Write out full resolution floats in flatbuffer to JSON
It turns out that we were missing a 1--2 digits of precision
necessary to fully represent floats/doubles. In most cases where we
would care about bit-perfect
This moves all of our flatbuffer->JSON conversion to use
std::format/std::to_chars, which specifically guarantee that you will be
able to read the number back fully, while truncating numbers reasonably
so that e.g. 0.1 actually renders reasonably to the human eye.
In doing so, this also reduces some of the inconsistencies between the
two FlatbufferToJson methods.
This forces updates in a variety of tests that make use of this code and
which check for exactly identical serialized JSON values.
Change-Id: Idbf6a5614043ce4d6c02f3104991e30fdca23334
Signed-off-by: James Kuszmaul <james.kuszmaul@bluerivertech.com>
diff --git a/aos/json_to_flatbuffer_test.cc b/aos/json_to_flatbuffer_test.cc
index 7c69acd..720c55d 100644
--- a/aos/json_to_flatbuffer_test.cc
+++ b/aos/json_to_flatbuffer_test.cc
@@ -93,10 +93,15 @@
EXPECT_TRUE(JsonAndBack("{ \"foo_long\": 5 }"));
EXPECT_TRUE(JsonAndBack("{ \"foo_ulong\": 5 }"));
- // TODO(james): Make FlatbufferToJson() always print out integer
- // floating-point numbers identically.
- EXPECT_TRUE(JsonAndBack("{ \"foo_float\": 5.0 }", TestReflection::kNo));
- EXPECT_TRUE(JsonAndBack("{ \"foo_double\": 5.0 }", TestReflection::kNo));
+ EXPECT_TRUE(JsonAndBack("{ \"foo_float\": 5 }"));
+ EXPECT_TRUE(JsonAndBack("{ \"foo_float\": 50 }"));
+ // Test that we can distinguish between floats that vary by a single bit.
+ EXPECT_TRUE(JsonAndBack("{ \"foo_float\": 1.1 }"));
+ EXPECT_TRUE(JsonAndBack("{ \"foo_float\": 1.0999999 }"));
+ EXPECT_TRUE(JsonAndBack("{ \"foo_double\": 5 }"));
+ // Check that we handle/distinguish between doubles that vary by a single bit.
+ EXPECT_TRUE(JsonAndBack("{ \"foo_double\": 1.561154546713 }"));
+ EXPECT_TRUE(JsonAndBack("{ \"foo_double\": 1.56115454671299 }"));
EXPECT_TRUE(JsonAndBack("{ \"foo_enum\": \"None\" }"));
EXPECT_TRUE(JsonAndBack("{ \"foo_enum\": \"UType\" }"));
@@ -116,24 +121,22 @@
"\"foo_byte\": 2 } } }"));
EXPECT_TRUE(JsonAndBack(
"{ \"foo_struct_scalars\": { \"foo_float\": 1.234, \"foo_double\": "
- "4.567, \"foo_int32\": -971, \"foo_uint32\": 4294967294, \"foo_int64\": "
- "-1030, \"foo_uint64\": 18446744073709551614 } }",
+ "4.567, \"foo_int32\": -971, \"foo_uint32\": 4294967294, "
+ "\"foo_int64\": -1030, \"foo_uint64\": 18446744073709551614 } }",
TestReflection::kNo));
// Confirm that we parse integers into floating point fields correctly.
EXPECT_TRUE(JsonAndBack(
"{ \"foo_struct_scalars\": { \"foo_float\": 1, \"foo_double\": "
"2, \"foo_int32\": 3, \"foo_uint32\": 4, \"foo_int64\": "
"5, \"foo_uint64\": 6 } }",
- "{ \"foo_struct_scalars\": { \"foo_float\": 1.0, \"foo_double\": "
- "2.0, \"foo_int32\": 3, \"foo_uint32\": 4, \"foo_int64\": "
- "5, \"foo_uint64\": 6 } }",
TestReflection::kNo));
EXPECT_TRUE(JsonAndBack(
"{ \"vector_foo_struct_scalars\": [ { \"foo_float\": 1.234, "
- "\"foo_double\": 4.567, \"foo_int32\": -971, \"foo_uint32\": 4294967294, "
- "\"foo_int64\": -1030, \"foo_uint64\": 18446744073709551614 }, { "
- "\"foo_float\": 2.0, \"foo_double\": 4.1, \"foo_int32\": 10, "
- "\"foo_uint32\": 13, \"foo_int64\": 15, \"foo_uint64\": 18 } ] }",
+ "\"foo_double\": 4.567, \"foo_int32\": -971, "
+ "\"foo_uint32\": 4294967294, \"foo_int64\": -1030, \"foo_uint64\": "
+ "18446744073709551614 }, { \"foo_float\": 2, \"foo_double\": "
+ "4.1, \"foo_int32\": 10, \"foo_uint32\": 13, "
+ "\"foo_int64\": 15, \"foo_uint64\": 18 } ] }",
TestReflection::kNo));
EXPECT_TRUE(
JsonAndBack("{ \"foo_struct_enum\": { \"foo_enum\": \"UByte\" } }"));
@@ -209,8 +212,8 @@
// Tests that we can handle decimal points.
TEST_F(JsonToFlatbufferTest, DecimalPoint) {
- EXPECT_TRUE(JsonAndBack("{ \"foo_float\": 5.1 }"));
- EXPECT_TRUE(JsonAndBack("{ \"foo_double\": 5.1 }"));
+ EXPECT_TRUE(JsonAndBack("{ \"foo_float\": 5.099999 }"));
+ EXPECT_TRUE(JsonAndBack("{ \"foo_double\": 5.099999999999 }"));
}
// Test what happens if you pass a field name that we don't know.
@@ -283,19 +286,15 @@
EXPECT_TRUE(JsonAndBack("{ \"vector_foo_ulong\": [ 9, 7, 1 ] }"));
EXPECT_TRUE(JsonAndBack("{ \"vector_foo_ulong\": [ ] }"));
- EXPECT_TRUE(JsonAndBack("{ \"vector_foo_float\": [ 9.0, 7.0, 1.0 ] }",
- TestReflection::kNo));
+ EXPECT_TRUE(JsonAndBack("{ \"vector_foo_float\": [ 9, 7, 1 ] }"));
EXPECT_TRUE(JsonAndBack("{ \"vector_foo_float\": [ ] }"));
- EXPECT_TRUE(JsonAndBack("{ \"vector_foo_double\": [ 9.0, 7.0, 1.0 ] }",
- TestReflection::kNo));
+ EXPECT_TRUE(JsonAndBack("{ \"vector_foo_double\": [ 9, 7, 1 ] }"));
EXPECT_TRUE(JsonAndBack("{ \"vector_foo_double\": [ ] }"));
- EXPECT_TRUE(JsonAndBack("{ \"vector_foo_float\": [ 9, 7, 1 ] }",
- "{ \"vector_foo_float\": [ 9.0, 7.0, 1.0 ] }",
- TestReflection::kNo));
- EXPECT_TRUE(JsonAndBack("{ \"vector_foo_double\": [ 9, 7, 1 ] }",
- "{ \"vector_foo_double\": [ 9.0, 7.0, 1.0 ] }",
- TestReflection::kNo));
+ EXPECT_TRUE(JsonAndBack("{ \"vector_foo_float\": [ 9.0, 7.0, 1.0 ] }",
+ "{ \"vector_foo_float\": [ 9, 7, 1 ] }"));
+ EXPