f):
+ case static_cast<uint8_t>(FormatConversionCharInternal::g):
+ case static_cast<uint8_t>(FormatConversionCharInternal::A):
+ case static_cast<uint8_t>(FormatConversionCharInternal::E):
+ case static_cast<uint8_t>(FormatConversionCharInternal::F):
+ case static_cast<uint8_t>(FormatConversionCharInternal::G):
+ return ConvertFloatImpl(static_cast<double>(v), conv, sink);
+
+ default:
+ ABSL_INTERNAL_ASSUME(false);
+ }
+
+ if (conv.is_basic()) {
+ sink->Append(as_digits.with_neg_and_zero());
return true;
}
- return ConvertIntImplInner(info, conv, sink);
+ return ConvertIntImplInnerSlow(as_digits, conv, sink);
}
template <typename T>
-bool ConvertIntArg(T v, const ConversionSpec conv, FormatSinkImpl *sink) {
- if (conv.conv().is_float()) {
- return FormatConvertImpl(static_cast<double>(v), conv, sink).value;
- }
- if (conv.conv().id() == ConversionChar::c)
- return ConvertCharImpl(static_cast<unsigned char>(v), conv, sink);
- if (!conv.conv().is_integral())
- return false;
- if (!conv.conv().is_signed() && IsSigned<T>::value) {
- using U = typename MakeUnsigned<T>::type;
- return FormatConvertImpl(static_cast<U>(v), conv, sink).value;
- }
- return ConvertIntImplInner(v, conv, sink);
+bool ConvertFloatArg(T v, const FormatConversionSpecImpl conv,
+ FormatSinkImpl *sink) {
+ return FormatConversionCharIsFloat(conv.conversion_char()) &&
+ ConvertFloatImpl(v, conv, sink);
}
-template <typename T>
-bool ConvertFloatArg(T v, const ConversionSpec conv, FormatSinkImpl *sink) {
- return conv.conv().is_float() && ConvertFloatImpl(v, conv, sink);
-}
-
-inline bool ConvertStringArg(string_view v, const ConversionSpec conv,
+inline bool ConvertStringArg(string_view v, const FormatConversionSpecImpl conv,
FormatSinkImpl *sink) {
- if (conv.conv().id() != ConversionChar::s)
- return false;
- if (conv.flags().basic) {
+ if (conv.is_basic()) {
sink->Append(v);
return true;
}
return sink->PutPaddedString(v, conv.width(), conv.precision(),
- conv.flags().left);
+ conv.has_left_flag());
}
} // namespace
// ==================== Strings ====================
-ConvertResult<Conv::s> FormatConvertImpl(const std::string &v,
- const ConversionSpec conv,
- FormatSinkImpl *sink) {
+StringConvertResult FormatConvertImpl(const std::string &v,
+ const FormatConversionSpecImpl conv,
+ FormatSinkImpl *sink) {
return {ConvertStringArg(v, conv, sink)};
}
-ConvertResult<Conv::s> FormatConvertImpl(string_view v,
- const ConversionSpec conv,
- FormatSinkImpl *sink) {
+StringConvertResult FormatConvertImpl(string_view v,
+ const FormatConversionSpecImpl conv,
+ FormatSinkImpl *sink) {
return {ConvertStringArg(v, conv, sink)};
}
-ConvertResult<Conv::s | Conv::p> FormatConvertImpl(const char *v,
- const ConversionSpec conv,
- FormatSinkImpl *sink) {
- if (conv.conv().id() == ConversionChar::p)
+ArgConvertResult<FormatConversionCharSetUnion(
+ FormatConversionCharSetInternal::s, FormatConversionCharSetInternal::p)>
+FormatConvertImpl(const char *v, const FormatConversionSpecImpl conv,
+ FormatSinkImpl *sink) {
+ if (conv.conversion_char() == FormatConversionCharInternal::p)
return {FormatConvertImpl(VoidPtr(v), conv, sink).value};
size_t len;
if (v == nullptr) {
@@ -275,96 +374,106 @@
} else if (conv.precision() < 0) {
len = std::strlen(v);
} else {
- // If precision is set, we look for the null terminator on the valid range.
+ // If precision is set, we look for the NUL-terminator on the valid range.
len = std::find(v, v + conv.precision(), '\0') - v;
}
return {ConvertStringArg(string_view(v, len), conv, sink)};
}
// ==================== Raw pointers ====================
-ConvertResult<Conv::p> FormatConvertImpl(VoidPtr v, const ConversionSpec conv,
- FormatSinkImpl *sink) {
- if (conv.conv().id() != ConversionChar::p)
- return {false};
+ArgConvertResult<FormatConversionCharSetInternal::p> FormatConvertImpl(
+ VoidPtr v, const FormatConversionSpecImpl conv, FormatSinkImpl *sink) {
if (!v.value) {
sink->Append("(nil)");
return {true};
}
- return {ConvertIntImplInner(v.value, conv, sink)};
+ IntDigits as_digits;
+ as_digits.PrintAsHexLower(v.value);
+ return {ConvertIntImplInnerSlow(as_digits, conv, sink)};
}
// ==================== Floats ====================
-FloatingConvertResult FormatConvertImpl(float v, const ConversionSpec conv,
+FloatingConvertResult FormatConvertImpl(float v,
+ const FormatConversionSpecImpl conv,
FormatSinkImpl *sink) {
return {ConvertFloatArg(v, conv, sink)};
}
-FloatingConvertResult FormatConvertImpl(double v, const ConversionSpec conv,
+FloatingConvertResult FormatConvertImpl(double v,
+ const FormatConversionSpecImpl conv,
FormatSinkImpl *sink) {
return {ConvertFloatArg(v, conv, sink)};
}
FloatingConvertResult FormatConvertImpl(long double v,
- const ConversionSpec conv,
+ const FormatConversionSpecImpl conv,
FormatSinkImpl *sink) {
return {ConvertFloatArg(v, conv, sink)};
}
// ==================== Chars ====================
-IntegralConvertResult FormatConvertImpl(char v, const ConversionSpec conv,
+IntegralConvertResult FormatConvertImpl(char v,
+ const FormatConversionSpecImpl conv,
FormatSinkImpl *sink) {
return {ConvertIntArg(v, conv, sink)};
}
IntegralConvertResult FormatConvertImpl(signed char v,
- const ConversionSpec conv,
+ const FormatConversionSpecImpl conv,
FormatSinkImpl *sink) {
return {ConvertIntArg(v, conv, sink)};
}
IntegralConvertResult FormatConvertImpl(unsigned char v,
- const ConversionSpec conv,
+ const FormatConversionSpecImpl conv,
FormatSinkImpl *sink) {
return {ConvertIntArg(v, conv, sink)};
}
// ==================== Ints ====================
IntegralConvertResult FormatConvertImpl(short v, // NOLINT
- const ConversionSpec conv,
+ const FormatConversionSpecImpl conv,
FormatSinkImpl *sink) {
return {ConvertIntArg(v, conv, sink)};
}
IntegralConvertResult FormatConvertImpl(unsigned short v, // NOLINT
- const ConversionSpec conv,
+ const FormatConversionSpecImpl conv,
FormatSinkImpl *sink) {
return {ConvertIntArg(v, conv, sink)};
}
-IntegralConvertResult FormatConvertImpl(int v, const ConversionSpec conv,
+IntegralConvertResult FormatConvertImpl(int v,
+ const FormatConversionSpecImpl conv,
FormatSinkImpl *sink) {
return {ConvertIntArg(v, conv, sink)};
}
-IntegralConvertResult FormatConvertImpl(unsigned v, const ConversionSpec conv,
+IntegralConvertResult FormatConvertImpl(unsigned v,
+ const FormatConversionSpecImpl conv,
FormatSinkImpl *sink) {
return {ConvertIntArg(v, conv, sink)};
}
IntegralConvertResult FormatConvertImpl(long v, // NOLINT
- const ConversionSpec conv,
+ const FormatConversionSpecImpl conv,
FormatSinkImpl *sink) {
return {ConvertIntArg(v, conv, sink)};
}
IntegralConvertResult FormatConvertImpl(unsigned long v, // NOLINT
- const ConversionSpec conv,
+ const FormatConversionSpecImpl conv,
FormatSinkImpl *sink) {
return {ConvertIntArg(v, conv, sink)};
}
IntegralConvertResult FormatConvertImpl(long long v, // NOLINT
- const ConversionSpec conv,
+ const FormatConversionSpecImpl conv,
FormatSinkImpl *sink) {
return {ConvertIntArg(v, conv, sink)};
}
IntegralConvertResult FormatConvertImpl(unsigned long long v, // NOLINT
- const ConversionSpec conv,
+ const FormatConversionSpecImpl conv,
+ FormatSinkImpl *sink) {
+ return {ConvertIntArg(v, conv, sink)};
+}
+IntegralConvertResult FormatConvertImpl(absl::int128 v,
+ const FormatConversionSpecImpl conv,
FormatSinkImpl *sink) {
return {ConvertIntArg(v, conv, sink)};
}
IntegralConvertResult FormatConvertImpl(absl::uint128 v,
- const ConversionSpec conv,
+ const FormatConversionSpecImpl conv,
FormatSinkImpl *sink) {
return {ConvertIntArg(v, conv, sink)};
}
@@ -372,6 +481,8 @@
ABSL_INTERNAL_FORMAT_DISPATCH_OVERLOADS_EXPAND_();
+
} // namespace str_format_internal
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/internal/str_format/arg.h b/absl/strings/internal/str_format/arg.h
index 5cb3a14..7040c86 100644
--- a/absl/strings/internal/str_format/arg.h
+++ b/absl/strings/internal/str_format/arg.h
@@ -1,3 +1,17 @@
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
#ifndef ABSL_STRINGS_INTERNAL_STR_FORMAT_ARG_H_
#define ABSL_STRINGS_INTERNAL_STR_FORMAT_ARG_H_
@@ -18,24 +32,44 @@
#include "absl/strings/internal/str_format/extension.h"
#include "absl/strings/string_view.h"
-class Cord;
-class CordReader;
-
namespace absl {
+ABSL_NAMESPACE_BEGIN
+class Cord;
class FormatCountCapture;
class FormatSink;
+template <absl::FormatConversionCharSet C>
+struct FormatConvertResult;
+class FormatConversionSpec;
+
namespace str_format_internal {
template <typename T, typename = void>
struct HasUserDefinedConvert : std::false_type {};
template <typename T>
-struct HasUserDefinedConvert<
- T, void_t<decltype(AbslFormatConvert(
- std::declval<const T&>(), std::declval<ConversionSpec>(),
- std::declval<FormatSink*>()))>> : std::true_type {};
+struct HasUserDefinedConvert<T, void_t<decltype(AbslFormatConvert(
+ std::declval<const T&>(),
+ std::declval<const FormatConversionSpec&>(),
+ std::declval<FormatSink*>()))>>
+ : std::true_type {};
+
+void AbslFormatConvert(); // Stops the lexical name lookup
+template <typename T>
+auto FormatConvertImpl(const T& v, FormatConversionSpecImpl conv,
+ FormatSinkImpl* sink)
+ -> decltype(AbslFormatConvert(v,
+ std::declval<const FormatConversionSpec&>(),
+ std::declval<FormatSink*>())) {
+ using FormatConversionSpecT =
+ absl::enable_if_t<sizeof(const T& (*)()) != 0, FormatConversionSpec>;
+ using FormatSinkT =
+ absl::enable_if_t<sizeof(const T& (*)()) != 0, FormatSink>;
+ auto fcs = conv.Wrap<FormatConversionSpecT>();
+ auto fs = sink->Wrap<FormatSinkT>();
+ return AbslFormatConvert(v, fcs, &fs);
+}
template <typename T>
class StreamedWrapper;
@@ -44,6 +78,13 @@
// then convert it, appending to `sink` and return `true`.
// Otherwise fail and return `false`.
+// AbslFormatConvert(v, conv, sink) is intended to be found by ADL on 'v'
+// as an extension mechanism. These FormatConvertImpl functions are the default
+// implementations.
+// The ADL search is augmented via the 'Sink*' parameter, which also
+// serves as a disambiguator to reject possible unintended 'AbslFormatConvert'
+// functions in the namespaces associated with 'v'.
+
// Raw pointers.
struct VoidPtr {
VoidPtr() = default;
@@ -53,28 +94,45 @@
: value(ptr ? reinterpret_cast<uintptr_t>(ptr) : 0) {}
uintptr_t value;
};
-ConvertResult<Conv::p> FormatConvertImpl(VoidPtr v, ConversionSpec conv,
- FormatSinkImpl* sink);
+
+template <FormatConversionCharSet C>
+struct ArgConvertResult {
+ bool value;
+};
+
+template <FormatConversionCharSet C>
+constexpr FormatConversionCharSet ExtractCharSet(FormatConvertResult<C>) {
+ return C;
+}
+
+template <FormatConversionCharSet C>
+constexpr FormatConversionCharSet ExtractCharSet(ArgConvertResult<C>) {
+ return C;
+}
+
+using StringConvertResult =
+ ArgConvertResult<FormatConversionCharSetInternal::s>;
+ArgConvertResult<FormatConversionCharSetInternal::p> FormatConvertImpl(
+ VoidPtr v, FormatConversionSpecImpl conv, FormatSinkImpl* sink);
// Strings.
-ConvertResult<Conv::s> FormatConvertImpl(const std::string& v,
- ConversionSpec conv,
- FormatSinkImpl* sink);
-ConvertResult<Conv::s> FormatConvertImpl(string_view v, ConversionSpec conv,
- FormatSinkImpl* sink);
-ConvertResult<Conv::s | Conv::p> FormatConvertImpl(const char* v,
- ConversionSpec conv,
- FormatSinkImpl* sink);
-template <class AbslCord,
- typename std::enable_if<
- std::is_same<AbslCord, ::Cord>::value>::type* = nullptr,
- class AbslCordReader = ::CordReader>
-ConvertResult<Conv::s> FormatConvertImpl(const AbslCord& value,
- ConversionSpec conv,
- FormatSinkImpl* sink) {
- if (conv.conv().id() != ConversionChar::s) return {false};
+StringConvertResult FormatConvertImpl(const std::string& v,
+ FormatConversionSpecImpl conv,
+ FormatSinkImpl* sink);
+StringConvertResult FormatConvertImpl(string_view v,
+ FormatConversionSpecImpl conv,
+ FormatSinkImpl* sink);
+ArgConvertResult<FormatConversionCharSetUnion(
+ FormatConversionCharSetInternal::s, FormatConversionCharSetInternal::p)>
+FormatConvertImpl(const char* v, const FormatConversionSpecImpl conv,
+ FormatSinkImpl* sink);
- bool is_left = conv.flags().left;
+template <class AbslCord, typename std::enable_if<std::is_same<
+ AbslCord, absl::Cord>::value>::type* = nullptr>
+StringConvertResult FormatConvertImpl(const AbslCord& value,
+ FormatConversionSpecImpl conv,
+ FormatSinkImpl* sink) {
+ bool is_left = conv.has_left_flag();
size_t space_remaining = 0;
int width = conv.width();
@@ -90,64 +148,80 @@
if (space_remaining > 0 && !is_left) sink->Append(space_remaining, ' ');
- string_view piece;
- for (AbslCordReader reader(value);
- to_write > 0 && reader.ReadFragment(&piece); to_write -= piece.size()) {
- if (piece.size() > to_write) piece.remove_suffix(piece.size() - to_write);
+ for (string_view piece : value.Chunks()) {
+ if (piece.size() > to_write) {
+ piece.remove_suffix(piece.size() - to_write);
+ to_write = 0;
+ } else {
+ to_write -= piece.size();
+ }
sink->Append(piece);
+ if (to_write == 0) {
+ break;
+ }
}
if (space_remaining > 0 && is_left) sink->Append(space_remaining, ' ');
return {true};
}
-using IntegralConvertResult =
- ConvertResult<Conv::c | Conv::numeric | Conv::star>;
-using FloatingConvertResult = ConvertResult<Conv::floating>;
+using IntegralConvertResult = ArgConvertResult<FormatConversionCharSetUnion(
+ FormatConversionCharSetInternal::c,
+ FormatConversionCharSetInternal::kNumeric,
+ FormatConversionCharSetInternal::kStar)>;
+using FloatingConvertResult =
+ ArgConvertResult<FormatConversionCharSetInternal::kFloating>;
// Floats.
-FloatingConvertResult FormatConvertImpl(float v, ConversionSpec conv,
+FloatingConvertResult FormatConvertImpl(float v, FormatConversionSpecImpl conv,
FormatSinkImpl* sink);
-FloatingConvertResult FormatConvertImpl(double v, ConversionSpec conv,
+FloatingConvertResult FormatConvertImpl(double v, FormatConversionSpecImpl conv,
FormatSinkImpl* sink);
-FloatingConvertResult FormatConvertImpl(long double v, ConversionSpec conv,
+FloatingConvertResult FormatConvertImpl(long double v,
+ FormatConversionSpecImpl conv,
FormatSinkImpl* sink);
// Chars.
-IntegralConvertResult FormatConvertImpl(char v, ConversionSpec conv,
+IntegralConvertResult FormatConvertImpl(char v, FormatConversionSpecImpl conv,
FormatSinkImpl* sink);
-IntegralConvertResult FormatConvertImpl(signed char v, ConversionSpec conv,
+IntegralConvertResult FormatConvertImpl(signed char v,
+ FormatConversionSpecImpl conv,
FormatSinkImpl* sink);
-IntegralConvertResult FormatConvertImpl(unsigned char v, ConversionSpec conv,
+IntegralConvertResult FormatConvertImpl(unsigned char v,
+ FormatConversionSpecImpl conv,
FormatSinkImpl* sink);
// Ints.
IntegralConvertResult FormatConvertImpl(short v, // NOLINT
- ConversionSpec conv,
+ FormatConversionSpecImpl conv,
FormatSinkImpl* sink);
IntegralConvertResult FormatConvertImpl(unsigned short v, // NOLINT
- ConversionSpec conv,
+ FormatConversionSpecImpl conv,
FormatSinkImpl* sink);
-IntegralConvertResult FormatConvertImpl(int v, ConversionSpec conv,
+IntegralConvertResult FormatConvertImpl(int v, FormatConversionSpecImpl conv,
FormatSinkImpl* sink);
-IntegralConvertResult FormatConvertImpl(unsigned v, ConversionSpec conv,
+IntegralConvertResult FormatConvertImpl(unsigned v,
+ FormatConversionSpecImpl conv,
FormatSinkImpl* sink);
IntegralConvertResult FormatConvertImpl(long v, // NOLINT
- ConversionSpec conv,
+ FormatConversionSpecImpl conv,
FormatSinkImpl* sink);
IntegralConvertResult FormatConvertImpl(unsigned long v, // NOLINT
- ConversionSpec conv,
+ FormatConversionSpecImpl conv,
FormatSinkImpl* sink);
IntegralConvertResult FormatConvertImpl(long long v, // NOLINT
- ConversionSpec conv,
+ FormatConversionSpecImpl conv,
FormatSinkImpl* sink);
IntegralConvertResult FormatConvertImpl(unsigned long long v, // NOLINT
- ConversionSpec conv,
+ FormatConversionSpecImpl conv,
FormatSinkImpl* sink);
-IntegralConvertResult FormatConvertImpl(uint128 v, ConversionSpec conv,
+IntegralConvertResult FormatConvertImpl(int128 v, FormatConversionSpecImpl conv,
+ FormatSinkImpl* sink);
+IntegralConvertResult FormatConvertImpl(uint128 v,
+ FormatConversionSpecImpl conv,
FormatSinkImpl* sink);
template <typename T, enable_if_t<std::is_same<T, bool>::value, int> = 0>
-IntegralConvertResult FormatConvertImpl(T v, ConversionSpec conv,
+IntegralConvertResult FormatConvertImpl(T v, FormatConversionSpecImpl conv,
FormatSinkImpl* sink) {
return FormatConvertImpl(static_cast<int>(v), conv, sink);
}
@@ -158,12 +232,12 @@
typename std::enable_if<std::is_enum<T>::value &&
!HasUserDefinedConvert<T>::value,
IntegralConvertResult>::type
-FormatConvertImpl(T v, ConversionSpec conv, FormatSinkImpl* sink);
+FormatConvertImpl(T v, FormatConversionSpecImpl conv, FormatSinkImpl* sink);
template <typename T>
-ConvertResult<Conv::s> FormatConvertImpl(const StreamedWrapper<T>& v,
- ConversionSpec conv,
- FormatSinkImpl* out) {
+StringConvertResult FormatConvertImpl(const StreamedWrapper<T>& v,
+ FormatConversionSpecImpl conv,
+ FormatSinkImpl* out) {
std::ostringstream oss;
oss << v.v_;
if (!oss) return {false};
@@ -174,22 +248,24 @@
// until after FormatCountCapture is fully defined.
struct FormatCountCaptureHelper {
template <class T = int>
- static ConvertResult<Conv::n> ConvertHelper(const FormatCountCapture& v,
- ConversionSpec conv,
- FormatSinkImpl* sink) {
+ static ArgConvertResult<FormatConversionCharSetInternal::n> ConvertHelper(
+ const FormatCountCapture& v, FormatConversionSpecImpl conv,
+ FormatSinkImpl* sink) {
const absl::enable_if_t<sizeof(T) != 0, FormatCountCapture>& v2 = v;
- if (conv.conv().id() != str_format_internal::ConversionChar::n)
+ if (conv.conversion_char() !=
+ str_format_internal::FormatConversionCharInternal::n) {
return {false};
+ }
*v2.p_ = static_cast<int>(sink->size());
return {true};
}
};
template <class T = int>
-ConvertResult<Conv::n> FormatConvertImpl(const FormatCountCapture& v,
- ConversionSpec conv,
- FormatSinkImpl* sink) {
+ArgConvertResult<FormatConversionCharSetInternal::n> FormatConvertImpl(
+ const FormatCountCapture& v, FormatConversionSpecImpl conv,
+ FormatSinkImpl* sink) {
return FormatCountCaptureHelper::ConvertHelper(v, conv, sink);
}
@@ -198,13 +274,13 @@
struct FormatArgImplFriend {
template <typename Arg>
static bool ToInt(Arg arg, int* out) {
- // A value initialized ConversionSpec has a `none` conv, which tells the
- // dispatcher to run the `int` conversion.
+ // A value initialized FormatConversionSpecImpl has a `none` conv, which
+ // tells the dispatcher to run the `int` conversion.
return arg.dispatcher_(arg.data_, {}, out);
}
template <typename Arg>
- static bool Convert(Arg arg, str_format_internal::ConversionSpec conv,
+ static bool Convert(Arg arg, FormatConversionSpecImpl conv,
FormatSinkImpl* out) {
return arg.dispatcher_(arg.data_, conv, out);
}
@@ -215,6 +291,15 @@
}
};
+template <typename Arg>
+constexpr FormatConversionCharSet ArgumentToConv() {
+ return absl::str_format_internal::ExtractCharSet(
+ decltype(str_format_internal::FormatConvertImpl(
+ std::declval<const Arg&>(),
+ std::declval<const FormatConversionSpecImpl&>(),
+ std::declval<FormatSinkImpl*>())){});
+}
+
// A type-erased handle to a format argument.
class FormatArgImpl {
private:
@@ -228,7 +313,7 @@
char buf[kInlinedSpace];
};
- using Dispatcher = bool (*)(Data, ConversionSpec, void* out);
+ using Dispatcher = bool (*)(Data, FormatConversionSpecImpl, void* out);
template <typename T>
struct store_by_value
@@ -370,15 +455,20 @@
}
template <typename T>
- static bool Dispatch(Data arg, ConversionSpec spec, void* out) {
+ static bool Dispatch(Data arg, FormatConversionSpecImpl spec, void* out) {
// A `none` conv indicates that we want the `int` conversion.
- if (ABSL_PREDICT_FALSE(spec.conv().id() == ConversionChar::none)) {
+ if (ABSL_PREDICT_FALSE(spec.conversion_char() ==
+ FormatConversionCharInternal::kNone)) {
return ToInt<T>(arg, static_cast<int*>(out), std::is_integral<T>(),
std::is_enum<T>());
}
-
+ if (ABSL_PREDICT_FALSE(!Contains(ArgumentToConv<T>(),
+ spec.conversion_char()))) {
+ return false;
+ }
return str_format_internal::FormatConvertImpl(
- Manager<T>::Value(arg), spec, static_cast<FormatSinkImpl*>(out))
+ Manager<T>::Value(arg), spec,
+ static_cast<FormatSinkImpl*>(out))
.value;
}
@@ -386,8 +476,9 @@
Dispatcher dispatcher_;
};
-#define ABSL_INTERNAL_FORMAT_DISPATCH_INSTANTIATE_(T, E) \
- E template bool FormatArgImpl::Dispatch<T>(Data, ConversionSpec, void*)
+#define ABSL_INTERNAL_FORMAT_DISPATCH_INSTANTIATE_(T, E) \
+ E template bool FormatArgImpl::Dispatch<T>(Data, FormatConversionSpecImpl, \
+ void*)
#define ABSL_INTERNAL_FORMAT_DISPATCH_OVERLOADS_EXPAND_(...) \
ABSL_INTERNAL_FORMAT_DISPATCH_INSTANTIATE_(str_format_internal::VoidPtr, \
@@ -408,6 +499,7 @@
__VA_ARGS__); \
ABSL_INTERNAL_FORMAT_DISPATCH_INSTANTIATE_(unsigned long long, /* NOLINT */ \
__VA_ARGS__); \
+ ABSL_INTERNAL_FORMAT_DISPATCH_INSTANTIATE_(int128, __VA_ARGS__); \
ABSL_INTERNAL_FORMAT_DISPATCH_INSTANTIATE_(uint128, __VA_ARGS__); \
ABSL_INTERNAL_FORMAT_DISPATCH_INSTANTIATE_(float, __VA_ARGS__); \
ABSL_INTERNAL_FORMAT_DISPATCH_INSTANTIATE_(double, __VA_ARGS__); \
@@ -418,7 +510,9 @@
ABSL_INTERNAL_FORMAT_DISPATCH_OVERLOADS_EXPAND_(extern);
+
} // namespace str_format_internal
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_INTERNAL_STR_FORMAT_ARG_H_
diff --git a/absl/strings/internal/str_format/arg_test.cc b/absl/strings/internal/str_format/arg_test.cc
index 3421fac..1261937 100644
--- a/absl/strings/internal/str_format/arg_test.cc
+++ b/absl/strings/internal/str_format/arg_test.cc
@@ -6,6 +6,12 @@
//
// https://www.apache.org/licenses/LICENSE-2.0
//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
#include "absl/strings/internal/str_format/arg.h"
#include <ostream>
@@ -14,6 +20,7 @@
#include "absl/strings/str_format.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace str_format_internal {
namespace {
@@ -22,8 +29,17 @@
enum Color { kRed, kGreen, kBlue };
static const char *hi() { return "hi"; }
+
+ struct X {};
+
+ X x_;
};
+inline FormatConvertResult<FormatConversionCharSet{}> AbslFormatConvert(
+ const FormatArgImplTest::X &, const FormatConversionSpec &, FormatSink *) {
+ return {false};
+}
+
TEST_F(FormatArgImplTest, ToInt) {
int out = 0;
EXPECT_TRUE(FormatArgImplFriend::ToInt(FormatArgImpl(1), &out));
@@ -58,6 +74,7 @@
FormatArgImpl(static_cast<int *>(nullptr)), &out));
EXPECT_FALSE(FormatArgImplFriend::ToInt(FormatArgImpl(hi()), &out));
EXPECT_FALSE(FormatArgImplFriend::ToInt(FormatArgImpl("hi"), &out));
+ EXPECT_FALSE(FormatArgImplFriend::ToInt(FormatArgImpl(x_), &out));
EXPECT_TRUE(FormatArgImplFriend::ToInt(FormatArgImpl(kBlue), &out));
EXPECT_EQ(2, out);
}
@@ -94,11 +111,12 @@
TEST_F(FormatArgImplTest, WorksWithCharArraysOfUnknownSize) {
std::string s;
FormatSinkImpl sink(&s);
- ConversionSpec conv;
- conv.set_conv(ConversionChar::FromChar('s'));
- conv.set_flags(Flags());
- conv.set_width(-1);
- conv.set_precision(-1);
+ FormatConversionSpecImpl conv;
+ FormatConversionSpecImplFriend::SetConversionChar(
+ FormatConversionCharInternal::s, &conv);
+ FormatConversionSpecImplFriend::SetFlags(Flags(), &conv);
+ FormatConversionSpecImplFriend::SetWidth(-1, &conv);
+ FormatConversionSpecImplFriend::SetPrecision(-1, &conv);
EXPECT_TRUE(
FormatArgImplFriend::Convert(FormatArgImpl(kMyArray), conv, &sink));
sink.Flush();
@@ -108,4 +126,5 @@
} // namespace
} // namespace str_format_internal
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/internal/str_format/bind.cc b/absl/strings/internal/str_format/bind.cc
index a31859e..4e68b90 100644
--- a/absl/strings/internal/str_format/bind.cc
+++ b/absl/strings/internal/str_format/bind.cc
@@ -1,3 +1,17 @@
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
#include "absl/strings/internal/str_format/bind.h"
#include <cerrno>
@@ -6,6 +20,7 @@
#include <string>
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace str_format_internal {
namespace {
@@ -65,19 +80,22 @@
return false;
}
- bound->set_width(width);
- bound->set_precision(precision);
- bound->set_flags(unbound->flags);
- if (force_left)
- bound->set_left(true);
- } else {
- bound->set_flags(unbound->flags);
- bound->set_width(-1);
- bound->set_precision(-1);
- }
+ FormatConversionSpecImplFriend::SetWidth(width, bound);
+ FormatConversionSpecImplFriend::SetPrecision(precision, bound);
- bound->set_length_mod(unbound->length_mod);
- bound->set_conv(unbound->conv);
+ if (force_left) {
+ Flags flags = unbound->flags;
+ flags.left = true;
+ FormatConversionSpecImplFriend::SetFlags(flags, bound);
+ } else {
+ FormatConversionSpecImplFriend::SetFlags(unbound->flags, bound);
+ }
+ } else {
+ FormatConversionSpecImplFriend::SetFlags(unbound->flags, bound);
+ FormatConversionSpecImplFriend::SetWidth(-1, bound);
+ FormatConversionSpecImplFriend::SetPrecision(-1, bound);
+ }
+ FormatConversionSpecImplFriend::SetConversionChar(unbound->conv, bound);
bound->set_arg(arg);
return true;
}
@@ -139,10 +157,11 @@
UntypedFormatSpecImpl spec("%d");
std::ostringstream ss;
- ss << "{" << Streamable(spec, {*bound.arg()}) << ":" << bound.flags();
+ ss << "{" << Streamable(spec, {*bound.arg()}) << ":"
+ << FormatConversionSpecImplFriend::FlagsToString(bound);
if (bound.width() >= 0) ss << bound.width();
if (bound.precision() >= 0) ss << "." << bound.precision();
- ss << bound.length_mod() << bound.conv() << "}";
+ ss << bound.conversion_char() << "}";
Append(ss.str());
return true;
}
@@ -196,6 +215,15 @@
return *out;
}
+std::string FormatPack(const UntypedFormatSpecImpl format,
+ absl::Span<const FormatArgImpl> args) {
+ std::string out;
+ if (ABSL_PREDICT_FALSE(!FormatUntyped(&out, format, args))) {
+ out.clear();
+ }
+ return out;
+}
+
int FprintF(std::FILE* output, const UntypedFormatSpecImpl format,
absl::Span<const FormatArgImpl> args) {
FILERawSink sink(output);
@@ -207,7 +235,7 @@
errno = sink.error();
return -1;
}
- if (sink.count() > std::numeric_limits<int>::max()) {
+ if (sink.count() > static_cast<size_t>(std::numeric_limits<int>::max())) {
errno = EFBIG;
return -1;
}
@@ -227,4 +255,5 @@
}
} // namespace str_format_internal
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/internal/str_format/bind.h b/absl/strings/internal/str_format/bind.h
index 7df140a..267cc0e 100644
--- a/absl/strings/internal/str_format/bind.h
+++ b/absl/strings/internal/str_format/bind.h
@@ -1,3 +1,17 @@
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
#ifndef ABSL_STRINGS_INTERNAL_STR_FORMAT_BIND_H_
#define ABSL_STRINGS_INTERNAL_STR_FORMAT_BIND_H_
@@ -13,12 +27,13 @@
#include "absl/types/span.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
class UntypedFormatSpec;
namespace str_format_internal {
-class BoundConversion : public ConversionSpec {
+class BoundConversion : public FormatConversionSpecImpl {
public:
const FormatArgImpl* arg() const { return arg_; }
void set_arg(const FormatArgImpl* a) { arg_ = a; }
@@ -59,7 +74,7 @@
size_t size_;
};
-template <typename T, typename...>
+template <typename T, FormatConversionCharSet...>
struct MakeDependent {
using type = T;
};
@@ -67,19 +82,19 @@
// Implicitly convertible from `const char*`, `string_view`, and the
// `ExtendedParsedFormat` type. This abstraction allows all format functions to
// operate on any without providing too many overloads.
-template <typename... Args>
+template <FormatConversionCharSet... Args>
class FormatSpecTemplate
: public MakeDependent<UntypedFormatSpec, Args...>::type {
using Base = typename MakeDependent<UntypedFormatSpec, Args...>::type;
public:
-#if ABSL_INTERNAL_ENABLE_FORMAT_CHECKER
+#ifdef ABSL_INTERNAL_ENABLE_FORMAT_CHECKER
- // Honeypot overload for when the std::string is not constexpr.
+ // Honeypot overload for when the string is not constexpr.
// We use the 'unavailable' attribute to give a better compiler error than
// just 'method is deleted'.
FormatSpecTemplate(...) // NOLINT
- __attribute__((unavailable("Format std::string is not constexpr.")));
+ __attribute__((unavailable("Format string is not constexpr.")));
// Honeypot overload for when the format is constexpr and invalid.
// We use the 'unavailable' attribute to give a better compiler error than
@@ -104,13 +119,11 @@
// Good format overload.
FormatSpecTemplate(const char* s) // NOLINT
- __attribute__((enable_if(ValidFormatImpl<ArgumentToConv<Args>()...>(s),
- "bad format trap")))
+ __attribute__((enable_if(ValidFormatImpl<Args...>(s), "bad format trap")))
: Base(s) {}
FormatSpecTemplate(string_view s) // NOLINT
- __attribute__((enable_if(ValidFormatImpl<ArgumentToConv<Args>()...>(s),
- "bad format trap")))
+ __attribute__((enable_if(ValidFormatImpl<Args...>(s), "bad format trap")))
: Base(s) {}
#else // ABSL_INTERNAL_ENABLE_FORMAT_CHECKER
@@ -120,19 +133,15 @@
#endif // ABSL_INTERNAL_ENABLE_FORMAT_CHECKER
- template <Conv... C, typename = typename std::enable_if<
- sizeof...(C) == sizeof...(Args) &&
- AllOf(Contains(ArgumentToConv<Args>(),
- C)...)>::type>
+ template <
+ FormatConversionCharSet... C,
+ typename = typename std::enable_if<sizeof...(C) == sizeof...(Args)>::type,
+ typename = typename std::enable_if<AllOf(Contains(Args,
+ C)...)>::type>
FormatSpecTemplate(const ExtendedParsedFormat<C...>& pc) // NOLINT
: Base(&pc) {}
};
-template <typename... Args>
-struct FormatSpecDeductionBarrier {
- using type = FormatSpecTemplate<Args...>;
-};
-
class Streamable {
public:
Streamable(const UntypedFormatSpecImpl& format,
@@ -178,12 +187,8 @@
std::string& AppendPack(std::string* out, UntypedFormatSpecImpl format,
absl::Span<const FormatArgImpl> args);
-inline std::string FormatPack(const UntypedFormatSpecImpl format,
- absl::Span<const FormatArgImpl> args) {
- std::string out;
- AppendPack(&out, format, args);
- return out;
-}
+std::string FormatPack(const UntypedFormatSpecImpl format,
+ absl::Span<const FormatArgImpl> args);
int FprintF(std::FILE* output, UntypedFormatSpecImpl format,
absl::Span<const FormatArgImpl> args);
@@ -199,13 +204,14 @@
private:
template <typename S>
- friend ConvertResult<Conv::s> FormatConvertImpl(const StreamedWrapper<S>& v,
- ConversionSpec conv,
- FormatSinkImpl* out);
+ friend ArgConvertResult<FormatConversionCharSetInternal::s> FormatConvertImpl(
+ const StreamedWrapper<S>& v, FormatConversionSpecImpl conv,
+ FormatSinkImpl* out);
const T& v_;
};
} // namespace str_format_internal
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_INTERNAL_STR_FORMAT_BIND_H_
diff --git a/absl/strings/internal/str_format/bind_test.cc b/absl/strings/internal/str_format/bind_test.cc
index 2574801..1eef9c4 100644
--- a/absl/strings/internal/str_format/bind_test.cc
+++ b/absl/strings/internal/str_format/bind_test.cc
@@ -1,3 +1,17 @@
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
#include "absl/strings/internal/str_format/bind.h"
#include <string.h>
@@ -6,6 +20,7 @@
#include "gtest/gtest.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace str_format_internal {
namespace {
@@ -112,18 +127,18 @@
FormatArgImpl(ia[2]), FormatArgImpl(ia[3]),
FormatArgImpl(ia[4])};
const Expectation kExpect[] = {
- {__LINE__, "a%4db%dc", "a{10:4d}b{20:d}c"},
- {__LINE__, "a%.4db%dc", "a{10:.4d}b{20:d}c"},
- {__LINE__, "a%4.5db%dc", "a{10:4.5d}b{20:d}c"},
- {__LINE__, "a%db%4.5dc", "a{10:d}b{20:4.5d}c"},
- {__LINE__, "a%db%*.*dc", "a{10:d}b{40:20.30d}c"},
- {__LINE__, "a%.*fb", "a{20:.10f}b"},
- {__LINE__, "a%1$db%2$*3$.*4$dc", "a{10:d}b{20:30.40d}c"},
- {__LINE__, "a%4$db%3$*2$.*1$dc", "a{40:d}b{30:20.10d}c"},
- {__LINE__, "a%04ldb", "a{10:04ld}b"},
- {__LINE__, "a%-#04lldb", "a{10:-#04lld}b"},
- {__LINE__, "a%1$*5$db", "a{10:-10d}b"},
- {__LINE__, "a%1$.*5$db", "a{10:d}b"},
+ {__LINE__, "a%4db%dc", "a{10:4d}b{20:d}c"},
+ {__LINE__, "a%.4db%dc", "a{10:.4d}b{20:d}c"},
+ {__LINE__, "a%4.5db%dc", "a{10:4.5d}b{20:d}c"},
+ {__LINE__, "a%db%4.5dc", "a{10:d}b{20:4.5d}c"},
+ {__LINE__, "a%db%*.*dc", "a{10:d}b{40:20.30d}c"},
+ {__LINE__, "a%.*fb", "a{20:.10f}b"},
+ {__LINE__, "a%1$db%2$*3$.*4$dc", "a{10:d}b{20:30.40d}c"},
+ {__LINE__, "a%4$db%3$*2$.*1$dc", "a{40:d}b{30:20.10d}c"},
+ {__LINE__, "a%04ldb", "a{10:04d}b"},
+ {__LINE__, "a%-#04lldb", "a{10:-#04d}b"},
+ {__LINE__, "a%1$*5$db", "a{10:-10d}b"},
+ {__LINE__, "a%1$.*5$db", "a{10:d}b"},
};
for (const Expectation &e : kExpect) {
absl::string_view fmt = e.fmt;
@@ -138,4 +153,5 @@
} // namespace
} // namespace str_format_internal
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/internal/str_format/checker.h b/absl/strings/internal/str_format/checker.h
index 8b594f2..2a2601e 100644
--- a/absl/strings/internal/str_format/checker.h
+++ b/absl/strings/internal/str_format/checker.h
@@ -1,20 +1,34 @@
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
#ifndef ABSL_STRINGS_INTERNAL_STR_FORMAT_CHECKER_H_
#define ABSL_STRINGS_INTERNAL_STR_FORMAT_CHECKER_H_
+#include "absl/base/attributes.h"
#include "absl/strings/internal/str_format/arg.h"
#include "absl/strings/internal/str_format/extension.h"
// Compile time check support for entry points.
#ifndef ABSL_INTERNAL_ENABLE_FORMAT_CHECKER
-#if defined(__clang__) && !defined(__native_client__)
-#if __has_attribute(enable_if)
+#if ABSL_HAVE_ATTRIBUTE(enable_if) && !defined(__native_client__)
#define ABSL_INTERNAL_ENABLE_FORMAT_CHECKER 1
-#endif // __has_attribute(enable_if)
-#endif // defined(__clang__) && !defined(__native_client__)
+#endif // ABSL_HAVE_ATTRIBUTE(enable_if) && !defined(__native_client__)
#endif // ABSL_INTERNAL_ENABLE_FORMAT_CHECKER
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace str_format_internal {
constexpr bool AllOf() { return true; }
@@ -24,14 +38,7 @@
return b && AllOf(t...);
}
-template <typename Arg>
-constexpr Conv ArgumentToConv() {
- return decltype(str_format_internal::FormatConvertImpl(
- std::declval<const Arg&>(), std::declval<const ConversionSpec&>(),
- std::declval<FormatSinkImpl*>()))::kConv;
-}
-
-#if ABSL_INTERNAL_ENABLE_FORMAT_CHECKER
+#ifdef ABSL_INTERNAL_ENABLE_FORMAT_CHECKER
constexpr bool ContainsChar(const char* chars, char c) {
return *chars == c || (*chars && ContainsChar(chars + 1, c));
@@ -39,14 +46,14 @@
// A constexpr compatible list of Convs.
struct ConvList {
- const Conv* array;
+ const FormatConversionCharSet* array;
int count;
// We do the bound check here to avoid having to do it on the callers.
- // Returning an empty Conv has the same effect as short circuiting because it
- // will never match any conversion.
- constexpr Conv operator[](int i) const {
- return i < count ? array[i] : Conv{};
+ // Returning an empty FormatConversionCharSet has the same effect as
+ // short circuiting because it will never match any conversion.
+ constexpr FormatConversionCharSet operator[](int i) const {
+ return i < count ? array[i] : FormatConversionCharSet{};
}
constexpr ConvList without_front() const {
@@ -57,7 +64,7 @@
template <size_t count>
struct ConvListT {
// Make sure the array has size > 0.
- Conv list[count ? count : 1];
+ FormatConversionCharSet list[count ? count : 1];
};
constexpr char GetChar(string_view str, size_t index) {
@@ -310,7 +317,7 @@
ConvList args_;
};
-template <Conv... C>
+template <FormatConversionCharSet... C>
constexpr bool ValidFormatImpl(string_view format) {
return FormatParser(format,
{ConvListT<sizeof...(C)>{{C...}}.list, sizeof...(C)})
@@ -320,6 +327,7 @@
#endif // ABSL_INTERNAL_ENABLE_FORMAT_CHECKER
} // namespace str_format_internal
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_INTERNAL_STR_FORMAT_CHECKER_H_
diff --git a/absl/strings/internal/str_format/checker_test.cc b/absl/strings/internal/str_format/checker_test.cc
index c1d8c76..7c70f47 100644
--- a/absl/strings/internal/str_format/checker_test.cc
+++ b/absl/strings/internal/str_format/checker_test.cc
@@ -1,3 +1,17 @@
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
#include <string>
#include "gmock/gmock.h"
@@ -5,21 +19,26 @@
#include "absl/strings/str_format.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace str_format_internal {
namespace {
-std::string ConvToString(Conv conv) {
+std::string ConvToString(FormatConversionCharSet conv) {
std::string out;
-#define CONV_SET_CASE(c) \
- if (Contains(conv, Conv::c)) out += #c;
- ABSL_CONVERSION_CHARS_EXPAND_(CONV_SET_CASE, )
+#define CONV_SET_CASE(c) \
+ if (Contains(conv, FormatConversionCharSetInternal::c)) { \
+ out += #c; \
+ }
+ ABSL_INTERNAL_CONVERSION_CHARS_EXPAND_(CONV_SET_CASE, )
#undef CONV_SET_CASE
- if (Contains(conv, Conv::star)) out += "*";
+ if (Contains(conv, FormatConversionCharSetInternal::kStar)) {
+ out += "*";
+ }
return out;
}
TEST(StrFormatChecker, ArgumentToConv) {
- Conv conv = ArgumentToConv<std::string>();
+ FormatConversionCharSet conv = ArgumentToConv<std::string>();
EXPECT_EQ(ConvToString(conv), "s");
conv = ArgumentToConv<const char*>();
@@ -35,7 +54,7 @@
EXPECT_EQ(ConvToString(conv), "p");
}
-#if ABSL_INTERNAL_ENABLE_FORMAT_CHECKER
+#ifdef ABSL_INTERNAL_ENABLE_FORMAT_CHECKER
struct Case {
bool result;
@@ -147,4 +166,5 @@
} // namespace
} // namespace str_format_internal
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/internal/str_format/convert_test.cc b/absl/strings/internal/str_format/convert_test.cc
index 814ccf4..375db0a 100644
--- a/absl/strings/internal/str_format/convert_test.cc
+++ b/absl/strings/internal/str_format/convert_test.cc
@@ -1,17 +1,46 @@
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
#include <errno.h>
#include <stdarg.h>
#include <stdio.h>
+
#include <cctype>
#include <cmath>
+#include <limits>
#include <string>
+#include <thread> // NOLINT
+#include "gmock/gmock.h"
#include "gtest/gtest.h"
+#include "absl/base/internal/raw_logging.h"
#include "absl/strings/internal/str_format/bind.h"
+#include "absl/strings/match.h"
+#include "absl/types/optional.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace str_format_internal {
namespace {
+struct NativePrintfTraits {
+ bool hex_float_has_glibc_rounding;
+ bool hex_float_prefers_denormal_repr;
+ bool hex_float_uses_minimal_precision_when_not_specified;
+ bool hex_float_optimizes_leading_digit_bit_count;
+};
+
template <typename T, size_t N>
size_t ArraySize(T (&)[N]) {
return N;
@@ -54,7 +83,7 @@
return oss.str();
}
-void StrAppend(std::string *dst, const char *format, va_list ap) {
+void StrAppendV(std::string *dst, const char *format, va_list ap) {
// First try with a small fixed size buffer
static const int kSpaceLength = 1024;
char space[kSpaceLength];
@@ -95,15 +124,79 @@
delete[] buf;
}
+void StrAppend(std::string *out, const char *format, ...) {
+ va_list ap;
+ va_start(ap, format);
+ StrAppendV(out, format, ap);
+ va_end(ap);
+}
+
std::string StrPrint(const char *format, ...) {
va_list ap;
va_start(ap, format);
std::string result;
- StrAppend(&result, format, ap);
+ StrAppendV(&result, format, ap);
va_end(ap);
return result;
}
+NativePrintfTraits VerifyNativeImplementationImpl() {
+ NativePrintfTraits result;
+
+ // >>> hex_float_has_glibc_rounding. To have glibc's rounding behavior we need
+ // to meet three requirements:
+ //
+ // - The threshold for rounding up is 8 (for e.g. MSVC uses 9).
+ // - If the digits lower than than the 8 are non-zero then we round up.
+ // - If the digits lower than the 8 are all zero then we round toward even.
+ //
+ // The numbers below represent all the cases covering {below,at,above} the
+ // threshold (8) with both {zero,non-zero} lower bits and both {even,odd}
+ // preceding digits.
+ const double d0079 = 65657.0; // 0x1.0079p+16
+ const double d0179 = 65913.0; // 0x1.0179p+16
+ const double d0080 = 65664.0; // 0x1.0080p+16
+ const double d0180 = 65920.0; // 0x1.0180p+16
+ const double d0081 = 65665.0; // 0x1.0081p+16
+ const double d0181 = 65921.0; // 0x1.0181p+16
+ result.hex_float_has_glibc_rounding =
+ StartsWith(StrPrint("%.2a", d0079), "0x1.00") &&
+ StartsWith(StrPrint("%.2a", d0179), "0x1.01") &&
+ StartsWith(StrPrint("%.2a", d0080), "0x1.00") &&
+ StartsWith(StrPrint("%.2a", d0180), "0x1.02") &&
+ StartsWith(StrPrint("%.2a", d0081), "0x1.01") &&
+ StartsWith(StrPrint("%.2a", d0181), "0x1.02");
+
+ // >>> hex_float_prefers_denormal_repr. Formatting `denormal` on glibc yields
+ // "0x0.0000000000001p-1022", whereas on std libs that don't use denormal
+ // representation it would either be 0x1p-1074 or 0x1.0000000000000-1074.
+ const double denormal = std::numeric_limits<double>::denorm_min();
+ result.hex_float_prefers_denormal_repr =
+ StartsWith(StrPrint("%a", denormal), "0x0.0000000000001");
+
+ // >>> hex_float_uses_minimal_precision_when_not_specified. Some (non-glibc)
+ // libs will format the following as "0x1.0079000000000p+16".
+ result.hex_float_uses_minimal_precision_when_not_specified =
+ (StrPrint("%a", d0079) == "0x1.0079p+16");
+
+ // >>> hex_float_optimizes_leading_digit_bit_count. The number 1.5, when
+ // formatted by glibc should yield "0x1.8p+0" for `double` and "0xcp-3" for
+ // `long double`, i.e., number of bits in the leading digit is adapted to the
+ // number of bits in the mantissa.
+ const double d_15 = 1.5;
+ const long double ld_15 = 1.5;
+ result.hex_float_optimizes_leading_digit_bit_count =
+ StartsWith(StrPrint("%a", d_15), "0x1.8") &&
+ StartsWith(StrPrint("%La", ld_15), "0xc");
+
+ return result;
+}
+
+const NativePrintfTraits &VerifyNativeImplementation() {
+ static NativePrintfTraits native_traits = VerifyNativeImplementationImpl();
+ return native_traits;
+}
+
class FormatConvertTest : public ::testing::Test { };
template <typename T>
@@ -151,18 +244,26 @@
UntypedFormatSpecImpl format("%.1s");
EXPECT_EQ("a", FormatPack(format, {FormatArgImpl(p)}));
- // We cap at the nul terminator.
+ // We cap at the NUL-terminator.
p = "ABC";
UntypedFormatSpecImpl format2("%.10s");
EXPECT_EQ("ABC", FormatPack(format2, {FormatArgImpl(p)}));
}
+// Pointer formatting is implementation defined. This checks that the argument
+// can be matched to `ptr`.
+MATCHER_P(MatchesPointerString, ptr, "") {
+ if (ptr == nullptr && arg == "(nil)") {
+ return true;
+ }
+ void* parsed = nullptr;
+ if (sscanf(arg.c_str(), "%p", &parsed) != 1) {
+ ABSL_RAW_LOG(FATAL, "Could not parse %s", arg.c_str());
+ }
+ return ptr == parsed;
+}
+
TEST_F(FormatConvertTest, Pointer) {
-#ifdef _MSC_VER
- // MSVC's printf implementation prints pointers differently. We can't easily
- // compare our implementation to theirs.
- return;
-#endif
static int x = 0;
const int *xp = &x;
char c = 'h';
@@ -173,48 +274,62 @@
using VoidF = void (*)();
VoidF fp = [] {}, fnil = nullptr;
volatile char vc;
- volatile char* vcp = &vc;
- volatile char* vcnil = nullptr;
- const FormatArgImpl args[] = {
+ volatile char *vcp = &vc;
+ volatile char *vcnil = nullptr;
+ const FormatArgImpl args_array[] = {
FormatArgImpl(xp), FormatArgImpl(cp), FormatArgImpl(inil),
FormatArgImpl(cnil), FormatArgImpl(mcp), FormatArgImpl(fp),
FormatArgImpl(fnil), FormatArgImpl(vcp), FormatArgImpl(vcnil),
};
- struct Expectation {
- std::string out;
- const char *fmt;
- };
- const Expectation kExpect[] = {
- {StrPrint("%p", &x), "%p"},
- {StrPrint("%20p", &x), "%20p"},
- {StrPrint("%.1p", &x), "%.1p"},
- {StrPrint("%.20p", &x), "%.20p"},
- {StrPrint("%30.20p", &x), "%30.20p"},
+ auto args = absl::MakeConstSpan(args_array);
- {StrPrint("%-p", &x), "%-p"},
- {StrPrint("%-20p", &x), "%-20p"},
- {StrPrint("%-.1p", &x), "%-.1p"},
- {StrPrint("%.20p", &x), "%.20p"},
- {StrPrint("%-30.20p", &x), "%-30.20p"},
+ EXPECT_THAT(FormatPack(UntypedFormatSpecImpl("%p"), args),
+ MatchesPointerString(&x));
+ EXPECT_THAT(FormatPack(UntypedFormatSpecImpl("%20p"), args),
+ MatchesPointerString(&x));
+ EXPECT_THAT(FormatPack(UntypedFormatSpecImpl("%.1p"), args),
+ MatchesPointerString(&x));
+ EXPECT_THAT(FormatPack(UntypedFormatSpecImpl("%.20p"), args),
+ MatchesPointerString(&x));
+ EXPECT_THAT(FormatPack(UntypedFormatSpecImpl("%30.20p"), args),
+ MatchesPointerString(&x));
- {StrPrint("%p", cp), "%2$p"}, // const char*
- {"(nil)", "%3$p"}, // null const char *
- {"(nil)", "%4$p"}, // null const int *
- {StrPrint("%p", mcp), "%5$p"}, // nonconst char*
+ EXPECT_THAT(FormatPack(UntypedFormatSpecImpl("%-p"), args),
+ MatchesPointerString(&x));
+ EXPECT_THAT(FormatPack(UntypedFormatSpecImpl("%-20p"), args),
+ MatchesPointerString(&x));
+ EXPECT_THAT(FormatPack(UntypedFormatSpecImpl("%-.1p"), args),
+ MatchesPointerString(&x));
+ EXPECT_THAT(FormatPack(UntypedFormatSpecImpl("%.20p"), args),
+ MatchesPointerString(&x));
+ EXPECT_THAT(FormatPack(UntypedFormatSpecImpl("%-30.20p"), args),
+ MatchesPointerString(&x));
- {StrPrint("%p", fp), "%6$p"}, // function pointer
- {StrPrint("%p", vcp), "%8$p"}, // function pointer
+ // const char*
+ EXPECT_THAT(FormatPack(UntypedFormatSpecImpl("%2$p"), args),
+ MatchesPointerString(cp));
+ // null const int*
+ EXPECT_THAT(FormatPack(UntypedFormatSpecImpl("%3$p"), args),
+ MatchesPointerString(nullptr));
+ // null const char*
+ EXPECT_THAT(FormatPack(UntypedFormatSpecImpl("%4$p"), args),
+ MatchesPointerString(nullptr));
+ // nonconst char*
+ EXPECT_THAT(FormatPack(UntypedFormatSpecImpl("%5$p"), args),
+ MatchesPointerString(mcp));
-#ifndef __APPLE__
- // Apple's printf differs here (0x0 vs. nil)
- {StrPrint("%p", fnil), "%7$p"}, // null function pointer
- {StrPrint("%p", vcnil), "%9$p"}, // null function pointer
-#endif
- };
- for (const Expectation &e : kExpect) {
- UntypedFormatSpecImpl format(e.fmt);
- EXPECT_EQ(e.out, FormatPack(format, absl::MakeSpan(args))) << e.fmt;
- }
+ // function pointers
+ EXPECT_THAT(FormatPack(UntypedFormatSpecImpl("%6$p"), args),
+ MatchesPointerString(reinterpret_cast<const void*>(fp)));
+ EXPECT_THAT(
+ FormatPack(UntypedFormatSpecImpl("%8$p"), args),
+ MatchesPointerString(reinterpret_cast<volatile const void *>(vcp)));
+
+ // null function pointers
+ EXPECT_THAT(FormatPack(UntypedFormatSpecImpl("%7$p"), args),
+ MatchesPointerString(nullptr));
+ EXPECT_THAT(FormatPack(UntypedFormatSpecImpl("%9$p"), args),
+ MatchesPointerString(nullptr));
}
struct Cardinal {
@@ -366,7 +481,6 @@
AllIntTypes;
INSTANTIATE_TYPED_TEST_CASE_P(TypedFormatConvertTestWithAllIntTypes,
TypedFormatConvertTest, AllIntTypes);
-
TEST_F(FormatConvertTest, VectorBool) {
// Make sure vector<bool>'s values behave as bools.
std::vector<bool> v = {true, false};
@@ -378,6 +492,42 @@
FormatArgImpl(cv[0]), FormatArgImpl(cv[1])})));
}
+
+TEST_F(FormatConvertTest, Int128) {
+ absl::int128 positive = static_cast<absl::int128>(0x1234567890abcdef) * 1979;
+ absl::int128 negative = -positive;
+ absl::int128 max = absl::Int128Max(), min = absl::Int128Min();
+ const FormatArgImpl args[] = {FormatArgImpl(positive),
+ FormatArgImpl(negative), FormatArgImpl(max),
+ FormatArgImpl(min)};
+
+ struct Case {
+ const char* format;
+ const char* expected;
+ } cases[] = {
+ {"%1$d", "2595989796776606496405"},
+ {"%1$30d", " 2595989796776606496405"},
+ {"%1$-30d", "2595989796776606496405 "},
+ {"%1$u", "2595989796776606496405"},
+ {"%1$x", "8cba9876066020f695"},
+ {"%2$d", "-2595989796776606496405"},
+ {"%2$30d", " -2595989796776606496405"},
+ {"%2$-30d", "-2595989796776606496405 "},
+ {"%2$u", "340282366920938460867384810655161715051"},
+ {"%2$x", "ffffffffffffff73456789f99fdf096b"},
+ {"%3$d", "170141183460469231731687303715884105727"},
+ {"%3$u", "170141183460469231731687303715884105727"},
+ {"%3$x", "7fffffffffffffffffffffffffffffff"},
+ {"%4$d", "-170141183460469231731687303715884105728"},
+ {"%4$x", "80000000000000000000000000000000"},
+ };
+
+ for (auto c : cases) {
+ UntypedFormatSpecImpl format(c.format);
+ EXPECT_EQ(c.expected, FormatPack(format, absl::MakeSpan(args)));
+ }
+}
+
TEST_F(FormatConvertTest, Uint128) {
absl::uint128 v = static_cast<absl::uint128>(0x1234567890abcdef) * 1979;
absl::uint128 max = absl::Uint128Max();
@@ -403,6 +553,68 @@
}
}
+template <typename Floating>
+void TestWithMultipleFormatsHelper(const std::vector<Floating> &floats) {
+ const NativePrintfTraits &native_traits = VerifyNativeImplementation();
+ // Reserve the space to ensure we don't allocate memory in the output itself.
+ std::string str_format_result;
+ str_format_result.reserve(1 << 20);
+ std::string string_printf_result;
+ string_printf_result.reserve(1 << 20);
+
+ const char *const kFormats[] = {
+ "%", "%.3", "%8.5", "%500", "%.5000", "%.60", "%.30", "%03",
+ "%+", "% ", "%-10", "%#15.3", "%#.0", "%.0", "%1$*2$", "%1$.*2$"};
+
+ for (const char *fmt : kFormats) {
+ for (char f : {'f', 'F', //
+ 'g', 'G', //
+ 'a', 'A', //
+ 'e', 'E'}) {
+ std::string fmt_str = std::string(fmt) + f;
+
+ if (fmt == absl::string_view("%.5000") && f != 'f' && f != 'F' &&
+ f != 'a' && f != 'A') {
+ // This particular test takes way too long with snprintf.
+ // Disable for the case we are not implementing natively.
+ continue;
+ }
+
+ if ((f == 'a' || f == 'A') &&
+ !native_traits.hex_float_has_glibc_rounding) {
+ continue;
+ }
+
+ for (Floating d : floats) {
+ if (!native_traits.hex_float_prefers_denormal_repr &&
+ (f == 'a' || f == 'A') && std::fpclassify(d) == FP_SUBNORMAL) {
+ continue;
+ }
+ int i = -10;
+ FormatArgImpl args[2] = {FormatArgImpl(d), FormatArgImpl(i)};
+ UntypedFormatSpecImpl format(fmt_str);
+
+ string_printf_result.clear();
+ StrAppend(&string_printf_result, fmt_str.c_str(), d, i);
+ str_format_result.clear();
+
+ {
+ AppendPack(&str_format_result, format, absl::MakeSpan(args));
+ }
+
+ if (string_printf_result != str_format_result) {
+ // We use ASSERT_EQ here because failures are usually correlated and a
+ // bug would print way too many failed expectations causing the test
+ // to time out.
+ ASSERT_EQ(string_printf_result, str_format_result)
+ << fmt_str << " " << StrPrint("%.18g", d) << " "
+ << StrPrint("%a", d) << " " << StrPrint("%.50f", d);
+ }
+ }
+ }
+ }
+}
+
TEST_F(FormatConvertTest, Float) {
#ifdef _MSC_VER
// MSVC has a different rounding policy than us so we can't test our
@@ -410,9 +622,62 @@
return;
#endif // _MSC_VER
- const char *const kFormats[] = {
- "%", "%.3", "%8.5", "%9", "%.60", "%.30", "%03", "%+",
- "% ", "%-10", "%#15.3", "%#.0", "%.0", "%1$*2$", "%1$.*2$"};
+ std::vector<float> floats = {0.0f,
+ -0.0f,
+ .9999999f,
+ 9999999.f,
+ std::numeric_limits<float>::max(),
+ -std::numeric_limits<float>::max(),
+ std::numeric_limits<float>::min(),
+ -std::numeric_limits<float>::min(),
+ std::numeric_limits<float>::lowest(),
+ -std::numeric_limits<float>::lowest(),
+ std::numeric_limits<float>::epsilon(),
+ std::numeric_limits<float>::epsilon() + 1.0f,
+ std::numeric_limits<float>::infinity(),
+ -std::numeric_limits<float>::infinity()};
+
+ // Some regression tests.
+ floats.push_back(0.999999989f);
+
+ if (std::numeric_limits<float>::has_denorm != std::denorm_absent) {
+ floats.push_back(std::numeric_limits<float>::denorm_min());
+ floats.push_back(-std::numeric_limits<float>::denorm_min());
+ }
+
+ for (float base :
+ {1.f, 12.f, 123.f, 1234.f, 12345.f, 123456.f, 1234567.f, 12345678.f,
+ 123456789.f, 1234567890.f, 12345678901.f, 12345678.f, 12345678.f}) {
+ for (int exp = -123; exp <= 123; ++exp) {
+ for (int sign : {1, -1}) {
+ floats.push_back(sign * std::ldexp(base, exp));
+ }
+ }
+ }
+
+ for (int exp = -300; exp <= 300; ++exp) {
+ const float all_ones_mantissa = 0xffffff;
+ floats.push_back(std::ldexp(all_ones_mantissa, exp));
+ }
+
+ // Remove duplicates to speed up the logic below.
+ std::sort(floats.begin(), floats.end());
+ floats.erase(std::unique(floats.begin(), floats.end()), floats.end());
+
+#ifndef __APPLE__
+ // Apple formats NaN differently (+nan) vs. (nan)
+ floats.push_back(std::nan(""));
+#endif
+
+ TestWithMultipleFormatsHelper(floats);
+}
+
+TEST_F(FormatConvertTest, Double) {
+#ifdef _MSC_VER
+ // MSVC has a different rounding policy than us so we can't test our
+ // implementation against the native one there.
+ return;
+#endif // _MSC_VER
std::vector<double> doubles = {0.0,
-0.0,
@@ -429,11 +694,6 @@
std::numeric_limits<double>::infinity(),
-std::numeric_limits<double>::infinity()};
-#ifndef __APPLE__
- // Apple formats NaN differently (+nan) vs. (nan)
- doubles.push_back(std::nan(""));
-#endif
-
// Some regression tests.
doubles.push_back(0.99999999999999989);
@@ -452,43 +712,375 @@
}
}
- for (const char *fmt : kFormats) {
- for (char f : {'f', 'F', //
- 'g', 'G', //
- 'a', 'A', //
- 'e', 'E'}) {
- std::string fmt_str = std::string(fmt) + f;
- for (double d : doubles) {
- int i = -10;
- FormatArgImpl args[2] = {FormatArgImpl(d), FormatArgImpl(i)};
- UntypedFormatSpecImpl format(fmt_str);
- // We use ASSERT_EQ here because failures are usually correlated and a
- // bug would print way too many failed expectations causing the test to
- // time out.
- ASSERT_EQ(StrPrint(fmt_str.c_str(), d, i),
- FormatPack(format, absl::MakeSpan(args)))
- << fmt_str << " " << StrPrint("%.18g", d) << " "
- << StrPrint("%.999f", d);
+ // Workaround libc bug.
+ // https://sourceware.org/bugzilla/show_bug.cgi?id=22142
+ const bool gcc_bug_22142 =
+ StrPrint("%f", std::numeric_limits<double>::max()) !=
+ "1797693134862315708145274237317043567980705675258449965989174768031"
+ "5726078002853876058955863276687817154045895351438246423432132688946"
+ "4182768467546703537516986049910576551282076245490090389328944075868"
+ "5084551339423045832369032229481658085593321233482747978262041447231"
+ "68738177180919299881250404026184124858368.000000";
+
+ if (!gcc_bug_22142) {
+ for (int exp = -300; exp <= 300; ++exp) {
+ const double all_ones_mantissa = 0x1fffffffffffff;
+ doubles.push_back(std::ldexp(all_ones_mantissa, exp));
+ }
+ }
+
+ if (gcc_bug_22142) {
+ for (auto &d : doubles) {
+ using L = std::numeric_limits<double>;
+ double d2 = std::abs(d);
+ if (d2 == L::max() || d2 == L::min() || d2 == L::denorm_min()) {
+ d = 0;
}
}
}
+
+ // Remove duplicates to speed up the logic below.
+ std::sort(doubles.begin(), doubles.end());
+ doubles.erase(std::unique(doubles.begin(), doubles.end()), doubles.end());
+
+#ifndef __APPLE__
+ // Apple formats NaN differently (+nan) vs. (nan)
+ doubles.push_back(std::nan(""));
+#endif
+
+ TestWithMultipleFormatsHelper(doubles);
+}
+
+TEST_F(FormatConvertTest, DoubleRound) {
+ std::string s;
+ const auto format = [&](const char *fmt, double d) -> std::string & {
+ s.clear();
+ FormatArgImpl args[1] = {FormatArgImpl(d)};
+ AppendPack(&s, UntypedFormatSpecImpl(fmt), absl::MakeSpan(args));
+#if !defined(_MSC_VER)
+ // MSVC has a different rounding policy than us so we can't test our
+ // implementation against the native one there.
+ EXPECT_EQ(StrPrint(fmt, d), s);
+#endif // _MSC_VER
+
+ return s;
+ };
+ // All of these values have to be exactly represented.
+ // Otherwise we might not be testing what we think we are testing.
+
+ // These values can fit in a 64bit "fast" representation.
+ const double exact_value = 0.00000000000005684341886080801486968994140625;
+ assert(exact_value == std::pow(2, -44));
+ // Round up at a 5xx.
+ EXPECT_EQ(format("%.13f", exact_value), "0.0000000000001");
+ // Round up at a >5
+ EXPECT_EQ(format("%.14f", exact_value), "0.00000000000006");
+ // Round down at a <5
+ EXPECT_EQ(format("%.16f", exact_value), "0.0000000000000568");
+ // Nine handling
+ EXPECT_EQ(format("%.35f", exact_value),
+ "0.00000000000005684341886080801486969");
+ EXPECT_EQ(format("%.36f", exact_value),
+ "0.000000000000056843418860808014869690");
+ // Round down the last nine.
+ EXPECT_EQ(format("%.37f", exact_value),
+ "0.0000000000000568434188608080148696899");
+ EXPECT_EQ(format("%.10f", 0.000003814697265625), "0.0000038147");
+ // Round up the last nine
+ EXPECT_EQ(format("%.11f", 0.000003814697265625), "0.00000381470");
+ EXPECT_EQ(format("%.12f", 0.000003814697265625), "0.000003814697");
+
+ // Round to even (down)
+ EXPECT_EQ(format("%.43f", exact_value),
+ "0.0000000000000568434188608080148696899414062");
+ // Exact
+ EXPECT_EQ(format("%.44f", exact_value),
+ "0.00000000000005684341886080801486968994140625");
+ // Round to even (up), let make the last digits 75 instead of 25
+ EXPECT_EQ(format("%.43f", exact_value + std::pow(2, -43)),
+ "0.0000000000001705302565824240446090698242188");
+ // Exact, just to check.
+ EXPECT_EQ(format("%.44f", exact_value + std::pow(2, -43)),
+ "0.00000000000017053025658242404460906982421875");
+
+ // This value has to be small enough that it won't fit in the uint128
+ // representation for printing.
+ const double small_exact_value =
+ 0.000000000000000000000000000000000000752316384526264005099991383822237233803945956334136013765601092018187046051025390625; // NOLINT
+ assert(small_exact_value == std::pow(2, -120));
+ // Round up at a 5xx.
+ EXPECT_EQ(format("%.37f", small_exact_value),
+ "0.0000000000000000000000000000000000008");
+ // Round down at a <5
+ EXPECT_EQ(format("%.38f", small_exact_value),
+ "0.00000000000000000000000000000000000075");
+ // Round up at a >5
+ EXPECT_EQ(format("%.41f", small_exact_value),
+ "0.00000000000000000000000000000000000075232");
+ // Nine handling
+ EXPECT_EQ(format("%.55f", small_exact_value),
+ "0.0000000000000000000000000000000000007523163845262640051");
+ EXPECT_EQ(format("%.56f", small_exact_value),
+ "0.00000000000000000000000000000000000075231638452626400510");
+ EXPECT_EQ(format("%.57f", small_exact_value),
+ "0.000000000000000000000000000000000000752316384526264005100");
+ EXPECT_EQ(format("%.58f", small_exact_value),
+ "0.0000000000000000000000000000000000007523163845262640051000");
+ // Round down the last nine
+ EXPECT_EQ(format("%.59f", small_exact_value),
+ "0.00000000000000000000000000000000000075231638452626400509999");
+ // Round up the last nine
+ EXPECT_EQ(format("%.79f", small_exact_value),
+ "0.000000000000000000000000000000000000"
+ "7523163845262640050999913838222372338039460");
+
+ // Round to even (down)
+ EXPECT_EQ(format("%.119f", small_exact_value),
+ "0.000000000000000000000000000000000000"
+ "75231638452626400509999138382223723380"
+ "394595633413601376560109201818704605102539062");
+ // Exact
+ EXPECT_EQ(format("%.120f", small_exact_value),
+ "0.000000000000000000000000000000000000"
+ "75231638452626400509999138382223723380"
+ "3945956334136013765601092018187046051025390625");
+ // Round to even (up), let make the last digits 75 instead of 25
+ EXPECT_EQ(format("%.119f", small_exact_value + std::pow(2, -119)),
+ "0.000000000000000000000000000000000002"
+ "25694915357879201529997415146671170141"
+ "183786900240804129680327605456113815307617188");
+ // Exact, just to check.
+ EXPECT_EQ(format("%.120f", small_exact_value + std::pow(2, -119)),
+ "0.000000000000000000000000000000000002"
+ "25694915357879201529997415146671170141"
+ "1837869002408041296803276054561138153076171875");
+}
+
+TEST_F(FormatConvertTest, DoubleRoundA) {
+ const NativePrintfTraits &native_traits = VerifyNativeImplementation();
+ std::string s;
+ const auto format = [&](const char *fmt, double d) -> std::string & {
+ s.clear();
+ FormatArgImpl args[1] = {FormatArgImpl(d)};
+ AppendPack(&s, UntypedFormatSpecImpl(fmt), absl::MakeSpan(args));
+ if (native_traits.hex_float_has_glibc_rounding) {
+ EXPECT_EQ(StrPrint(fmt, d), s);
+ }
+ return s;
+ };
+
+ // 0x1.00018000p+100
+ const double on_boundary_odd = 1267679614447900152596896153600.0;
+ EXPECT_EQ(format("%.0a", on_boundary_odd), "0x1p+100");
+ EXPECT_EQ(format("%.1a", on_boundary_odd), "0x1.0p+100");
+ EXPECT_EQ(format("%.2a", on_boundary_odd), "0x1.00p+100");
+ EXPECT_EQ(format("%.3a", on_boundary_odd), "0x1.000p+100");
+ EXPECT_EQ(format("%.4a", on_boundary_odd), "0x1.0002p+100"); // round
+ EXPECT_EQ(format("%.5a", on_boundary_odd), "0x1.00018p+100");
+ EXPECT_EQ(format("%.6a", on_boundary_odd), "0x1.000180p+100");
+
+ // 0x1.00028000p-2
+ const double on_boundary_even = 0.250009536743164062500;
+ EXPECT_EQ(format("%.0a", on_boundary_even), "0x1p-2");
+ EXPECT_EQ(format("%.1a", on_boundary_even), "0x1.0p-2");
+ EXPECT_EQ(format("%.2a", on_boundary_even), "0x1.00p-2");
+ EXPECT_EQ(format("%.3a", on_boundary_even), "0x1.000p-2");
+ EXPECT_EQ(format("%.4a", on_boundary_even), "0x1.0002p-2"); // no round
+ EXPECT_EQ(format("%.5a", on_boundary_even), "0x1.00028p-2");
+ EXPECT_EQ(format("%.6a", on_boundary_even), "0x1.000280p-2");
+
+ // 0x1.00018001p+1
+ const double slightly_over = 2.00004577683284878730773925781250;
+ EXPECT_EQ(format("%.0a", slightly_over), "0x1p+1");
+ EXPECT_EQ(format("%.1a", slightly_over), "0x1.0p+1");
+ EXPECT_EQ(format("%.2a", slightly_over), "0x1.00p+1");
+ EXPECT_EQ(format("%.3a", slightly_over), "0x1.000p+1");
+ EXPECT_EQ(format("%.4a", slightly_over), "0x1.0002p+1");
+ EXPECT_EQ(format("%.5a", slightly_over), "0x1.00018p+1");
+ EXPECT_EQ(format("%.6a", slightly_over), "0x1.000180p+1");
+
+ // 0x1.00017fffp+0
+ const double slightly_under = 1.000022887950763106346130371093750;
+ EXPECT_EQ(format("%.0a", slightly_under), "0x1p+0");
+ EXPECT_EQ(format("%.1a", slightly_under), "0x1.0p+0");
+ EXPECT_EQ(format("%.2a", slightly_under), "0x1.00p+0");
+ EXPECT_EQ(format("%.3a", slightly_under), "0x1.000p+0");
+ EXPECT_EQ(format("%.4a", slightly_under), "0x1.0001p+0");
+ EXPECT_EQ(format("%.5a", slightly_under), "0x1.00018p+0");
+ EXPECT_EQ(format("%.6a", slightly_under), "0x1.000180p+0");
+ EXPECT_EQ(format("%.7a", slightly_under), "0x1.0001800p+0");
+
+ // 0x1.1b3829ac28058p+3
+ const double hex_value = 8.85060580848964661981881363317370414733886718750;
+ EXPECT_EQ(format("%.0a", hex_value), "0x1p+3");
+ EXPECT_EQ(format("%.1a", hex_value), "0x1.2p+3");
+ EXPECT_EQ(format("%.2a", hex_value), "0x1.1bp+3");
+ EXPECT_EQ(format("%.3a", hex_value), "0x1.1b4p+3");
+ EXPECT_EQ(format("%.4a", hex_value), "0x1.1b38p+3");
+ EXPECT_EQ(format("%.5a", hex_value), "0x1.1b383p+3");
+ EXPECT_EQ(format("%.6a", hex_value), "0x1.1b382ap+3");
+ EXPECT_EQ(format("%.7a", hex_value), "0x1.1b3829bp+3");
+ EXPECT_EQ(format("%.8a", hex_value), "0x1.1b3829acp+3");
+ EXPECT_EQ(format("%.9a", hex_value), "0x1.1b3829ac3p+3");
+ EXPECT_EQ(format("%.10a", hex_value), "0x1.1b3829ac28p+3");
+ EXPECT_EQ(format("%.11a", hex_value), "0x1.1b3829ac280p+3");
+ EXPECT_EQ(format("%.12a", hex_value), "0x1.1b3829ac2806p+3");
+ EXPECT_EQ(format("%.13a", hex_value), "0x1.1b3829ac28058p+3");
+ EXPECT_EQ(format("%.14a", hex_value), "0x1.1b3829ac280580p+3");
+ EXPECT_EQ(format("%.15a", hex_value), "0x1.1b3829ac2805800p+3");
+ EXPECT_EQ(format("%.16a", hex_value), "0x1.1b3829ac28058000p+3");
+ EXPECT_EQ(format("%.17a", hex_value), "0x1.1b3829ac280580000p+3");
+ EXPECT_EQ(format("%.18a", hex_value), "0x1.1b3829ac2805800000p+3");
+ EXPECT_EQ(format("%.19a", hex_value), "0x1.1b3829ac28058000000p+3");
+ EXPECT_EQ(format("%.20a", hex_value), "0x1.1b3829ac280580000000p+3");
+ EXPECT_EQ(format("%.21a", hex_value), "0x1.1b3829ac2805800000000p+3");
+
+ // 0x1.0818283848586p+3
+ const double hex_value2 = 8.2529488658208371987257123691961169242858886718750;
+ EXPECT_EQ(format("%.0a", hex_value2), "0x1p+3");
+ EXPECT_EQ(format("%.1a", hex_value2), "0x1.1p+3");
+ EXPECT_EQ(format("%.2a", hex_value2), "0x1.08p+3");
+ EXPECT_EQ(format("%.3a", hex_value2), "0x1.082p+3");
+ EXPECT_EQ(format("%.4a", hex_value2), "0x1.0818p+3");
+ EXPECT_EQ(format("%.5a", hex_value2), "0x1.08183p+3");
+ EXPECT_EQ(format("%.6a", hex_value2), "0x1.081828p+3");
+ EXPECT_EQ(format("%.7a", hex_value2), "0x1.0818284p+3");
+ EXPECT_EQ(format("%.8a", hex_value2), "0x1.08182838p+3");
+ EXPECT_EQ(format("%.9a", hex_value2), "0x1.081828385p+3");
+ EXPECT_EQ(format("%.10a", hex_value2), "0x1.0818283848p+3");
+ EXPECT_EQ(format("%.11a", hex_value2), "0x1.08182838486p+3");
+ EXPECT_EQ(format("%.12a", hex_value2), "0x1.081828384858p+3");
+ EXPECT_EQ(format("%.13a", hex_value2), "0x1.0818283848586p+3");
+ EXPECT_EQ(format("%.14a", hex_value2), "0x1.08182838485860p+3");
+ EXPECT_EQ(format("%.15a", hex_value2), "0x1.081828384858600p+3");
+ EXPECT_EQ(format("%.16a", hex_value2), "0x1.0818283848586000p+3");
+ EXPECT_EQ(format("%.17a", hex_value2), "0x1.08182838485860000p+3");
+ EXPECT_EQ(format("%.18a", hex_value2), "0x1.081828384858600000p+3");
+ EXPECT_EQ(format("%.19a", hex_value2), "0x1.0818283848586000000p+3");
+ EXPECT_EQ(format("%.20a", hex_value2), "0x1.08182838485860000000p+3");
+ EXPECT_EQ(format("%.21a", hex_value2), "0x1.081828384858600000000p+3");
+}
+
+TEST_F(FormatConvertTest, LongDoubleRoundA) {
+ if (std::numeric_limits<long double>::digits % 4 != 0) {
+ // This test doesn't really make sense to run on platforms where a long
+ // double has a different mantissa size (mod 4) than Prod, since then the
+ // leading digit will be formatted differently.
+ return;
+ }
+ const NativePrintfTraits &native_traits = VerifyNativeImplementation();
+ std::string s;
+ const auto format = [&](const char *fmt, long double d) -> std::string & {
+ s.clear();
+ FormatArgImpl args[1] = {FormatArgImpl(d)};
+ AppendPack(&s, UntypedFormatSpecImpl(fmt), absl::MakeSpan(args));
+ if (native_traits.hex_float_has_glibc_rounding &&
+ native_traits.hex_float_optimizes_leading_digit_bit_count) {
+ EXPECT_EQ(StrPrint(fmt, d), s);
+ }
+ return s;
+ };
+
+ // 0x8.8p+4
+ const long double on_boundary_even = 136.0;
+ EXPECT_EQ(format("%.0La", on_boundary_even), "0x8p+4");
+ EXPECT_EQ(format("%.1La", on_boundary_even), "0x8.8p+4");
+ EXPECT_EQ(format("%.2La", on_boundary_even), "0x8.80p+4");
+ EXPECT_EQ(format("%.3La", on_boundary_even), "0x8.800p+4");
+ EXPECT_EQ(format("%.4La", on_boundary_even), "0x8.8000p+4");
+ EXPECT_EQ(format("%.5La", on_boundary_even), "0x8.80000p+4");
+ EXPECT_EQ(format("%.6La", on_boundary_even), "0x8.800000p+4");
+
+ // 0x9.8p+4
+ const long double on_boundary_odd = 152.0;
+ EXPECT_EQ(format("%.0La", on_boundary_odd), "0xap+4");
+ EXPECT_EQ(format("%.1La", on_boundary_odd), "0x9.8p+4");
+ EXPECT_EQ(format("%.2La", on_boundary_odd), "0x9.80p+4");
+ EXPECT_EQ(format("%.3La", on_boundary_odd), "0x9.800p+4");
+ EXPECT_EQ(format("%.4La", on_boundary_odd), "0x9.8000p+4");
+ EXPECT_EQ(format("%.5La", on_boundary_odd), "0x9.80000p+4");
+ EXPECT_EQ(format("%.6La", on_boundary_odd), "0x9.800000p+4");
+
+ // 0x8.80001p+24
+ const long double slightly_over = 142606352.0;
+ EXPECT_EQ(format("%.0La", slightly_over), "0x9p+24");
+ EXPECT_EQ(format("%.1La", slightly_over), "0x8.8p+24");
+ EXPECT_EQ(format("%.2La", slightly_over), "0x8.80p+24");
+ EXPECT_EQ(format("%.3La", slightly_over), "0x8.800p+24");
+ EXPECT_EQ(format("%.4La", slightly_over), "0x8.8000p+24");
+ EXPECT_EQ(format("%.5La", slightly_over), "0x8.80001p+24");
+ EXPECT_EQ(format("%.6La", slightly_over), "0x8.800010p+24");
+
+ // 0x8.7ffffp+24
+ const long double slightly_under = 142606320.0;
+ EXPECT_EQ(format("%.0La", slightly_under), "0x8p+24");
+ EXPECT_EQ(format("%.1La", slightly_under), "0x8.8p+24");
+ EXPECT_EQ(format("%.2La", slightly_under), "0x8.80p+24");
+ EXPECT_EQ(format("%.3La", slightly_under), "0x8.800p+24");
+ EXPECT_EQ(format("%.4La", slightly_under), "0x8.8000p+24");
+ EXPECT_EQ(format("%.5La", slightly_under), "0x8.7ffffp+24");
+ EXPECT_EQ(format("%.6La", slightly_under), "0x8.7ffff0p+24");
+ EXPECT_EQ(format("%.7La", slightly_under), "0x8.7ffff00p+24");
+
+ // 0xc.0828384858688000p+128
+ const long double eights = 4094231060438608800781871108094404067328.0;
+ EXPECT_EQ(format("%.0La", eights), "0xcp+128");
+ EXPECT_EQ(format("%.1La", eights), "0xc.1p+128");
+ EXPECT_EQ(format("%.2La", eights), "0xc.08p+128");
+ EXPECT_EQ(format("%.3La", eights), "0xc.083p+128");
+ EXPECT_EQ(format("%.4La", eights), "0xc.0828p+128");
+ EXPECT_EQ(format("%.5La", eights), "0xc.08284p+128");
+ EXPECT_EQ(format("%.6La", eights), "0xc.082838p+128");
+ EXPECT_EQ(format("%.7La", eights), "0xc.0828385p+128");
+ EXPECT_EQ(format("%.8La", eights), "0xc.08283848p+128");
+ EXPECT_EQ(format("%.9La", eights), "0xc.082838486p+128");
+ EXPECT_EQ(format("%.10La", eights), "0xc.0828384858p+128");
+ EXPECT_EQ(format("%.11La", eights), "0xc.08283848587p+128");
+ EXPECT_EQ(format("%.12La", eights), "0xc.082838485868p+128");
+ EXPECT_EQ(format("%.13La", eights), "0xc.0828384858688p+128");
+ EXPECT_EQ(format("%.14La", eights), "0xc.08283848586880p+128");
+ EXPECT_EQ(format("%.15La", eights), "0xc.082838485868800p+128");
+ EXPECT_EQ(format("%.16La", eights), "0xc.0828384858688000p+128");
+}
+
+// We don't actually store the results. This is just to exercise the rest of the
+// machinery.
+struct NullSink {
+ friend void AbslFormatFlush(NullSink *sink, string_view str) {}
+};
+
+template <typename... T>
+bool FormatWithNullSink(absl::string_view fmt, const T &... a) {
+ NullSink sink;
+ FormatArgImpl args[] = {FormatArgImpl(a)...};
+ return FormatUntyped(&sink, UntypedFormatSpecImpl(fmt), absl::MakeSpan(args));
+}
+
+TEST_F(FormatConvertTest, ExtremeWidthPrecision) {
+ for (const char *fmt : {"f"}) {
+ for (double d : {1e-100, 1.0, 1e100}) {
+ constexpr int max = std::numeric_limits<int>::max();
+ EXPECT_TRUE(FormatWithNullSink(std::string("%.*") + fmt, max, d));
+ EXPECT_TRUE(FormatWithNullSink(std::string("%1.*") + fmt, max, d));
+ EXPECT_TRUE(FormatWithNullSink(std::string("%*") + fmt, max, d));
+ EXPECT_TRUE(FormatWithNullSink(std::string("%*.*") + fmt, max, max, d));
+ }
+ }
}
TEST_F(FormatConvertTest, LongDouble) {
- const char *const kFormats[] = {"%", "%.3", "%8.5", "%9",
+#ifdef _MSC_VER
+ // MSVC has a different rounding policy than us so we can't test our
+ // implementation against the native one there.
+ return;
+#endif // _MSC_VER
+ const NativePrintfTraits &native_traits = VerifyNativeImplementation();
+ const char *const kFormats[] = {"%", "%.3", "%8.5", "%9", "%.5000",
"%.60", "%+", "% ", "%-10"};
- // This value is not representable in double, but it is in long double that
- // uses the extended format.
- // This is to verify that we are not truncating the value mistakenly through a
- // double.
- long double very_precise = 10000000000000000.25L;
-
std::vector<long double> doubles = {
0.0,
-0.0,
- very_precise,
- 1 / very_precise,
std::numeric_limits<long double>::max(),
-std::numeric_limits<long double>::max(),
std::numeric_limits<long double>::min(),
@@ -496,28 +1088,65 @@
std::numeric_limits<long double>::infinity(),
-std::numeric_limits<long double>::infinity()};
+ for (long double base : {1.L, 12.L, 123.L, 1234.L, 12345.L, 123456.L,
+ 1234567.L, 12345678.L, 123456789.L, 1234567890.L,
+ 12345678901.L, 123456789012.L, 1234567890123.L,
+ // This value is not representable in double, but it
+ // is in long double that uses the extended format.
+ // This is to verify that we are not truncating the
+ // value mistakenly through a double.
+ 10000000000000000.25L}) {
+ for (int exp : {-1000, -500, 0, 500, 1000}) {
+ for (int sign : {1, -1}) {
+ doubles.push_back(sign * std::ldexp(base, exp));
+ doubles.push_back(sign / std::ldexp(base, exp));
+ }
+ }
+ }
+
+ // Regression tests
+ //
+ // Using a string literal because not all platforms support hex literals or it
+ // might be out of range.
+ doubles.push_back(std::strtold("-0xf.ffffffb5feafffbp-16324L", nullptr));
+
for (const char *fmt : kFormats) {
for (char f : {'f', 'F', //
'g', 'G', //
'a', 'A', //
'e', 'E'}) {
std::string fmt_str = std::string(fmt) + 'L' + f;
+
+ if (fmt == absl::string_view("%.5000") && f != 'f' && f != 'F' &&
+ f != 'a' && f != 'A') {
+ // This particular test takes way too long with snprintf.
+ // Disable for the case we are not implementing natively.
+ continue;
+ }
+
+ if (f == 'a' || f == 'A') {
+ if (!native_traits.hex_float_has_glibc_rounding ||
+ !native_traits.hex_float_optimizes_leading_digit_bit_count) {
+ continue;
+ }
+ }
+
for (auto d : doubles) {
FormatArgImpl arg(d);
UntypedFormatSpecImpl format(fmt_str);
// We use ASSERT_EQ here because failures are usually correlated and a
// bug would print way too many failed expectations causing the test to
// time out.
- ASSERT_EQ(StrPrint(fmt_str.c_str(), d),
- FormatPack(format, {&arg, 1}))
+ ASSERT_EQ(StrPrint(fmt_str.c_str(), d), FormatPack(format, {&arg, 1}))
<< fmt_str << " " << StrPrint("%.18Lg", d) << " "
- << StrPrint("%.999Lf", d);
+ << StrPrint("%La", d) << " " << StrPrint("%.1080Lf", d);
}
}
}
}
-TEST_F(FormatConvertTest, IntAsFloat) {
+TEST_F(FormatConvertTest, IntAsDouble) {
+ const NativePrintfTraits &native_traits = VerifyNativeImplementation();
const int kMin = std::numeric_limits<int>::min();
const int kMax = std::numeric_limits<int>::max();
const int ia[] = {
@@ -533,14 +1162,17 @@
const char *fmt;
};
const double dx = static_cast<double>(fx);
- const Expectation kExpect[] = {
- { __LINE__, StrPrint("%f", dx), "%f" },
- { __LINE__, StrPrint("%12f", dx), "%12f" },
- { __LINE__, StrPrint("%.12f", dx), "%.12f" },
- { __LINE__, StrPrint("%12a", dx), "%12a" },
- { __LINE__, StrPrint("%.12a", dx), "%.12a" },
+ std::vector<Expectation> expect = {
+ {__LINE__, StrPrint("%f", dx), "%f"},
+ {__LINE__, StrPrint("%12f", dx), "%12f"},
+ {__LINE__, StrPrint("%.12f", dx), "%.12f"},
+ {__LINE__, StrPrint("%.12a", dx), "%.12a"},
};
- for (const Expectation &e : kExpect) {
+ if (native_traits.hex_float_uses_minimal_precision_when_not_specified) {
+ Expectation ex = {__LINE__, StrPrint("%12a", dx), "%12a"};
+ expect.push_back(ex);
+ }
+ for (const Expectation &e : expect) {
SCOPED_TRACE(e.line);
SCOPED_TRACE(e.fmt);
UntypedFormatSpecImpl format(e.fmt);
@@ -585,6 +1217,26 @@
EXPECT_TRUE(FormatFails("%*d", ""));
}
+// Sanity check to make sure that we are testing what we think we're testing on
+// e.g. the x86_64+glibc platform.
+TEST_F(FormatConvertTest, GlibcHasCorrectTraits) {
+#if !defined(__GLIBC__) || !defined(__x86_64__)
+ return;
+#endif
+ const NativePrintfTraits &native_traits = VerifyNativeImplementation();
+ // If one of the following tests break then it is either because the above PP
+ // macro guards failed to exclude a new platform (likely) or because something
+ // has changed in the implemention of glibc sprintf float formatting behavior.
+ // If the latter, then the code that computes these flags needs to be
+ // revisited and/or possibly the StrFormat implementation.
+ EXPECT_TRUE(native_traits.hex_float_has_glibc_rounding);
+ EXPECT_TRUE(native_traits.hex_float_prefers_denormal_repr);
+ EXPECT_TRUE(
+ native_traits.hex_float_uses_minimal_precision_when_not_specified);
+ EXPECT_TRUE(native_traits.hex_float_optimizes_leading_digit_bit_count);
+}
+
} // namespace
} // namespace str_format_internal
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/internal/str_format/extension.cc b/absl/strings/internal/str_format/extension.cc
index d7f5815..bb0d96c 100644
--- a/absl/strings/internal/str_format/extension.cc
+++ b/absl/strings/internal/str_format/extension.cc
@@ -20,38 +20,8 @@
#include <string>
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace str_format_internal {
-namespace {
-// clang-format off
-#define ABSL_LENGTH_MODS_EXPAND_ \
- X_VAL(h) X_SEP \
- X_VAL(hh) X_SEP \
- X_VAL(l) X_SEP \
- X_VAL(ll) X_SEP \
- X_VAL(L) X_SEP \
- X_VAL(j) X_SEP \
- X_VAL(z) X_SEP \
- X_VAL(t) X_SEP \
- X_VAL(q)
-// clang-format on
-} // namespace
-
-const LengthMod::Spec LengthMod::kSpecs[] = {
-#define X_VAL(id) { LengthMod::id, #id, strlen(#id) }
-#define X_SEP ,
- ABSL_LENGTH_MODS_EXPAND_, {LengthMod::none, "", 0}
-#undef X_VAL
-#undef X_SEP
-};
-
-const ConversionChar::Spec ConversionChar::kSpecs[] = {
-#define X_VAL(id) { ConversionChar::id, #id[0] }
-#define X_SEP ,
- ABSL_CONVERSION_CHARS_EXPAND_(X_VAL, X_SEP),
- {ConversionChar::none, '\0'},
-#undef X_VAL
-#undef X_SEP
-};
std::string Flags::ToString() const {
std::string s;
@@ -63,22 +33,43 @@
return s;
}
-const size_t LengthMod::kNumValues;
+#define ABSL_INTERNAL_X_VAL(id) \
+ constexpr absl::FormatConversionChar FormatConversionCharInternal::id;
+ABSL_INTERNAL_CONVERSION_CHARS_EXPAND_(ABSL_INTERNAL_X_VAL, )
+#undef ABSL_INTERNAL_X_VAL
+// NOLINTNEXTLINE(readability-redundant-declaration)
+constexpr absl::FormatConversionChar FormatConversionCharInternal::kNone;
-const size_t ConversionChar::kNumValues;
+#define ABSL_INTERNAL_CHAR_SET_CASE(c) \
+ constexpr FormatConversionCharSet FormatConversionCharSetInternal::c;
+ABSL_INTERNAL_CONVERSION_CHARS_EXPAND_(ABSL_INTERNAL_CHAR_SET_CASE, )
+#undef ABSL_INTERNAL_CHAR_SET_CASE
-bool FormatSinkImpl::PutPaddedString(string_view v, int w, int p, bool l) {
+// NOLINTNEXTLINE(readability-redundant-declaration)
+constexpr FormatConversionCharSet FormatConversionCharSetInternal::kStar;
+// NOLINTNEXTLINE(readability-redundant-declaration)
+constexpr FormatConversionCharSet FormatConversionCharSetInternal::kIntegral;
+// NOLINTNEXTLINE(readability-redundant-declaration)
+constexpr FormatConversionCharSet FormatConversionCharSetInternal::kFloating;
+// NOLINTNEXTLINE(readability-redundant-declaration)
+constexpr FormatConversionCharSet FormatConversionCharSetInternal::kNumeric;
+// NOLINTNEXTLINE(readability-redundant-declaration)
+constexpr FormatConversionCharSet FormatConversionCharSetInternal::kPointer;
+
+bool FormatSinkImpl::PutPaddedString(string_view value, int width,
+ int precision, bool left) {
size_t space_remaining = 0;
- if (w >= 0) space_remaining = w;
- size_t n = v.size();
- if (p >= 0) n = std::min(n, static_cast<size_t>(p));
- string_view shown(v.data(), n);
+ if (width >= 0) space_remaining = width;
+ size_t n = value.size();
+ if (precision >= 0) n = std::min(n, static_cast<size_t>(precision));
+ string_view shown(value.data(), n);
space_remaining = Excess(shown.size(), space_remaining);
- if (!l) Append(space_remaining, ' ');
+ if (!left) Append(space_remaining, ' ');
Append(shown);
- if (l) Append(space_remaining, ' ');
+ if (left) Append(space_remaining, ' ');
return true;
}
} // namespace str_format_internal
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/internal/str_format/extension.h b/absl/strings/internal/str_format/extension.h
index eb81f8a..a9b9e13 100644
--- a/absl/strings/internal/str_format/extension.h
+++ b/absl/strings/internal/str_format/extension.h
@@ -17,17 +17,22 @@
#define ABSL_STRINGS_INTERNAL_STR_FORMAT_EXTENSION_H_
#include <limits.h>
+
#include <cstddef>
#include <cstring>
#include <ostream>
+#include "absl/base/config.h"
#include "absl/base/port.h"
+#include "absl/meta/type_traits.h"
#include "absl/strings/internal/str_format/output.h"
#include "absl/strings/string_view.h"
-class Cord;
-
namespace absl {
+ABSL_NAMESPACE_BEGIN
+
+enum class FormatConversionChar : uint8_t;
+enum class FormatConversionCharSet : uint64_t;
namespace str_format_internal {
@@ -102,7 +107,7 @@
size_t size() const { return size_; }
// Put 'v' to 'sink' with specified width, precision, and left flag.
- bool PutPaddedString(string_view v, int w, int p, bool l);
+ bool PutPaddedString(string_view v, int width, int precision, bool left);
template <typename T>
T Wrap() {
@@ -136,58 +141,10 @@
}
};
-struct LengthMod {
- public:
- enum Id : uint8_t {
- h, hh, l, ll, L, j, z, t, q, none
- };
- static const size_t kNumValues = none + 1;
-
- LengthMod() : id_(none) {}
-
- // Index into the opaque array of LengthMod enums.
- // Requires: i < kNumValues
- static LengthMod FromIndex(size_t i) {
- return LengthMod(kSpecs[i].value);
- }
-
- static LengthMod FromId(Id id) { return LengthMod(id); }
-
- // The length modifier std::string associated with a specified LengthMod.
- string_view name() const {
- const Spec& spec = kSpecs[id_];
- return {spec.name, spec.name_length};
- }
-
- Id id() const { return id_; }
-
- friend bool operator==(const LengthMod& a, const LengthMod& b) {
- return a.id() == b.id();
- }
- friend bool operator!=(const LengthMod& a, const LengthMod& b) {
- return !(a == b);
- }
- friend std::ostream& operator<<(std::ostream& os, const LengthMod& v) {
- return os << v.name();
- }
-
- private:
- struct Spec {
- Id value;
- const char *name;
- size_t name_length;
- };
- static const Spec kSpecs[];
-
- explicit LengthMod(Id id) : id_(id) {}
-
- Id id_;
-};
-
// clang-format off
-#define ABSL_CONVERSION_CHARS_EXPAND_(X_VAL, X_SEP) \
+#define ABSL_INTERNAL_CONVERSION_CHARS_EXPAND_(X_VAL, X_SEP) \
/* text */ \
- X_VAL(c) X_SEP X_VAL(C) X_SEP X_VAL(s) X_SEP X_VAL(S) X_SEP \
+ X_VAL(c) X_SEP X_VAL(s) X_SEP \
/* ints */ \
X_VAL(d) X_SEP X_VAL(i) X_SEP X_VAL(o) X_SEP \
X_VAL(u) X_SEP X_VAL(x) X_SEP X_VAL(X) X_SEP \
@@ -198,119 +155,119 @@
X_VAL(n) X_SEP X_VAL(p)
// clang-format on
-struct ConversionChar {
- public:
- enum Id : uint8_t {
- c, C, s, S, // text
+// This type should not be referenced, it exists only to provide labels
+// internally that match the values declared in FormatConversionChar in
+// str_format.h. This is meant to allow internal libraries to use the same
+// declared interface type as the public interface
+// (absl::StrFormatConversionChar) while keeping the definition in a public
+// header.
+// Internal libraries should use the form
+// `FormatConversionCharInternal::c`, `FormatConversionCharInternal::kNone` for
+// comparisons. Use in switch statements is not recommended due to a bug in how
+// gcc 4.9 -Wswitch handles declared but undefined enums.
+struct FormatConversionCharInternal {
+ FormatConversionCharInternal() = delete;
+
+ private:
+ // clang-format off
+ enum class Enum : uint8_t {
+ c, s, // text
d, i, o, u, x, X, // int
f, F, e, E, g, G, a, A, // float
n, p, // misc
- none
+ kNone
};
- static const size_t kNumValues = none + 1;
-
- ConversionChar() : id_(none) {}
-
+ // clang-format on
public:
- // Index into the opaque array of ConversionChar enums.
- // Requires: i < kNumValues
- static ConversionChar FromIndex(size_t i) {
- return ConversionChar(kSpecs[i].value);
- }
-
- static ConversionChar FromChar(char c) {
- ConversionChar::Id out_id = ConversionChar::none;
- switch (c) {
-#define X_VAL(id) \
- case #id[0]: \
- out_id = ConversionChar::id; \
- break;
- ABSL_CONVERSION_CHARS_EXPAND_(X_VAL, )
-#undef X_VAL
- default:
- break;
- }
- return ConversionChar(out_id);
- }
-
- static ConversionChar FromId(Id id) { return ConversionChar(id); }
- Id id() const { return id_; }
-
- int radix() const {
- switch (id()) {
- case x: case X: case a: case A: case p: return 16;
- case o: return 8;
- default: return 10;
- }
- }
-
- bool upper() const {
- switch (id()) {
- case X: case F: case E: case G: case A: return true;
- default: return false;
- }
- }
-
- bool is_signed() const {
- switch (id()) {
- case d: case i: return true;
- default: return false;
- }
- }
-
- bool is_integral() const {
- switch (id()) {
- case d: case i: case u: case o: case x: case X:
- return true;
- default: return false;
- }
- }
-
- bool is_float() const {
- switch (id()) {
- case a: case e: case f: case g: case A: case E: case F: case G:
- return true;
- default: return false;
- }
- }
-
- bool IsValid() const { return id() != none; }
-
- // The associated char.
- char Char() const { return kSpecs[id_].name; }
-
- friend bool operator==(const ConversionChar& a, const ConversionChar& b) {
- return a.id() == b.id();
- }
- friend bool operator!=(const ConversionChar& a, const ConversionChar& b) {
- return !(a == b);
- }
- friend std::ostream& operator<<(std::ostream& os, const ConversionChar& v) {
- char c = v.Char();
- if (!c) c = '?';
- return os << c;
- }
-
- private:
- struct Spec {
- Id value;
- char name;
- };
- static const Spec kSpecs[];
-
- explicit ConversionChar(Id id) : id_(id) {}
-
- Id id_;
+#define ABSL_INTERNAL_X_VAL(id) \
+ static constexpr FormatConversionChar id = \
+ static_cast<FormatConversionChar>(Enum::id);
+ ABSL_INTERNAL_CONVERSION_CHARS_EXPAND_(ABSL_INTERNAL_X_VAL, )
+#undef ABSL_INTERNAL_X_VAL
+ static constexpr FormatConversionChar kNone =
+ static_cast<FormatConversionChar>(Enum::kNone);
};
+// clang-format on
-class ConversionSpec {
+inline FormatConversionChar FormatConversionCharFromChar(char c) {
+ switch (c) {
+#define ABSL_INTERNAL_X_VAL(id) \
+ case #id[0]: \
+ return FormatConversionCharInternal::id;
+ ABSL_INTERNAL_CONVERSION_CHARS_EXPAND_(ABSL_INTERNAL_X_VAL, )
+#undef ABSL_INTERNAL_X_VAL
+ }
+ return FormatConversionCharInternal::kNone;
+}
+
+inline bool FormatConversionCharIsUpper(FormatConversionChar c) {
+ if (c == FormatConversionCharInternal::X ||
+ c == FormatConversionCharInternal::F ||
+ c == FormatConversionCharInternal::E ||
+ c == FormatConversionCharInternal::G ||
+ c == FormatConversionCharInternal::A) {
+ return true;
+ } else {
+ return false;
+ }
+}
+
+inline bool FormatConversionCharIsFloat(FormatConversionChar c) {
+ if (c == FormatConversionCharInternal::a ||
+ c == FormatConversionCharInternal::e ||
+ c == FormatConversionCharInternal::f ||
+ c == FormatConversionCharInternal::g ||
+ c == FormatConversionCharInternal::A ||
+ c == FormatConversionCharInternal::E ||
+ c == FormatConversionCharInternal::F ||
+ c == FormatConversionCharInternal::G) {
+ return true;
+ } else {
+ return false;
+ }
+}
+
+inline char FormatConversionCharToChar(FormatConversionChar c) {
+ if (c == FormatConversionCharInternal::kNone) {
+ return '\0';
+
+#define ABSL_INTERNAL_X_VAL(e) \
+ } else if (c == FormatConversionCharInternal::e) { \
+ return #e[0];
+#define ABSL_INTERNAL_X_SEP
+ ABSL_INTERNAL_CONVERSION_CHARS_EXPAND_(ABSL_INTERNAL_X_VAL,
+ ABSL_INTERNAL_X_SEP)
+ } else {
+ return '\0';
+ }
+
+#undef ABSL_INTERNAL_X_VAL
+#undef ABSL_INTERNAL_X_SEP
+}
+
+// The associated char.
+inline std::ostream& operator<<(std::ostream& os, FormatConversionChar v) {
+ char c = FormatConversionCharToChar(v);
+ if (!c) c = '?';
+ return os << c;
+}
+
+struct FormatConversionSpecImplFriend;
+
+class FormatConversionSpecImpl {
public:
- Flags flags() const { return flags_; }
- LengthMod length_mod() const { return length_mod_; }
- ConversionChar conv() const {
+ // Width and precison are not specified, no flags are set.
+ bool is_basic() const { return flags_.basic; }
+ bool has_left_flag() const { return flags_.left; }
+ bool has_show_pos_flag() const { return flags_.show_pos; }
+ bool has_sign_col_flag() const { return flags_.sign_col; }
+ bool has_alt_flag() const { return flags_.alt; }
+ bool has_zero_flag() const { return flags_.zero; }
+
+ FormatConversionChar conversion_char() const {
// Keep this field first in the struct . It generates better code when
// accessing it when ConversionSpec is passed by value in registers.
- static_assert(offsetof(ConversionSpec, conv_) == 0, "");
+ static_assert(offsetof(FormatConversionSpecImpl, conv_) == 0, "");
return conv_;
}
@@ -321,46 +278,36 @@
// negative value.
int precision() const { return precision_; }
- void set_flags(Flags f) { flags_ = f; }
- void set_length_mod(LengthMod lm) { length_mod_ = lm; }
- void set_conv(ConversionChar c) { conv_ = c; }
- void set_width(int w) { width_ = w; }
- void set_precision(int p) { precision_ = p; }
- void set_left(bool b) { flags_.left = b; }
+ template <typename T>
+ T Wrap() {
+ return T(*this);
+ }
private:
- ConversionChar conv_;
+ friend struct str_format_internal::FormatConversionSpecImplFriend;
+ FormatConversionChar conv_ = FormatConversionCharInternal::kNone;
Flags flags_;
- LengthMod length_mod_;
int width_;
int precision_;
};
-constexpr uint64_t ConversionCharToConvValue(char conv) {
- return
-#define CONV_SET_CASE(c) \
- conv == #c[0] ? (uint64_t{1} << (1 + ConversionChar::Id::c)):
- ABSL_CONVERSION_CHARS_EXPAND_(CONV_SET_CASE, )
-#undef CONV_SET_CASE
- conv == '*'
- ? 1
- : 0;
-}
-
-enum class Conv : uint64_t {
-#define CONV_SET_CASE(c) c = ConversionCharToConvValue(#c[0]),
- ABSL_CONVERSION_CHARS_EXPAND_(CONV_SET_CASE, )
-#undef CONV_SET_CASE
-
- // Used for width/precision '*' specification.
- star = ConversionCharToConvValue('*'),
-
- // Some predefined values:
- integral = d | i | u | o | x | X,
- floating = a | e | f | g | A | E | F | G,
- numeric = integral | floating,
- string = s,
- pointer = p
+struct FormatConversionSpecImplFriend final {
+ static void SetFlags(Flags f, FormatConversionSpecImpl* conv) {
+ conv->flags_ = f;
+ }
+ static void SetConversionChar(FormatConversionChar c,
+ FormatConversionSpecImpl* conv) {
+ conv->conv_ = c;
+ }
+ static void SetWidth(int w, FormatConversionSpecImpl* conv) {
+ conv->width_ = w;
+ }
+ static void SetPrecision(int p, FormatConversionSpecImpl* conv) {
+ conv->precision_ = p;
+ }
+ static std::string FlagsToString(const FormatConversionSpecImpl& spec) {
+ return spec.flags_.ToString();
+ }
};
// Type safe OR operator.
@@ -368,36 +315,104 @@
// 1. operator| on enums makes them decay to integers and the result is an
// integer. We need the result to stay as an enum.
// 2. We use "enum class" which would not work even if we accepted the decay.
-constexpr Conv operator|(Conv a, Conv b) {
- return Conv(static_cast<uint64_t>(a) | static_cast<uint64_t>(b));
+constexpr FormatConversionCharSet FormatConversionCharSetUnion(
+ FormatConversionCharSet a) {
+ return a;
+}
+
+template <typename... CharSet>
+constexpr FormatConversionCharSet FormatConversionCharSetUnion(
+ FormatConversionCharSet a, CharSet... rest) {
+ return static_cast<FormatConversionCharSet>(
+ static_cast<uint64_t>(a) |
+ static_cast<uint64_t>(FormatConversionCharSetUnion(rest...)));
+}
+
+constexpr uint64_t FormatConversionCharToConvInt(FormatConversionChar c) {
+ return uint64_t{1} << (1 + static_cast<uint8_t>(c));
+}
+
+constexpr uint64_t FormatConversionCharToConvInt(char conv) {
+ return
+#define ABSL_INTERNAL_CHAR_SET_CASE(c) \
+ conv == #c[0] \
+ ? FormatConversionCharToConvInt(FormatConversionCharInternal::c) \
+ :
+ ABSL_INTERNAL_CONVERSION_CHARS_EXPAND_(ABSL_INTERNAL_CHAR_SET_CASE, )
+#undef ABSL_INTERNAL_CHAR_SET_CASE
+ conv == '*'
+ ? 1
+ : 0;
+}
+
+constexpr FormatConversionCharSet FormatConversionCharToConvValue(char conv) {
+ return static_cast<FormatConversionCharSet>(
+ FormatConversionCharToConvInt(conv));
+}
+
+struct FormatConversionCharSetInternal {
+#define ABSL_INTERNAL_CHAR_SET_CASE(c) \
+ static constexpr FormatConversionCharSet c = \
+ FormatConversionCharToConvValue(#c[0]);
+ ABSL_INTERNAL_CONVERSION_CHARS_EXPAND_(ABSL_INTERNAL_CHAR_SET_CASE, )
+#undef ABSL_INTERNAL_CHAR_SET_CASE
+
+ // Used for width/precision '*' specification.
+ static constexpr FormatConversionCharSet kStar =
+ FormatConversionCharToConvValue('*');
+
+ static constexpr FormatConversionCharSet kIntegral =
+ FormatConversionCharSetUnion(d, i, u, o, x, X);
+ static constexpr FormatConversionCharSet kFloating =
+ FormatConversionCharSetUnion(a, e, f, g, A, E, F, G);
+ static constexpr FormatConversionCharSet kNumeric =
+ FormatConversionCharSetUnion(kIntegral, kFloating);
+ static constexpr FormatConversionCharSet kPointer = p;
+};
+
+// Type safe OR operator.
+// We need this for two reasons:
+// 1. operator| on enums makes them decay to integers and the result is an
+// integer. We need the result to stay as an enum.
+// 2. We use "enum class" which would not work even if we accepted the decay.
+constexpr FormatConversionCharSet operator|(FormatConversionCharSet a,
+ FormatConversionCharSet b) {
+ return FormatConversionCharSetUnion(a, b);
+}
+
+// Overloaded conversion functions to support absl::ParsedFormat.
+// Get a conversion with a single character in it.
+constexpr FormatConversionCharSet ToFormatConversionCharSet(char c) {
+ return static_cast<FormatConversionCharSet>(
+ FormatConversionCharToConvValue(c));
}
// Get a conversion with a single character in it.
-constexpr Conv ConversionCharToConv(char c) {
- return Conv(ConversionCharToConvValue(c));
+constexpr FormatConversionCharSet ToFormatConversionCharSet(
+ FormatConversionCharSet c) {
+ return c;
}
+template <typename T>
+void ToFormatConversionCharSet(T) = delete;
+
// Checks whether `c` exists in `set`.
-constexpr bool Contains(Conv set, char c) {
- return (static_cast<uint64_t>(set) & ConversionCharToConvValue(c)) != 0;
+constexpr bool Contains(FormatConversionCharSet set, char c) {
+ return (static_cast<uint64_t>(set) &
+ static_cast<uint64_t>(FormatConversionCharToConvValue(c))) != 0;
}
// Checks whether all the characters in `c` are contained in `set`
-constexpr bool Contains(Conv set, Conv c) {
+constexpr bool Contains(FormatConversionCharSet set,
+ FormatConversionCharSet c) {
return (static_cast<uint64_t>(set) & static_cast<uint64_t>(c)) ==
static_cast<uint64_t>(c);
}
-// Return type of the AbslFormatConvert() functions.
-// The Conv template parameter is used to inform the framework of what
-// conversion characters are supported by that AbslFormatConvert routine.
-template <Conv C>
-struct ConvertResult {
- static constexpr Conv kConv = C;
- bool value;
-};
-template <Conv C>
-constexpr Conv ConvertResult<C>::kConv;
+// Checks whether all the characters in `c` are contained in `set`
+constexpr bool Contains(FormatConversionCharSet set, FormatConversionChar c) {
+ return (static_cast<uint64_t>(set) & FormatConversionCharToConvInt(c)) != 0;
+}
// Return capacity - used, clipped to a minimum of 0.
inline size_t Excess(size_t used, size_t capacity) {
@@ -406,6 +421,7 @@
} // namespace str_format_internal
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_INTERNAL_STR_FORMAT_EXTENSION_H_
diff --git a/absl/strings/internal/str_format/extension_test.cc b/absl/strings/internal/str_format/extension_test.cc
index 4e23fef..1c93fdb 100644
--- a/absl/strings/internal/str_format/extension_test.cc
+++ b/absl/strings/internal/str_format/extension_test.cc
@@ -19,9 +19,26 @@
#include <random>
#include <string>
-#include "absl/strings/str_format.h"
-
#include "gtest/gtest.h"
+#include "absl/strings/str_format.h"
+#include "absl/strings/string_view.h"
+
+namespace my_namespace {
+class UserDefinedType {
+ public:
+ UserDefinedType() = default;
+
+ void Append(absl::string_view str) { value_.append(str.data(), str.size()); }
+ const std::string& Value() const { return value_; }
+
+ friend void AbslFormatFlush(UserDefinedType* x, absl::string_view str) {
+ x->Append(str);
+ }
+
+ private:
+ std::string value_;
+};
+} // namespace my_namespace
namespace {
@@ -63,4 +80,19 @@
EXPECT_EQ(actual, expected);
}
}
+
+TEST(FormatExtensionTest, VerifyEnumEquality) {
+#define X_VAL(id) \
+ EXPECT_EQ(absl::FormatConversionChar::id, \
+ absl::str_format_internal::FormatConversionCharInternal::id);
+ ABSL_INTERNAL_CONVERSION_CHARS_EXPAND_(X_VAL, );
+#undef X_VAL
+
+#define X_VAL(id) \
+ EXPECT_EQ(absl::FormatConversionCharSet::id, \
+ absl::str_format_internal::FormatConversionCharSetInternal::id);
+ ABSL_INTERNAL_CONVERSION_CHARS_EXPAND_(X_VAL, );
+#undef X_VAL
+}
+
} // namespace
diff --git a/absl/strings/internal/str_format/float_conversion.cc b/absl/strings/internal/str_format/float_conversion.cc
index 9236acd..2aa41aa 100644
--- a/absl/strings/internal/str_format/float_conversion.cc
+++ b/absl/strings/internal/str_format/float_conversion.cc
@@ -1,25 +1,938 @@
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
#include "absl/strings/internal/str_format/float_conversion.h"
#include <string.h>
+
#include <algorithm>
#include <cassert>
#include <cmath>
+#include <limits>
#include <string>
+#include "absl/base/attributes.h"
#include "absl/base/config.h"
+#include "absl/base/optimization.h"
+#include "absl/functional/function_ref.h"
+#include "absl/meta/type_traits.h"
+#include "absl/numeric/bits.h"
+#include "absl/numeric/int128.h"
+#include "absl/strings/numbers.h"
+#include "absl/types/optional.h"
+#include "absl/types/span.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace str_format_internal {
namespace {
-char *CopyStringTo(string_view v, char *out) {
+// The code below wants to avoid heap allocations.
+// To do so it needs to allocate memory on the stack.
+// `StackArray` will allocate memory on the stack in the form of a uint32_t
+// array and call the provided callback with said memory.
+// It will allocate memory in increments of 512 bytes. We could allocate the
+// largest needed unconditionally, but that is more than we need in most of
+// cases. This way we use less stack in the common cases.
+class StackArray {
+ using Func = absl::FunctionRef<void(absl::Span<uint32_t>)>;
+ static constexpr size_t kStep = 512 / sizeof(uint32_t);
+ // 5 steps is 2560 bytes, which is enough to hold a long double with the
+ // largest/smallest exponents.
+ // The operations below will static_assert their particular maximum.
+ static constexpr size_t kNumSteps = 5;
+
+ // We do not want this function to be inlined.
+ // Otherwise the caller will allocate the stack space unnecessarily for all
+ // the variants even though it only calls one.
+ template <size_t steps>
+ ABSL_ATTRIBUTE_NOINLINE static void RunWithCapacityImpl(Func f) {
+ uint32_t values[steps * kStep]{};
+ f(absl::MakeSpan(values));
+ }
+
+ public:
+ static constexpr size_t kMaxCapacity = kStep * kNumSteps;
+
+ static void RunWithCapacity(size_t capacity, Func f) {
+ assert(capacity <= kMaxCapacity);
+ const size_t step = (capacity + kStep - 1) / kStep;
+ assert(step <= kNumSteps);
+ switch (step) {
+ case 1:
+ return RunWithCapacityImpl<1>(f);
+ case 2:
+ return RunWithCapacityImpl<2>(f);
+ case 3:
+ return RunWithCapacityImpl<3>(f);
+ case 4:
+ return RunWithCapacityImpl<4>(f);
+ case 5:
+ return RunWithCapacityImpl<5>(f);
+ }
+
+ assert(false && "Invalid capacity");
+ }
+};
+
+// Calculates `10 * (*v) + carry` and stores the result in `*v` and returns
+// the carry.
+template <typename Int>
+inline Int MultiplyBy10WithCarry(Int *v, Int carry) {
+ using BiggerInt = absl::conditional_t<sizeof(Int) == 4, uint64_t, uint128>;
+ BiggerInt tmp = 10 * static_cast<BiggerInt>(*v) + carry;
+ *v = static_cast<Int>(tmp);
+ return static_cast<Int>(tmp >> (sizeof(Int) * 8));
+}
+
+// Calculates `(2^64 * carry + *v) / 10`.
+// Stores the quotient in `*v` and returns the remainder.
+// Requires: `0 <= carry <= 9`
+inline uint64_t DivideBy10WithCarry(uint64_t *v, uint64_t carry) {
+ constexpr uint64_t divisor = 10;
+ // 2^64 / divisor = chunk_quotient + chunk_remainder / divisor
+ constexpr uint64_t chunk_quotient = (uint64_t{1} << 63) / (divisor / 2);
+ constexpr uint64_t chunk_remainder = uint64_t{} - chunk_quotient * divisor;
+
+ const uint64_t mod = *v % divisor;
+ const uint64_t next_carry = chunk_remainder * carry + mod;
+ *v = *v / divisor + carry * chunk_quotient + next_carry / divisor;
+ return next_carry % divisor;
+}
+
+// Generates the decimal representation for an integer of the form `v * 2^exp`,
+// where `v` and `exp` are both positive integers.
+// It generates the digits from the left (ie the most significant digit first)
+// to allow for direct printing into the sink.
+//
+// Requires `0 <= exp` and `exp <= numeric_limits<long double>::max_exponent`.
+class BinaryToDecimal {
+ static constexpr int ChunksNeeded(int exp) {
+ // We will left shift a uint128 by `exp` bits, so we need `128+exp` total
+ // bits. Round up to 32.
+ // See constructor for details about adding `10%` to the value.
+ return (128 + exp + 31) / 32 * 11 / 10;
+ }
+
+ public:
+ // Run the conversion for `v * 2^exp` and call `f(binary_to_decimal)`.
+ // This function will allocate enough stack space to perform the conversion.
+ static void RunConversion(uint128 v, int exp,
+ absl::FunctionRef<void(BinaryToDecimal)> f) {
+ assert(exp > 0);
+ assert(exp <= std::numeric_limits<long double>::max_exponent);
+ static_assert(
+ static_cast<int>(StackArray::kMaxCapacity) >=
+ ChunksNeeded(std::numeric_limits<long double>::max_exponent),
+ "");
+
+ StackArray::RunWithCapacity(
+ ChunksNeeded(exp),
+ [=](absl::Span<uint32_t> input) { f(BinaryToDecimal(input, v, exp)); });
+ }
+
+ int TotalDigits() const {
+ return static_cast<int>((decimal_end_ - decimal_start_) * kDigitsPerChunk +
+ CurrentDigits().size());
+ }
+
+ // See the current block of digits.
+ absl::string_view CurrentDigits() const {
+ return absl::string_view(digits_ + kDigitsPerChunk - size_, size_);
+ }
+
+ // Advance the current view of digits.
+ // Returns `false` when no more digits are available.
+ bool AdvanceDigits() {
+ if (decimal_start_ >= decimal_end_) return false;
+
+ uint32_t w = data_[decimal_start_++];
+ for (size_ = 0; size_ < kDigitsPerChunk; w /= 10) {
+ digits_[kDigitsPerChunk - ++size_] = w % 10 + '0';
+ }
+ return true;
+ }
+
+ private:
+ BinaryToDecimal(absl::Span<uint32_t> data, uint128 v, int exp) : data_(data) {
+ // We need to print the digits directly into the sink object without
+ // buffering them all first. To do this we need two things:
+ // - to know the total number of digits to do padding when necessary
+ // - to generate the decimal digits from the left.
+ //
+ // In order to do this, we do a two pass conversion.
+ // On the first pass we convert the binary representation of the value into
+ // a decimal representation in which each uint32_t chunk holds up to 9
+ // decimal digits. In the second pass we take each decimal-holding-uint32_t
+ // value and generate the ascii decimal digits into `digits_`.
+ //
+ // The binary and decimal representations actually share the same memory
+ // region. As we go converting the chunks from binary to decimal we free
+ // them up and reuse them for the decimal representation. One caveat is that
+ // the decimal representation is around 7% less efficient in space than the
+ // binary one. We allocate an extra 10% memory to account for this. See
+ // ChunksNeeded for this calculation.
+ int chunk_index = exp / 32;
+ decimal_start_ = decimal_end_ = ChunksNeeded(exp);
+ const int offset = exp % 32;
+ // Left shift v by exp bits.
+ data_[chunk_index] = static_cast<uint32_t>(v << offset);
+ for (v >>= (32 - offset); v; v >>= 32)
+ data_[++chunk_index] = static_cast<uint32_t>(v);
+
+ while (chunk_index >= 0) {
+ // While we have more than one chunk available, go in steps of 1e9.
+ // `data_[chunk_index]` holds the highest non-zero binary chunk, so keep
+ // the variable updated.
+ uint32_t carry = 0;
+ for (int i = chunk_index; i >= 0; --i) {
+ uint64_t tmp = uint64_t{data_[i]} + (uint64_t{carry} << 32);
+ data_[i] = static_cast<uint32_t>(tmp / uint64_t{1000000000});
+ carry = static_cast<uint32_t>(tmp % uint64_t{1000000000});
+ }
+
+ // If the highest chunk is now empty, remove it from view.
+ if (data_[chunk_index] == 0) --chunk_index;
+
+ --decimal_start_;
+ assert(decimal_start_ != chunk_index);
+ data_[decimal_start_] = carry;
+ }
+
+ // Fill the first set of digits. The first chunk might not be complete, so
+ // handle differently.
+ for (uint32_t first = data_[decimal_start_++]; first != 0; first /= 10) {
+ digits_[kDigitsPerChunk - ++size_] = first % 10 + '0';
+ }
+ }
+
+ private:
+ static constexpr int kDigitsPerChunk = 9;
+
+ int decimal_start_;
+ int decimal_end_;
+
+ char digits_[kDigitsPerChunk];
+ int size_ = 0;
+
+ absl::Span<uint32_t> data_;
+};
+
+// Converts a value of the form `x * 2^-exp` into a sequence of decimal digits.
+// Requires `-exp < 0` and
+// `-exp >= limits<long double>::min_exponent - limits<long double>::digits`.
+class FractionalDigitGenerator {
+ public:
+ // Run the conversion for `v * 2^exp` and call `f(generator)`.
+ // This function will allocate enough stack space to perform the conversion.
+ static void RunConversion(
+ uint128 v, int exp, absl::FunctionRef<void(FractionalDigitGenerator)> f) {
+ using Limits = std::numeric_limits<long double>;
+ assert(-exp < 0);
+ assert(-exp >= Limits::min_exponent - 128);
+ static_assert(StackArray::kMaxCapacity >=
+ (Limits::digits + 128 - Limits::min_exponent + 31) / 32,
+ "");
+ StackArray::RunWithCapacity((Limits::digits + exp + 31) / 32,
+ [=](absl::Span<uint32_t> input) {
+ f(FractionalDigitGenerator(input, v, exp));
+ });
+ }
+
+ // Returns true if there are any more non-zero digits left.
+ bool HasMoreDigits() const { return next_digit_ != 0 || chunk_index_ >= 0; }
+
+ // Returns true if the remainder digits are greater than 5000...
+ bool IsGreaterThanHalf() const {
+ return next_digit_ > 5 || (next_digit_ == 5 && chunk_index_ >= 0);
+ }
+ // Returns true if the remainder digits are exactly 5000...
+ bool IsExactlyHalf() const { return next_digit_ == 5 && chunk_index_ < 0; }
+
+ struct Digits {
+ int digit_before_nine;
+ int num_nines;
+ };
+
+ // Get the next set of digits.
+ // They are composed by a non-9 digit followed by a runs of zero or more 9s.
+ Digits GetDigits() {
+ Digits digits{next_digit_, 0};
+
+ next_digit_ = GetOneDigit();
+ while (next_digit_ == 9) {
+ ++digits.num_nines;
+ next_digit_ = GetOneDigit();
+ }
+
+ return digits;
+ }
+
+ private:
+ // Return the next digit.
+ int GetOneDigit() {
+ if (chunk_index_ < 0) return 0;
+
+ uint32_t carry = 0;
+ for (int i = chunk_index_; i >= 0; --i) {
+ carry = MultiplyBy10WithCarry(&data_[i], carry);
+ }
+ // If the lowest chunk is now empty, remove it from view.
+ if (data_[chunk_index_] == 0) --chunk_index_;
+ return carry;
+ }
+
+ FractionalDigitGenerator(absl::Span<uint32_t> data, uint128 v, int exp)
+ : chunk_index_(exp / 32), data_(data) {
+ const int offset = exp % 32;
+ // Right shift `v` by `exp` bits.
+ data_[chunk_index_] = static_cast<uint32_t>(v << (32 - offset));
+ v >>= offset;
+ // Make sure we don't overflow the data. We already calculated that
+ // non-zero bits fit, so we might not have space for leading zero bits.
+ for (int pos = chunk_index_; v; v >>= 32)
+ data_[--pos] = static_cast<uint32_t>(v);
+
+ // Fill next_digit_, as GetDigits expects it to be populated always.
+ next_digit_ = GetOneDigit();
+ }
+
+ int next_digit_;
+ int chunk_index_;
+ absl::Span<uint32_t> data_;
+};
+
+// Count the number of leading zero bits.
+int LeadingZeros(uint64_t v) { return countl_zero(v); }
+int LeadingZeros(uint128 v) {
+ auto high = static_cast<uint64_t>(v >> 64);
+ auto low = static_cast<uint64_t>(v);
+ return high != 0 ? countl_zero(high) : 64 + countl_zero(low);
+}
+
+// Round up the text digits starting at `p`.
+// The buffer must have an extra digit that is known to not need rounding.
+// This is done below by having an extra '0' digit on the left.
+void RoundUp(char *p) {
+ while (*p == '9' || *p == '.') {
+ if (*p == '9') *p = '0';
+ --p;
+ }
+ ++*p;
+}
+
+// Check the previous digit and round up or down to follow the round-to-even
+// policy.
+void RoundToEven(char *p) {
+ if (*p == '.') --p;
+ if (*p % 2 == 1) RoundUp(p);
+}
+
+// Simple integral decimal digit printing for values that fit in 64-bits.
+// Returns the pointer to the last written digit.
+char *PrintIntegralDigitsFromRightFast(uint64_t v, char *p) {
+ do {
+ *--p = DivideBy10WithCarry(&v, 0) + '0';
+ } while (v != 0);
+ return p;
+}
+
+// Simple integral decimal digit printing for values that fit in 128-bits.
+// Returns the pointer to the last written digit.
+char *PrintIntegralDigitsFromRightFast(uint128 v, char *p) {
+ auto high = static_cast<uint64_t>(v >> 64);
+ auto low = static_cast<uint64_t>(v);
+
+ while (high != 0) {
+ uint64_t carry = DivideBy10WithCarry(&high, 0);
+ carry = DivideBy10WithCarry(&low, carry);
+ *--p = carry + '0';
+ }
+ return PrintIntegralDigitsFromRightFast(low, p);
+}
+
+// Simple fractional decimal digit printing for values that fir in 64-bits after
+// shifting.
+// Performs rounding if necessary to fit within `precision`.
+// Returns the pointer to one after the last character written.
+char *PrintFractionalDigitsFast(uint64_t v, char *start, int exp,
+ int precision) {
+ char *p = start;
+ v <<= (64 - exp);
+ while (precision > 0) {
+ if (!v) return p;
+ *p++ = MultiplyBy10WithCarry(&v, uint64_t{0}) + '0';
+ --precision;
+ }
+
+ // We need to round.
+ if (v < 0x8000000000000000) {
+ // We round down, so nothing to do.
+ } else if (v > 0x8000000000000000) {
+ // We round up.
+ RoundUp(p - 1);
+ } else {
+ RoundToEven(p - 1);
+ }
+
+ assert(precision == 0);
+ // Precision can only be zero here.
+ return p;
+}
+
+// Simple fractional decimal digit printing for values that fir in 128-bits
+// after shifting.
+// Performs rounding if necessary to fit within `precision`.
+// Returns the pointer to one after the last character written.
+char *PrintFractionalDigitsFast(uint128 v, char *start, int exp,
+ int precision) {
+ char *p = start;
+ v <<= (128 - exp);
+ auto high = static_cast<uint64_t>(v >> 64);
+ auto low = static_cast<uint64_t>(v);
+
+ // While we have digits to print and `low` is not empty, do the long
+ // multiplication.
+ while (precision > 0 && low != 0) {
+ uint64_t carry = MultiplyBy10WithCarry(&low, uint64_t{0});
+ carry = MultiplyBy10WithCarry(&high, carry);
+
+ *p++ = carry + '0';
+ --precision;
+ }
+
+ // Now `low` is empty, so use a faster approach for the rest of the digits.
+ // This block is pretty much the same as the main loop for the 64-bit case
+ // above.
+ while (precision > 0) {
+ if (!high) return p;
+ *p++ = MultiplyBy10WithCarry(&high, uint64_t{0}) + '0';
+ --precision;
+ }
+
+ // We need to round.
+ if (high < 0x8000000000000000) {
+ // We round down, so nothing to do.
+ } else if (high > 0x8000000000000000 || low != 0) {
+ // We round up.
+ RoundUp(p - 1);
+ } else {
+ RoundToEven(p - 1);
+ }
+
+ assert(precision == 0);
+ // Precision can only be zero here.
+ return p;
+}
+
+struct FormatState {
+ char sign_char;
+ int precision;
+ const FormatConversionSpecImpl &conv;
+ FormatSinkImpl *sink;
+
+ // In `alt` mode (flag #) we keep the `.` even if there are no fractional
+ // digits. In non-alt mode, we strip it.
+ bool ShouldPrintDot() const { return precision != 0 || conv.has_alt_flag(); }
+};
+
+struct Padding {
+ int left_spaces;
+ int zeros;
+ int right_spaces;
+};
+
+Padding ExtraWidthToPadding(size_t total_size, const FormatState &state) {
+ if (state.conv.width() < 0 ||
+ static_cast<size_t>(state.conv.width()) <= total_size) {
+ return {0, 0, 0};
+ }
+ int missing_chars = state.conv.width() - total_size;
+ if (state.conv.has_left_flag()) {
+ return {0, 0, missing_chars};
+ } else if (state.conv.has_zero_flag()) {
+ return {0, missing_chars, 0};
+ } else {
+ return {missing_chars, 0, 0};
+ }
+}
+
+void FinalPrint(const FormatState &state, absl::string_view data,
+ int padding_offset, int trailing_zeros,
+ absl::string_view data_postfix) {
+ if (state.conv.width() < 0) {
+ // No width specified. Fast-path.
+ if (state.sign_char != '\0') state.sink->Append(1, state.sign_char);
+ state.sink->Append(data);
+ state.sink->Append(trailing_zeros, '0');
+ state.sink->Append(data_postfix);
+ return;
+ }
+
+ auto padding = ExtraWidthToPadding((state.sign_char != '\0' ? 1 : 0) +
+ data.size() + data_postfix.size() +
+ static_cast<size_t>(trailing_zeros),
+ state);
+
+ state.sink->Append(padding.left_spaces, ' ');
+ if (state.sign_char != '\0') state.sink->Append(1, state.sign_char);
+ // Padding in general needs to be inserted somewhere in the middle of `data`.
+ state.sink->Append(data.substr(0, padding_offset));
+ state.sink->Append(padding.zeros, '0');
+ state.sink->Append(data.substr(padding_offset));
+ state.sink->Append(trailing_zeros, '0');
+ state.sink->Append(data_postfix);
+ state.sink->Append(padding.right_spaces, ' ');
+}
+
+// Fastpath %f formatter for when the shifted value fits in a simple integral
+// type.
+// Prints `v*2^exp` with the options from `state`.
+template <typename Int>
+void FormatFFast(Int v, int exp, const FormatState &state) {
+ constexpr int input_bits = sizeof(Int) * 8;
+
+ static constexpr size_t integral_size =
+ /* in case we need to round up an extra digit */ 1 +
+ /* decimal digits for uint128 */ 40 + 1;
+ char buffer[integral_size + /* . */ 1 + /* max digits uint128 */ 128];
+ buffer[integral_size] = '.';
+ char *const integral_digits_end = buffer + integral_size;
+ char *integral_digits_start;
+ char *const fractional_digits_start = buffer + integral_size + 1;
+ char *fractional_digits_end = fractional_digits_start;
+
+ if (exp >= 0) {
+ const int total_bits = input_bits - LeadingZeros(v) + exp;
+ integral_digits_start =
+ total_bits <= 64
+ ? PrintIntegralDigitsFromRightFast(static_cast<uint64_t>(v) << exp,
+ integral_digits_end)
+ : PrintIntegralDigitsFromRightFast(static_cast<uint128>(v) << exp,
+ integral_digits_end);
+ } else {
+ exp = -exp;
+
+ integral_digits_start = PrintIntegralDigitsFromRightFast(
+ exp < input_bits ? v >> exp : 0, integral_digits_end);
+ // PrintFractionalDigits may pull a carried 1 all the way up through the
+ // integral portion.
+ integral_digits_start[-1] = '0';
+
+ fractional_digits_end =
+ exp <= 64 ? PrintFractionalDigitsFast(v, fractional_digits_start, exp,
+ state.precision)
+ : PrintFractionalDigitsFast(static_cast<uint128>(v),
+ fractional_digits_start, exp,
+ state.precision);
+ // There was a carry, so include the first digit too.
+ if (integral_digits_start[-1] != '0') --integral_digits_start;
+ }
+
+ size_t size = fractional_digits_end - integral_digits_start;
+
+ // In `alt` mode (flag #) we keep the `.` even if there are no fractional
+ // digits. In non-alt mode, we strip it.
+ if (!state.ShouldPrintDot()) --size;
+ FinalPrint(state, absl::string_view(integral_digits_start, size),
+ /*padding_offset=*/0,
+ static_cast<int>(state.precision - (fractional_digits_end -
+ fractional_digits_start)),
+ /*data_postfix=*/"");
+}
+
+// Slow %f formatter for when the shifted value does not fit in a uint128, and
+// `exp > 0`.
+// Prints `v*2^exp` with the options from `state`.
+// This one is guaranteed to not have fractional digits, so we don't have to
+// worry about anything after the `.`.
+void FormatFPositiveExpSlow(uint128 v, int exp, const FormatState &state) {
+ BinaryToDecimal::RunConversion(v, exp, [&](BinaryToDecimal btd) {
+ const size_t total_digits =
+ btd.TotalDigits() +
+ (state.ShouldPrintDot() ? static_cast<size_t>(state.precision) + 1 : 0);
+
+ const auto padding = ExtraWidthToPadding(
+ total_digits + (state.sign_char != '\0' ? 1 : 0), state);
+
+ state.sink->Append(padding.left_spaces, ' ');
+ if (state.sign_char != '\0') state.sink->Append(1, state.sign_char);
+ state.sink->Append(padding.zeros, '0');
+
+ do {
+ state.sink->Append(btd.CurrentDigits());
+ } while (btd.AdvanceDigits());
+
+ if (state.ShouldPrintDot()) state.sink->Append(1, '.');
+ state.sink->Append(state.precision, '0');
+ state.sink->Append(padding.right_spaces, ' ');
+ });
+}
+
+// Slow %f formatter for when the shifted value does not fit in a uint128, and
+// `exp < 0`.
+// Prints `v*2^exp` with the options from `state`.
+// This one is guaranteed to be < 1.0, so we don't have to worry about integral
+// digits.
+void FormatFNegativeExpSlow(uint128 v, int exp, const FormatState &state) {
+ const size_t total_digits =
+ /* 0 */ 1 +
+ (state.ShouldPrintDot() ? static_cast<size_t>(state.precision) + 1 : 0);
+ auto padding =
+ ExtraWidthToPadding(total_digits + (state.sign_char ? 1 : 0), state);
+ padding.zeros += 1;
+ state.sink->Append(padding.left_spaces, ' ');
+ if (state.sign_char != '\0') state.sink->Append(1, state.sign_char);
+ state.sink->Append(padding.zeros, '0');
+
+ if (state.ShouldPrintDot()) state.sink->Append(1, '.');
+
+ // Print digits
+ int digits_to_go = state.precision;
+
+ FractionalDigitGenerator::RunConversion(
+ v, exp, [&](FractionalDigitGenerator digit_gen) {
+ // There are no digits to print here.
+ if (state.precision == 0) return;
+
+ // We go one digit at a time, while keeping track of runs of nines.
+ // The runs of nines are used to perform rounding when necessary.
+
+ while (digits_to_go > 0 && digit_gen.HasMoreDigits()) {
+ auto digits = digit_gen.GetDigits();
+
+ // Now we have a digit and a run of nines.
+ // See if we can print them all.
+ if (digits.num_nines + 1 < digits_to_go) {
+ // We don't have to round yet, so print them.
+ state.sink->Append(1, digits.digit_before_nine + '0');
+ state.sink->Append(digits.num_nines, '9');
+ digits_to_go -= digits.num_nines + 1;
+
+ } else {
+ // We can't print all the nines, see where we have to truncate.
+
+ bool round_up = false;
+ if (digits.num_nines + 1 > digits_to_go) {
+ // We round up at a nine. No need to print them.
+ round_up = true;
+ } else {
+ // We can fit all the nines, but truncate just after it.
+ if (digit_gen.IsGreaterThanHalf()) {
+ round_up = true;
+ } else if (digit_gen.IsExactlyHalf()) {
+ // Round to even
+ round_up =
+ digits.num_nines != 0 || digits.digit_before_nine % 2 == 1;
+ }
+ }
+
+ if (round_up) {
+ state.sink->Append(1, digits.digit_before_nine + '1');
+ --digits_to_go;
+ // The rest will be zeros.
+ } else {
+ state.sink->Append(1, digits.digit_before_nine + '0');
+ state.sink->Append(digits_to_go - 1, '9');
+ digits_to_go = 0;
+ }
+ return;
+ }
+ }
+ });
+
+ state.sink->Append(digits_to_go, '0');
+ state.sink->Append(padding.right_spaces, ' ');
+}
+
+template <typename Int>
+void FormatF(Int mantissa, int exp, const FormatState &state) {
+ if (exp >= 0) {
+ const int total_bits = sizeof(Int) * 8 - LeadingZeros(mantissa) + exp;
+
+ // Fallback to the slow stack-based approach if we can't do it in a 64 or
+ // 128 bit state.
+ if (ABSL_PREDICT_FALSE(total_bits > 128)) {
+ return FormatFPositiveExpSlow(mantissa, exp, state);
+ }
+ } else {
+ // Fallback to the slow stack-based approach if we can't do it in a 64 or
+ // 128 bit state.
+ if (ABSL_PREDICT_FALSE(exp < -128)) {
+ return FormatFNegativeExpSlow(mantissa, -exp, state);
+ }
+ }
+ return FormatFFast(mantissa, exp, state);
+}
+
+// Grab the group of four bits (nibble) from `n`. E.g., nibble 1 corresponds to
+// bits 4-7.
+template <typename Int>
+uint8_t GetNibble(Int n, int nibble_index) {
+ constexpr Int mask_low_nibble = Int{0xf};
+ int shift = nibble_index * 4;
+ n &= mask_low_nibble << shift;
+ return static_cast<uint8_t>((n >> shift) & 0xf);
+}
+
+// Add one to the given nibble, applying carry to higher nibbles. Returns true
+// if overflow, false otherwise.
+template <typename Int>
+bool IncrementNibble(int nibble_index, Int *n) {
+ constexpr int kShift = sizeof(Int) * 8 - 1;
+ constexpr int kNumNibbles = sizeof(Int) * 8 / 4;
+ Int before = *n >> kShift;
+ // Here we essentially want to take the number 1 and move it into the requsted
+ // nibble, then add it to *n to effectively increment the nibble. However,
+ // ASan will complain if we try to shift the 1 beyond the limits of the Int,
+ // i.e., if the nibble_index is out of range. So therefore we check for this
+ // and if we are out of range we just add 0 which leaves *n unchanged, which
+ // seems like the reasonable thing to do in that case.
+ *n += ((nibble_index >= kNumNibbles) ? 0 : (Int{1} << (nibble_index * 4)));
+ Int after = *n >> kShift;
+ return (before && !after) || (nibble_index >= kNumNibbles);
+}
+
+// Return a mask with 1's in the given nibble and all lower nibbles.
+template <typename Int>
+Int MaskUpToNibbleInclusive(int nibble_index) {
+ constexpr int kNumNibbles = sizeof(Int) * 8 / 4;
+ static const Int ones = ~Int{0};
+ return ones >> std::max(0, 4 * (kNumNibbles - nibble_index - 1));
+}
+
+// Return a mask with 1's below the given nibble.
+template <typename Int>
+Int MaskUpToNibbleExclusive(int nibble_index) {
+ return nibble_index <= 0 ? 0 : MaskUpToNibbleInclusive<Int>(nibble_index - 1);
+}
+
+template <typename Int>
+Int MoveToNibble(uint8_t nibble, int nibble_index) {
+ return Int{nibble} << (4 * nibble_index);
+}
+
+// Given mantissa size, find optimal # of mantissa bits to put in initial digit.
+//
+// In the hex representation we keep a single hex digit to the left of the dot.
+// However, the question as to how many bits of the mantissa should be put into
+// that hex digit in theory is arbitrary, but in practice it is optimal to
+// choose based on the size of the mantissa. E.g., for a `double`, there are 53
+// mantissa bits, so that means that we should put 1 bit to the left of the dot,
+// thereby leaving 52 bits to the right, which is evenly divisible by four and
+// thus all fractional digits represent actual precision. For a `long double`,
+// on the other hand, there are 64 bits of mantissa, thus we can use all four
+// bits for the initial hex digit and still have a number left over (60) that is
+// a multiple of four. Once again, the goal is to have all fractional digits
+// represent real precision.
+template <typename Float>
+constexpr int HexFloatLeadingDigitSizeInBits() {
+ return std::numeric_limits<Float>::digits % 4 > 0
+ ? std::numeric_limits<Float>::digits % 4
+ : 4;
+}
+
+// This function captures the rounding behavior of glibc for hex float
+// representations. E.g. when rounding 0x1.ab800000 to a precision of .2
+// ("%.2a") glibc will round up because it rounds toward the even number (since
+// 0xb is an odd number, it will round up to 0xc). However, when rounding at a
+// point that is not followed by 800000..., it disregards the parity and rounds
+// up if > 8 and rounds down if < 8.
+template <typename Int>
+bool HexFloatNeedsRoundUp(Int mantissa, int final_nibble_displayed,
+ uint8_t leading) {
+ // If the last nibble (hex digit) to be displayed is the lowest on in the
+ // mantissa then that means that we don't have any further nibbles to inform
+ // rounding, so don't round.
+ if (final_nibble_displayed <= 0) {
+ return false;
+ }
+ int rounding_nibble_idx = final_nibble_displayed - 1;
+ constexpr int kTotalNibbles = sizeof(Int) * 8 / 4;
+ assert(final_nibble_displayed <= kTotalNibbles);
+ Int mantissa_up_to_rounding_nibble_inclusive =
+ mantissa & MaskUpToNibbleInclusive<Int>(rounding_nibble_idx);
+ Int eight = MoveToNibble<Int>(8, rounding_nibble_idx);
+ if (mantissa_up_to_rounding_nibble_inclusive != eight) {
+ return mantissa_up_to_rounding_nibble_inclusive > eight;
+ }
+ // Nibble in question == 8.
+ uint8_t round_if_odd = (final_nibble_displayed == kTotalNibbles)
+ ? leading
+ : GetNibble(mantissa, final_nibble_displayed);
+ return round_if_odd % 2 == 1;
+}
+
+// Stores values associated with a Float type needed by the FormatA
+// implementation in order to avoid templatizing that function by the Float
+// type.
+struct HexFloatTypeParams {
+ template <typename Float>
+ explicit HexFloatTypeParams(Float)
+ : min_exponent(std::numeric_limits<Float>::min_exponent - 1),
+ leading_digit_size_bits(HexFloatLeadingDigitSizeInBits<Float>()) {
+ assert(leading_digit_size_bits >= 1 && leading_digit_size_bits <= 4);
+ }
+
+ int min_exponent;
+ int leading_digit_size_bits;
+};
+
+// Hex Float Rounding. First check if we need to round; if so, then we do that
+// by manipulating (incrementing) the mantissa, that way we can later print the
+// mantissa digits by iterating through them in the same way regardless of
+// whether a rounding happened.
+template <typename Int>
+void FormatARound(bool precision_specified, const FormatState &state,
+ uint8_t *leading, Int *mantissa, int *exp) {
+ constexpr int kTotalNibbles = sizeof(Int) * 8 / 4;
+ // Index of the last nibble that we could display given precision.
+ int final_nibble_displayed =
+ precision_specified ? std::max(0, (kTotalNibbles - state.precision)) : 0;
+ if (HexFloatNeedsRoundUp(*mantissa, final_nibble_displayed, *leading)) {
+ // Need to round up.
+ bool overflow = IncrementNibble(final_nibble_displayed, mantissa);
+ *leading += (overflow ? 1 : 0);
+ if (ABSL_PREDICT_FALSE(*leading > 15)) {
+ // We have overflowed the leading digit. This would mean that we would
+ // need two hex digits to the left of the dot, which is not allowed. So
+ // adjust the mantissa and exponent so that the result is always 1.0eXXX.
+ *leading = 1;
+ *mantissa = 0;
+ *exp += 4;
+ }
+ }
+ // Now that we have handled a possible round-up we can go ahead and zero out
+ // all the nibbles of the mantissa that we won't need.
+ if (precision_specified) {
+ *mantissa &= ~MaskUpToNibbleExclusive<Int>(final_nibble_displayed);
+ }
+}
+
+template <typename Int>
+void FormatANormalize(const HexFloatTypeParams float_traits, uint8_t *leading,
+ Int *mantissa, int *exp) {
+ constexpr int kIntBits = sizeof(Int) * 8;
+ static const Int kHighIntBit = Int{1} << (kIntBits - 1);
+ const int kLeadDigitBitsCount = float_traits.leading_digit_size_bits;
+ // Normalize mantissa so that highest bit set is in MSB position, unless we
+ // get interrupted by the exponent threshold.
+ while (*mantissa && !(*mantissa & kHighIntBit)) {
+ if (ABSL_PREDICT_FALSE(*exp - 1 < float_traits.min_exponent)) {
+ *mantissa >>= (float_traits.min_exponent - *exp);
+ *exp = float_traits.min_exponent;
+ return;
+ }
+ *mantissa <<= 1;
+ --*exp;
+ }
+ // Extract bits for leading digit then shift them away leaving the
+ // fractional part.
+ *leading =
+ static_cast<uint8_t>(*mantissa >> (kIntBits - kLeadDigitBitsCount));
+ *exp -= (*mantissa != 0) ? kLeadDigitBitsCount : *exp;
+ *mantissa <<= kLeadDigitBitsCount;
+}
+
+template <typename Int>
+void FormatA(const HexFloatTypeParams float_traits, Int mantissa, int exp,
+ bool uppercase, const FormatState &state) {
+ // Int properties.
+ constexpr int kIntBits = sizeof(Int) * 8;
+ constexpr int kTotalNibbles = sizeof(Int) * 8 / 4;
+ // Did the user specify a precision explicitly?
+ const bool precision_specified = state.conv.precision() >= 0;
+
+ // ========== Normalize/Denormalize ==========
+ exp += kIntBits; // make all digits fractional digits.
+ // This holds the (up to four) bits of leading digit, i.e., the '1' in the
+ // number 0x1.e6fp+2. It's always > 0 unless number is zero or denormal.
+ uint8_t leading = 0;
+ FormatANormalize(float_traits, &leading, &mantissa, &exp);
+
+ // =============== Rounding ==================
+ // Check if we need to round; if so, then we do that by manipulating
+ // (incrementing) the mantissa before beginning to print characters.
+ FormatARound(precision_specified, state, &leading, &mantissa, &exp);
+
+ // ============= Format Result ===============
+ // This buffer holds the "0x1.ab1de3" portion of "0x1.ab1de3pe+2". Compute the
+ // size with long double which is the largest of the floats.
+ constexpr size_t kBufSizeForHexFloatRepr =
+ 2 // 0x
+ + std::numeric_limits<long double>::digits / 4 // number of hex digits
+ + 1 // round up
+ + 1; // "." (dot)
+ char digits_buffer[kBufSizeForHexFloatRepr];
+ char *digits_iter = digits_buffer;
+ const char *const digits =
+ static_cast<const char *>("0123456789ABCDEF0123456789abcdef") +
+ (uppercase ? 0 : 16);
+
+ // =============== Hex Prefix ================
+ *digits_iter++ = '0';
+ *digits_iter++ = uppercase ? 'X' : 'x';
+
+ // ========== Non-Fractional Digit ===========
+ *digits_iter++ = digits[leading];
+
+ // ================== Dot ====================
+ // There are three reasons we might need a dot. Keep in mind that, at this
+ // point, the mantissa holds only the fractional part.
+ if ((precision_specified && state.precision > 0) ||
+ (!precision_specified && mantissa > 0) || state.conv.has_alt_flag()) {
+ *digits_iter++ = '.';
+ }
+
+ // ============ Fractional Digits ============
+ int digits_emitted = 0;
+ while (mantissa > 0) {
+ *digits_iter++ = digits[GetNibble(mantissa, kTotalNibbles - 1)];
+ mantissa <<= 4;
+ ++digits_emitted;
+ }
+ int trailing_zeros =
+ precision_specified ? state.precision - digits_emitted : 0;
+ assert(trailing_zeros >= 0);
+ auto digits_result = string_view(digits_buffer, digits_iter - digits_buffer);
+
+ // =============== Exponent ==================
+ constexpr size_t kBufSizeForExpDecRepr =
+ numbers_internal::kFastToBufferSize // requred for FastIntToBuffer
+ + 1 // 'p' or 'P'
+ + 1; // '+' or '-'
+ char exp_buffer[kBufSizeForExpDecRepr];
+ exp_buffer[0] = uppercase ? 'P' : 'p';
+ exp_buffer[1] = exp >= 0 ? '+' : '-';
+ numbers_internal::FastIntToBuffer(exp < 0 ? -exp : exp, exp_buffer + 2);
+
+ // ============ Assemble Result ==============
+ FinalPrint(state, //
+ digits_result, // 0xN.NNN...
+ 2, // offset in `data` to start padding if needed.
+ trailing_zeros, // num remaining mantissa padding zeros
+ exp_buffer); // exponent
+}
+
+char *CopyStringTo(absl::string_view v, char *out) {
std::memcpy(out, v.data(), v.size());
return out + v.size();
}
template <typename Float>
-bool FallbackToSnprintf(const Float v, const ConversionSpec &conv,
+bool FallbackToSnprintf(const Float v, const FormatConversionSpecImpl &conv,
FormatSinkImpl *sink) {
int w = conv.width() >= 0 ? conv.width() : 0;
int p = conv.precision() >= 0 ? conv.precision() : -1;
@@ -27,22 +940,22 @@
{
char *fp = fmt;
*fp++ = '%';
- fp = CopyStringTo(conv.flags().ToString(), fp);
+ fp = CopyStringTo(FormatConversionSpecImplFriend::FlagsToString(conv), fp);
fp = CopyStringTo("*.*", fp);
if (std::is_same<long double, Float>()) {
*fp++ = 'L';
}
- *fp++ = conv.conv().Char();
+ *fp++ = FormatConversionCharToChar(conv.conversion_char());
*fp = 0;
assert(fp < fmt + sizeof(fmt));
}
std::string space(512, '\0');
- string_view result;
+ absl::string_view result;
while (true) {
int n = snprintf(&space[0], space.size(), fmt, w, p, v);
if (n < 0) return false;
if (static_cast<size_t>(n) < space.size()) {
- result = string_view(space.data(), n);
+ result = absl::string_view(space.data(), n);
break;
}
space.resize(n + 1);
@@ -95,19 +1008,24 @@
// Otherwise, return false.
template <typename Float>
bool ConvertNonNumericFloats(char sign_char, Float v,
- const ConversionSpec &conv, FormatSinkImpl *sink) {
+ const FormatConversionSpecImpl &conv,
+ FormatSinkImpl *sink) {
char text[4], *ptr = text;
- if (sign_char) *ptr++ = sign_char;
+ if (sign_char != '\0') *ptr++ = sign_char;
if (std::isnan(v)) {
- ptr = std::copy_n(conv.conv().upper() ? "NAN" : "nan", 3, ptr);
+ ptr = std::copy_n(
+ FormatConversionCharIsUpper(conv.conversion_char()) ? "NAN" : "nan", 3,
+ ptr);
} else if (std::isinf(v)) {
- ptr = std::copy_n(conv.conv().upper() ? "INF" : "inf", 3, ptr);
+ ptr = std::copy_n(
+ FormatConversionCharIsUpper(conv.conversion_char()) ? "INF" : "inf", 3,
+ ptr);
} else {
return false;
}
return sink->PutPaddedString(string_view(text, ptr - text), conv.width(), -1,
- conv.flags().left);
+ conv.has_left_flag());
}
// Round up the last digit of the value.
@@ -167,7 +1085,12 @@
template <typename Float>
struct Decomposed {
- Float mantissa;
+ using MantissaType =
+ absl::conditional_t<std::is_same<long double, Float>::value, uint128,
+ uint64_t>;
+ static_assert(std::numeric_limits<Float>::digits <= sizeof(MantissaType) * 8,
+ "");
+ MantissaType mantissa;
int exponent;
};
@@ -178,7 +1101,8 @@
Float m = std::frexp(v, &exp);
m = std::ldexp(m, std::numeric_limits<Float>::digits);
exp -= std::numeric_limits<Float>::digits;
- return {m, exp};
+
+ return {static_cast<typename Decomposed<Float>::MantissaType>(m), exp};
}
// Print 'digits' as decimal.
@@ -347,31 +1271,32 @@
return false;
}
-void WriteBufferToSink(char sign_char, string_view str,
- const ConversionSpec &conv, FormatSinkImpl *sink) {
+void WriteBufferToSink(char sign_char, absl::string_view str,
+ const FormatConversionSpecImpl &conv,
+ FormatSinkImpl *sink) {
int left_spaces = 0, zeros = 0, right_spaces = 0;
int missing_chars =
conv.width() >= 0 ? std::max(conv.width() - static_cast<int>(str.size()) -
static_cast<int>(sign_char != 0),
0)
: 0;
- if (conv.flags().left) {
+ if (conv.has_left_flag()) {
right_spaces = missing_chars;
- } else if (conv.flags().zero) {
+ } else if (conv.has_zero_flag()) {
zeros = missing_chars;
} else {
left_spaces = missing_chars;
}
sink->Append(left_spaces, ' ');
- if (sign_char) sink->Append(1, sign_char);
+ if (sign_char != '\0') sink->Append(1, sign_char);
sink->Append(zeros, '0');
sink->Append(str);
sink->Append(right_spaces, ' ');
}
template <typename Float>
-bool FloatToSink(const Float v, const ConversionSpec &conv,
+bool FloatToSink(const Float v, const FormatConversionSpecImpl &conv,
FormatSinkImpl *sink) {
// Print the sign or the sign column.
Float abs_v = v;
@@ -379,9 +1304,9 @@
if (std::signbit(abs_v)) {
sign_char = '-';
abs_v = -abs_v;
- } else if (conv.flags().show_pos) {
+ } else if (conv.has_show_pos_flag()) {
sign_char = '+';
- } else if (conv.flags().sign_col) {
+ } else if (conv.has_sign_col_flag()) {
sign_char = ' ';
}
@@ -398,88 +1323,96 @@
Buffer buffer;
- switch (conv.conv().id()) {
- case ConversionChar::f:
- case ConversionChar::F:
- if (!FloatToBuffer<FormatStyle::Fixed>(decomposed, precision, &buffer,
- nullptr)) {
- return FallbackToSnprintf(v, conv, sink);
- }
- if (!conv.flags().alt && buffer.back() == '.') buffer.pop_back();
- break;
+ FormatConversionChar c = conv.conversion_char();
- case ConversionChar::e:
- case ConversionChar::E:
- if (!FloatToBuffer<FormatStyle::Precision>(decomposed, precision, &buffer,
- &exp)) {
- return FallbackToSnprintf(v, conv, sink);
- }
- if (!conv.flags().alt && buffer.back() == '.') buffer.pop_back();
- PrintExponent(exp, conv.conv().upper() ? 'E' : 'e', &buffer);
- break;
-
- case ConversionChar::g:
- case ConversionChar::G:
- precision = std::max(0, precision - 1);
- if (!FloatToBuffer<FormatStyle::Precision>(decomposed, precision, &buffer,
- &exp)) {
- return FallbackToSnprintf(v, conv, sink);
- }
- if (precision + 1 > exp && exp >= -4) {
- if (exp < 0) {
- // Have 1.23456, needs 0.00123456
- // Move the first digit
- buffer.begin[1] = *buffer.begin;
- // Add some zeros
- for (; exp < -1; ++exp) *buffer.begin-- = '0';
- *buffer.begin-- = '.';
- *buffer.begin = '0';
- } else if (exp > 0) {
- // Have 1.23456, needs 1234.56
- // Move the '.' exp positions to the right.
- std::rotate(buffer.begin + 1, buffer.begin + 2,
- buffer.begin + exp + 2);
- }
- exp = 0;
- }
- if (!conv.flags().alt) {
- while (buffer.back() == '0') buffer.pop_back();
- if (buffer.back() == '.') buffer.pop_back();
- }
- if (exp) PrintExponent(exp, conv.conv().upper() ? 'E' : 'e', &buffer);
- break;
-
- case ConversionChar::a:
- case ConversionChar::A:
+ if (c == FormatConversionCharInternal::f ||
+ c == FormatConversionCharInternal::F) {
+ FormatF(decomposed.mantissa, decomposed.exponent,
+ {sign_char, precision, conv, sink});
+ return true;
+ } else if (c == FormatConversionCharInternal::e ||
+ c == FormatConversionCharInternal::E) {
+ if (!FloatToBuffer<FormatStyle::Precision>(decomposed, precision, &buffer,
+ &exp)) {
return FallbackToSnprintf(v, conv, sink);
-
- default:
- return false;
+ }
+ if (!conv.has_alt_flag() && buffer.back() == '.') buffer.pop_back();
+ PrintExponent(
+ exp, FormatConversionCharIsUpper(conv.conversion_char()) ? 'E' : 'e',
+ &buffer);
+ } else if (c == FormatConversionCharInternal::g ||
+ c == FormatConversionCharInternal::G) {
+ precision = std::max(0, precision - 1);
+ if (!FloatToBuffer<FormatStyle::Precision>(decomposed, precision, &buffer,
+ &exp)) {
+ return FallbackToSnprintf(v, conv, sink);
+ }
+ if (precision + 1 > exp && exp >= -4) {
+ if (exp < 0) {
+ // Have 1.23456, needs 0.00123456
+ // Move the first digit
+ buffer.begin[1] = *buffer.begin;
+ // Add some zeros
+ for (; exp < -1; ++exp) *buffer.begin-- = '0';
+ *buffer.begin-- = '.';
+ *buffer.begin = '0';
+ } else if (exp > 0) {
+ // Have 1.23456, needs 1234.56
+ // Move the '.' exp positions to the right.
+ std::rotate(buffer.begin + 1, buffer.begin + 2, buffer.begin + exp + 2);
+ }
+ exp = 0;
+ }
+ if (!conv.has_alt_flag()) {
+ while (buffer.back() == '0') buffer.pop_back();
+ if (buffer.back() == '.') buffer.pop_back();
+ }
+ if (exp) {
+ PrintExponent(
+ exp, FormatConversionCharIsUpper(conv.conversion_char()) ? 'E' : 'e',
+ &buffer);
+ }
+ } else if (c == FormatConversionCharInternal::a ||
+ c == FormatConversionCharInternal::A) {
+ bool uppercase = (c == FormatConversionCharInternal::A);
+ FormatA(HexFloatTypeParams(Float{}), decomposed.mantissa,
+ decomposed.exponent, uppercase, {sign_char, precision, conv, sink});
+ return true;
+ } else {
+ return false;
}
WriteBufferToSink(sign_char,
- string_view(buffer.begin, buffer.end - buffer.begin), conv,
- sink);
+ absl::string_view(buffer.begin, buffer.end - buffer.begin),
+ conv, sink);
return true;
}
} // namespace
-bool ConvertFloatImpl(long double v, const ConversionSpec &conv,
+bool ConvertFloatImpl(long double v, const FormatConversionSpecImpl &conv,
FormatSinkImpl *sink) {
+ if (std::numeric_limits<long double>::digits ==
+ 2 * std::numeric_limits<double>::digits) {
+ // This is the `double-double` representation of `long double`.
+ // We do not handle it natively. Fallback to snprintf.
+ return FallbackToSnprintf(v, conv, sink);
+ }
+
return FloatToSink(v, conv, sink);
}
-bool ConvertFloatImpl(float v, const ConversionSpec &conv,
+bool ConvertFloatImpl(float v, const FormatConversionSpecImpl &conv,
FormatSinkImpl *sink) {
- return FloatToSink(v, conv, sink);
+ return FloatToSink(static_cast<double>(v), conv, sink);
}
-bool ConvertFloatImpl(double v, const ConversionSpec &conv,
+bool ConvertFloatImpl(double v, const FormatConversionSpecImpl &conv,
FormatSinkImpl *sink) {
return FloatToSink(v, conv, sink);
}
} // namespace str_format_internal
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/internal/str_format/float_conversion.h b/absl/strings/internal/str_format/float_conversion.h
index 8ba5566..71100e7 100644
--- a/absl/strings/internal/str_format/float_conversion.h
+++ b/absl/strings/internal/str_format/float_conversion.h
@@ -1,21 +1,37 @@
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
#ifndef ABSL_STRINGS_INTERNAL_STR_FORMAT_FLOAT_CONVERSION_H_
#define ABSL_STRINGS_INTERNAL_STR_FORMAT_FLOAT_CONVERSION_H_
#include "absl/strings/internal/str_format/extension.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace str_format_internal {
-bool ConvertFloatImpl(float v, const ConversionSpec &conv,
+bool ConvertFloatImpl(float v, const FormatConversionSpecImpl &conv,
FormatSinkImpl *sink);
-bool ConvertFloatImpl(double v, const ConversionSpec &conv,
+bool ConvertFloatImpl(double v, const FormatConversionSpecImpl &conv,
FormatSinkImpl *sink);
-bool ConvertFloatImpl(long double v, const ConversionSpec &conv,
+bool ConvertFloatImpl(long double v, const FormatConversionSpecImpl &conv,
FormatSinkImpl *sink);
} // namespace str_format_internal
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_INTERNAL_STR_FORMAT_FLOAT_CONVERSION_H_
diff --git a/absl/strings/internal/str_format/output.cc b/absl/strings/internal/str_format/output.cc
index 38987b6..c4b2470 100644
--- a/absl/strings/internal/str_format/output.cc
+++ b/absl/strings/internal/str_format/output.cc
@@ -18,6 +18,7 @@
#include <cstring>
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace str_format_internal {
namespace {
@@ -67,4 +68,5 @@
}
} // namespace str_format_internal
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/internal/str_format/output.h b/absl/strings/internal/str_format/output.h
index 42da641..8030dae 100644
--- a/absl/strings/internal/str_format/output.h
+++ b/absl/strings/internal/str_format/output.h
@@ -28,9 +28,8 @@
#include "absl/base/port.h"
#include "absl/strings/string_view.h"
-class Cord;
-
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace str_format_internal {
// RawSink implementation that writes into a char* buffer.
@@ -75,12 +74,6 @@
out->write(s.data(), s.size());
}
-template <class AbslCord, typename = typename std::enable_if<
- std::is_same<AbslCord, ::Cord>::value>::type>
-inline void AbslFormatFlush(AbslCord* out, string_view s) {
- out->Append(s);
-}
-
inline void AbslFormatFlush(FILERawSink* sink, string_view v) {
sink->Write(v);
}
@@ -89,13 +82,15 @@
sink->Write(v);
}
+// This is a SFINAE to get a better compiler error message when the type
+// is not supported.
template <typename T>
-auto InvokeFlush(T* out, string_view s)
- -> decltype(str_format_internal::AbslFormatFlush(out, s)) {
- str_format_internal::AbslFormatFlush(out, s);
+auto InvokeFlush(T* out, string_view s) -> decltype(AbslFormatFlush(out, s)) {
+ AbslFormatFlush(out, s);
}
} // namespace str_format_internal
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_INTERNAL_STR_FORMAT_OUTPUT_H_
diff --git a/absl/strings/internal/str_format/output_test.cc b/absl/strings/internal/str_format/output_test.cc
index 6e04abe..ce2e91a 100644
--- a/absl/strings/internal/str_format/output_test.cc
+++ b/absl/strings/internal/str_format/output_test.cc
@@ -19,8 +19,10 @@
#include "gmock/gmock.h"
#include "gtest/gtest.h"
+#include "absl/strings/cord.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace {
TEST(InvokeFlush, String) {
@@ -36,6 +38,12 @@
EXPECT_EQ(str.str(), "ABCDEF");
}
+TEST(InvokeFlush, Cord) {
+ absl::Cord str("ABC");
+ str_format_internal::InvokeFlush(&str, "DEF");
+ EXPECT_EQ(str, "ABCDEF");
+}
+
TEST(BufferRawSink, Limits) {
char buf[16];
{
@@ -67,5 +75,5 @@
}
} // namespace
+ABSL_NAMESPACE_END
} // namespace absl
-
diff --git a/absl/strings/internal/str_format/parser.cc b/absl/strings/internal/str_format/parser.cc
index 9ef5615..f308d02 100644
--- a/absl/strings/internal/str_format/parser.cc
+++ b/absl/strings/internal/str_format/parser.cc
@@ -1,3 +1,17 @@
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
#include "absl/strings/internal/str_format/parser.h"
#include <assert.h>
@@ -14,10 +28,12 @@
#include <unordered_set>
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace str_format_internal {
-using CC = ConversionChar::Id;
-using LM = LengthMod::Id;
+using CC = FormatConversionCharInternal;
+using LM = LengthMod;
+
ABSL_CONST_INIT const ConvTag kTags[256] = {
{}, {}, {}, {}, {}, {}, {}, {}, // 00-07
{}, {}, {}, {}, {}, {}, {}, {}, // 08-0f
@@ -27,9 +43,9 @@
{}, {}, {}, {}, {}, {}, {}, {}, // 28-2f
{}, {}, {}, {}, {}, {}, {}, {}, // 30-37
{}, {}, {}, {}, {}, {}, {}, {}, // 38-3f
- {}, CC::A, {}, CC::C, {}, CC::E, CC::F, CC::G, // @ABCDEFG
+ {}, CC::A, {}, {}, {}, CC::E, CC::F, CC::G, // @ABCDEFG
{}, {}, {}, {}, LM::L, {}, {}, {}, // HIJKLMNO
- {}, {}, {}, CC::S, {}, {}, {}, {}, // PQRSTUVW
+ {}, {}, {}, {}, {}, {}, {}, {}, // PQRSTUVW
CC::X, {}, {}, {}, {}, {}, {}, {}, // XYZ[\]^_
{}, CC::a, {}, CC::c, CC::d, CC::e, CC::f, CC::g, // `abcdefg
LM::h, CC::i, LM::j, {}, LM::l, {}, CC::n, CC::o, // hijklmno
@@ -204,11 +220,11 @@
using str_format_internal::LengthMod;
LengthMod length_mod = tag.as_length();
ABSL_FORMAT_PARSER_INTERNAL_GET_CHAR();
- if (c == 'h' && length_mod.id() == LengthMod::h) {
- conv->length_mod = LengthMod::FromId(LengthMod::hh);
+ if (c == 'h' && length_mod == LengthMod::h) {
+ conv->length_mod = LengthMod::hh;
ABSL_FORMAT_PARSER_INTERNAL_GET_CHAR();
- } else if (c == 'l' && length_mod.id() == LengthMod::l) {
- conv->length_mod = LengthMod::FromId(LengthMod::ll);
+ } else if (c == 'l' && length_mod == LengthMod::l) {
+ conv->length_mod = LengthMod::ll;
ABSL_FORMAT_PARSER_INTERNAL_GET_CHAR();
} else {
conv->length_mod = length_mod;
@@ -227,6 +243,32 @@
} // namespace
+std::string LengthModToString(LengthMod v) {
+ switch (v) {
+ case LengthMod::h:
+ return "h";
+ case LengthMod::hh:
+ return "hh";
+ case LengthMod::l:
+ return "l";
+ case LengthMod::ll:
+ return "ll";
+ case LengthMod::L:
+ return "L";
+ case LengthMod::j:
+ return "j";
+ case LengthMod::z:
+ return "z";
+ case LengthMod::t:
+ return "t";
+ case LengthMod::q:
+ return "q";
+ case LengthMod::none:
+ return "";
+ }
+ return "";
+}
+
const char *ConsumeUnboundConversion(const char *p, const char *end,
UnboundConversion *conv, int *next_arg) {
if (*next_arg < 0) return ConsumeConversion<true>(p, end, conv, next_arg);
@@ -268,15 +310,17 @@
char* data_pos;
};
-ParsedFormatBase::ParsedFormatBase(string_view format, bool allow_ignored,
- std::initializer_list<Conv> convs)
+ParsedFormatBase::ParsedFormatBase(
+ string_view format, bool allow_ignored,
+ std::initializer_list<FormatConversionCharSet> convs)
: data_(format.empty() ? nullptr : new char[format.size()]) {
has_error_ = !ParseFormatString(format, ParsedFormatConsumer(this)) ||
!MatchesConversions(allow_ignored, convs);
}
bool ParsedFormatBase::MatchesConversions(
- bool allow_ignored, std::initializer_list<Conv> convs) const {
+ bool allow_ignored,
+ std::initializer_list<FormatConversionCharSet> convs) const {
std::unordered_set<int> used;
auto add_if_valid_conv = [&](int pos, char c) {
if (static_cast<size_t>(pos) > convs.size() ||
@@ -294,10 +338,13 @@
if (conv.width.is_from_arg() &&
!add_if_valid_conv(conv.width.get_from_arg(), '*'))
return false;
- if (!add_if_valid_conv(conv.arg_position, conv.conv.Char())) return false;
+ if (!add_if_valid_conv(conv.arg_position,
+ FormatConversionCharToChar(conv.conv)))
+ return false;
}
return used.size() == convs.size() || allow_ignored;
}
} // namespace str_format_internal
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/internal/str_format/parser.h b/absl/strings/internal/str_format/parser.h
index b7614a0..6504dd3 100644
--- a/absl/strings/internal/str_format/parser.h
+++ b/absl/strings/internal/str_format/parser.h
@@ -1,3 +1,17 @@
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
#ifndef ABSL_STRINGS_INTERNAL_STR_FORMAT_PARSER_H_
#define ABSL_STRINGS_INTERNAL_STR_FORMAT_PARSER_H_
@@ -6,18 +20,25 @@
#include <stdlib.h>
#include <cassert>
+#include <cstdint>
#include <initializer_list>
#include <iosfwd>
#include <iterator>
#include <memory>
+#include <string>
#include <vector>
#include "absl/strings/internal/str_format/checker.h"
#include "absl/strings/internal/str_format/extension.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace str_format_internal {
+enum class LengthMod : std::uint8_t { h, hh, l, ll, L, j, z, t, q, none };
+
+std::string LengthModToString(LengthMod v);
+
// The analyzed properties of a single specified conversion.
struct UnboundConversion {
UnboundConversion()
@@ -59,8 +80,8 @@
InputValue precision;
Flags flags;
- LengthMod length_mod;
- ConversionChar conv;
+ LengthMod length_mod = LengthMod::none;
+ FormatConversionChar conv = FormatConversionCharInternal::kNone;
};
// Consume conversion spec prefix (not including '%') of [p, end) if valid.
@@ -72,25 +93,28 @@
UnboundConversion* conv, int* next_arg);
// Helper tag class for the table below.
-// It allows fast `char -> ConversionChar/LengthMod` checking and conversions.
+// It allows fast `char -> ConversionChar/LengthMod` checking and
+// conversions.
class ConvTag {
public:
- constexpr ConvTag(ConversionChar::Id id) : tag_(id) {} // NOLINT
+ constexpr ConvTag(FormatConversionChar conversion_char) // NOLINT
+ : tag_(static_cast<int8_t>(conversion_char)) {}
// We invert the length modifiers to make them negative so that we can easily
// test for them.
- constexpr ConvTag(LengthMod::Id id) : tag_(~id) {} // NOLINT
+ constexpr ConvTag(LengthMod length_mod) // NOLINT
+ : tag_(~static_cast<std::int8_t>(length_mod)) {}
// Everything else is -128, which is negative to make is_conv() simpler.
constexpr ConvTag() : tag_(-128) {}
bool is_conv() const { return tag_ >= 0; }
bool is_length() const { return tag_ < 0 && tag_ != -128; }
- ConversionChar as_conv() const {
+ FormatConversionChar as_conv() const {
assert(is_conv());
- return ConversionChar::FromId(static_cast<ConversionChar::Id>(tag_));
+ return static_cast<FormatConversionChar>(tag_);
}
LengthMod as_length() const {
assert(is_length());
- return LengthMod::FromId(static_cast<LengthMod::Id>(~tag_));
+ return static_cast<LengthMod>(~tag_);
}
private:
@@ -133,7 +157,7 @@
auto tag = GetTagForChar(percent[1]);
if (tag.is_conv()) {
if (ABSL_PREDICT_FALSE(next_arg < 0)) {
- // This indicates an error in the format std::string.
+ // This indicates an error in the format string.
// The only way to get `next_arg < 0` here is to have a positional
// argument first which sets next_arg to -1 and then a non-positional
// argument.
@@ -176,8 +200,9 @@
class ParsedFormatBase {
public:
- explicit ParsedFormatBase(string_view format, bool allow_ignored,
- std::initializer_list<Conv> convs);
+ explicit ParsedFormatBase(
+ string_view format, bool allow_ignored,
+ std::initializer_list<FormatConversionCharSet> convs);
ParsedFormatBase(const ParsedFormatBase& other) { *this = other; }
@@ -224,8 +249,9 @@
private:
// Returns whether the conversions match and if !allow_ignored it verifies
// that all conversions are used by the format.
- bool MatchesConversions(bool allow_ignored,
- std::initializer_list<Conv> convs) const;
+ bool MatchesConversions(
+ bool allow_ignored,
+ std::initializer_list<FormatConversionCharSet> convs) const;
struct ParsedFormatConsumer;
@@ -270,14 +296,14 @@
// This is the only API that allows the user to pass a runtime specified format
// string. These factory functions will return NULL if the format does not match
// the conversions requested by the user.
-template <str_format_internal::Conv... C>
+template <FormatConversionCharSet... C>
class ExtendedParsedFormat : public str_format_internal::ParsedFormatBase {
public:
explicit ExtendedParsedFormat(string_view format)
-#if ABSL_INTERNAL_ENABLE_FORMAT_CHECKER
+#ifdef ABSL_INTERNAL_ENABLE_FORMAT_CHECKER
__attribute__((
enable_if(str_format_internal::EnsureConstexpr(format),
- "Format std::string is not constexpr."),
+ "Format string is not constexpr."),
enable_if(str_format_internal::ValidFormatImpl<C...>(format),
"Format specified does not match the template arguments.")))
#endif // ABSL_INTERNAL_ENABLE_FORMAT_CHECKER
@@ -317,6 +343,7 @@
: ParsedFormatBase(s, allow_ignored, {C...}) {}
};
} // namespace str_format_internal
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_INTERNAL_STR_FORMAT_PARSER_H_
diff --git a/absl/strings/internal/str_format/parser_test.cc b/absl/strings/internal/str_format/parser_test.cc
index 6d356093..a5fa1c7 100644
--- a/absl/strings/internal/str_format/parser_test.cc
+++ b/absl/strings/internal/str_format/parser_test.cc
@@ -1,3 +1,17 @@
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
#include "absl/strings/internal/str_format/parser.h"
#include <string.h>
@@ -7,6 +21,7 @@
#include "absl/base/macros.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace str_format_internal {
namespace {
@@ -16,7 +31,7 @@
TEST(LengthModTest, Names) {
struct Expectation {
int line;
- LengthMod::Id id;
+ LengthMod mod;
const char *name;
};
const Expectation kExpect[] = {
@@ -31,37 +46,33 @@
{__LINE__, LengthMod::t, "t" },
{__LINE__, LengthMod::q, "q" },
};
- EXPECT_EQ(ABSL_ARRAYSIZE(kExpect), LengthMod::kNumValues);
+ EXPECT_EQ(ABSL_ARRAYSIZE(kExpect), 10);
for (auto e : kExpect) {
SCOPED_TRACE(e.line);
- LengthMod mod = LengthMod::FromId(e.id);
- EXPECT_EQ(e.id, mod.id());
- EXPECT_EQ(e.name, mod.name());
+ EXPECT_EQ(e.name, LengthModToString(e.mod));
}
}
TEST(ConversionCharTest, Names) {
struct Expectation {
- ConversionChar::Id id;
+ FormatConversionChar id;
char name;
};
// clang-format off
const Expectation kExpect[] = {
-#define X(c) {ConversionChar::c, #c[0]}
- X(c), X(C), X(s), X(S), // text
+#define X(c) {FormatConversionCharInternal::c, #c[0]}
+ X(c), X(s), // text
X(d), X(i), X(o), X(u), X(x), X(X), // int
X(f), X(F), X(e), X(E), X(g), X(G), X(a), X(A), // float
X(n), X(p), // misc
#undef X
- {ConversionChar::none, '\0'},
+ {FormatConversionCharInternal::kNone, '\0'},
};
// clang-format on
- EXPECT_EQ(ABSL_ARRAYSIZE(kExpect), ConversionChar::kNumValues);
for (auto e : kExpect) {
SCOPED_TRACE(e.name);
- ConversionChar v = ConversionChar::FromId(e.id);
- EXPECT_EQ(e.id, v.id());
- EXPECT_EQ(e.name, v.Char());
+ FormatConversionChar v = e.id;
+ EXPECT_EQ(e.name, FormatConversionCharToChar(v));
}
}
@@ -120,13 +131,12 @@
EXPECT_FALSE(Run("dd")); // no excess allowed
EXPECT_TRUE(Run("d"));
- EXPECT_EQ('d', o.conv.Char());
+ EXPECT_EQ('d', FormatConversionCharToChar(o.conv));
EXPECT_FALSE(o.width.is_from_arg());
EXPECT_LT(o.width.value(), 0);
EXPECT_FALSE(o.precision.is_from_arg());
EXPECT_LT(o.precision.value(), 0);
EXPECT_EQ(1, o.arg_position);
- EXPECT_EQ(LengthMod::none, o.length_mod.id());
}
TEST_F(ConsumeUnboundConversionTest, ArgPosition) {
@@ -162,7 +172,7 @@
TEST_F(ConsumeUnboundConversionTest, WidthAndPrecision) {
EXPECT_TRUE(Run("14d"));
- EXPECT_EQ('d', o.conv.Char());
+ EXPECT_EQ('d', FormatConversionCharToChar(o.conv));
EXPECT_FALSE(o.width.is_from_arg());
EXPECT_EQ(14, o.width.value());
EXPECT_FALSE(o.precision.is_from_arg());
@@ -289,6 +299,29 @@
}
}
+TEST_F(ConsumeUnboundConversionTest, LengthMod) {
+ EXPECT_TRUE(Run("d"));
+ EXPECT_EQ(LengthMod::none, o.length_mod);
+ EXPECT_TRUE(Run("hd"));
+ EXPECT_EQ(LengthMod::h, o.length_mod);
+ EXPECT_TRUE(Run("hhd"));
+ EXPECT_EQ(LengthMod::hh, o.length_mod);
+ EXPECT_TRUE(Run("ld"));
+ EXPECT_EQ(LengthMod::l, o.length_mod);
+ EXPECT_TRUE(Run("lld"));
+ EXPECT_EQ(LengthMod::ll, o.length_mod);
+ EXPECT_TRUE(Run("Lf"));
+ EXPECT_EQ(LengthMod::L, o.length_mod);
+ EXPECT_TRUE(Run("qf"));
+ EXPECT_EQ(LengthMod::q, o.length_mod);
+ EXPECT_TRUE(Run("jd"));
+ EXPECT_EQ(LengthMod::j, o.length_mod);
+ EXPECT_TRUE(Run("zd"));
+ EXPECT_EQ(LengthMod::z, o.length_mod);
+ EXPECT_TRUE(Run("td"));
+ EXPECT_EQ(LengthMod::t, o.length_mod);
+}
+
struct SummarizeConsumer {
std::string* out;
explicit SummarizeConsumer(std::string* out) : out(out) {}
@@ -309,7 +342,7 @@
if (conv.precision.is_from_arg()) {
*out += "." + std::to_string(conv.precision.get_from_arg()) + "$*";
}
- *out += conv.conv.Char();
+ *out += FormatConversionCharToChar(conv.conv);
*out += "}";
return true;
}
@@ -330,7 +363,8 @@
ParsedFormatBase p2 = p1; // copy construct (empty)
EXPECT_EQ(SummarizeParsedFormat(p1), SummarizeParsedFormat(p2));
- p1 = ParsedFormatBase("hello%s", true, {Conv::s}); // move assign
+ p1 = ParsedFormatBase("hello%s", true,
+ {FormatConversionCharSetInternal::s}); // move assign
EXPECT_EQ("[hello]{s:1$s}", SummarizeParsedFormat(p1));
ParsedFormatBase p3 = p1; // copy construct (nonempty)
@@ -348,7 +382,7 @@
struct ExpectParse {
const char* in;
- std::initializer_list<Conv> conv_set;
+ std::initializer_list<FormatConversionCharSet> conv_set;
const char* out;
};
@@ -358,9 +392,9 @@
const ExpectParse kExpect[] = {
{"", {}, ""},
{"ab", {}, "[ab]"},
- {"a%d", {Conv::d}, "[a]{d:1$d}"},
- {"a%+d", {Conv::d}, "[a]{+d:1$d}"},
- {"a% d", {Conv::d}, "[a]{ d:1$d}"},
+ {"a%d", {FormatConversionCharSetInternal::d}, "[a]{d:1$d}"},
+ {"a%+d", {FormatConversionCharSetInternal::d}, "[a]{+d:1$d}"},
+ {"a% d", {FormatConversionCharSetInternal::d}, "[a]{ d:1$d}"},
{"a%b %d", {}, "[a]!"}, // stop after error
};
for (const auto& e : kExpect) {
@@ -372,13 +406,13 @@
TEST_F(ParsedFormatTest, ParsingFlagOrder) {
const ExpectParse kExpect[] = {
- {"a%+ 0d", {Conv::d}, "[a]{+ 0d:1$d}"},
- {"a%+0 d", {Conv::d}, "[a]{+0 d:1$d}"},
- {"a%0+ d", {Conv::d}, "[a]{0+ d:1$d}"},
- {"a% +0d", {Conv::d}, "[a]{ +0d:1$d}"},
- {"a%0 +d", {Conv::d}, "[a]{0 +d:1$d}"},
- {"a% 0+d", {Conv::d}, "[a]{ 0+d:1$d}"},
- {"a%+ 0+d", {Conv::d}, "[a]{+ 0+d:1$d}"},
+ {"a%+ 0d", {FormatConversionCharSetInternal::d}, "[a]{+ 0d:1$d}"},
+ {"a%+0 d", {FormatConversionCharSetInternal::d}, "[a]{+0 d:1$d}"},
+ {"a%0+ d", {FormatConversionCharSetInternal::d}, "[a]{0+ d:1$d}"},
+ {"a% +0d", {FormatConversionCharSetInternal::d}, "[a]{ +0d:1$d}"},
+ {"a%0 +d", {FormatConversionCharSetInternal::d}, "[a]{0 +d:1$d}"},
+ {"a% 0+d", {FormatConversionCharSetInternal::d}, "[a]{ 0+d:1$d}"},
+ {"a%+ 0+d", {FormatConversionCharSetInternal::d}, "[a]{+ 0+d:1$d}"},
};
for (const auto& e : kExpect) {
SCOPED_TRACE(e.in);
@@ -389,4 +423,5 @@
} // namespace
} // namespace str_format_internal
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/internal/str_join_internal.h b/absl/strings/internal/str_join_internal.h
index 7c35f4d..31dbf67 100644
--- a/absl/strings/internal/str_join_internal.h
+++ b/absl/strings/internal/str_join_internal.h
@@ -43,6 +43,7 @@
#include "absl/strings/str_cat.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace strings_internal {
//
@@ -307,6 +308,7 @@
}
} // namespace strings_internal
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_INTERNAL_STR_JOIN_INTERNAL_H_
diff --git a/absl/strings/internal/str_split_internal.h b/absl/strings/internal/str_split_internal.h
index 52f6222..a2f41c1 100644
--- a/absl/strings/internal/str_split_internal.h
+++ b/absl/strings/internal/str_split_internal.h
@@ -47,12 +47,13 @@
#endif // _GLIBCXX_DEBUG
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace strings_internal {
// This class is implicitly constructible from everything that absl::string_view
-// is implicitly constructible from. If it's constructed from a temporary
-// string, the data is moved into a data member so its lifetime matches that of
-// the ConvertibleToStringView instance.
+// is implicitly constructible from, except for rvalue strings. This means it
+// can be used as a function parameter in places where passing a temporary
+// string might cause memory lifetime issues.
class ConvertibleToStringView {
public:
ConvertibleToStringView(const char* s) // NOLINT(runtime/explicit)
@@ -64,41 +65,12 @@
: value_(s) {}
// Matches rvalue strings and moves their data to a member.
-ConvertibleToStringView(std::string&& s) // NOLINT(runtime/explicit)
- : copy_(std::move(s)), value_(copy_) {}
-
- ConvertibleToStringView(const ConvertibleToStringView& other)
- : copy_(other.copy_),
- value_(other.IsSelfReferential() ? copy_ : other.value_) {}
-
- ConvertibleToStringView(ConvertibleToStringView&& other) {
- StealMembers(std::move(other));
- }
-
- ConvertibleToStringView& operator=(ConvertibleToStringView other) {
- StealMembers(std::move(other));
- return *this;
- }
+ ConvertibleToStringView(std::string&& s) = delete;
+ ConvertibleToStringView(const std::string&& s) = delete;
absl::string_view value() const { return value_; }
private:
- // Returns true if ctsp's value refers to its internal copy_ member.
- bool IsSelfReferential() const { return value_.data() == copy_.data(); }
-
- void StealMembers(ConvertibleToStringView&& other) {
- if (other.IsSelfReferential()) {
- copy_ = std::move(other.copy_);
- value_ = copy_;
- other.value_ = other.copy_;
- } else {
- value_ = other.value_;
- }
- }
-
- // Holds the data moved from temporary std::string arguments. Declared first
- // so that 'value' can refer to 'copy_'.
- std::string copy_;
absl::string_view value_;
};
@@ -272,7 +244,11 @@
// the split strings: only strings for which the predicate returns true will be
// kept. A Predicate object is any unary functor that takes an absl::string_view
// and returns bool.
-template <typename Delimiter, typename Predicate>
+//
+// The StringType parameter can be either string_view or string, depending on
+// whether the Splitter refers to a string stored elsewhere, or if the string
+// resides inside the Splitter itself.
+template <typename Delimiter, typename Predicate, typename StringType>
class Splitter {
public:
using DelimiterType = Delimiter;
@@ -280,12 +256,12 @@
using const_iterator = strings_internal::SplitIterator<Splitter>;
using value_type = typename std::iterator_traits<const_iterator>::value_type;
- Splitter(ConvertibleToStringView input_text, Delimiter d, Predicate p)
+ Splitter(StringType input_text, Delimiter d, Predicate p)
: text_(std::move(input_text)),
delimiter_(std::move(d)),
predicate_(std::move(p)) {}
- absl::string_view text() const { return text_.value(); }
+ absl::string_view text() const { return text_; }
const Delimiter& delimiter() const { return delimiter_; }
const Predicate& predicate() const { return predicate_; }
@@ -335,7 +311,7 @@
Container operator()(const Splitter& splitter) const {
Container c;
auto it = std::inserter(c, c.end());
- for (const auto sp : splitter) {
+ for (const auto& sp : splitter) {
*it++ = ValueType(sp);
}
return c;
@@ -400,7 +376,7 @@
Container m;
typename Container::iterator it;
bool insert = true;
- for (const auto sp : splitter) {
+ for (const auto& sp : splitter) {
if (insert) {
it = Inserter<Container>::Insert(&m, First(sp), Second());
} else {
@@ -442,12 +418,13 @@
};
};
- ConvertibleToStringView text_;
+ StringType text_;
Delimiter delimiter_;
Predicate predicate_;
};
} // namespace strings_internal
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_INTERNAL_STR_SPLIT_INTERNAL_H_
diff --git a/absl/strings/internal/string_constant.h b/absl/strings/internal/string_constant.h
new file mode 100644
index 0000000..a11336b
--- /dev/null
+++ b/absl/strings/internal/string_constant.h
@@ -0,0 +1,64 @@
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#ifndef ABSL_STRINGS_INTERNAL_STRING_CONSTANT_H_
+#define ABSL_STRINGS_INTERNAL_STRING_CONSTANT_H_
+
+#include "absl/meta/type_traits.h"
+#include "absl/strings/string_view.h"
+
+namespace absl {
+ABSL_NAMESPACE_BEGIN
+namespace strings_internal {
+
+// StringConstant<T> represents a compile time string constant.
+// It can be accessed via its `absl::string_view value` static member.
+// It is guaranteed that the `string_view` returned has constant `.data()`,
+// constant `.size()` and constant `value[i]` for all `0 <= i < .size()`
+//
+// The `T` is an opaque type. It is guaranteed that different string constants
+// will have different values of `T`. This allows users to associate the string
+// constant with other static state at compile time.
+//
+// Instances should be made using the `MakeStringConstant()` factory function
+// below.
+template <typename T>
+struct StringConstant {
+ static constexpr absl::string_view value = T{}();
+ constexpr absl::string_view operator()() const { return value; }
+
+ // Check to be sure `view` points to constant data.
+ // Otherwise, it can't be constant evaluated.
+ static_assert(value.empty() || 2 * value[0] != 1,
+ "The input string_view must point to constant data.");
+};
+
+template <typename T>
+constexpr absl::string_view StringConstant<T>::value; // NOLINT
+
+// Factory function for `StringConstant` instances.
+// It supports callables that have a constexpr default constructor and a
+// constexpr operator().
+// It must return an `absl::string_view` or `const char*` pointing to constant
+// data. This is validated at compile time.
+template <typename T>
+constexpr StringConstant<T> MakeStringConstant(T) {
+ return {};
+}
+
+} // namespace strings_internal
+ABSL_NAMESPACE_END
+} // namespace absl
+
+#endif // ABSL_STRINGS_INTERNAL_STRING_CONSTANT_H_
diff --git a/absl/strings/internal/string_constant_test.cc b/absl/strings/internal/string_constant_test.cc
new file mode 100644
index 0000000..392833c
--- /dev/null
+++ b/absl/strings/internal/string_constant_test.cc
@@ -0,0 +1,60 @@
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "absl/strings/internal/string_constant.h"
+
+#include "absl/meta/type_traits.h"
+#include "gmock/gmock.h"
+#include "gtest/gtest.h"
+
+namespace {
+
+using absl::strings_internal::MakeStringConstant;
+
+struct Callable {
+ constexpr absl::string_view operator()() const {
+ return absl::string_view("Callable", 8);
+ }
+};
+
+TEST(StringConstant, Traits) {
+ constexpr auto str = MakeStringConstant(Callable{});
+ using T = decltype(str);
+
+ EXPECT_TRUE(std::is_empty<T>::value);
+ EXPECT_TRUE(std::is_trivial<T>::value);
+ EXPECT_TRUE(absl::is_trivially_default_constructible<T>::value);
+ EXPECT_TRUE(absl::is_trivially_copy_constructible<T>::value);
+ EXPECT_TRUE(absl::is_trivially_move_constructible<T>::value);
+ EXPECT_TRUE(absl::is_trivially_destructible<T>::value);
+}
+
+TEST(StringConstant, MakeFromCallable) {
+ constexpr auto str = MakeStringConstant(Callable{});
+ using T = decltype(str);
+ EXPECT_EQ(Callable{}(), T::value);
+ EXPECT_EQ(Callable{}(), str());
+}
+
+TEST(StringConstant, MakeFromStringConstant) {
+ // We want to make sure the StringConstant itself is a valid input to the
+ // factory function.
+ constexpr auto str = MakeStringConstant(Callable{});
+ constexpr auto str2 = MakeStringConstant(str);
+ using T = decltype(str2);
+ EXPECT_EQ(Callable{}(), T::value);
+ EXPECT_EQ(Callable{}(), str2());
+}
+
+} // namespace
diff --git a/absl/strings/internal/utf8.cc b/absl/strings/internal/utf8.cc
index 82d36c2..8fd8edc 100644
--- a/absl/strings/internal/utf8.cc
+++ b/absl/strings/internal/utf8.cc
@@ -17,6 +17,7 @@
#include "absl/strings/internal/utf8.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace strings_internal {
size_t EncodeUTF8Char(char *buffer, char32_t utf8_char) {
@@ -48,4 +49,5 @@
}
} // namespace strings_internal
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/internal/utf8.h b/absl/strings/internal/utf8.h
index 0423630..32fb109 100644
--- a/absl/strings/internal/utf8.h
+++ b/absl/strings/internal/utf8.h
@@ -20,7 +20,10 @@
#include <cstddef>
#include <cstdint>
+#include "absl/base/config.h"
+
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace strings_internal {
// For Unicode code points 0 through 0x10FFFF, EncodeUTF8Char writes
@@ -41,6 +44,7 @@
size_t EncodeUTF8Char(char *buffer, char32_t utf8_char);
} // namespace strings_internal
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_INTERNAL_UTF8_H_
diff --git a/absl/strings/match.cc b/absl/strings/match.cc
index 7b24241..2d67250 100644
--- a/absl/strings/match.cc
+++ b/absl/strings/match.cc
@@ -17,22 +17,27 @@
#include "absl/strings/internal/memutil.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
-bool EqualsIgnoreCase(absl::string_view piece1, absl::string_view piece2) {
+bool EqualsIgnoreCase(absl::string_view piece1,
+ absl::string_view piece2) noexcept {
return (piece1.size() == piece2.size() &&
0 == absl::strings_internal::memcasecmp(piece1.data(), piece2.data(),
piece1.size()));
// memcasecmp uses absl::ascii_tolower().
}
-bool StartsWithIgnoreCase(absl::string_view text, absl::string_view prefix) {
+bool StartsWithIgnoreCase(absl::string_view text,
+ absl::string_view prefix) noexcept {
return (text.size() >= prefix.size()) &&
EqualsIgnoreCase(text.substr(0, prefix.size()), prefix);
}
-bool EndsWithIgnoreCase(absl::string_view text, absl::string_view suffix) {
+bool EndsWithIgnoreCase(absl::string_view text,
+ absl::string_view suffix) noexcept {
return (text.size() >= suffix.size()) &&
EqualsIgnoreCase(text.substr(text.size() - suffix.size()), suffix);
}
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/match.h b/absl/strings/match.h
index 762f359..038cbb3 100644
--- a/absl/strings/match.h
+++ b/absl/strings/match.h
@@ -20,7 +20,7 @@
// This file contains simple utilities for performing string matching checks.
// All of these function parameters are specified as `absl::string_view`,
// meaning that these functions can accept `std::string`, `absl::string_view` or
-// nul-terminated C-style strings.
+// NUL-terminated C-style strings.
//
// Examples:
// std::string s = "foo";
@@ -38,18 +38,25 @@
#include "absl/strings/string_view.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
// StrContains()
//
// Returns whether a given string `haystack` contains the substring `needle`.
-inline bool StrContains(absl::string_view haystack, absl::string_view needle) {
+inline bool StrContains(absl::string_view haystack,
+ absl::string_view needle) noexcept {
return haystack.find(needle, 0) != haystack.npos;
}
+inline bool StrContains(absl::string_view haystack, char needle) noexcept {
+ return haystack.find(needle) != haystack.npos;
+}
+
// StartsWith()
//
// Returns whether a given string `text` begins with `prefix`.
-inline bool StartsWith(absl::string_view text, absl::string_view prefix) {
+inline bool StartsWith(absl::string_view text,
+ absl::string_view prefix) noexcept {
return prefix.empty() ||
(text.size() >= prefix.size() &&
memcmp(text.data(), prefix.data(), prefix.size()) == 0);
@@ -58,7 +65,8 @@
// EndsWith()
//
// Returns whether a given string `text` ends with `suffix`.
-inline bool EndsWith(absl::string_view text, absl::string_view suffix) {
+inline bool EndsWith(absl::string_view text,
+ absl::string_view suffix) noexcept {
return suffix.empty() ||
(text.size() >= suffix.size() &&
memcmp(text.data() + (text.size() - suffix.size()), suffix.data(),
@@ -69,20 +77,24 @@
//
// Returns whether given ASCII strings `piece1` and `piece2` are equal, ignoring
// case in the comparison.
-bool EqualsIgnoreCase(absl::string_view piece1, absl::string_view piece2);
+bool EqualsIgnoreCase(absl::string_view piece1,
+ absl::string_view piece2) noexcept;
// StartsWithIgnoreCase()
//
// Returns whether a given ASCII string `text` starts with `prefix`,
// ignoring case in the comparison.
-bool StartsWithIgnoreCase(absl::string_view text, absl::string_view prefix);
+bool StartsWithIgnoreCase(absl::string_view text,
+ absl::string_view prefix) noexcept;
// EndsWithIgnoreCase()
//
// Returns whether a given ASCII string `text` ends with `suffix`, ignoring
// case in the comparison.
-bool EndsWithIgnoreCase(absl::string_view text, absl::string_view suffix);
+bool EndsWithIgnoreCase(absl::string_view text,
+ absl::string_view suffix) noexcept;
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_MATCH_H_
diff --git a/absl/strings/match_test.cc b/absl/strings/match_test.cc
index 4c313dd..5841bc1 100644
--- a/absl/strings/match_test.cc
+++ b/absl/strings/match_test.cc
@@ -66,6 +66,23 @@
EXPECT_FALSE(absl::StrContains("", "a"));
}
+TEST(MatchTest, ContainsChar) {
+ absl::string_view a("abcdefg");
+ absl::string_view b("abcd");
+ EXPECT_TRUE(absl::StrContains(a, 'a'));
+ EXPECT_TRUE(absl::StrContains(a, 'b'));
+ EXPECT_TRUE(absl::StrContains(a, 'e'));
+ EXPECT_FALSE(absl::StrContains(a, 'h'));
+
+ EXPECT_TRUE(absl::StrContains(b, 'a'));
+ EXPECT_TRUE(absl::StrContains(b, 'b'));
+ EXPECT_FALSE(absl::StrContains(b, 'e'));
+ EXPECT_FALSE(absl::StrContains(b, 'h'));
+
+ EXPECT_FALSE(absl::StrContains("", 'a'));
+ EXPECT_FALSE(absl::StrContains("", 'a'));
+}
+
TEST(MatchTest, ContainsNull) {
const std::string s = "foo";
const char* cs = "foo";
diff --git a/absl/strings/numbers.cc b/absl/strings/numbers.cc
index 38d1486..e6bf44c 100644
--- a/absl/strings/numbers.cc
+++ b/absl/strings/numbers.cc
@@ -19,8 +19,8 @@
#include <algorithm>
#include <cassert>
-#include <cfloat> // for DBL_DIG and FLT_DIG
-#include <cmath> // for HUGE_VAL
+#include <cfloat> // for DBL_DIG and FLT_DIG
+#include <cmath> // for HUGE_VAL
#include <cstdint>
#include <cstdio>
#include <cstdlib>
@@ -30,15 +30,18 @@
#include <memory>
#include <utility>
-#include "absl/base/internal/bits.h"
+#include "absl/base/attributes.h"
#include "absl/base/internal/raw_logging.h"
+#include "absl/numeric/bits.h"
#include "absl/strings/ascii.h"
#include "absl/strings/charconv.h"
+#include "absl/strings/escaping.h"
#include "absl/strings/internal/memutil.h"
#include "absl/strings/match.h"
#include "absl/strings/str_cat.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
bool SimpleAtof(absl::string_view str, float* out) {
*out = 0.0;
@@ -92,43 +95,6 @@
return true;
}
-namespace {
-
-// Writes a two-character representation of 'i' to 'buf'. 'i' must be in the
-// range 0 <= i < 100, and buf must have space for two characters. Example:
-// char buf[2];
-// PutTwoDigits(42, buf);
-// // buf[0] == '4'
-// // buf[1] == '2'
-inline void PutTwoDigits(size_t i, char* buf) {
- static const char two_ASCII_digits[100][2] = {
- {'0', '0'}, {'0', '1'}, {'0', '2'}, {'0', '3'}, {'0', '4'},
- {'0', '5'}, {'0', '6'}, {'0', '7'}, {'0', '8'}, {'0', '9'},
- {'1', '0'}, {'1', '1'}, {'1', '2'}, {'1', '3'}, {'1', '4'},
- {'1', '5'}, {'1', '6'}, {'1', '7'}, {'1', '8'}, {'1', '9'},
- {'2', '0'}, {'2', '1'}, {'2', '2'}, {'2', '3'}, {'2', '4'},
- {'2', '5'}, {'2', '6'}, {'2', '7'}, {'2', '8'}, {'2', '9'},
- {'3', '0'}, {'3', '1'}, {'3', '2'}, {'3', '3'}, {'3', '4'},
- {'3', '5'}, {'3', '6'}, {'3', '7'}, {'3', '8'}, {'3', '9'},
- {'4', '0'}, {'4', '1'}, {'4', '2'}, {'4', '3'}, {'4', '4'},
- {'4', '5'}, {'4', '6'}, {'4', '7'}, {'4', '8'}, {'4', '9'},
- {'5', '0'}, {'5', '1'}, {'5', '2'}, {'5', '3'}, {'5', '4'},
- {'5', '5'}, {'5', '6'}, {'5', '7'}, {'5', '8'}, {'5', '9'},
- {'6', '0'}, {'6', '1'}, {'6', '2'}, {'6', '3'}, {'6', '4'},
- {'6', '5'}, {'6', '6'}, {'6', '7'}, {'6', '8'}, {'6', '9'},
- {'7', '0'}, {'7', '1'}, {'7', '2'}, {'7', '3'}, {'7', '4'},
- {'7', '5'}, {'7', '6'}, {'7', '7'}, {'7', '8'}, {'7', '9'},
- {'8', '0'}, {'8', '1'}, {'8', '2'}, {'8', '3'}, {'8', '4'},
- {'8', '5'}, {'8', '6'}, {'8', '7'}, {'8', '8'}, {'8', '9'},
- {'9', '0'}, {'9', '1'}, {'9', '2'}, {'9', '3'}, {'9', '4'},
- {'9', '5'}, {'9', '6'}, {'9', '7'}, {'9', '8'}, {'9', '9'}
- };
- assert(i < 100);
- memcpy(buf, two_ASCII_digits[i], 2);
-}
-
-} // namespace
-
bool SimpleAtob(absl::string_view str, bool* out) {
ABSL_RAW_CHECK(out != nullptr, "Output pointer must not be nullptr.");
if (EqualsIgnoreCase(str, "true") || EqualsIgnoreCase(str, "t") ||
@@ -337,7 +303,7 @@
uint64_t bits128_up = (bits96_127 >> 32) + (bits64_127 < bits64_95);
if (bits128_up == 0) return {bits64_127, bits0_63};
- int shift = 64 - base_internal::CountLeadingZeros64(bits128_up);
+ auto shift = static_cast<unsigned>(bit_width(bits128_up));
uint64_t lo = (bits0_63 >> shift) + (bits64_127 << (64 - shift));
uint64_t hi = (bits64_127 >> shift) + (bits128_up << (64 - shift));
return {hi, lo};
@@ -368,7 +334,7 @@
5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5,
5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5};
result = Mul32(result, powers_of_five[expfive & 15]);
- int shift = base_internal::CountLeadingZeros64(result.first);
+ int shift = countl_zero(result.first);
if (shift != 0) {
result.first = (result.first << shift) + (result.second >> (64 - shift));
result.second = (result.second << shift);
@@ -495,13 +461,13 @@
int two_digits = dddddd / 10000;
dddddd -= two_digits * 10000;
- PutTwoDigits(two_digits, &exp_dig.digits[0]);
+ numbers_internal::PutTwoDigits(two_digits, &exp_dig.digits[0]);
two_digits = dddddd / 100;
dddddd -= two_digits * 100;
- PutTwoDigits(two_digits, &exp_dig.digits[2]);
+ numbers_internal::PutTwoDigits(two_digits, &exp_dig.digits[2]);
- PutTwoDigits(dddddd, &exp_dig.digits[4]);
+ numbers_internal::PutTwoDigits(dddddd, &exp_dig.digits[4]);
return exp_dig;
}
@@ -754,8 +720,8 @@
// commonly used bases.
template <typename IntType>
struct LookupTables {
- static const IntType kVmaxOverBase[];
- static const IntType kVminOverBase[];
+ ABSL_CONST_INIT static const IntType kVmaxOverBase[];
+ ABSL_CONST_INIT static const IntType kVminOverBase[];
};
// An array initializer macro for X/base where base in [0, 36].
@@ -770,6 +736,163 @@
X / 35, X / 36, \
}
+// This kVmaxOverBase is generated with
+// for (int base = 2; base < 37; ++base) {
+// absl::uint128 max = std::numeric_limits<absl::uint128>::max();
+// auto result = max / base;
+// std::cout << " MakeUint128(" << absl::Uint128High64(result) << "u, "
+// << absl::Uint128Low64(result) << "u),\n";
+// }
+// See https://godbolt.org/z/aneYsb
+//
+// uint128& operator/=(uint128) is not constexpr, so hardcode the resulting
+// array to avoid a static initializer.
+template<>
+const uint128 LookupTables<uint128>::kVmaxOverBase[] = {
+ 0,
+ 0,
+ MakeUint128(9223372036854775807u, 18446744073709551615u),
+ MakeUint128(6148914691236517205u, 6148914691236517205u),
+ MakeUint128(4611686018427387903u, 18446744073709551615u),
+ MakeUint128(3689348814741910323u, 3689348814741910323u),
+ MakeUint128(3074457345618258602u, 12297829382473034410u),
+ MakeUint128(2635249153387078802u, 5270498306774157604u),
+ MakeUint128(2305843009213693951u, 18446744073709551615u),
+ MakeUint128(2049638230412172401u, 14347467612885206812u),
+ MakeUint128(1844674407370955161u, 11068046444225730969u),
+ MakeUint128(1676976733973595601u, 8384883669867978007u),
+ MakeUint128(1537228672809129301u, 6148914691236517205u),
+ MakeUint128(1418980313362273201u, 4256940940086819603u),
+ MakeUint128(1317624576693539401u, 2635249153387078802u),
+ MakeUint128(1229782938247303441u, 1229782938247303441u),
+ MakeUint128(1152921504606846975u, 18446744073709551615u),
+ MakeUint128(1085102592571150095u, 1085102592571150095u),
+ MakeUint128(1024819115206086200u, 16397105843297379214u),
+ MakeUint128(970881267037344821u, 16504981539634861972u),
+ MakeUint128(922337203685477580u, 14757395258967641292u),
+ MakeUint128(878416384462359600u, 14054662151397753612u),
+ MakeUint128(838488366986797800u, 13415813871788764811u),
+ MakeUint128(802032351030850070u, 4812194106185100421u),
+ MakeUint128(768614336404564650u, 12297829382473034410u),
+ MakeUint128(737869762948382064u, 11805916207174113034u),
+ MakeUint128(709490156681136600u, 11351842506898185609u),
+ MakeUint128(683212743470724133u, 17080318586768103348u),
+ MakeUint128(658812288346769700u, 10540996613548315209u),
+ MakeUint128(636094623231363848u, 15266270957552732371u),
+ MakeUint128(614891469123651720u, 9838263505978427528u),
+ MakeUint128(595056260442243600u, 9520900167075897608u),
+ MakeUint128(576460752303423487u, 18446744073709551615u),
+ MakeUint128(558992244657865200u, 8943875914525843207u),
+ MakeUint128(542551296285575047u, 9765923333140350855u),
+ MakeUint128(527049830677415760u, 8432797290838652167u),
+ MakeUint128(512409557603043100u, 8198552921648689607u),
+};
+
+// This kVmaxOverBase generated with
+// for (int base = 2; base < 37; ++base) {
+// absl::int128 max = std::numeric_limits<absl::int128>::max();
+// auto result = max / base;
+// std::cout << "\tMakeInt128(" << absl::Int128High64(result) << ", "
+// << absl::Int128Low64(result) << "u),\n";
+// }
+// See https://godbolt.org/z/7djYWz
+//
+// int128& operator/=(int128) is not constexpr, so hardcode the resulting array
+// to avoid a static initializer.
+template<>
+const int128 LookupTables<int128>::kVmaxOverBase[] = {
+ 0,
+ 0,
+ MakeInt128(4611686018427387903, 18446744073709551615u),
+ MakeInt128(3074457345618258602, 12297829382473034410u),
+ MakeInt128(2305843009213693951, 18446744073709551615u),
+ MakeInt128(1844674407370955161, 11068046444225730969u),
+ MakeInt128(1537228672809129301, 6148914691236517205u),
+ MakeInt128(1317624576693539401, 2635249153387078802u),
+ MakeInt128(1152921504606846975, 18446744073709551615u),
+ MakeInt128(1024819115206086200, 16397105843297379214u),
+ MakeInt128(922337203685477580, 14757395258967641292u),
+ MakeInt128(838488366986797800, 13415813871788764811u),
+ MakeInt128(768614336404564650, 12297829382473034410u),
+ MakeInt128(709490156681136600, 11351842506898185609u),
+ MakeInt128(658812288346769700, 10540996613548315209u),
+ MakeInt128(614891469123651720, 9838263505978427528u),
+ MakeInt128(576460752303423487, 18446744073709551615u),
+ MakeInt128(542551296285575047, 9765923333140350855u),
+ MakeInt128(512409557603043100, 8198552921648689607u),
+ MakeInt128(485440633518672410, 17475862806672206794u),
+ MakeInt128(461168601842738790, 7378697629483820646u),
+ MakeInt128(439208192231179800, 7027331075698876806u),
+ MakeInt128(419244183493398900, 6707906935894382405u),
+ MakeInt128(401016175515425035, 2406097053092550210u),
+ MakeInt128(384307168202282325, 6148914691236517205u),
+ MakeInt128(368934881474191032, 5902958103587056517u),
+ MakeInt128(354745078340568300, 5675921253449092804u),
+ MakeInt128(341606371735362066, 17763531330238827482u),
+ MakeInt128(329406144173384850, 5270498306774157604u),
+ MakeInt128(318047311615681924, 7633135478776366185u),
+ MakeInt128(307445734561825860, 4919131752989213764u),
+ MakeInt128(297528130221121800, 4760450083537948804u),
+ MakeInt128(288230376151711743, 18446744073709551615u),
+ MakeInt128(279496122328932600, 4471937957262921603u),
+ MakeInt128(271275648142787523, 14106333703424951235u),
+ MakeInt128(263524915338707880, 4216398645419326083u),
+ MakeInt128(256204778801521550, 4099276460824344803u),
+};
+
+// This kVminOverBase generated with
+// for (int base = 2; base < 37; ++base) {
+// absl::int128 min = std::numeric_limits<absl::int128>::min();
+// auto result = min / base;
+// std::cout << "\tMakeInt128(" << absl::Int128High64(result) << ", "
+// << absl::Int128Low64(result) << "u),\n";
+// }
+//
+// See https://godbolt.org/z/7djYWz
+//
+// int128& operator/=(int128) is not constexpr, so hardcode the resulting array
+// to avoid a static initializer.
+template<>
+const int128 LookupTables<int128>::kVminOverBase[] = {
+ 0,
+ 0,
+ MakeInt128(-4611686018427387904, 0u),
+ MakeInt128(-3074457345618258603, 6148914691236517206u),
+ MakeInt128(-2305843009213693952, 0u),
+ MakeInt128(-1844674407370955162, 7378697629483820647u),
+ MakeInt128(-1537228672809129302, 12297829382473034411u),
+ MakeInt128(-1317624576693539402, 15811494920322472814u),
+ MakeInt128(-1152921504606846976, 0u),
+ MakeInt128(-1024819115206086201, 2049638230412172402u),
+ MakeInt128(-922337203685477581, 3689348814741910324u),
+ MakeInt128(-838488366986797801, 5030930201920786805u),
+ MakeInt128(-768614336404564651, 6148914691236517206u),
+ MakeInt128(-709490156681136601, 7094901566811366007u),
+ MakeInt128(-658812288346769701, 7905747460161236407u),
+ MakeInt128(-614891469123651721, 8608480567731124088u),
+ MakeInt128(-576460752303423488, 0u),
+ MakeInt128(-542551296285575048, 8680820740569200761u),
+ MakeInt128(-512409557603043101, 10248191152060862009u),
+ MakeInt128(-485440633518672411, 970881267037344822u),
+ MakeInt128(-461168601842738791, 11068046444225730970u),
+ MakeInt128(-439208192231179801, 11419412998010674810u),
+ MakeInt128(-419244183493398901, 11738837137815169211u),
+ MakeInt128(-401016175515425036, 16040647020617001406u),
+ MakeInt128(-384307168202282326, 12297829382473034411u),
+ MakeInt128(-368934881474191033, 12543785970122495099u),
+ MakeInt128(-354745078340568301, 12770822820260458812u),
+ MakeInt128(-341606371735362067, 683212743470724134u),
+ MakeInt128(-329406144173384851, 13176245766935394012u),
+ MakeInt128(-318047311615681925, 10813608594933185431u),
+ MakeInt128(-307445734561825861, 13527612320720337852u),
+ MakeInt128(-297528130221121801, 13686293990171602812u),
+ MakeInt128(-288230376151711744, 0u),
+ MakeInt128(-279496122328932601, 13974806116446630013u),
+ MakeInt128(-271275648142787524, 4340410370284600381u),
+ MakeInt128(-263524915338707881, 14230345428290225533u),
+ MakeInt128(-256204778801521551, 14347467612885206813u),
+};
+
template <typename IntType>
const IntType LookupTables<IntType>::kVmaxOverBase[] =
X_OVER_BASE_INITIALIZER(std::numeric_limits<IntType>::max());
@@ -789,6 +912,8 @@
assert(base >= 0);
assert(vmax >= static_cast<IntType>(base));
const IntType vmax_over_base = LookupTables<IntType>::kVmaxOverBase[base];
+ assert(base < 2 ||
+ std::numeric_limits<IntType>::max() / base == vmax_over_base);
const char* start = text.data();
const char* end = start + text.size();
// loop over digits
@@ -822,6 +947,8 @@
assert(vmin < 0);
assert(vmin <= 0 - base);
IntType vmin_over_base = LookupTables<IntType>::kVminOverBase[base];
+ assert(base < 2 ||
+ std::numeric_limits<IntType>::min() / base == vmin_over_base);
// 2003 c++ standard [expr.mul]
// "... the sign of the remainder is implementation-defined."
// Although (vmin/base)*base + vmin%base is always vmin.
@@ -885,6 +1012,48 @@
} // anonymous namespace
namespace numbers_internal {
+
+// Digit conversion.
+ABSL_CONST_INIT ABSL_DLL const char kHexChar[] =
+ "0123456789abcdef";
+
+ABSL_CONST_INIT ABSL_DLL const char kHexTable[513] =
+ "000102030405060708090a0b0c0d0e0f"
+ "101112131415161718191a1b1c1d1e1f"
+ "202122232425262728292a2b2c2d2e2f"
+ "303132333435363738393a3b3c3d3e3f"
+ "404142434445464748494a4b4c4d4e4f"
+ "505152535455565758595a5b5c5d5e5f"
+ "606162636465666768696a6b6c6d6e6f"
+ "707172737475767778797a7b7c7d7e7f"
+ "808182838485868788898a8b8c8d8e8f"
+ "909192939495969798999a9b9c9d9e9f"
+ "a0a1a2a3a4a5a6a7a8a9aaabacadaeaf"
+ "b0b1b2b3b4b5b6b7b8b9babbbcbdbebf"
+ "c0c1c2c3c4c5c6c7c8c9cacbcccdcecf"
+ "d0d1d2d3d4d5d6d7d8d9dadbdcdddedf"
+ "e0e1e2e3e4e5e6e7e8e9eaebecedeeef"
+ "f0f1f2f3f4f5f6f7f8f9fafbfcfdfeff";
+
+ABSL_CONST_INIT ABSL_DLL const char two_ASCII_digits[100][2] = {
+ {'0', '0'}, {'0', '1'}, {'0', '2'}, {'0', '3'}, {'0', '4'}, {'0', '5'},
+ {'0', '6'}, {'0', '7'}, {'0', '8'}, {'0', '9'}, {'1', '0'}, {'1', '1'},
+ {'1', '2'}, {'1', '3'}, {'1', '4'}, {'1', '5'}, {'1', '6'}, {'1', '7'},
+ {'1', '8'}, {'1', '9'}, {'2', '0'}, {'2', '1'}, {'2', '2'}, {'2', '3'},
+ {'2', '4'}, {'2', '5'}, {'2', '6'}, {'2', '7'}, {'2', '8'}, {'2', '9'},
+ {'3', '0'}, {'3', '1'}, {'3', '2'}, {'3', '3'}, {'3', '4'}, {'3', '5'},
+ {'3', '6'}, {'3', '7'}, {'3', '8'}, {'3', '9'}, {'4', '0'}, {'4', '1'},
+ {'4', '2'}, {'4', '3'}, {'4', '4'}, {'4', '5'}, {'4', '6'}, {'4', '7'},
+ {'4', '8'}, {'4', '9'}, {'5', '0'}, {'5', '1'}, {'5', '2'}, {'5', '3'},
+ {'5', '4'}, {'5', '5'}, {'5', '6'}, {'5', '7'}, {'5', '8'}, {'5', '9'},
+ {'6', '0'}, {'6', '1'}, {'6', '2'}, {'6', '3'}, {'6', '4'}, {'6', '5'},
+ {'6', '6'}, {'6', '7'}, {'6', '8'}, {'6', '9'}, {'7', '0'}, {'7', '1'},
+ {'7', '2'}, {'7', '3'}, {'7', '4'}, {'7', '5'}, {'7', '6'}, {'7', '7'},
+ {'7', '8'}, {'7', '9'}, {'8', '0'}, {'8', '1'}, {'8', '2'}, {'8', '3'},
+ {'8', '4'}, {'8', '5'}, {'8', '6'}, {'8', '7'}, {'8', '8'}, {'8', '9'},
+ {'9', '0'}, {'9', '1'}, {'9', '2'}, {'9', '3'}, {'9', '4'}, {'9', '5'},
+ {'9', '6'}, {'9', '7'}, {'9', '8'}, {'9', '9'}};
+
bool safe_strto32_base(absl::string_view text, int32_t* value, int base) {
return safe_int_internal<int32_t>(text, value, base);
}
@@ -893,6 +1062,10 @@
return safe_int_internal<int64_t>(text, value, base);
}
+bool safe_strto128_base(absl::string_view text, int128* value, int base) {
+ return safe_int_internal<absl::int128>(text, value, base);
+}
+
bool safe_strtou32_base(absl::string_view text, uint32_t* value, int base) {
return safe_uint_internal<uint32_t>(text, value, base);
}
@@ -900,6 +1073,11 @@
bool safe_strtou64_base(absl::string_view text, uint64_t* value, int base) {
return safe_uint_internal<uint64_t>(text, value, base);
}
-} // namespace numbers_internal
+bool safe_strtou128_base(absl::string_view text, uint128* value, int base) {
+ return safe_uint_internal<absl::uint128>(text, value, base);
+}
+
+} // namespace numbers_internal
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/numbers.h b/absl/strings/numbers.h
index 100839b..ffc738f 100644
--- a/absl/strings/numbers.h
+++ b/absl/strings/numbers.h
@@ -1,4 +1,3 @@
-//
// Copyright 2017 The Abseil Authors.
//
// Licensed under the Apache License, Version 2.0 (the "License");
@@ -24,6 +23,10 @@
#ifndef ABSL_STRINGS_NUMBERS_H_
#define ABSL_STRINGS_NUMBERS_H_
+#ifdef __SSE4_2__
+#include <x86intrin.h>
+#endif
+
#include <cstddef>
#include <cstdlib>
#include <cstring>
@@ -32,38 +35,54 @@
#include <string>
#include <type_traits>
+#include "absl/base/config.h"
+#ifdef __SSE4_2__
+// TODO(jorg): Remove this when we figure out the right way
+// to swap bytes on SSE 4.2 that works with the compilers
+// we claim to support. Also, add tests for the compiler
+// that doesn't support the Intel _bswap64 intrinsic but
+// does support all the SSE 4.2 intrinsics
+#include "absl/base/internal/endian.h"
+#endif
#include "absl/base/macros.h"
#include "absl/base/port.h"
+#include "absl/numeric/bits.h"
#include "absl/numeric/int128.h"
#include "absl/strings/string_view.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
// SimpleAtoi()
//
-// Converts the given string into an integer value, returning `true` if
-// successful. The string must reflect a base-10 integer (optionally followed or
-// preceded by ASCII whitespace) whose value falls within the range of the
-// integer type. If any errors are encountered, this function returns `false`,
-// leaving `out` in an unspecified state.
+// Converts the given string (optionally followed or preceded by ASCII
+// whitespace) into an integer value, returning `true` if successful. The string
+// must reflect a base-10 integer whose value falls within the range of the
+// integer type (optionally preceded by a `+` or `-`). If any errors are
+// encountered, this function returns `false`, leaving `out` in an unspecified
+// state.
template <typename int_type>
ABSL_MUST_USE_RESULT bool SimpleAtoi(absl::string_view str, int_type* out);
// SimpleAtof()
//
// Converts the given string (optionally followed or preceded by ASCII
-// whitespace) into a float, which may be rounded on overflow or underflow.
+// whitespace) into a float, which may be rounded on overflow or underflow,
+// returning `true` if successful.
// See https://en.cppreference.com/w/c/string/byte/strtof for details about the
-// allowed formats for `str`. If any errors are encountered, this function
+// allowed formats for `str`, except SimpleAtof() is locale-independent and will
+// always use the "C" locale. If any errors are encountered, this function
// returns `false`, leaving `out` in an unspecified state.
ABSL_MUST_USE_RESULT bool SimpleAtof(absl::string_view str, float* out);
// SimpleAtod()
//
// Converts the given string (optionally followed or preceded by ASCII
-// whitespace) into a double, which may be rounded on overflow or underflow.
+// whitespace) into a double, which may be rounded on overflow or underflow,
+// returning `true` if successful.
// See https://en.cppreference.com/w/c/string/byte/strtof for details about the
-// allowed formats for `str`. If any errors are encountered, this function
+// allowed formats for `str`, except SimpleAtod is locale-independent and will
+// always use the "C" locale. If any errors are encountered, this function
// returns `false`, leaving `out` in an unspecified state.
ABSL_MUST_USE_RESULT bool SimpleAtod(absl::string_view str, double* out);
@@ -77,18 +96,42 @@
// unspecified state.
ABSL_MUST_USE_RESULT bool SimpleAtob(absl::string_view str, bool* out);
+ABSL_NAMESPACE_END
} // namespace absl
// End of public API. Implementation details follow.
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace numbers_internal {
+// Digit conversion.
+ABSL_DLL extern const char kHexChar[17]; // 0123456789abcdef
+ABSL_DLL extern const char
+ kHexTable[513]; // 000102030405060708090a0b0c0d0e0f1011...
+ABSL_DLL extern const char
+ two_ASCII_digits[100][2]; // 00, 01, 02, 03...
+
+// Writes a two-character representation of 'i' to 'buf'. 'i' must be in the
+// range 0 <= i < 100, and buf must have space for two characters. Example:
+// char buf[2];
+// PutTwoDigits(42, buf);
+// // buf[0] == '4'
+// // buf[1] == '2'
+inline void PutTwoDigits(size_t i, char* buf) {
+ assert(i < 100);
+ memcpy(buf, two_ASCII_digits[i], 2);
+}
+
// safe_strto?() functions for implementing SimpleAtoi()
bool safe_strto32_base(absl::string_view text, int32_t* value, int base);
bool safe_strto64_base(absl::string_view text, int64_t* value, int base);
+bool safe_strto128_base(absl::string_view text, absl::int128* value,
+ int base);
bool safe_strtou32_base(absl::string_view text, uint32_t* value, int base);
bool safe_strtou64_base(absl::string_view text, uint64_t* value, int base);
+bool safe_strtou128_base(absl::string_view text, absl::uint128* value,
+ int base);
static const int kFastToBufferSize = 32;
static const int kSixDigitsToBufferSize = 16;
@@ -170,20 +213,53 @@
return parsed;
}
+// FastHexToBufferZeroPad16()
+//
+// Outputs `val` into `out` as if by `snprintf(out, 17, "%016x", val)` but
+// without the terminating null character. Thus `out` must be of length >= 16.
+// Returns the number of non-pad digits of the output (it can never be zero
+// since 0 has one digit).
+inline size_t FastHexToBufferZeroPad16(uint64_t val, char* out) {
+#ifdef __SSE4_2__
+ uint64_t be = absl::big_endian::FromHost64(val);
+ const auto kNibbleMask = _mm_set1_epi8(0xf);
+ const auto kHexDigits = _mm_setr_epi8('0', '1', '2', '3', '4', '5', '6', '7',
+ '8', '9', 'a', 'b', 'c', 'd', 'e', 'f');
+ auto v = _mm_loadl_epi64(reinterpret_cast<__m128i*>(&be)); // load lo dword
+ auto v4 = _mm_srli_epi64(v, 4); // shift 4 right
+ auto il = _mm_unpacklo_epi8(v4, v); // interleave bytes
+ auto m = _mm_and_si128(il, kNibbleMask); // mask out nibbles
+ auto hexchars = _mm_shuffle_epi8(kHexDigits, m); // hex chars
+ _mm_storeu_si128(reinterpret_cast<__m128i*>(out), hexchars);
+#else
+ for (int i = 0; i < 8; ++i) {
+ auto byte = (val >> (56 - 8 * i)) & 0xFF;
+ auto* hex = &absl::numbers_internal::kHexTable[byte * 2];
+ std::memcpy(out + 2 * i, hex, 2);
+ }
+#endif
+ // | 0x1 so that even 0 has 1 digit.
+ return 16 - countl_zero(val | 0x1) / 4;
+}
+
} // namespace numbers_internal
-// SimpleAtoi()
-//
-// Converts a string to an integer, using `safe_strto?()` functions for actual
-// parsing, returning `true` if successful. The `safe_strto?()` functions apply
-// strict checking; the string must be a base-10 integer, optionally followed or
-// preceded by ASCII whitespace, with a value in the range of the corresponding
-// integer type.
template <typename int_type>
ABSL_MUST_USE_RESULT bool SimpleAtoi(absl::string_view str, int_type* out) {
return numbers_internal::safe_strtoi_base(str, out, 10);
}
+ABSL_MUST_USE_RESULT inline bool SimpleAtoi(absl::string_view str,
+ absl::int128* out) {
+ return numbers_internal::safe_strto128_base(str, out, 10);
+}
+
+ABSL_MUST_USE_RESULT inline bool SimpleAtoi(absl::string_view str,
+ absl::uint128* out) {
+ return numbers_internal::safe_strtou128_base(str, out, 10);
+}
+
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_NUMBERS_H_
diff --git a/absl/strings/numbers_benchmark.cc b/absl/strings/numbers_benchmark.cc
index 54dbedd..6e79b3e 100644
--- a/absl/strings/numbers_benchmark.cc
+++ b/absl/strings/numbers_benchmark.cc
@@ -20,6 +20,8 @@
#include "benchmark/benchmark.h"
#include "absl/base/internal/raw_logging.h"
+#include "absl/random/distributions.h"
+#include "absl/random/random.h"
#include "absl/strings/numbers.h"
namespace {
@@ -260,4 +262,25 @@
->ArgPair(10, 4)
->ArgPair(10, 8);
+void BM_FastHexToBufferZeroPad16(benchmark::State& state) {
+ absl::BitGen rng;
+ std::vector<uint64_t> nums;
+ nums.resize(1000);
+ auto min = std::numeric_limits<uint64_t>::min();
+ auto max = std::numeric_limits<uint64_t>::max();
+ for (auto& num : nums) {
+ num = absl::LogUniform(rng, min, max);
+ }
+
+ char buf[16];
+ while (state.KeepRunningBatch(nums.size())) {
+ for (auto num : nums) {
+ auto digits = absl::numbers_internal::FastHexToBufferZeroPad16(num, buf);
+ benchmark::DoNotOptimize(digits);
+ benchmark::DoNotOptimize(buf);
+ }
+ }
+}
+BENCHMARK(BM_FastHexToBufferZeroPad16);
+
} // namespace
diff --git a/absl/strings/numbers_test.cc b/absl/strings/numbers_test.cc
index 77d7e78..27616bf 100644
--- a/absl/strings/numbers_test.cc
+++ b/absl/strings/numbers_test.cc
@@ -17,6 +17,7 @@
#include "absl/strings/numbers.h"
#include <sys/types.h>
+
#include <cfenv> // NOLINT(build/c++11)
#include <cinttypes>
#include <climits>
@@ -36,13 +37,16 @@
#include "gmock/gmock.h"
#include "gtest/gtest.h"
#include "absl/base/internal/raw_logging.h"
-#include "absl/strings/str_cat.h"
-
+#include "absl/random/distributions.h"
+#include "absl/random/random.h"
#include "absl/strings/internal/numbers_test_common.h"
+#include "absl/strings/internal/ostringstream.h"
#include "absl/strings/internal/pow10_helper.h"
+#include "absl/strings/str_cat.h"
namespace {
+using absl::SimpleAtoi;
using absl::numbers_internal::kSixDigitsToBufferSize;
using absl::numbers_internal::safe_strto32_base;
using absl::numbers_internal::safe_strto64_base;
@@ -52,7 +56,6 @@
using absl::strings_internal::Itoa;
using absl::strings_internal::strtouint32_test_cases;
using absl::strings_internal::strtouint64_test_cases;
-using absl::SimpleAtoi;
using testing::Eq;
using testing::MatchesRegex;
@@ -204,6 +207,9 @@
std::string actual = absl::StrCat(absl::Hex(v, absl::kZeroPad16));
snprintf(expected, sizeof(expected), "%016" PRIx64, static_cast<uint64_t>(v));
EXPECT_EQ(expected, actual) << " Input " << v;
+ actual = absl::StrCat(absl::Hex(v, absl::kSpacePad16));
+ snprintf(expected, sizeof(expected), "%16" PRIx64, static_cast<uint64_t>(v));
+ EXPECT_EQ(expected, actual) << " Input " << v;
}
TEST(Numbers, TestFastPrints) {
@@ -244,7 +250,9 @@
template <typename int_type, typename in_val_type>
void VerifySimpleAtoiGood(in_val_type in_value, int_type exp_value) {
- std::string s = absl::StrCat(in_value);
+ std::string s;
+ // (u)int128 can be streamed but not StrCat'd.
+ absl::strings_internal::OStringStream(&s) << in_value;
int_type x = static_cast<int_type>(~exp_value);
EXPECT_TRUE(SimpleAtoi(s, &x))
<< "in_value=" << in_value << " s=" << s << " x=" << x;
@@ -256,7 +264,9 @@
template <typename int_type, typename in_val_type>
void VerifySimpleAtoiBad(in_val_type in_value) {
- std::string s = absl::StrCat(in_value);
+ std::string s;
+ // (u)int128 can be streamed but not StrCat'd.
+ absl::strings_internal::OStringStream(&s) << in_value;
int_type x;
EXPECT_FALSE(SimpleAtoi(s, &x));
EXPECT_FALSE(SimpleAtoi(s.c_str(), &x));
@@ -320,18 +330,68 @@
VerifySimpleAtoiGood<uint64_t>(std::numeric_limits<uint64_t>::max(),
std::numeric_limits<uint64_t>::max());
+ // SimpleAtoi(absl::string_view, absl::uint128)
+ VerifySimpleAtoiGood<absl::uint128>(0, 0);
+ VerifySimpleAtoiGood<absl::uint128>(42, 42);
+ VerifySimpleAtoiBad<absl::uint128>(-42);
+
+ VerifySimpleAtoiBad<absl::uint128>(std::numeric_limits<int32_t>::min());
+ VerifySimpleAtoiGood<absl::uint128>(std::numeric_limits<int32_t>::max(),
+ std::numeric_limits<int32_t>::max());
+ VerifySimpleAtoiGood<absl::uint128>(std::numeric_limits<uint32_t>::max(),
+ std::numeric_limits<uint32_t>::max());
+ VerifySimpleAtoiBad<absl::uint128>(std::numeric_limits<int64_t>::min());
+ VerifySimpleAtoiGood<absl::uint128>(std::numeric_limits<int64_t>::max(),
+ std::numeric_limits<int64_t>::max());
+ VerifySimpleAtoiGood<absl::uint128>(std::numeric_limits<uint64_t>::max(),
+ std::numeric_limits<uint64_t>::max());
+ VerifySimpleAtoiGood<absl::uint128>(
+ std::numeric_limits<absl::uint128>::max(),
+ std::numeric_limits<absl::uint128>::max());
+
+ // SimpleAtoi(absl::string_view, absl::int128)
+ VerifySimpleAtoiGood<absl::int128>(0, 0);
+ VerifySimpleAtoiGood<absl::int128>(42, 42);
+ VerifySimpleAtoiGood<absl::int128>(-42, -42);
+
+ VerifySimpleAtoiGood<absl::int128>(std::numeric_limits<int32_t>::min(),
+ std::numeric_limits<int32_t>::min());
+ VerifySimpleAtoiGood<absl::int128>(std::numeric_limits<int32_t>::max(),
+ std::numeric_limits<int32_t>::max());
+ VerifySimpleAtoiGood<absl::int128>(std::numeric_limits<uint32_t>::max(),
+ std::numeric_limits<uint32_t>::max());
+ VerifySimpleAtoiGood<absl::int128>(std::numeric_limits<int64_t>::min(),
+ std::numeric_limits<int64_t>::min());
+ VerifySimpleAtoiGood<absl::int128>(std::numeric_limits<int64_t>::max(),
+ std::numeric_limits<int64_t>::max());
+ VerifySimpleAtoiGood<absl::int128>(std::numeric_limits<uint64_t>::max(),
+ std::numeric_limits<uint64_t>::max());
+ VerifySimpleAtoiGood<absl::int128>(
+ std::numeric_limits<absl::int128>::min(),
+ std::numeric_limits<absl::int128>::min());
+ VerifySimpleAtoiGood<absl::int128>(
+ std::numeric_limits<absl::int128>::max(),
+ std::numeric_limits<absl::int128>::max());
+ VerifySimpleAtoiBad<absl::int128>(std::numeric_limits<absl::uint128>::max());
+
// Some other types
VerifySimpleAtoiGood<int>(-42, -42);
VerifySimpleAtoiGood<int32_t>(-42, -42);
VerifySimpleAtoiGood<uint32_t>(42, 42);
VerifySimpleAtoiGood<unsigned int>(42, 42);
VerifySimpleAtoiGood<int64_t>(-42, -42);
- VerifySimpleAtoiGood<long>(-42, -42); // NOLINT(runtime/int)
+ VerifySimpleAtoiGood<long>(-42, -42); // NOLINT: runtime-int
VerifySimpleAtoiGood<uint64_t>(42, 42);
VerifySimpleAtoiGood<size_t>(42, 42);
VerifySimpleAtoiGood<std::string::size_type>(42, 42);
}
+TEST(NumbersTest, Atod) {
+ double d;
+ EXPECT_TRUE(absl::SimpleAtod("nan", &d));
+ EXPECT_TRUE(std::isnan(d));
+}
+
TEST(NumbersTest, Atoenum) {
enum E01 {
E01_zero = 0,
@@ -455,7 +515,7 @@
EXPECT_TRUE(safe_strto32_base(std::string("0x1234"), &value, 16));
EXPECT_EQ(0x1234, value);
- // Base-10 std::string version.
+ // Base-10 string version.
EXPECT_TRUE(safe_strto32_base("1234", &value, 10));
EXPECT_EQ(1234, value);
}
@@ -596,7 +656,7 @@
EXPECT_TRUE(safe_strto64_base(std::string("0x1234"), &value, 16));
EXPECT_EQ(0x1234, value);
- // Base-10 std::string version.
+ // Base-10 string version.
EXPECT_TRUE(safe_strto64_base("1234", &value, 10));
EXPECT_EQ(1234, value);
}
@@ -652,6 +712,91 @@
TEST(stringtest, safe_strtou64_random) {
test_random_integer_parse_base<uint64_t>(&safe_strtou64_base);
}
+TEST(stringtest, safe_strtou128_random) {
+ // random number generators don't work for uint128, and
+ // uint128 can be streamed but not StrCat'd, so this code must be custom
+ // implemented for uint128, but is generally the same as what's above.
+ // test_random_integer_parse_base<absl::uint128>(
+ // &absl::numbers_internal::safe_strtou128_base);
+ using RandomEngine = std::minstd_rand0;
+ using IntType = absl::uint128;
+ constexpr auto parse_func = &absl::numbers_internal::safe_strtou128_base;
+
+ std::random_device rd;
+ RandomEngine rng(rd());
+ std::uniform_int_distribution<uint64_t> random_uint64(
+ std::numeric_limits<uint64_t>::min());
+ std::uniform_int_distribution<int> random_base(2, 35);
+
+ for (size_t i = 0; i < kNumRandomTests; i++) {
+ IntType value = random_uint64(rng);
+ value = (value << 64) + random_uint64(rng);
+ int base = random_base(rng);
+ std::string str_value;
+ EXPECT_TRUE(Itoa<IntType>(value, base, &str_value));
+ IntType parsed_value;
+
+ // Test successful parse
+ EXPECT_TRUE(parse_func(str_value, &parsed_value, base));
+ EXPECT_EQ(parsed_value, value);
+
+ // Test overflow
+ std::string s;
+ absl::strings_internal::OStringStream(&s)
+ << std::numeric_limits<IntType>::max() << value;
+ EXPECT_FALSE(parse_func(s, &parsed_value, base));
+
+ // Test underflow
+ s.clear();
+ absl::strings_internal::OStringStream(&s) << "-" << value;
+ EXPECT_FALSE(parse_func(s, &parsed_value, base));
+ }
+}
+TEST(stringtest, safe_strto128_random) {
+ // random number generators don't work for int128, and
+ // int128 can be streamed but not StrCat'd, so this code must be custom
+ // implemented for int128, but is generally the same as what's above.
+ // test_random_integer_parse_base<absl::int128>(
+ // &absl::numbers_internal::safe_strto128_base);
+ using RandomEngine = std::minstd_rand0;
+ using IntType = absl::int128;
+ constexpr auto parse_func = &absl::numbers_internal::safe_strto128_base;
+
+ std::random_device rd;
+ RandomEngine rng(rd());
+ std::uniform_int_distribution<int64_t> random_int64(
+ std::numeric_limits<int64_t>::min());
+ std::uniform_int_distribution<uint64_t> random_uint64(
+ std::numeric_limits<uint64_t>::min());
+ std::uniform_int_distribution<int> random_base(2, 35);
+
+ for (size_t i = 0; i < kNumRandomTests; ++i) {
+ int64_t high = random_int64(rng);
+ uint64_t low = random_uint64(rng);
+ IntType value = absl::MakeInt128(high, low);
+
+ int base = random_base(rng);
+ std::string str_value;
+ EXPECT_TRUE(Itoa<IntType>(value, base, &str_value));
+ IntType parsed_value;
+
+ // Test successful parse
+ EXPECT_TRUE(parse_func(str_value, &parsed_value, base));
+ EXPECT_EQ(parsed_value, value);
+
+ // Test overflow
+ std::string s;
+ absl::strings_internal::OStringStream(&s)
+ << std::numeric_limits<IntType>::max() << value;
+ EXPECT_FALSE(parse_func(s, &parsed_value, base));
+
+ // Test underflow
+ s.clear();
+ absl::strings_internal::OStringStream(&s)
+ << std::numeric_limits<IntType>::min() << value;
+ EXPECT_FALSE(parse_func(s, &parsed_value, base));
+ }
+}
TEST(stringtest, safe_strtou32_base) {
for (int i = 0; strtouint32_test_cases()[i].str != nullptr; ++i) {
@@ -713,11 +858,11 @@
}
}
-// feenableexcept() and fedisableexcept() are missing on macOS, MSVC,
-// and WebAssembly.
-#if defined(_MSC_VER) || defined(__APPLE__) || defined(__EMSCRIPTEN__)
-#define ABSL_MISSING_FEENABLEEXCEPT 1
-#define ABSL_MISSING_FEDISABLEEXCEPT 1
+// feenableexcept() and fedisableexcept() are extensions supported by some libc
+// implementations.
+#if defined(__GLIBC__) || defined(__BIONIC__)
+#define ABSL_HAVE_FEENABLEEXCEPT 1
+#define ABSL_HAVE_FEDISABLEEXCEPT 1
#endif
class SimpleDtoaTest : public testing::Test {
@@ -725,7 +870,7 @@
void SetUp() override {
// Store the current floating point env & clear away any pending exceptions.
feholdexcept(&fp_env_);
-#ifndef ABSL_MISSING_FEENABLEEXCEPT
+#ifdef ABSL_HAVE_FEENABLEEXCEPT
// Turn on floating point exceptions.
feenableexcept(FE_DIVBYZERO | FE_INVALID | FE_OVERFLOW);
#endif
@@ -735,7 +880,7 @@
// Restore the floating point environment to the original state.
// In theory fedisableexcept is unnecessary; fesetenv will also do it.
// In practice, our toolchains have subtle bugs.
-#ifndef ABSL_MISSING_FEDISABLEEXCEPT
+#ifdef ABSL_HAVE_FEDISABLEEXCEPT
fedisableexcept(FE_DIVBYZERO | FE_INVALID | FE_OVERFLOW);
#endif
fesetenv(&fp_env_);
@@ -1184,4 +1329,28 @@
}
}
+void TestFastHexToBufferZeroPad16(uint64_t v) {
+ char buf[16];
+ auto digits = absl::numbers_internal::FastHexToBufferZeroPad16(v, buf);
+ absl::string_view res(buf, 16);
+ char buf2[17];
+ snprintf(buf2, sizeof(buf2), "%016" PRIx64, v);
+ EXPECT_EQ(res, buf2) << v;
+ size_t expected_digits = snprintf(buf2, sizeof(buf2), "%" PRIx64, v);
+ EXPECT_EQ(digits, expected_digits) << v;
+}
+
+TEST(FastHexToBufferZeroPad16, Smoke) {
+ TestFastHexToBufferZeroPad16(std::numeric_limits<uint64_t>::min());
+ TestFastHexToBufferZeroPad16(std::numeric_limits<uint64_t>::max());
+ TestFastHexToBufferZeroPad16(std::numeric_limits<int64_t>::min());
+ TestFastHexToBufferZeroPad16(std::numeric_limits<int64_t>::max());
+ absl::BitGen rng;
+ for (int i = 0; i < 100000; ++i) {
+ TestFastHexToBufferZeroPad16(
+ absl::LogUniform(rng, std::numeric_limits<uint64_t>::min(),
+ std::numeric_limits<uint64_t>::max()));
+ }
+}
+
} // namespace
diff --git a/absl/strings/str_cat.cc b/absl/strings/str_cat.cc
index 9961944..dd5d25b 100644
--- a/absl/strings/str_cat.cc
+++ b/absl/strings/str_cat.cc
@@ -15,34 +15,34 @@
#include "absl/strings/str_cat.h"
#include <assert.h>
+
#include <algorithm>
#include <cstdint>
#include <cstring>
#include "absl/strings/ascii.h"
#include "absl/strings/internal/resize_uninitialized.h"
+#include "absl/strings/numbers.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
AlphaNum::AlphaNum(Hex hex) {
+ static_assert(numbers_internal::kFastToBufferSize >= 32,
+ "This function only works when output buffer >= 32 bytes long");
char* const end = &digits_[numbers_internal::kFastToBufferSize];
- char* writer = end;
- uint64_t value = hex.value;
- static const char hexdigits[] = "0123456789abcdef";
- do {
- *--writer = hexdigits[value & 0xF];
- value >>= 4;
- } while (value != 0);
-
- char* beg;
- if (end - writer < hex.width) {
- beg = end - hex.width;
- std::fill_n(beg, writer - beg, hex.fill);
+ auto real_width =
+ absl::numbers_internal::FastHexToBufferZeroPad16(hex.value, end - 16);
+ if (real_width >= hex.width) {
+ piece_ = absl::string_view(end - real_width, real_width);
} else {
- beg = writer;
+ // Pad first 16 chars because FastHexToBufferZeroPad16 pads only to 16 and
+ // max pad width can be up to 20.
+ std::memset(end - 32, hex.fill, 16);
+ // Patch up everything else up to the real_width.
+ std::memset(end - real_width - 16, hex.fill, 16);
+ piece_ = absl::string_view(end - hex.width, hex.width);
}
-
- piece_ = absl::string_view(beg, end - beg);
}
AlphaNum::AlphaNum(Dec dec) {
@@ -141,12 +141,12 @@
std::string CatPieces(std::initializer_list<absl::string_view> pieces) {
std::string result;
size_t total_size = 0;
- for (const absl::string_view piece : pieces) total_size += piece.size();
+ for (const absl::string_view& piece : pieces) total_size += piece.size();
strings_internal::STLStringResizeUninitialized(&result, total_size);
char* const begin = &result[0];
char* out = begin;
- for (const absl::string_view piece : pieces) {
+ for (const absl::string_view& piece : pieces) {
const size_t this_size = piece.size();
if (this_size != 0) {
memcpy(out, piece.data(), this_size);
@@ -170,7 +170,7 @@
std::initializer_list<absl::string_view> pieces) {
size_t old_size = dest->size();
size_t total_size = old_size;
- for (const absl::string_view piece : pieces) {
+ for (const absl::string_view& piece : pieces) {
ASSERT_NO_OVERLAP(*dest, piece);
total_size += piece.size();
}
@@ -178,7 +178,7 @@
char* const begin = &(*dest)[0];
char* out = begin + old_size;
- for (const absl::string_view piece : pieces) {
+ for (const absl::string_view& piece : pieces) {
const size_t this_size = piece.size();
if (this_size != 0) {
memcpy(out, piece.data(), this_size);
@@ -242,4 +242,5 @@
assert(out == begin + dest->size());
}
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/str_cat.h b/absl/strings/str_cat.h
index a99aac0..a8a85c7 100644
--- a/absl/strings/str_cat.h
+++ b/absl/strings/str_cat.h
@@ -64,6 +64,7 @@
#include "absl/strings/string_view.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace strings_internal {
// AlphaNumBuffer allows a way to pass a string to StrCat without having to do
@@ -252,7 +253,7 @@
const std::basic_string<char, std::char_traits<char>, Allocator>& str)
: piece_(str) {}
- // Use std::string literals ":" instead of character literals ':'.
+ // Use string literals ":" instead of character literals ':'.
AlphaNum(char c) = delete; // NOLINT(runtime/explicit)
AlphaNum(const AlphaNum&) = delete;
@@ -290,7 +291,8 @@
// StrCat()
// -----------------------------------------------------------------------------
//
-// Merges given strings or numbers, using no delimiter(s).
+// Merges given strings or numbers, using no delimiter(s), returning the merged
+// result as a string.
//
// `StrCat()` is designed to be the fastest possible way to construct a string
// out of a mix of raw C strings, string_views, strings, bool values,
@@ -400,6 +402,7 @@
return result;
}
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_STR_CAT_H_
diff --git a/absl/strings/str_cat_benchmark.cc b/absl/strings/str_cat_benchmark.cc
index 14c63b3..02c4dbe 100644
--- a/absl/strings/str_cat_benchmark.cc
+++ b/absl/strings/str_cat_benchmark.cc
@@ -23,7 +23,7 @@
namespace {
const char kStringOne[] = "Once Upon A Time, ";
-const char kStringTwo[] = "There was a std::string benchmark";
+const char kStringTwo[] = "There was a string benchmark";
// We want to include negative numbers in the benchmark, so this function
// is used to count 0, 1, -1, 2, -2, 3, -3, ...
@@ -137,4 +137,51 @@
}
BENCHMARK(BM_DoubleToString_By_SixDigits);
+template <typename... Chunks>
+void BM_StrAppendImpl(benchmark::State& state, size_t total_bytes,
+ Chunks... chunks) {
+ for (auto s : state) {
+ std::string result;
+ while (result.size() < total_bytes) {
+ absl::StrAppend(&result, chunks...);
+ benchmark::DoNotOptimize(result);
+ }
+ }
+}
+
+void BM_StrAppend(benchmark::State& state) {
+ const int total_bytes = state.range(0);
+ const int chunks_at_a_time = state.range(1);
+ const absl::string_view kChunk = "0123456789";
+
+ switch (chunks_at_a_time) {
+ case 1:
+ return BM_StrAppendImpl(state, total_bytes, kChunk);
+ case 2:
+ return BM_StrAppendImpl(state, total_bytes, kChunk, kChunk);
+ case 4:
+ return BM_StrAppendImpl(state, total_bytes, kChunk, kChunk, kChunk,
+ kChunk);
+ case 8:
+ return BM_StrAppendImpl(state, total_bytes, kChunk, kChunk, kChunk,
+ kChunk, kChunk, kChunk, kChunk, kChunk);
+ default:
+ std::abort();
+ }
+}
+
+template <typename B>
+void StrAppendConfig(B* benchmark) {
+ for (int bytes : {10, 100, 1000, 10000}) {
+ for (int chunks : {1, 2, 4, 8}) {
+ // Only add the ones that divide properly. Otherwise we are over counting.
+ if (bytes % (10 * chunks) == 0) {
+ benchmark->Args({bytes, chunks});
+ }
+ }
+ }
+}
+
+BENCHMARK(BM_StrAppend)->Apply(StrAppendConfig);
+
} // namespace
diff --git a/absl/strings/str_cat_test.cc b/absl/strings/str_cat_test.cc
index be39880..f3770dc 100644
--- a/absl/strings/str_cat_test.cc
+++ b/absl/strings/str_cat_test.cc
@@ -162,7 +162,7 @@
EXPECT_EQ(result, "12345678910, 10987654321!");
std::string one =
- "1"; // Actually, it's the size of this std::string that we want; a
+ "1"; // Actually, it's the size of this string that we want; a
// 64-bit build distinguishes between size_t and uint64_t,
// even though they're both unsigned 64-bit values.
result = absl::StrCat("And a ", one.size(), " and a ",
@@ -375,7 +375,7 @@
EXPECT_EQ(result.substr(old_size), "12345678910, 10987654321!");
std::string one =
- "1"; // Actually, it's the size of this std::string that we want; a
+ "1"; // Actually, it's the size of this string that we want; a
// 64-bit build distinguishes between size_t and uint64_t,
// even though they're both unsigned 64-bit values.
old_size = result.size();
@@ -463,7 +463,7 @@
}
TEST(StrAppend, CornerCasesNonEmptyAppend) {
- for (std::string result : {"hello", "a std::string too long to fit in the SSO"}) {
+ for (std::string result : {"hello", "a string too long to fit in the SSO"}) {
const std::string expected = result;
absl::StrAppend(&result, "");
EXPECT_EQ(result, expected);
diff --git a/absl/strings/str_format.h b/absl/strings/str_format.h
index 607e2bc..0146510 100644
--- a/absl/strings/str_format.h
+++ b/absl/strings/str_format.h
@@ -19,7 +19,7 @@
//
// The `str_format` library is a typesafe replacement for the family of
// `printf()` string formatting routines within the `<cstdio>` standard library
-// header. Like the `printf` family, the `str_format` uses a "format string" to
+// header. Like the `printf` family, `str_format` uses a "format string" to
// perform argument substitutions based on types. See the `FormatSpec` section
// below for format string documentation.
//
@@ -57,8 +57,7 @@
// arbitrary sink types:
//
// * A generic `Format()` function to write outputs to arbitrary sink types,
-// which must implement a `RawSinkFormat` interface. (See
-// `str_format_sink.h` for more information.)
+// which must implement a `FormatRawSink` interface.
//
// * A `FormatUntyped()` function that is similar to `Format()` except it is
// loosely typed. `FormatUntyped()` is not a template and does not perform
@@ -66,8 +65,7 @@
// boolean from a runtime check.
//
// In addition, the `str_format` library provides extension points for
-// augmenting formatting to new types. These extensions are fully documented
-// within the `str_format_extension.h` header file.
+// augmenting formatting to new types. See "StrFormat Extensions" below.
#ifndef ABSL_STRINGS_STR_FORMAT_H_
#define ABSL_STRINGS_STR_FORMAT_H_
@@ -82,6 +80,7 @@
#include "absl/strings/internal/str_format/parser.h" // IWYU pragma: export
namespace absl {
+ABSL_NAMESPACE_BEGIN
// UntypedFormatSpec
//
@@ -254,8 +253,8 @@
// argument, etc.
template <typename... Args>
-using FormatSpec =
- typename str_format_internal::FormatSpecDeductionBarrier<Args...>::type;
+using FormatSpec = str_format_internal::FormatSpecTemplate<
+ str_format_internal::ArgumentToConv<Args>()...>;
// ParsedFormat
//
@@ -282,9 +281,36 @@
// } else {
// ... error case ...
// }
+
+#if defined(__cpp_nontype_template_parameter_auto)
+// If C++17 is available, an 'extended' format is also allowed that can specify
+// multiple conversion characters per format argument, using a combination of
+// `absl::FormatConversionCharSet` enum values (logically a set union)
+// via the `|` operator. (Single character-based arguments are still accepted,
+// but cannot be combined). Some common conversions also have predefined enum
+// values, such as `absl::FormatConversionCharSet::kIntegral`.
+//
+// Example:
+// // Extended format supports multiple conversion characters per argument,
+// // specified via a combination of `FormatConversionCharSet` enums.
+// using MyFormat = absl::ParsedFormat<absl::FormatConversionCharSet::d |
+// absl::FormatConversionCharSet::x>;
+// MyFormat GetFormat(bool use_hex) {
+// if (use_hex) return MyFormat("foo %x bar");
+// return MyFormat("foo %d bar");
+// }
+// // `format` can be used with any value that supports 'd' and 'x',
+// // like `int`.
+// auto format = GetFormat(use_hex);
+// value = StringF(format, i);
+template <auto... Conv>
+using ParsedFormat = absl::str_format_internal::ExtendedParsedFormat<
+ absl::str_format_internal::ToFormatConversionCharSet(Conv)...>;
+#else
template <char... Conv>
using ParsedFormat = str_format_internal::ExtendedParsedFormat<
- str_format_internal::ConversionCharToConv(Conv)...>;
+ absl::str_format_internal::ToFormatConversionCharSet(Conv)...>;
+#endif // defined(__cpp_nontype_template_parameter_auto)
// StrFormat()
//
@@ -400,6 +426,12 @@
// This function is functionally equivalent to `std::snprintf()` (and
// type-safe); prefer `absl::SNPrintF()` over `std::snprintf()`.
//
+// In particular, a successful call to `absl::SNPrintF()` writes at most `size`
+// bytes of the formatted output to `output`, including a NUL-terminator, and
+// returns the number of bytes that would have been written if truncation did
+// not occur. In the event of an error, a negative value is returned and `errno`
+// is set.
+//
// Example:
//
// std::string_view s = "Ulaanbaatar";
@@ -425,6 +457,16 @@
//
// FormatRawSink is a type erased wrapper around arbitrary sink objects
// specifically used as an argument to `Format()`.
+//
+// All the object has to do define an overload of `AbslFormatFlush()` for the
+// sink, usually by adding a ADL-based free function in the same namespace as
+// the sink:
+//
+// void AbslFormatFlush(MySink* dest, absl::string_view part);
+//
+// where `dest` is the pointer passed to `absl::Format()`. The function should
+// append `part` to `dest`.
+//
// FormatRawSink does not own the passed sink object. The passed object must
// outlive the FormatRawSink.
class FormatRawSink {
@@ -448,12 +490,13 @@
// `absl::FormatRawSink` interface), using a format string and zero or more
// additional arguments.
//
-// By default, `std::string` and `std::ostream` are supported as destination
-// objects. If a `std::string` is used the formatted string is appended to it.
+// By default, `std::string`, `std::ostream`, and `absl::Cord` are supported as
+// destination objects. If a `std::string` is used the formatted string is
+// appended to it.
//
-// `absl::Format()` is a generic version of `absl::StrFormat(), for custom
-// sinks. The format string, like format strings for `StrFormat()`, is checked
-// at compile-time.
+// `absl::Format()` is a generic version of `absl::StrAppendFormat()`, for
+// custom sinks. The format string, like format strings for `StrFormat()`, is
+// checked at compile-time.
//
// On failure, this function returns `false` and the state of the sink is
// unspecified.
@@ -524,6 +567,247 @@
str_format_internal::UntypedFormatSpecImpl::Extract(format), args);
}
+//------------------------------------------------------------------------------
+// StrFormat Extensions
+//------------------------------------------------------------------------------
+//
+// AbslFormatConvert()
+//
+// The StrFormat library provides a customization API for formatting
+// user-defined types using absl::StrFormat(). The API relies on detecting an
+// overload in the user-defined type's namespace of a free (non-member)
+// `AbslFormatConvert()` function, usually as a friend definition with the
+// following signature:
+//
+// absl::FormatConvertResult<...> AbslFormatConvert(
+// const X& value,
+// const absl::FormatConversionSpec& spec,
+// absl::FormatSink *sink);
+//
+// An `AbslFormatConvert()` overload for a type should only be declared in the
+// same file and namespace as said type.
+//
+// The abstractions within this definition include:
+//
+// * An `absl::FormatConversionSpec` to specify the fields to pull from a
+// user-defined type's format string
+// * An `absl::FormatSink` to hold the converted string data during the
+// conversion process.
+// * An `absl::FormatConvertResult` to hold the status of the returned
+// formatting operation
+//
+// The return type encodes all the conversion characters that your
+// AbslFormatConvert() routine accepts. The return value should be {true}.
+// A return value of {false} will result in `StrFormat()` returning
+// an empty string. This result will be propagated to the result of
+// `FormatUntyped`.
+//
+// Example:
+//
+// struct Point {
+// // To add formatting support to `Point`, we simply need to add a free
+// // (non-member) function `AbslFormatConvert()`. This method interprets
+// // `spec` to print in the request format. The allowed conversion characters
+// // can be restricted via the type of the result, in this example
+// // string and integral formatting are allowed (but not, for instance
+// // floating point characters like "%f"). You can add such a free function
+// // using a friend declaration within the body of the class:
+// friend absl::FormatConvertResult<absl::FormatConversionCharSet::kString |
+// absl::FormatConversionCharSet::kIntegral>
+// AbslFormatConvert(const Point& p, const absl::FormatConversionSpec& spec,
+// absl::FormatSink* s) {
+// if (spec.conversion_char() == absl::FormatConversionChar::s) {
+// s->Append(absl::StrCat("x=", p.x, " y=", p.y));
+// } else {
+// s->Append(absl::StrCat(p.x, ",", p.y));
+// }
+// return {true};
+// }
+//
+// int x;
+// int y;
+// };
+
+// clang-format off
+
+// FormatConversionChar
+//
+// Specifies the formatting character provided in the format string
+// passed to `StrFormat()`.
+enum class FormatConversionChar : uint8_t {
+ c, s, // text
+ d, i, o, u, x, X, // int
+ f, F, e, E, g, G, a, A, // float
+ n, p // misc
+};
+// clang-format on
+
+// FormatConversionSpec
+//
+// Specifies modifications to the conversion of the format string, through use
+// of one or more format flags in the source format string.
+class FormatConversionSpec {
+ public:
+ // FormatConversionSpec::is_basic()
+ //
+ // Indicates that width and precision are not specified, and no additional
+ // flags are set for this conversion character in the format string.
+ bool is_basic() const { return impl_.is_basic(); }
+
+ // FormatConversionSpec::has_left_flag()
+ //
+ // Indicates whether the result should be left justified for this conversion
+ // character in the format string. This flag is set through use of a '-'
+ // character in the format string. E.g. "%-s"
+ bool has_left_flag() const { return impl_.has_left_flag(); }
+
+ // FormatConversionSpec::has_show_pos_flag()
+ //
+ // Indicates whether a sign column is prepended to the result for this
+ // conversion character in the format string, even if the result is positive.
+ // This flag is set through use of a '+' character in the format string.
+ // E.g. "%+d"
+ bool has_show_pos_flag() const { return impl_.has_show_pos_flag(); }
+
+ // FormatConversionSpec::has_sign_col_flag()
+ //
+ // Indicates whether a mandatory sign column is added to the result for this
+ // conversion character. This flag is set through use of a space character
+ // (' ') in the format string. E.g. "% i"
+ bool has_sign_col_flag() const { return impl_.has_sign_col_flag(); }
+
+ // FormatConversionSpec::has_alt_flag()
+ //
+ // Indicates whether an "alternate" format is applied to the result for this
+ // conversion character. Alternative forms depend on the type of conversion
+ // character, and unallowed alternatives are undefined. This flag is set
+ // through use of a '#' character in the format string. E.g. "%#h"
+ bool has_alt_flag() const { return impl_.has_alt_flag(); }
+
+ // FormatConversionSpec::has_zero_flag()
+ //
+ // Indicates whether zeroes should be prepended to the result for this
+ // conversion character instead of spaces. This flag is set through use of the
+ // '0' character in the format string. E.g. "%0f"
+ bool has_zero_flag() const { return impl_.has_zero_flag(); }
+
+ // FormatConversionSpec::conversion_char()
+ //
+ // Returns the underlying conversion character.
+ FormatConversionChar conversion_char() const {
+ return impl_.conversion_char();
+ }
+
+ // FormatConversionSpec::width()
+ //
+ // Returns the specified width (indicated through use of a non-zero integer
+ // value or '*' character) of the conversion character. If width is
+ // unspecified, it returns a negative value.
+ int width() const { return impl_.width(); }
+
+ // FormatConversionSpec::precision()
+ //
+ // Returns the specified precision (through use of the '.' character followed
+ // by a non-zero integer value or '*' character) of the conversion character.
+ // If precision is unspecified, it returns a negative value.
+ int precision() const { return impl_.precision(); }
+
+ private:
+ explicit FormatConversionSpec(
+ str_format_internal::FormatConversionSpecImpl impl)
+ : impl_(impl) {}
+
+ friend str_format_internal::FormatConversionSpecImpl;
+
+ absl::str_format_internal::FormatConversionSpecImpl impl_;
+};
+
+// Type safe OR operator for FormatConversionCharSet to allow accepting multiple
+// conversion chars in custom format converters.
+constexpr FormatConversionCharSet operator|(FormatConversionCharSet a,
+ FormatConversionCharSet b) {
+ return static_cast<FormatConversionCharSet>(static_cast<uint64_t>(a) |
+ static_cast<uint64_t>(b));
+}
+
+// FormatConversionCharSet
+//
+// Specifies the _accepted_ conversion types as a template parameter to
+// FormatConvertResult for custom implementations of `AbslFormatConvert`.
+// Note the helper predefined alias definitions (kIntegral, etc.) below.
+enum class FormatConversionCharSet : uint64_t {
+ // text
+ c = str_format_internal::FormatConversionCharToConvInt('c'),
+ s = str_format_internal::FormatConversionCharToConvInt('s'),
+ // integer
+ d = str_format_internal::FormatConversionCharToConvInt('d'),
+ i = str_format_internal::FormatConversionCharToConvInt('i'),
+ o = str_format_internal::FormatConversionCharToConvInt('o'),
+ u = str_format_internal::FormatConversionCharToConvInt('u'),
+ x = str_format_internal::FormatConversionCharToConvInt('x'),
+ X = str_format_internal::FormatConversionCharToConvInt('X'),
+ // Float
+ f = str_format_internal::FormatConversionCharToConvInt('f'),
+ F = str_format_internal::FormatConversionCharToConvInt('F'),
+ e = str_format_internal::FormatConversionCharToConvInt('e'),
+ E = str_format_internal::FormatConversionCharToConvInt('E'),
+ g = str_format_internal::FormatConversionCharToConvInt('g'),
+ G = str_format_internal::FormatConversionCharToConvInt('G'),
+ a = str_format_internal::FormatConversionCharToConvInt('a'),
+ A = str_format_internal::FormatConversionCharToConvInt('A'),
+ // misc
+ n = str_format_internal::FormatConversionCharToConvInt('n'),
+ p = str_format_internal::FormatConversionCharToConvInt('p'),
+
+ // Used for width/precision '*' specification.
+ kStar = static_cast<uint64_t>(
+ absl::str_format_internal::FormatConversionCharSetInternal::kStar),
+ // Some predefined values:
+ kIntegral = d | i | u | o | x | X,
+ kFloating = a | e | f | g | A | E | F | G,
+ kNumeric = kIntegral | kFloating,
+ kString = s,
+ kPointer = p,
+};
+
+// FormatSink
+//
+// An abstraction to which conversions write their string data.
+//
+class FormatSink {
+ public:
+ // Appends `count` copies of `ch`.
+ void Append(size_t count, char ch) { sink_->Append(count, ch); }
+
+ void Append(string_view v) { sink_->Append(v); }
+
+ // Appends the first `precision` bytes of `v`. If this is less than
+ // `width`, spaces will be appended first (if `left` is false), or
+ // after (if `left` is true) to ensure the total amount appended is
+ // at least `width`.
+ bool PutPaddedString(string_view v, int width, int precision, bool left) {
+ return sink_->PutPaddedString(v, width, precision, left);
+ }
+
+ private:
+ friend str_format_internal::FormatSinkImpl;
+ explicit FormatSink(str_format_internal::FormatSinkImpl* s) : sink_(s) {}
+ str_format_internal::FormatSinkImpl* sink_;
+};
+
+// FormatConvertResult
+//
+// Indicates whether a call to AbslFormatConvert() was successful.
+// This return type informs the StrFormat extension framework (through
+// ADL but using the return type) of what conversion characters are supported.
+// It is strongly discouraged to return {false}, as this will result in an
+// empty string in StrFormat.
+template <FormatConversionCharSet C>
+struct FormatConvertResult {
+ bool value;
+};
+
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_STR_FORMAT_H_
diff --git a/absl/strings/str_format_test.cc b/absl/strings/str_format_test.cc
index cfd81bb..c60027a 100644
--- a/absl/strings/str_format_test.cc
+++ b/absl/strings/str_format_test.cc
@@ -1,3 +1,18 @@
+// Copyright 2020 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "absl/strings/str_format.h"
#include <cstdarg>
#include <cstdint>
@@ -6,10 +21,12 @@
#include "gmock/gmock.h"
#include "gtest/gtest.h"
-#include "absl/strings/str_format.h"
+#include "absl/strings/cord.h"
+#include "absl/strings/str_cat.h"
#include "absl/strings/string_view.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace {
using str_format_internal::FormatArgImpl;
@@ -241,7 +258,7 @@
std::FILE* file() const { return file_; }
- // Read the file into a std::string.
+ // Read the file into a string.
std::string ReadFile() {
std::fseek(file_, 0, SEEK_END);
int size = std::ftell(file_);
@@ -344,11 +361,12 @@
EXPECT_EQ(StrFormat("%c", int{'a'}), "a");
EXPECT_EQ(StrFormat("%c", long{'a'}), "a"); // NOLINT
EXPECT_EQ(StrFormat("%c", uint64_t{'a'}), "a");
- // "s" - std::string Eg: "C" -> "C", std::string("C++") -> "C++"
+ // "s" - string Eg: "C" -> "C", std::string("C++") -> "C++"
// Formats std::string, char*, string_view, and Cord.
EXPECT_EQ(StrFormat("%s", "C"), "C");
EXPECT_EQ(StrFormat("%s", std::string("C++")), "C++");
EXPECT_EQ(StrFormat("%s", string_view("view")), "view");
+ EXPECT_EQ(StrFormat("%s", absl::Cord("cord")), "cord");
// Integral Conversion
// These format integral types: char, int, long, uint64_t, etc.
EXPECT_EQ(StrFormat("%d", char{10}), "10");
@@ -449,7 +467,7 @@
if (conv.precision.is_from_arg()) {
*out += "." + std::to_string(conv.precision.get_from_arg()) + "$*";
}
- *out += conv.conv.Char();
+ *out += str_format_internal::FormatConversionCharToChar(conv.conv);
*out += "}";
return true;
}
@@ -531,76 +549,152 @@
EXPECT_FALSE((ParsedFormat<'s', 'd', 'g'>::New(format)));
}
-using str_format_internal::Conv;
+#if defined(__cpp_nontype_template_parameter_auto)
+
+template <auto T>
+std::true_type IsValidParsedFormatArgTest(ParsedFormat<T>*);
+
+template <auto T>
+std::false_type IsValidParsedFormatArgTest(...);
+
+template <auto T>
+using IsValidParsedFormatArg = decltype(IsValidParsedFormatArgTest<T>(nullptr));
+
+TEST_F(ParsedFormatTest, OnlyValidTypesAllowed) {
+ ASSERT_TRUE(IsValidParsedFormatArg<'c'>::value);
+
+ ASSERT_TRUE(IsValidParsedFormatArg<FormatConversionCharSet::d>::value);
+
+ ASSERT_TRUE(IsValidParsedFormatArg<absl::FormatConversionCharSet::d |
+ absl::FormatConversionCharSet::x>::value);
+ ASSERT_TRUE(
+ IsValidParsedFormatArg<absl::FormatConversionCharSet::kIntegral>::value);
+
+ // This is an easy mistake to make, however, this will reduce to an integer
+ // which has no meaning, so we need to ensure it doesn't compile.
+ ASSERT_FALSE(IsValidParsedFormatArg<'x' | 'd'>::value);
+
+ // For now, we disallow construction based on ConversionChar (rather than
+ // CharSet)
+ ASSERT_FALSE(IsValidParsedFormatArg<absl::FormatConversionChar::d>::value);
+}
+
+TEST_F(ParsedFormatTest, ExtendedTyping) {
+ EXPECT_FALSE(ParsedFormat<FormatConversionCharSet::d>::New(""));
+ ASSERT_TRUE(ParsedFormat<absl::FormatConversionCharSet::d>::New("%d"));
+ auto v1 = ParsedFormat<'d', absl::FormatConversionCharSet::s>::New("%d%s");
+ ASSERT_TRUE(v1);
+ auto v2 = ParsedFormat<absl::FormatConversionCharSet::d, 's'>::New("%d%s");
+ ASSERT_TRUE(v2);
+ auto v3 = ParsedFormat<absl::FormatConversionCharSet::d |
+ absl::FormatConversionCharSet::s,
+ 's'>::New("%d%s");
+ ASSERT_TRUE(v3);
+ auto v4 = ParsedFormat<absl::FormatConversionCharSet::d |
+ absl::FormatConversionCharSet::s,
+ 's'>::New("%s%s");
+ ASSERT_TRUE(v4);
+}
+#endif
TEST_F(ParsedFormatTest, UncheckedCorrect) {
- auto f = ExtendedParsedFormat<Conv::d>::New("ABC%dDEF");
+ auto f =
+ ExtendedParsedFormat<absl::FormatConversionCharSet::d>::New("ABC%dDEF");
ASSERT_TRUE(f);
EXPECT_EQ("[ABC]{d:1$d}[DEF]", SummarizeParsedFormat(*f));
std::string format = "%sFFF%dZZZ%f";
- auto f2 =
- ExtendedParsedFormat<Conv::string, Conv::d, Conv::floating>::New(format);
+ auto f2 = ExtendedParsedFormat<
+ absl::FormatConversionCharSet::kString, absl::FormatConversionCharSet::d,
+ absl::FormatConversionCharSet::kFloating>::New(format);
ASSERT_TRUE(f2);
EXPECT_EQ("{s:1$s}[FFF]{d:2$d}[ZZZ]{f:3$f}", SummarizeParsedFormat(*f2));
- f2 = ExtendedParsedFormat<Conv::string, Conv::d, Conv::floating>::New(
- "%s %d %f");
+ f2 = ExtendedParsedFormat<
+ absl::FormatConversionCharSet::kString, absl::FormatConversionCharSet::d,
+ absl::FormatConversionCharSet::kFloating>::New("%s %d %f");
ASSERT_TRUE(f2);
EXPECT_EQ("{s:1$s}[ ]{d:2$d}[ ]{f:3$f}", SummarizeParsedFormat(*f2));
- auto star = ExtendedParsedFormat<Conv::star, Conv::d>::New("%*d");
+ auto star =
+ ExtendedParsedFormat<absl::FormatConversionCharSet::kStar,
+ absl::FormatConversionCharSet::d>::New("%*d");
ASSERT_TRUE(star);
EXPECT_EQ("{*d:2$1$*d}", SummarizeParsedFormat(*star));
- auto dollar = ExtendedParsedFormat<Conv::d, Conv::s>::New("%2$s %1$d");
+ auto dollar =
+ ExtendedParsedFormat<absl::FormatConversionCharSet::d,
+ absl::FormatConversionCharSet::s>::New("%2$s %1$d");
ASSERT_TRUE(dollar);
EXPECT_EQ("{2$s:2$s}[ ]{1$d:1$d}", SummarizeParsedFormat(*dollar));
// with reuse
- dollar = ExtendedParsedFormat<Conv::d, Conv::s>::New("%2$s %1$d %1$d");
+ dollar = ExtendedParsedFormat<
+ absl::FormatConversionCharSet::d,
+ absl::FormatConversionCharSet::s>::New("%2$s %1$d %1$d");
ASSERT_TRUE(dollar);
EXPECT_EQ("{2$s:2$s}[ ]{1$d:1$d}[ ]{1$d:1$d}",
SummarizeParsedFormat(*dollar));
}
TEST_F(ParsedFormatTest, UncheckedIgnoredArgs) {
- EXPECT_FALSE((ExtendedParsedFormat<Conv::d, Conv::s>::New("ABC")));
- EXPECT_FALSE((ExtendedParsedFormat<Conv::d, Conv::s>::New("%dABC")));
- EXPECT_FALSE((ExtendedParsedFormat<Conv::d, Conv::s>::New("ABC%2$s")));
- auto f = ExtendedParsedFormat<Conv::d, Conv::s>::NewAllowIgnored("ABC");
+ EXPECT_FALSE(
+ (ExtendedParsedFormat<absl::FormatConversionCharSet::d,
+ absl::FormatConversionCharSet::s>::New("ABC")));
+ EXPECT_FALSE(
+ (ExtendedParsedFormat<absl::FormatConversionCharSet::d,
+ absl::FormatConversionCharSet::s>::New("%dABC")));
+ EXPECT_FALSE(
+ (ExtendedParsedFormat<absl::FormatConversionCharSet::d,
+ absl::FormatConversionCharSet::s>::New("ABC%2$s")));
+ auto f = ExtendedParsedFormat<
+ absl::FormatConversionCharSet::d,
+ absl::FormatConversionCharSet::s>::NewAllowIgnored("ABC");
ASSERT_TRUE(f);
EXPECT_EQ("[ABC]", SummarizeParsedFormat(*f));
- f = ExtendedParsedFormat<Conv::d, Conv::s>::NewAllowIgnored("%dABC");
+ f = ExtendedParsedFormat<
+ absl::FormatConversionCharSet::d,
+ absl::FormatConversionCharSet::s>::NewAllowIgnored("%dABC");
ASSERT_TRUE(f);
EXPECT_EQ("{d:1$d}[ABC]", SummarizeParsedFormat(*f));
- f = ExtendedParsedFormat<Conv::d, Conv::s>::NewAllowIgnored("ABC%2$s");
+ f = ExtendedParsedFormat<
+ absl::FormatConversionCharSet::d,
+ absl::FormatConversionCharSet::s>::NewAllowIgnored("ABC%2$s");
ASSERT_TRUE(f);
EXPECT_EQ("[ABC]{2$s:2$s}", SummarizeParsedFormat(*f));
}
TEST_F(ParsedFormatTest, UncheckedMultipleTypes) {
- auto dx = ExtendedParsedFormat<Conv::d | Conv::x>::New("%1$d %1$x");
+ auto dx =
+ ExtendedParsedFormat<absl::FormatConversionCharSet::d |
+ absl::FormatConversionCharSet::x>::New("%1$d %1$x");
EXPECT_TRUE(dx);
EXPECT_EQ("{1$d:1$d}[ ]{1$x:1$x}", SummarizeParsedFormat(*dx));
- dx = ExtendedParsedFormat<Conv::d | Conv::x>::New("%1$d");
+ dx = ExtendedParsedFormat<absl::FormatConversionCharSet::d |
+ absl::FormatConversionCharSet::x>::New("%1$d");
EXPECT_TRUE(dx);
EXPECT_EQ("{1$d:1$d}", SummarizeParsedFormat(*dx));
}
TEST_F(ParsedFormatTest, UncheckedIncorrect) {
- EXPECT_FALSE(ExtendedParsedFormat<Conv::d>::New(""));
+ EXPECT_FALSE(ExtendedParsedFormat<absl::FormatConversionCharSet::d>::New(""));
- EXPECT_FALSE(ExtendedParsedFormat<Conv::d>::New("ABC%dDEF%d"));
+ EXPECT_FALSE(ExtendedParsedFormat<absl::FormatConversionCharSet::d>::New(
+ "ABC%dDEF%d"));
std::string format = "%sFFF%dZZZ%f";
- EXPECT_FALSE((ExtendedParsedFormat<Conv::s, Conv::d, Conv::g>::New(format)));
+ EXPECT_FALSE(
+ (ExtendedParsedFormat<absl::FormatConversionCharSet::s,
+ absl::FormatConversionCharSet::d,
+ absl::FormatConversionCharSet::g>::New(format)));
}
TEST_F(ParsedFormatTest, RegressionMixPositional) {
- EXPECT_FALSE((ExtendedParsedFormat<Conv::d, Conv::o>::New("%1$d %o")));
+ EXPECT_FALSE(
+ (ExtendedParsedFormat<absl::FormatConversionCharSet::d,
+ absl::FormatConversionCharSet::o>::New("%1$d %o")));
}
using FormatWrapperTest = ::testing::Test;
@@ -622,8 +716,41 @@
}
} // namespace
+ABSL_NAMESPACE_END
} // namespace absl
+using FormatExtensionTest = ::testing::Test;
+
+struct Point {
+ friend absl::FormatConvertResult<absl::FormatConversionCharSet::kString |
+ absl::FormatConversionCharSet::kIntegral>
+ AbslFormatConvert(const Point& p, const absl::FormatConversionSpec& spec,
+ absl::FormatSink* s) {
+ if (spec.conversion_char() == absl::FormatConversionChar::s) {
+ s->Append(absl::StrCat("x=", p.x, " y=", p.y));
+ } else {
+ s->Append(absl::StrCat(p.x, ",", p.y));
+ }
+ return {true};
+ }
+
+ int x = 10;
+ int y = 20;
+};
+
+TEST_F(FormatExtensionTest, AbslFormatConvertExample) {
+ Point p;
+ EXPECT_EQ(absl::StrFormat("a %s z", p), "a x=10 y=20 z");
+ EXPECT_EQ(absl::StrFormat("a %d z", p), "a 10,20 z");
+
+ // Typed formatting will fail to compile an invalid format.
+ // StrFormat("%f", p); // Does not compile.
+ std::string actual;
+ absl::UntypedFormatSpec f1("%f");
+ // FormatUntyped will return false for bad character.
+ EXPECT_FALSE(absl::FormatUntyped(&actual, f1, {absl::FormatArg(p)}));
+}
+
// Some codegen thunks that we can use to easily dump the generated assembly for
// different StrFormat calls.
diff --git a/absl/strings/str_join.h b/absl/strings/str_join.h
index 4772f5d..ae5731a 100644
--- a/absl/strings/str_join.h
+++ b/absl/strings/str_join.h
@@ -60,6 +60,7 @@
#include "absl/strings/string_view.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
// -----------------------------------------------------------------------------
// Concept: Formatter
@@ -286,6 +287,7 @@
return strings_internal::JoinAlgorithm(value, separator, AlphaNumFormatter());
}
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_STR_JOIN_H_
diff --git a/absl/strings/str_join_test.cc b/absl/strings/str_join_test.cc
index 921d9c2..2be6256 100644
--- a/absl/strings/str_join_test.cc
+++ b/absl/strings/str_join_test.cc
@@ -134,26 +134,26 @@
//
{
- // Empty range yields an empty std::string.
+ // Empty range yields an empty string.
std::vector<std::string> v;
EXPECT_EQ("", absl::StrJoin(v, "-"));
}
{
- // A range of 1 element gives a std::string with that element but no
+ // A range of 1 element gives a string with that element but no
// separator.
std::vector<std::string> v = {"foo"};
EXPECT_EQ("foo", absl::StrJoin(v, "-"));
}
{
- // A range with a single empty std::string element
+ // A range with a single empty string element
std::vector<std::string> v = {""};
EXPECT_EQ("", absl::StrJoin(v, "-"));
}
{
- // A range with 2 elements, one of which is an empty std::string
+ // A range with 2 elements, one of which is an empty string
std::vector<std::string> v = {"a", ""};
EXPECT_EQ("a-", absl::StrJoin(v, "-"));
}
diff --git a/absl/strings/str_replace.cc b/absl/strings/str_replace.cc
index 280f63d..2bd5fa9 100644
--- a/absl/strings/str_replace.cc
+++ b/absl/strings/str_replace.cc
@@ -17,6 +17,7 @@
#include "absl/strings/str_cat.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace strings_internal {
using FixedMapping =
@@ -77,4 +78,5 @@
return StrReplaceAll<strings_internal::FixedMapping>(replacements, target);
}
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/str_replace.h b/absl/strings/str_replace.h
index 30540d0..273c707 100644
--- a/absl/strings/str_replace.h
+++ b/absl/strings/str_replace.h
@@ -46,6 +46,7 @@
#include "absl/strings/string_view.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
// StrReplaceAll()
//
@@ -212,6 +213,7 @@
return substitutions;
}
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_STR_REPLACE_H_
diff --git a/absl/strings/str_replace_benchmark.cc b/absl/strings/str_replace_benchmark.cc
index 95b2dc1..01331da 100644
--- a/absl/strings/str_replace_benchmark.cc
+++ b/absl/strings/str_replace_benchmark.cc
@@ -62,7 +62,7 @@
}
}
// big_string->resize(50);
- // OK, we've set up the std::string, now let's set up expectations - first by
+ // OK, we've set up the string, now let's set up expectations - first by
// just replacing "the" with "box"
after_replacing_the = new std::string(*big_string);
for (size_t pos = 0;
diff --git a/absl/strings/str_replace_test.cc b/absl/strings/str_replace_test.cc
index 1ca23af..9d8c7f7 100644
--- a/absl/strings/str_replace_test.cc
+++ b/absl/strings/str_replace_test.cc
@@ -25,7 +25,7 @@
TEST(StrReplaceAll, OneReplacement) {
std::string s;
- // Empty std::string.
+ // Empty string.
s = absl::StrReplaceAll(s, {{"", ""}});
EXPECT_EQ(s, "");
s = absl::StrReplaceAll(s, {{"x", ""}});
@@ -47,7 +47,7 @@
s = absl::StrReplaceAll("abc", {{"xyz", "123"}});
EXPECT_EQ(s, "abc");
- // Replace entire std::string.
+ // Replace entire string.
s = absl::StrReplaceAll("abc", {{"abc", "xyz"}});
EXPECT_EQ(s, "xyz");
@@ -88,7 +88,7 @@
TEST(StrReplaceAll, ManyReplacements) {
std::string s;
- // Empty std::string.
+ // Empty string.
s = absl::StrReplaceAll("", {{"", ""}, {"x", ""}, {"", "y"}, {"x", "y"}});
EXPECT_EQ(s, "");
@@ -96,7 +96,7 @@
s = absl::StrReplaceAll("abc", {{"", ""}, {"", "y"}, {"x", ""}});
EXPECT_EQ(s, "abc");
- // Replace entire std::string, one char at a time
+ // Replace entire string, one char at a time
s = absl::StrReplaceAll("abc", {{"a", "x"}, {"b", "y"}, {"c", "z"}});
EXPECT_EQ(s, "xyz");
s = absl::StrReplaceAll("zxy", {{"z", "x"}, {"x", "y"}, {"y", "z"}});
@@ -264,7 +264,7 @@
std::string s;
int reps;
- // Empty std::string.
+ // Empty string.
s = "";
reps = absl::StrReplaceAll({{"", ""}, {"x", ""}, {"", "y"}, {"x", "y"}}, &s);
EXPECT_EQ(reps, 0);
@@ -276,7 +276,7 @@
EXPECT_EQ(reps, 0);
EXPECT_EQ(s, "abc");
- // Replace entire std::string, one char at a time
+ // Replace entire string, one char at a time
s = "abc";
reps = absl::StrReplaceAll({{"a", "x"}, {"b", "y"}, {"c", "z"}}, &s);
EXPECT_EQ(reps, 3);
diff --git a/absl/strings/str_split.cc b/absl/strings/str_split.cc
index 2593130..e08c26b 100644
--- a/absl/strings/str_split.cc
+++ b/absl/strings/str_split.cc
@@ -27,6 +27,7 @@
#include "absl/strings/ascii.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace {
@@ -41,7 +42,7 @@
absl::string_view delimiter, size_t pos,
FindPolicy find_policy) {
if (delimiter.empty() && text.length() > 0) {
- // Special case for empty std::string delimiters: always return a zero-length
+ // Special case for empty string delimiters: always return a zero-length
// absl::string_view referring to the item at position 1 past pos.
return absl::string_view(text.data() + pos + 1, 0);
}
@@ -126,7 +127,7 @@
size_t pos) const {
pos = std::min(pos, text.size()); // truncate `pos`
absl::string_view substr = text.substr(pos);
- // If the std::string is shorter than the chunk size we say we
+ // If the string is shorter than the chunk size we say we
// "can't find the delimiter" so this will be the last chunk.
if (substr.length() <= static_cast<size_t>(length_))
return absl::string_view(text.data() + text.size(), 0);
@@ -134,4 +135,5 @@
return absl::string_view(substr.data() + length_, 0);
}
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/str_split.h b/absl/strings/str_split.h
index 7333078..bfbca42 100644
--- a/absl/strings/str_split.h
+++ b/absl/strings/str_split.h
@@ -44,11 +44,13 @@
#include <vector>
#include "absl/base/internal/raw_logging.h"
+#include "absl/base/macros.h"
#include "absl/strings/internal/str_split_internal.h"
#include "absl/strings/string_view.h"
#include "absl/strings/strip.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
//------------------------------------------------------------------------------
// Delimiters
@@ -367,6 +369,12 @@
}
};
+template <typename T>
+using EnableSplitIfString =
+ typename std::enable_if<std::is_same<T, std::string>::value ||
+ std::is_same<T, const std::string>::value,
+ int>::type;
+
//------------------------------------------------------------------------------
// StrSplit()
//------------------------------------------------------------------------------
@@ -487,25 +495,54 @@
// Try not to depend on this distinction because the bug may one day be fixed.
template <typename Delimiter>
strings_internal::Splitter<
- typename strings_internal::SelectDelimiter<Delimiter>::type, AllowEmpty>
+ typename strings_internal::SelectDelimiter<Delimiter>::type, AllowEmpty,
+ absl::string_view>
StrSplit(strings_internal::ConvertibleToStringView text, Delimiter d) {
using DelimiterType =
typename strings_internal::SelectDelimiter<Delimiter>::type;
- return strings_internal::Splitter<DelimiterType, AllowEmpty>(
+ return strings_internal::Splitter<DelimiterType, AllowEmpty,
+ absl::string_view>(
+ text.value(), DelimiterType(d), AllowEmpty());
+}
+
+template <typename Delimiter, typename StringType,
+ EnableSplitIfString<StringType> = 0>
+strings_internal::Splitter<
+ typename strings_internal::SelectDelimiter<Delimiter>::type, AllowEmpty,
+ std::string>
+StrSplit(StringType&& text, Delimiter d) {
+ using DelimiterType =
+ typename strings_internal::SelectDelimiter<Delimiter>::type;
+ return strings_internal::Splitter<DelimiterType, AllowEmpty, std::string>(
std::move(text), DelimiterType(d), AllowEmpty());
}
template <typename Delimiter, typename Predicate>
strings_internal::Splitter<
- typename strings_internal::SelectDelimiter<Delimiter>::type, Predicate>
+ typename strings_internal::SelectDelimiter<Delimiter>::type, Predicate,
+ absl::string_view>
StrSplit(strings_internal::ConvertibleToStringView text, Delimiter d,
Predicate p) {
using DelimiterType =
typename strings_internal::SelectDelimiter<Delimiter>::type;
- return strings_internal::Splitter<DelimiterType, Predicate>(
+ return strings_internal::Splitter<DelimiterType, Predicate,
+ absl::string_view>(
+ text.value(), DelimiterType(d), std::move(p));
+}
+
+template <typename Delimiter, typename Predicate, typename StringType,
+ EnableSplitIfString<StringType> = 0>
+strings_internal::Splitter<
+ typename strings_internal::SelectDelimiter<Delimiter>::type, Predicate,
+ std::string>
+StrSplit(StringType&& text, Delimiter d, Predicate p) {
+ using DelimiterType =
+ typename strings_internal::SelectDelimiter<Delimiter>::type;
+ return strings_internal::Splitter<DelimiterType, Predicate, std::string>(
std::move(text), DelimiterType(d), std::move(p));
}
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_STR_SPLIT_H_
diff --git a/absl/strings/str_split_test.cc b/absl/strings/str_split_test.cc
index 02f27bc..7f7c097 100644
--- a/absl/strings/str_split_test.cc
+++ b/absl/strings/str_split_test.cc
@@ -27,8 +27,10 @@
#include "gmock/gmock.h"
#include "gtest/gtest.h"
-#include "absl/base/dynamic_annotations.h" // for RunningOnValgrind
+#include "absl/base/dynamic_annotations.h"
#include "absl/base/macros.h"
+#include "absl/container/flat_hash_map.h"
+#include "absl/container/node_hash_map.h"
#include "absl/strings/numbers.h"
namespace {
@@ -71,7 +73,7 @@
// namespaces just like callers will need to use.
TEST(Split, APIExamples) {
{
- // Passes std::string delimiter. Assumes the default of ByString.
+ // Passes string delimiter. Assumes the default of ByString.
std::vector<std::string> v = absl::StrSplit("a,b,c", ","); // NOLINT
EXPECT_THAT(v, ElementsAre("a", "b", "c"));
@@ -97,7 +99,7 @@
}
{
- // Uses the Literal std::string "=>" as the delimiter.
+ // Uses the Literal string "=>" as the delimiter.
const std::vector<std::string> v = absl::StrSplit("a=>b=>c", "=>");
EXPECT_THAT(v, ElementsAre("a", "b", "c"));
}
@@ -121,17 +123,17 @@
}
{
- // Splits the input std::string into individual characters by using an empty
- // std::string as the delimiter.
+ // Splits the input string into individual characters by using an empty
+ // string as the delimiter.
std::vector<std::string> v = absl::StrSplit("abc", "");
EXPECT_THAT(v, ElementsAre("a", "b", "c"));
}
{
- // Splits std::string data with embedded NUL characters, using NUL as the
+ // Splits string data with embedded NUL characters, using NUL as the
// delimiter. A simple delimiter of "\0" doesn't work because strlen() will
- // say that's the empty std::string when constructing the absl::string_view
- // delimiter. Instead, a non-empty std::string containing NUL can be used as the
+ // say that's the empty string when constructing the absl::string_view
+ // delimiter. Instead, a non-empty string containing NUL can be used as the
// delimiter.
std::string embedded_nulls("a\0b\0c", 5);
std::string null_delim("\0", 1);
@@ -365,7 +367,7 @@
TEST(Splitter, RangeIterators) {
auto splitter = absl::StrSplit("a,b,c", ',');
std::vector<absl::string_view> output;
- for (const absl::string_view p : splitter) {
+ for (const absl::string_view& p : splitter) {
output.push_back(p);
}
EXPECT_THAT(output, ElementsAre("a", "b", "c"));
@@ -421,6 +423,18 @@
TestMapConversionOperator<std::multimap<std::string, std::string>>(splitter);
TestMapConversionOperator<std::unordered_map<std::string, std::string>>(
splitter);
+ TestMapConversionOperator<
+ absl::node_hash_map<absl::string_view, absl::string_view>>(splitter);
+ TestMapConversionOperator<
+ absl::node_hash_map<absl::string_view, std::string>>(splitter);
+ TestMapConversionOperator<
+ absl::node_hash_map<std::string, absl::string_view>>(splitter);
+ TestMapConversionOperator<
+ absl::flat_hash_map<absl::string_view, absl::string_view>>(splitter);
+ TestMapConversionOperator<
+ absl::flat_hash_map<absl::string_view, std::string>>(splitter);
+ TestMapConversionOperator<
+ absl::flat_hash_map<std::string, absl::string_view>>(splitter);
// Tests conversion to std::pair
@@ -436,7 +450,7 @@
// less-than, equal-to, and more-than 2 strings.
TEST(Splitter, ToPair) {
{
- // Empty std::string
+ // Empty string
std::pair<std::string, std::string> p = absl::StrSplit("", ',');
EXPECT_EQ("", p.first);
EXPECT_EQ("", p.second);
@@ -565,7 +579,7 @@
TEST(Split, Temporary) {
// Use a std::string longer than the SSO length, so that when the temporary is
- // destroyed, if the splitter keeps a reference to the std::string's contents,
+ // destroyed, if the splitter keeps a reference to the string's contents,
// it'll reference freed memory instead of just dead on-stack memory.
const char input[] = "a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u";
EXPECT_LT(sizeof(std::string), ABSL_ARRAYSIZE(input))
@@ -651,14 +665,14 @@
// Tests splitting utf8 strings and utf8 delimiters.
std::string utf8_string = u8"\u03BA\u1F79\u03C3\u03BC\u03B5";
{
- // A utf8 input std::string with an ascii delimiter.
+ // A utf8 input string with an ascii delimiter.
std::string to_split = "a," + utf8_string;
std::vector<absl::string_view> v = absl::StrSplit(to_split, ',');
EXPECT_THAT(v, ElementsAre("a", utf8_string));
}
{
- // A utf8 input std::string and a utf8 delimiter.
+ // A utf8 input string and a utf8 delimiter.
std::string to_split = "a," + utf8_string + ",b";
std::string unicode_delimiter = "," + utf8_string + ",";
std::vector<absl::string_view> v =
@@ -667,7 +681,7 @@
}
{
- // A utf8 input std::string and ByAnyChar with ascii chars.
+ // A utf8 input string and ByAnyChar with ascii chars.
std::vector<absl::string_view> v =
absl::StrSplit(u8"Foo h\u00E4llo th\u4E1Ere", absl::ByAnyChar(" \t"));
EXPECT_THAT(v, ElementsAre("Foo", u8"h\u00E4llo", u8"th\u4E1Ere"));
@@ -814,10 +828,10 @@
ByString comma_string(",");
TestComma(comma_string);
- // The first occurrence of empty std::string ("") in a std::string is at position 0.
+ // The first occurrence of empty string ("") in a string is at position 0.
// There is a test below that demonstrates this for absl::string_view::find().
// If the ByString delimiter returned position 0 for this, there would
- // be an infinite loop in the SplitIterator code. To avoid this, empty std::string
+ // be an infinite loop in the SplitIterator code. To avoid this, empty string
// is a special case in that it always returns the item at position 1.
absl::string_view abc("abc");
EXPECT_EQ(0, abc.find("")); // "" is found at position 0
@@ -876,7 +890,7 @@
EXPECT_FALSE(IsFoundAt("=", two_delims, -1));
// ByAnyChar behaves just like ByString when given a delimiter of empty
- // std::string. That is, it always returns a zero-length absl::string_view
+ // string. That is, it always returns a zero-length absl::string_view
// referring to the item at position 1, not position 0.
ByAnyChar empty("");
EXPECT_FALSE(IsFoundAt("", empty, 0));
@@ -913,7 +927,7 @@
std::vector<absl::string_view> v = absl::StrSplit(s, '-');
EXPECT_EQ(2, v.size());
// The first element will contain 2G of 'x's.
- // testing::StartsWith is too slow with a 2G std::string.
+ // testing::StartsWith is too slow with a 2G string.
EXPECT_EQ('x', v[0][0]);
EXPECT_EQ('x', v[0][1]);
EXPECT_EQ('x', v[0][3]);
diff --git a/absl/strings/string_view.cc b/absl/strings/string_view.cc
index dc034a8..c5f5de9 100644
--- a/absl/strings/string_view.cc
+++ b/absl/strings/string_view.cc
@@ -14,7 +14,7 @@
#include "absl/strings/string_view.h"
-#ifndef ABSL_HAVE_STD_STRING_VIEW
+#ifndef ABSL_USES_STD_STRING_VIEW
#include <algorithm>
#include <climits>
@@ -24,6 +24,7 @@
#include "absl/strings/internal/memutil.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace {
void WritePadding(std::ostream& o, size_t pad) {
@@ -228,6 +229,7 @@
ABSL_STRING_VIEW_SELECTANY
constexpr string_view::size_type string_view::kMaxSize;
+ABSL_NAMESPACE_END
} // namespace absl
-#endif // ABSL_HAVE_STD_STRING_VIEW
+#endif // ABSL_USES_STD_STRING_VIEW
diff --git a/absl/strings/string_view.h b/absl/strings/string_view.h
index 68b90aa..5260b5b 100644
--- a/absl/strings/string_view.h
+++ b/absl/strings/string_view.h
@@ -28,18 +28,6 @@
#define ABSL_STRINGS_STRING_VIEW_H_
#include <algorithm>
-#include "absl/base/config.h"
-
-#ifdef ABSL_HAVE_STD_STRING_VIEW
-
-#include <string_view> // IWYU pragma: export
-
-namespace absl {
-using std::string_view;
-} // namespace absl
-
-#else // ABSL_HAVE_STD_STRING_VIEW
-
#include <cassert>
#include <cstddef>
#include <cstring>
@@ -48,12 +36,33 @@
#include <limits>
#include <string>
+#include "absl/base/config.h"
#include "absl/base/internal/throw_delegate.h"
#include "absl/base/macros.h"
#include "absl/base/optimization.h"
#include "absl/base/port.h"
+#ifdef ABSL_USES_STD_STRING_VIEW
+
+#include <string_view> // IWYU pragma: export
+
namespace absl {
+ABSL_NAMESPACE_BEGIN
+using string_view = std::string_view;
+ABSL_NAMESPACE_END
+} // namespace absl
+
+#else // ABSL_USES_STD_STRING_VIEW
+
+#if ABSL_HAVE_BUILTIN(__builtin_memcmp) || \
+ (defined(__GNUC__) && !defined(__clang__))
+#define ABSL_INTERNAL_STRING_VIEW_MEMCMP __builtin_memcmp
+#else // ABSL_HAVE_BUILTIN(__builtin_memcmp)
+#define ABSL_INTERNAL_STRING_VIEW_MEMCMP memcmp
+#endif // ABSL_HAVE_BUILTIN(__builtin_memcmp)
+
+namespace absl {
+ABSL_NAMESPACE_BEGIN
// absl::string_view
//
@@ -102,17 +111,22 @@
// example, when splitting a string, `std::vector<absl::string_view>` is a
// natural data type for the output.
//
-// When constructed from a source which is nul-terminated, the `string_view`
-// itself will not include the nul-terminator unless a specific size (including
-// the nul) is passed to the constructor. As a result, common idioms that work
-// on nul-terminated strings do not work on `string_view` objects. If you write
+// For another example, a Cord is a non-contiguous, potentially very
+// long string-like object. The Cord class has an interface that iteratively
+// provides string_view objects that point to the successive pieces of a Cord
+// object.
+//
+// When constructed from a source which is NUL-terminated, the `string_view`
+// itself will not include the NUL-terminator unless a specific size (including
+// the NUL) is passed to the constructor. As a result, common idioms that work
+// on NUL-terminated strings do not work on `string_view` objects. If you write
// code that scans a `string_view`, you must check its length rather than test
// for nul, for example. Note, however, that nuls may still be embedded within
// a `string_view` explicitly.
//
// You may create a null `string_view` in two ways:
//
-// absl::string_view sv();
+// absl::string_view sv;
// absl::string_view sv(nullptr, 0);
//
// For the above, `sv.data() == nullptr`, `sv.length() == 0`, and
@@ -168,9 +182,11 @@
string_view( // NOLINT(runtime/explicit)
const std::basic_string<char, std::char_traits<char>, Allocator>&
str) noexcept
- : ptr_(str.data()), length_(CheckLengthInternal(str.size())) {}
+ // This is implemented in terms of `string_view(p, n)` so `str.size()`
+ // doesn't need to be reevaluated after `ptr_` is set.
+ : string_view(str.data(), str.size()) {}
- // Implicit constructor of a `string_view` from nul-terminated `str`. When
+ // Implicit constructor of a `string_view` from NUL-terminated `str`. When
// accepting possibly null strings, use `absl::NullSafeStringView(str)`
// instead (see below).
constexpr string_view(const char* str) // NOLINT(runtime/explicit)
@@ -271,7 +287,9 @@
//
// Returns the ith element of the `string_view` using the array operator.
// Note that this operator does not perform any bounds checking.
- constexpr const_reference operator[](size_type i) const { return ptr_[i]; }
+ constexpr const_reference operator[](size_type i) const {
+ return ABSL_HARDENING_ASSERT(i < size()), ptr_[i];
+ }
// string_view::at()
//
@@ -281,27 +299,32 @@
constexpr const_reference at(size_type i) const {
return ABSL_PREDICT_TRUE(i < size())
? ptr_[i]
- : (base_internal::ThrowStdOutOfRange("absl::string_view::at"),
+ : ((void)base_internal::ThrowStdOutOfRange(
+ "absl::string_view::at"),
ptr_[i]);
}
// string_view::front()
//
// Returns the first element of a `string_view`.
- constexpr const_reference front() const { return ptr_[0]; }
+ constexpr const_reference front() const {
+ return ABSL_HARDENING_ASSERT(!empty()), ptr_[0];
+ }
// string_view::back()
//
// Returns the last element of a `string_view`.
- constexpr const_reference back() const { return ptr_[size() - 1]; }
+ constexpr const_reference back() const {
+ return ABSL_HARDENING_ASSERT(!empty()), ptr_[size() - 1];
+ }
// string_view::data()
//
// Returns a pointer to the underlying character array (which is of course
// stored elsewhere). Note that `string_view::data()` may contain embedded nul
- // characters, but the returned buffer may or may not be nul-terminated;
- // therefore, do not pass `data()` to a routine that expects a nul-terminated
- // std::string.
+ // characters, but the returned buffer may or may not be NUL-terminated;
+ // therefore, do not pass `data()` to a routine that expects a NUL-terminated
+ // string.
constexpr const_pointer data() const noexcept { return ptr_; }
// Modifiers
@@ -309,9 +332,9 @@
// string_view::remove_prefix()
//
// Removes the first `n` characters from the `string_view`. Note that the
- // underlying std::string is not changed, only the view.
+ // underlying string is not changed, only the view.
void remove_prefix(size_type n) {
- assert(n <= length_);
+ ABSL_HARDENING_ASSERT(n <= length_);
ptr_ += n;
length_ -= n;
}
@@ -319,9 +342,9 @@
// string_view::remove_suffix()
//
// Removes the last `n` characters from the `string_view`. Note that the
- // underlying std::string is not changed, only the view.
+ // underlying string is not changed, only the view.
void remove_suffix(size_type n) {
- assert(n <= length_);
+ ABSL_HARDENING_ASSERT(n <= length_);
length_ -= n;
}
@@ -364,31 +387,29 @@
// Returns a "substring" of the `string_view` (at offset `pos` and length
// `n`) as another string_view. This function throws `std::out_of_bounds` if
// `pos > size`.
- string_view substr(size_type pos, size_type n = npos) const {
- if (ABSL_PREDICT_FALSE(pos > length_))
- base_internal::ThrowStdOutOfRange("absl::string_view::substr");
- n = (std::min)(n, length_ - pos);
- return string_view(ptr_ + pos, n);
+ // Use absl::ClippedSubstr if you need a truncating substr operation.
+ constexpr string_view substr(size_type pos, size_type n = npos) const {
+ return ABSL_PREDICT_FALSE(pos > length_)
+ ? (base_internal::ThrowStdOutOfRange(
+ "absl::string_view::substr"),
+ string_view())
+ : string_view(ptr_ + pos, Min(n, length_ - pos));
}
// string_view::compare()
//
// Performs a lexicographical comparison between the `string_view` and
// another `absl::string_view`, returning -1 if `this` is less than, 0 if
- // `this` is equal to, and 1 if `this` is greater than the passed std::string
+ // `this` is equal to, and 1 if `this` is greater than the passed string
// view. Note that in the case of data equality, a further comparison is made
// on the respective sizes of the two `string_view`s to determine which is
// smaller, equal, or greater.
- int compare(string_view x) const noexcept {
- auto min_length = (std::min)(length_, x.length_);
- if (min_length > 0) {
- int r = memcmp(ptr_, x.ptr_, min_length);
- if (r < 0) return -1;
- if (r > 0) return 1;
- }
- if (length_ < x.length_) return -1;
- if (length_ > x.length_) return 1;
- return 0;
+ constexpr int compare(string_view x) const noexcept {
+ return CompareImpl(length_, x.length_,
+ Min(length_, x.length_) == 0
+ ? 0
+ : ABSL_INTERNAL_STRING_VIEW_MEMCMP(
+ ptr_, x.ptr_, Min(length_, x.length_)));
}
// Overload of `string_view::compare()` for comparing a substring of the
@@ -405,17 +426,17 @@
}
// Overload of `string_view::compare()` for comparing a `string_view` and a
- // a different C-style std::string `s`.
+ // a different C-style string `s`.
int compare(const char* s) const { return compare(string_view(s)); }
// Overload of `string_view::compare()` for comparing a substring of the
- // `string_view` and a different std::string C-style std::string `s`.
+ // `string_view` and a different string C-style string `s`.
int compare(size_type pos1, size_type count1, const char* s) const {
return substr(pos1, count1).compare(string_view(s));
}
// Overload of `string_view::compare()` for comparing a substring of the
- // `string_view` and a substring of a different C-style std::string `s`.
+ // `string_view` and a substring of a different C-style string `s`.
int compare(size_type pos1, size_type count1, const char* s,
size_type count2) const {
return substr(pos1, count1).compare(string_view(s, count2));
@@ -505,7 +526,7 @@
(std::numeric_limits<difference_type>::max)();
static constexpr size_type CheckLengthInternal(size_type len) {
- return ABSL_ASSERT(len <= kMaxSize), len;
+ return ABSL_HARDENING_ASSERT(len <= kMaxSize), len;
}
static constexpr size_type StrlenInternal(const char* str) {
@@ -526,6 +547,17 @@
#endif
}
+ static constexpr size_t Min(size_type length_a, size_type length_b) {
+ return length_a < length_b ? length_a : length_b;
+ }
+
+ static constexpr int CompareImpl(size_type length_a, size_type length_b,
+ int compare_result) {
+ return compare_result == 0 ? static_cast<int>(length_a > length_b) -
+ static_cast<int>(length_a < length_b)
+ : (compare_result < 0 ? -1 : 1);
+ }
+
const char* ptr_;
size_type length_;
};
@@ -533,44 +565,44 @@
// This large function is defined inline so that in a fairly common case where
// one of the arguments is a literal, the compiler can elide a lot of the
// following comparisons.
-inline bool operator==(string_view x, string_view y) noexcept {
- auto len = x.size();
- if (len != y.size()) {
- return false;
- }
-
- return x.data() == y.data() || len <= 0 ||
- memcmp(x.data(), y.data(), len) == 0;
+constexpr bool operator==(string_view x, string_view y) noexcept {
+ return x.size() == y.size() &&
+ (x.empty() ||
+ ABSL_INTERNAL_STRING_VIEW_MEMCMP(x.data(), y.data(), x.size()) == 0);
}
-inline bool operator!=(string_view x, string_view y) noexcept {
+constexpr bool operator!=(string_view x, string_view y) noexcept {
return !(x == y);
}
-inline bool operator<(string_view x, string_view y) noexcept {
- auto min_size = (std::min)(x.size(), y.size());
- const int r = min_size == 0 ? 0 : memcmp(x.data(), y.data(), min_size);
- return (r < 0) || (r == 0 && x.size() < y.size());
+constexpr bool operator<(string_view x, string_view y) noexcept {
+ return x.compare(y) < 0;
}
-inline bool operator>(string_view x, string_view y) noexcept { return y < x; }
+constexpr bool operator>(string_view x, string_view y) noexcept {
+ return y < x;
+}
-inline bool operator<=(string_view x, string_view y) noexcept {
+constexpr bool operator<=(string_view x, string_view y) noexcept {
return !(y < x);
}
-inline bool operator>=(string_view x, string_view y) noexcept {
+constexpr bool operator>=(string_view x, string_view y) noexcept {
return !(x < y);
}
// IO Insertion Operator
std::ostream& operator<<(std::ostream& o, string_view piece);
+ABSL_NAMESPACE_END
} // namespace absl
-#endif // ABSL_HAVE_STD_STRING_VIEW
+#undef ABSL_INTERNAL_STRING_VIEW_MEMCMP
+
+#endif // ABSL_USES_STD_STRING_VIEW
namespace absl {
+ABSL_NAMESPACE_BEGIN
// ClippedSubstr()
//
@@ -587,10 +619,11 @@
// Creates an `absl::string_view` from a pointer `p` even if it's null-valued.
// This function should be used where an `absl::string_view` can be created from
// a possibly-null pointer.
-inline string_view NullSafeStringView(const char* p) {
+constexpr string_view NullSafeStringView(const char* p) {
return p ? string_view(p) : string_view();
}
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_STRING_VIEW_H_
diff --git a/absl/strings/string_view_benchmark.cc b/absl/strings/string_view_benchmark.cc
index 46909cb..0d74e23 100644
--- a/absl/strings/string_view_benchmark.cc
+++ b/absl/strings/string_view_benchmark.cc
@@ -30,6 +30,24 @@
namespace {
+void BM_StringViewFromString(benchmark::State& state) {
+ std::string s(state.range(0), 'x');
+ std::string* ps = &s;
+ struct SV {
+ SV() = default;
+ explicit SV(const std::string& s) : sv(s) {}
+ absl::string_view sv;
+ } sv;
+ SV* psv = &sv;
+ benchmark::DoNotOptimize(ps);
+ benchmark::DoNotOptimize(psv);
+ for (auto _ : state) {
+ new (psv) SV(*ps);
+ benchmark::DoNotOptimize(sv);
+ }
+}
+BENCHMARK(BM_StringViewFromString)->Arg(12)->Arg(128);
+
// Provide a forcibly out-of-line wrapper for operator== that can be used in
// benchmarks to measure the impact of inlining.
ABSL_ATTRIBUTE_NOINLINE
@@ -142,11 +160,45 @@
absl::string_view b = y;
for (auto _ : state) {
+ benchmark::DoNotOptimize(a);
+ benchmark::DoNotOptimize(b);
benchmark::DoNotOptimize(a.compare(b));
}
}
BENCHMARK(BM_CompareSame)->DenseRange(0, 3)->Range(4, 1 << 10);
+void BM_CompareFirstOneLess(benchmark::State& state) {
+ const int len = state.range(0);
+ std::string x(len, 'a');
+ std::string y = x;
+ y.back() = 'b';
+ absl::string_view a = x;
+ absl::string_view b = y;
+
+ for (auto _ : state) {
+ benchmark::DoNotOptimize(a);
+ benchmark::DoNotOptimize(b);
+ benchmark::DoNotOptimize(a.compare(b));
+ }
+}
+BENCHMARK(BM_CompareFirstOneLess)->DenseRange(1, 3)->Range(4, 1 << 10);
+
+void BM_CompareSecondOneLess(benchmark::State& state) {
+ const int len = state.range(0);
+ std::string x(len, 'a');
+ std::string y = x;
+ x.back() = 'b';
+ absl::string_view a = x;
+ absl::string_view b = y;
+
+ for (auto _ : state) {
+ benchmark::DoNotOptimize(a);
+ benchmark::DoNotOptimize(b);
+ benchmark::DoNotOptimize(a.compare(b));
+ }
+}
+BENCHMARK(BM_CompareSecondOneLess)->DenseRange(1, 3)->Range(4, 1 << 10);
+
void BM_find_string_view_len_one(benchmark::State& state) {
std::string haystack(state.range(0), '0');
absl::string_view s(haystack);
diff --git a/absl/strings/string_view_test.cc b/absl/strings/string_view_test.cc
index eb8b170..643af8f 100644
--- a/absl/strings/string_view_test.cc
+++ b/absl/strings/string_view_test.cc
@@ -28,6 +28,7 @@
#include "gtest/gtest.h"
#include "absl/base/config.h"
#include "absl/base/dynamic_annotations.h"
+#include "absl/base/options.h"
#if defined(ABSL_HAVE_STD_STRING_VIEW) || defined(__ANDROID__)
// We don't control the death messaging when using std::string_view.
@@ -410,7 +411,7 @@
EXPECT_EQ(a.find(e, 17), 17);
absl::string_view g("xx not found bb");
EXPECT_EQ(a.find(g), absl::string_view::npos);
- // empty std::string nonsense
+ // empty string nonsense
EXPECT_EQ(d.find(b), absl::string_view::npos);
EXPECT_EQ(e.find(b), absl::string_view::npos);
EXPECT_EQ(d.find(b, 4), absl::string_view::npos);
@@ -438,7 +439,7 @@
EXPECT_EQ(g.find('o', 4), 4);
EXPECT_EQ(g.find('o', 5), 8);
EXPECT_EQ(a.find('b', 5), absl::string_view::npos);
- // empty std::string nonsense
+ // empty string nonsense
EXPECT_EQ(d.find('\0'), absl::string_view::npos);
EXPECT_EQ(e.find('\0'), absl::string_view::npos);
EXPECT_EQ(d.find('\0', 4), absl::string_view::npos);
@@ -465,7 +466,7 @@
EXPECT_EQ(e.rfind(b), absl::string_view::npos);
EXPECT_EQ(d.rfind(b, 4), absl::string_view::npos);
EXPECT_EQ(e.rfind(b, 7), absl::string_view::npos);
- // empty std::string nonsense
+ // empty string nonsense
EXPECT_EQ(d.rfind(d, 4), std::string().rfind(std::string()));
EXPECT_EQ(e.rfind(d, 7), std::string().rfind(std::string()));
EXPECT_EQ(d.rfind(e, 4), std::string().rfind(std::string()));
@@ -484,7 +485,7 @@
EXPECT_EQ(f.rfind('\0', 12), 3);
EXPECT_EQ(f.rfind('3'), 2);
EXPECT_EQ(f.rfind('5'), 5);
- // empty std::string nonsense
+ // empty string nonsense
EXPECT_EQ(d.rfind('o'), absl::string_view::npos);
EXPECT_EQ(e.rfind('o'), absl::string_view::npos);
EXPECT_EQ(d.rfind('o', 4), absl::string_view::npos);
@@ -520,7 +521,7 @@
EXPECT_EQ(g.find_first_of(c), 0);
EXPECT_EQ(a.find_first_of(f), absl::string_view::npos);
EXPECT_EQ(f.find_first_of(a), absl::string_view::npos);
- // empty std::string nonsense
+ // empty string nonsense
EXPECT_EQ(a.find_first_of(d), absl::string_view::npos);
EXPECT_EQ(a.find_first_of(e), absl::string_view::npos);
EXPECT_EQ(d.find_first_of(b), absl::string_view::npos);
@@ -538,7 +539,7 @@
EXPECT_EQ(a.find_first_not_of(f), 0);
EXPECT_EQ(a.find_first_not_of(d), 0);
EXPECT_EQ(a.find_first_not_of(e), 0);
- // empty std::string nonsense
+ // empty string nonsense
EXPECT_EQ(a.find_first_not_of(d), 0);
EXPECT_EQ(a.find_first_not_of(e), 0);
EXPECT_EQ(a.find_first_not_of(d, 1), 1);
@@ -566,7 +567,7 @@
EXPECT_EQ(f.find_first_not_of('\0'), 0);
EXPECT_EQ(f.find_first_not_of('\0', 3), 4);
EXPECT_EQ(f.find_first_not_of('\0', 2), 2);
- // empty std::string nonsense
+ // empty string nonsense
EXPECT_EQ(d.find_first_not_of('x'), absl::string_view::npos);
EXPECT_EQ(e.find_first_not_of('x'), absl::string_view::npos);
EXPECT_EQ(d.find_first_not_of('\0'), absl::string_view::npos);
@@ -606,7 +607,7 @@
EXPECT_EQ(f.find_last_of(i, 5), 5);
EXPECT_EQ(f.find_last_of(i, 6), 6);
EXPECT_EQ(f.find_last_of(a, 4), absl::string_view::npos);
- // empty std::string nonsense
+ // empty string nonsense
EXPECT_EQ(f.find_last_of(d), absl::string_view::npos);
EXPECT_EQ(f.find_last_of(e), absl::string_view::npos);
EXPECT_EQ(f.find_last_of(d, 4), absl::string_view::npos);
@@ -632,7 +633,7 @@
EXPECT_EQ(a.find_last_not_of(c, 24), 22);
EXPECT_EQ(a.find_last_not_of(b, 3), 3);
EXPECT_EQ(a.find_last_not_of(b, 2), absl::string_view::npos);
- // empty std::string nonsense
+ // empty string nonsense
EXPECT_EQ(f.find_last_not_of(d), f.size()-1);
EXPECT_EQ(f.find_last_not_of(e), f.size()-1);
EXPECT_EQ(f.find_last_not_of(d, 4), 4);
@@ -656,7 +657,7 @@
EXPECT_EQ(h.find_last_not_of('x', 2), 2);
EXPECT_EQ(h.find_last_not_of('=', 2), absl::string_view::npos);
EXPECT_EQ(b.find_last_not_of('b', 1), 0);
- // empty std::string nonsense
+ // empty string nonsense
EXPECT_EQ(d.find_last_not_of('x'), absl::string_view::npos);
EXPECT_EQ(e.find_last_not_of('x'), absl::string_view::npos);
EXPECT_EQ(d.find_last_not_of('\0'), absl::string_view::npos);
@@ -678,7 +679,7 @@
EXPECT_EQ(a.substr(23, 99), c);
EXPECT_EQ(a.substr(0), a);
EXPECT_EQ(a.substr(3, 2), "de");
- // empty std::string nonsense
+ // empty string nonsense
EXPECT_EQ(d.substr(0, 99), e);
// use of npos
EXPECT_EQ(a.substr(0, absl::string_view::npos), a);
@@ -818,6 +819,18 @@
EXPECT_EQ(&c, &csp.back());
}
+TEST(StringViewTest, FrontBackEmpty) {
+#ifndef ABSL_USES_STD_STRING_VIEW
+#if !defined(NDEBUG) || ABSL_OPTION_HARDENED
+ // Abseil's string_view implementation has debug assertions that check that
+ // front() and back() are not called on an empty string_view.
+ absl::string_view sv;
+ ABSL_EXPECT_DEATH_IF_SUPPORTED(sv.front(), "");
+ ABSL_EXPECT_DEATH_IF_SUPPORTED(sv.back(), "");
+#endif
+#endif
+}
+
// `std::string_view::string_view(const char*)` calls
// `std::char_traits<char>::length(const char*)` to get the string length. In
// libc++, it doesn't allow `nullptr` in the constexpr context, with the error
@@ -830,7 +843,7 @@
// to the standard, but `absl::string_view` implements a different
// behavior for historical reasons. We work around tests that construct
// `string_view` from `nullptr` when using libc++.
-#if !defined(ABSL_HAVE_STD_STRING_VIEW) || \
+#if !defined(ABSL_USES_STD_STRING_VIEW) || \
(!(defined(_GLIBCXX_RELEASE) && _GLIBCXX_RELEASE >= 9) && \
!defined(_LIBCPP_VERSION) && !defined(_MSC_VER))
#define ABSL_HAVE_STRING_VIEW_FROM_NULLPTR 1
@@ -847,7 +860,7 @@
EXPECT_EQ(s.size(), 0);
// .ToString() on a absl::string_view with nullptr should produce the empty
- // std::string.
+ // string.
EXPECT_EQ("", std::string(s));
#endif // ABSL_HAVE_STRING_VIEW_FROM_NULLPTR
}
@@ -902,9 +915,9 @@
EXPECT_EQ(abc.at(1), 'b');
EXPECT_EQ(abc.at(2), 'c');
#ifdef ABSL_HAVE_EXCEPTIONS
- EXPECT_THROW(abc.at(3), std::out_of_range);
+ EXPECT_THROW((void)abc.at(3), std::out_of_range);
#else
- ABSL_EXPECT_DEATH_IF_SUPPORTED(abc.at(3), "absl::string_view::at");
+ ABSL_EXPECT_DEATH_IF_SUPPORTED((void)abc.at(3), "absl::string_view::at");
#endif
}
@@ -931,6 +944,31 @@
}
}
+TEST(StringViewTest, ConstexprNullSafeStringView) {
+ {
+ constexpr absl::string_view s = absl::NullSafeStringView(nullptr);
+ EXPECT_EQ(nullptr, s.data());
+ EXPECT_EQ(0, s.size());
+ EXPECT_EQ(absl::string_view(), s);
+ }
+#if !defined(_MSC_VER) || _MSC_VER >= 1910
+ // MSVC 2017+ is required for good constexpr string_view support.
+ // See the implementation of `absl::string_view::StrlenInternal()`.
+ {
+ static constexpr char kHi[] = "hi";
+ absl::string_view s = absl::NullSafeStringView(kHi);
+ EXPECT_EQ(kHi, s.data());
+ EXPECT_EQ(strlen(kHi), s.size());
+ EXPECT_EQ(absl::string_view("hi"), s);
+ }
+ {
+ constexpr absl::string_view s = absl::NullSafeStringView("hello");
+ EXPECT_EQ(s.size(), 5);
+ EXPECT_EQ("hello", s);
+ }
+#endif
+}
+
TEST(StringViewTest, ConstexprCompiles) {
constexpr absl::string_view sp;
#ifdef ABSL_HAVE_STRING_VIEW_FROM_NULLPTR
@@ -938,9 +976,9 @@
#endif
constexpr absl::string_view cstr_len("cstr", 4);
-#if defined(ABSL_HAVE_STD_STRING_VIEW)
+#if defined(ABSL_USES_STD_STRING_VIEW)
// In libstdc++ (as of 7.2), `std::string_view::string_view(const char*)`
- // calls `std::char_traits<char>::length(const char*)` to get the std::string
+ // calls `std::char_traits<char>::length(const char*)` to get the string
// length, but it is not marked constexpr yet. See GCC bug:
// https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78156
// Also, there is a LWG issue that adds constexpr to length() which was just
@@ -952,7 +990,7 @@
#define ABSL_HAVE_CONSTEXPR_STRING_VIEW_FROM_CSTR 1
#endif // !__GLIBCXX__
-#else // ABSL_HAVE_STD_STRING_VIEW
+#else // ABSL_USES_STD_STRING_VIEW
// This duplicates the check for __builtin_strlen in the header.
#if ABSL_HAVE_BUILTIN(__builtin_strlen) || \
@@ -967,13 +1005,36 @@
#define ABSL_HAVE_CONSTEXPR_STRING_VIEW_FROM_CSTR 1
#endif
-#endif // ABSL_HAVE_STD_STRING_VIEW
+#endif // ABSL_USES_STD_STRING_VIEW
#ifdef ABSL_HAVE_CONSTEXPR_STRING_VIEW_FROM_CSTR
constexpr absl::string_view cstr_strlen("foo");
EXPECT_EQ(cstr_strlen.length(), 3);
constexpr absl::string_view cstr_strlen2 = "bar";
EXPECT_EQ(cstr_strlen2, "bar");
+
+#if ABSL_HAVE_BUILTIN(__builtin_memcmp) || \
+ (defined(__GNUC__) && !defined(__clang__))
+#define ABSL_HAVE_CONSTEXPR_STRING_VIEW_COMPARISON 1
+#endif
+#ifdef ABSL_HAVE_CONSTEXPR_STRING_VIEW_COMPARISON
+ constexpr absl::string_view foo = "foo";
+ constexpr absl::string_view bar = "bar";
+ constexpr bool foo_eq_bar = foo == bar;
+ constexpr bool foo_ne_bar = foo != bar;
+ constexpr bool foo_lt_bar = foo < bar;
+ constexpr bool foo_le_bar = foo <= bar;
+ constexpr bool foo_gt_bar = foo > bar;
+ constexpr bool foo_ge_bar = foo >= bar;
+ constexpr int foo_compare_bar = foo.compare(bar);
+ EXPECT_FALSE(foo_eq_bar);
+ EXPECT_TRUE(foo_ne_bar);
+ EXPECT_FALSE(foo_lt_bar);
+ EXPECT_FALSE(foo_le_bar);
+ EXPECT_TRUE(foo_gt_bar);
+ EXPECT_TRUE(foo_ge_bar);
+ EXPECT_GT(foo_compare_bar, 0);
+#endif
#endif
#if !defined(__clang__) || 3 < __clang_major__ || \
@@ -1026,6 +1087,14 @@
EXPECT_EQ(sp_npos, -1);
}
+TEST(StringViewTest, ConstexprSubstr) {
+ constexpr absl::string_view foobar("foobar", 6);
+ constexpr absl::string_view foo = foobar.substr(0, 3);
+ constexpr absl::string_view bar = foobar.substr(3);
+ EXPECT_EQ(foo, "foo");
+ EXPECT_EQ(bar, "bar");
+}
+
TEST(StringViewTest, Noexcept) {
EXPECT_TRUE((std::is_nothrow_constructible<absl::string_view,
const std::string&>::value));
@@ -1060,6 +1129,17 @@
EXPECT_TRUE(noexcept(sp.find_last_not_of('f')));
}
+TEST(StringViewTest, BoundsCheck) {
+#ifndef ABSL_USES_STD_STRING_VIEW
+#if !defined(NDEBUG) || ABSL_OPTION_HARDENED
+ // Abseil's string_view implementation has bounds-checking in debug mode.
+ absl::string_view h = "hello";
+ ABSL_EXPECT_DEATH_IF_SUPPORTED(h[5], "");
+ ABSL_EXPECT_DEATH_IF_SUPPORTED(h[-1], "");
+#endif
+#endif
+}
+
TEST(ComparisonOpsTest, StringCompareNotAmbiguous) {
EXPECT_EQ("hello", std::string("hello"));
EXPECT_LT("hello", std::string("world"));
@@ -1097,11 +1177,11 @@
EXPECT_EQ(absl::string_view::npos, a.rfind('x'));
}
-#ifndef THREAD_SANITIZER // Allocates too much memory for tsan.
+#ifndef ABSL_HAVE_THREAD_SANITIZER // Allocates too much memory for tsan.
TEST(HugeStringView, TwoPointTwoGB) {
- if (sizeof(size_t) <= 4 || RunningOnValgrind())
+ if (sizeof(size_t) <= 4)
return;
- // Try a huge std::string piece.
+ // Try a huge string piece.
const size_t size = size_t{2200} * 1000 * 1000;
std::string s(size, 'a');
absl::string_view sp(s);
@@ -1111,9 +1191,9 @@
sp.remove_suffix(2);
EXPECT_EQ(size - 1 - 2, sp.length());
}
-#endif // THREAD_SANITIZER
+#endif // ABSL_HAVE_THREAD_SANITIZER
-#if !defined(NDEBUG) && !defined(ABSL_HAVE_STD_STRING_VIEW)
+#if !defined(NDEBUG) && !defined(ABSL_USES_STD_STRING_VIEW)
TEST(NonNegativeLenTest, NonNegativeLen) {
ABSL_EXPECT_DEATH_IF_SUPPORTED(absl::string_view("xyz", -1),
"len <= kMaxSize");
@@ -1129,7 +1209,7 @@
ABSL_EXPECT_DEATH_IF_SUPPORTED(absl::string_view("", max_size + 1),
"len <= kMaxSize");
}
-#endif // !defined(NDEBUG) && !defined(ABSL_HAVE_STD_STRING_VIEW)
+#endif // !defined(NDEBUG) && !defined(ABSL_USES_STD_STRING_VIEW)
class StringViewStreamTest : public ::testing::Test {
public:
diff --git a/absl/strings/strip.h b/absl/strings/strip.h
index e1341e0..111872c 100644
--- a/absl/strings/strip.h
+++ b/absl/strings/strip.h
@@ -30,6 +30,7 @@
#include "absl/strings/string_view.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
// ConsumePrefix()
//
@@ -84,6 +85,7 @@
return str;
}
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_STRIP_H_
diff --git a/absl/strings/substitute.cc b/absl/strings/substitute.cc
index bc17695..1f3c740 100644
--- a/absl/strings/substitute.cc
+++ b/absl/strings/substitute.cc
@@ -23,6 +23,7 @@
#include "absl/strings/string_view.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace substitute_internal {
void SubstituteAndAppendArray(std::string* output, absl::string_view format,
@@ -35,7 +36,7 @@
if (i + 1 >= format.size()) {
#ifndef NDEBUG
ABSL_RAW_LOG(FATAL,
- "Invalid strings::Substitute() format std::string: \"%s\".",
+ "Invalid absl::Substitute() format string: \"%s\".",
absl::CEscape(format).c_str());
#endif
return;
@@ -45,8 +46,8 @@
#ifndef NDEBUG
ABSL_RAW_LOG(
FATAL,
- "Invalid strings::Substitute() format std::string: asked for \"$"
- "%d\", but only %d args were given. Full format std::string was: "
+ "Invalid absl::Substitute() format string: asked for \"$"
+ "%d\", but only %d args were given. Full format string was: "
"\"%s\".",
index, static_cast<int>(num_args), absl::CEscape(format).c_str());
#endif
@@ -60,7 +61,7 @@
} else {
#ifndef NDEBUG
ABSL_RAW_LOG(FATAL,
- "Invalid strings::Substitute() format std::string: \"%s\".",
+ "Invalid absl::Substitute() format string: \"%s\".",
absl::CEscape(format).c_str());
#endif
return;
@@ -72,7 +73,7 @@
if (size == 0) return;
- // Build the std::string.
+ // Build the string.
size_t original_size = output->size();
strings_internal::STLStringResizeUninitialized(output, original_size + size);
char* target = &(*output)[original_size];
@@ -94,7 +95,6 @@
assert(target == output->data() + output->size());
}
-static const char kHexDigits[] = "0123456789abcdef";
Arg::Arg(const void* value) {
static_assert(sizeof(scratch_) >= sizeof(value) * 2 + 2,
"fix sizeof(scratch_)");
@@ -104,7 +104,7 @@
char* ptr = scratch_ + sizeof(scratch_);
uintptr_t num = reinterpret_cast<uintptr_t>(value);
do {
- *--ptr = kHexDigits[num & 0xf];
+ *--ptr = absl::numbers_internal::kHexChar[num & 0xf];
num >>= 4;
} while (num != 0);
*--ptr = 'x';
@@ -119,7 +119,7 @@
char* writer = end;
uint64_t value = hex.value;
do {
- *--writer = kHexDigits[value & 0xF];
+ *--writer = absl::numbers_internal::kHexChar[value & 0xF];
value >>= 4;
} while (value != 0);
@@ -167,4 +167,5 @@
}
} // namespace substitute_internal
+ABSL_NAMESPACE_END
} // namespace absl
diff --git a/absl/strings/substitute.h b/absl/strings/substitute.h
index 233e9dc..c6da4dc 100644
--- a/absl/strings/substitute.h
+++ b/absl/strings/substitute.h
@@ -50,7 +50,7 @@
//
// Supported types:
// * absl::string_view, std::string, const char* (null is equivalent to "")
-// * int32_t, int64_t, uint32_t, uint64
+// * int32_t, int64_t, uint32_t, uint64_t
// * float, double
// * bool (Printed as "true" or "false")
// * pointer types other than char* (Printed as "0x<lower case hex string>",
@@ -86,6 +86,7 @@
#include "absl/strings/strip.h"
namespace absl {
+ABSL_NAMESPACE_BEGIN
namespace substitute_internal {
// Arg
@@ -98,7 +99,7 @@
// This class has implicit constructors.
class Arg {
public:
- // Overloads for std::string-y things
+ // Overloads for string-y things
//
// Explicitly overload `const char*` so the compiler doesn't cast to `bool`.
Arg(const char* value) // NOLINT(runtime/explicit)
@@ -119,7 +120,9 @@
// representation. However, we can't really know, so we make the caller decide
// what to do.
Arg(char value) // NOLINT(runtime/explicit)
- : piece_(scratch_, 1) { scratch_[0] = value; }
+ : piece_(scratch_, 1) {
+ scratch_[0] = value;
+ }
Arg(short value) // NOLINT(*)
: piece_(scratch_,
numbers_internal::FastIntToBuffer(value, scratch_) - scratch_) {}
@@ -189,7 +192,12 @@
#if defined(ABSL_BAD_CALL_IF)
constexpr int CalculateOneBit(const char* format) {
- return (*format < '0' || *format > '9') ? 0 : (1 << (*format - '0'));
+ // Returns:
+ // * 2^N for '$N' when N is in [0-9]
+ // * 0 for correct '$' escaping: '$$'.
+ // * -1 otherwise.
+ return (*format < '0' || *format > '9') ? (*format == '$' ? 0 : -1)
+ : (1 << (*format - '0'));
}
constexpr const char* SkipNumber(const char* format) {
@@ -197,10 +205,11 @@
}
constexpr int PlaceholderBitmask(const char* format) {
- return !*format ? 0 : *format != '$'
- ? PlaceholderBitmask(format + 1)
- : (CalculateOneBit(format + 1) |
- PlaceholderBitmask(SkipNumber(format + 1)));
+ return !*format
+ ? 0
+ : *format != '$' ? PlaceholderBitmask(format + 1)
+ : (CalculateOneBit(format + 1) |
+ PlaceholderBitmask(SkipNumber(format + 1)));
}
#endif // ABSL_BAD_CALL_IF
@@ -354,13 +363,13 @@
void SubstituteAndAppend(std::string* output, const char* format)
ABSL_BAD_CALL_IF(substitute_internal::PlaceholderBitmask(format) != 0,
"There were no substitution arguments "
- "but this format std::string has a $[0-9] in it");
+ "but this format string has a $[0-9] in it");
void SubstituteAndAppend(std::string* output, const char* format,
const substitute_internal::Arg& a0)
ABSL_BAD_CALL_IF(substitute_internal::PlaceholderBitmask(format) != 1,
"There was 1 substitution argument given, but "
- "this format std::string is either missing its $0, or "
+ "this format string is either missing its $0, or "
"contains one of $1-$9");
void SubstituteAndAppend(std::string* output, const char* format,
@@ -368,7 +377,7 @@
const substitute_internal::Arg& a1)
ABSL_BAD_CALL_IF(substitute_internal::PlaceholderBitmask(format) != 3,
"There were 2 substitution arguments given, but "
- "this format std::string is either missing its $0/$1, or "
+ "this format string is either missing its $0/$1, or "
"contains one of $2-$9");
void SubstituteAndAppend(std::string* output, const char* format,
@@ -377,7 +386,7 @@
const substitute_internal::Arg& a2)
ABSL_BAD_CALL_IF(substitute_internal::PlaceholderBitmask(format) != 7,
"There were 3 substitution arguments given, but "
- "this format std::string is either missing its $0/$1/$2, or "
+ "this format string is either missing its $0/$1/$2, or "
"contains one of $3-$9");
void SubstituteAndAppend(std::string* output, const char* format,
@@ -387,7 +396,7 @@
const substitute_internal::Arg& a3)
ABSL_BAD_CALL_IF(substitute_internal::PlaceholderBitmask(format) != 15,
"There were 4 substitution arguments given, but "
- "this format std::string is either missing its $0-$3, or "
+ "this format string is either missing its $0-$3, or "
"contains one of $4-$9");
void SubstituteAndAppend(std::string* output, const char* format,
@@ -398,7 +407,7 @@
const substitute_internal::Arg& a4)
ABSL_BAD_CALL_IF(substitute_internal::PlaceholderBitmask(format) != 31,
"There were 5 substitution arguments given, but "
- "this format std::string is either missing its $0-$4, or "
+ "this format string is either missing its $0-$4, or "
"contains one of $5-$9");
void SubstituteAndAppend(std::string* output, const char* format,
@@ -410,7 +419,7 @@
const substitute_internal::Arg& a5)
ABSL_BAD_CALL_IF(substitute_internal::PlaceholderBitmask(format) != 63,
"There were 6 substitution arguments given, but "
- "this format std::string is either missing its $0-$5, or "
+ "this format string is either missing its $0-$5, or "
"contains one of $6-$9");
void SubstituteAndAppend(
@@ -420,7 +429,7 @@
const substitute_internal::Arg& a5, const substitute_internal::Arg& a6)
ABSL_BAD_CALL_IF(substitute_internal::PlaceholderBitmask(format) != 127,
"There were 7 substitution arguments given, but "
- "this format std::string is either missing its $0-$6, or "
+ "this format string is either missing its $0-$6, or "
"contains one of $7-$9");
void SubstituteAndAppend(
@@ -431,7 +440,7 @@
const substitute_internal::Arg& a7)
ABSL_BAD_CALL_IF(substitute_internal::PlaceholderBitmask(format) != 255,
"There were 8 substitution arguments given, but "
- "this format std::string is either missing its $0-$7, or "
+ "this format string is either missing its $0-$7, or "
"contains one of $8-$9");
void SubstituteAndAppend(
@@ -443,7 +452,7 @@
ABSL_BAD_CALL_IF(
substitute_internal::PlaceholderBitmask(format) != 511,
"There were 9 substitution arguments given, but "
- "this format std::string is either missing its $0-$8, or contains a $9");
+ "this format string is either missing its $0-$8, or contains a $9");
void SubstituteAndAppend(
std::string* output, const char* format, const substitute_internal::Arg& a0,
@@ -454,7 +463,7 @@
const substitute_internal::Arg& a9)
ABSL_BAD_CALL_IF(substitute_internal::PlaceholderBitmask(format) != 1023,
"There were 10 substitution arguments given, but this "
- "format std::string doesn't contain all of $0 through $9");
+ "format string doesn't contain all of $0 through $9");
#endif // ABSL_BAD_CALL_IF
// Substitute()
@@ -580,19 +589,19 @@
std::string Substitute(const char* format)
ABSL_BAD_CALL_IF(substitute_internal::PlaceholderBitmask(format) != 0,
"There were no substitution arguments "
- "but this format std::string has a $[0-9] in it");
+ "but this format string has a $[0-9] in it");
std::string Substitute(const char* format, const substitute_internal::Arg& a0)
ABSL_BAD_CALL_IF(substitute_internal::PlaceholderBitmask(format) != 1,
"There was 1 substitution argument given, but "
- "this format std::string is either missing its $0, or "
+ "this format string is either missing its $0, or "
"contains one of $1-$9");
std::string Substitute(const char* format, const substitute_internal::Arg& a0,
const substitute_internal::Arg& a1)
ABSL_BAD_CALL_IF(substitute_internal::PlaceholderBitmask(format) != 3,
"There were 2 substitution arguments given, but "
- "this format std::string is either missing its $0/$1, or "
+ "this format string is either missing its $0/$1, or "
"contains one of $2-$9");
std::string Substitute(const char* format, const substitute_internal::Arg& a0,
@@ -600,7 +609,7 @@
const substitute_internal::Arg& a2)
ABSL_BAD_CALL_IF(substitute_internal::PlaceholderBitmask(format) != 7,
"There were 3 substitution arguments given, but "
- "this format std::string is either missing its $0/$1/$2, or "
+ "this format string is either missing its $0/$1/$2, or "
"contains one of $3-$9");
std::string Substitute(const char* format, const substitute_internal::Arg& a0,
@@ -609,7 +618,7 @@
const substitute_internal::Arg& a3)
ABSL_BAD_CALL_IF(substitute_internal::PlaceholderBitmask(format) != 15,
"There were 4 substitution arguments given, but "
- "this format std::string is either missing its $0-$3, or "
+ "this format string is either missing its $0-$3, or "
"contains one of $4-$9");
std::string Substitute(const char* format, const substitute_internal::Arg& a0,
@@ -619,7 +628,7 @@
const substitute_internal::Arg& a4)
ABSL_BAD_CALL_IF(substitute_internal::PlaceholderBitmask(format) != 31,
"There were 5 substitution arguments given, but "
- "this format std::string is either missing its $0-$4, or "
+ "this format string is either missing its $0-$4, or "
"contains one of $5-$9");
std::string Substitute(const char* format, const substitute_internal::Arg& a0,
@@ -630,7 +639,7 @@
const substitute_internal::Arg& a5)
ABSL_BAD_CALL_IF(substitute_internal::PlaceholderBitmask(format) != 63,
"There were 6 substitution arguments given, but "
- "this format std::string is either missing its $0-$5, or "
+ "this format string is either missing its $0-$5, or "
"contains one of $6-$9");
std::string Substitute(const char* format, const substitute_internal::Arg& a0,
@@ -642,7 +651,7 @@
const substitute_internal::Arg& a6)
ABSL_BAD_CALL_IF(substitute_internal::PlaceholderBitmask(format) != 127,
"There were 7 substitution arguments given, but "
- "this format std::string is either missing its $0-$6, or "
+ "this format string is either missing its $0-$6, or "
"contains one of $7-$9");
std::string Substitute(const char* format, const substitute_internal::Arg& a0,
@@ -655,7 +664,7 @@
const substitute_internal::Arg& a7)
ABSL_BAD_CALL_IF(substitute_internal::PlaceholderBitmask(format) != 255,
"There were 8 substitution arguments given, but "
- "this format std::string is either missing its $0-$7, or "
+ "this format string is either missing its $0-$7, or "
"contains one of $8-$9");
std::string Substitute(
@@ -667,7 +676,7 @@
ABSL_BAD_CALL_IF(
substitute_internal::PlaceholderBitmask(format) != 511,
"There were 9 substitution arguments given, but "
- "this format std::string is either missing its $0-$8, or contains a $9");
+ "this format string is either missing its $0-$8, or contains a $9");
std::string Substitute(
const char* format, const substitute_internal::Arg& a0,
@@ -678,9 +687,10 @@
const substitute_internal::Arg& a9)
ABSL_BAD_CALL_IF(substitute_internal::PlaceholderBitmask(format) != 1023,
"There were 10 substitution arguments given, but this "
- "format std::string doesn't contain all of $0 through $9");
+ "format string doesn't contain all of $0 through $9");
#endif // ABSL_BAD_CALL_IF
+ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_STRINGS_SUBSTITUTE_H_
diff --git a/absl/strings/substitute_test.cc b/absl/strings/substitute_test.cc
index e27abb1..442c921 100644
--- a/absl/strings/substitute_test.cc
+++ b/absl/strings/substitute_test.cc
@@ -89,7 +89,7 @@
str = absl::Substitute("$0", char_buf);
EXPECT_EQ("print me too", str);
- // null char* is "doubly" special. Represented as the empty std::string.
+ // null char* is "doubly" special. Represented as the empty string.
char_p = nullptr;
str = absl::Substitute("$0", char_p);
EXPECT_EQ("", str);
@@ -189,14 +189,14 @@
TEST(SubstituteDeathTest, SubstituteDeath) {
EXPECT_DEBUG_DEATH(
static_cast<void>(absl::Substitute(absl::string_view("-$2"), "a", "b")),
- "Invalid strings::Substitute\\(\\) format std::string: asked for \"\\$2\", "
+ "Invalid absl::Substitute\\(\\) format string: asked for \"\\$2\", "
"but only 2 args were given.");
EXPECT_DEBUG_DEATH(
- static_cast<void>(absl::Substitute("-$z-")),
- "Invalid strings::Substitute\\(\\) format std::string: \"-\\$z-\"");
+ static_cast<void>(absl::Substitute(absl::string_view("-$z-"))),
+ "Invalid absl::Substitute\\(\\) format string: \"-\\$z-\"");
EXPECT_DEBUG_DEATH(
- static_cast<void>(absl::Substitute("-$")),
- "Invalid strings::Substitute\\(\\) format std::string: \"-\\$\"");
+ static_cast<void>(absl::Substitute(absl::string_view("-$"))),
+ "Invalid absl::Substitute\\(\\) format string: \"-\\$\"");
}
#endif // GTEST_HAS_DEATH_TEST
diff --git a/absl/strings/testdata/getline-1.txt b/absl/strings/testdata/getline-1.txt
deleted file mode 100644
index 19b9097..0000000
--- a/absl/strings/testdata/getline-1.txt
+++ /dev/null
@@ -1,3 +0,0 @@
-alpha
-
-beta gamma
diff --git a/absl/strings/testdata/getline-2.txt b/absl/strings/testdata/getline-2.txt
deleted file mode 100644
index d6842d8..0000000
--- a/absl/strings/testdata/getline-2.txt
+++ /dev/null
@@ -1 +0,0 @@
-one.two.three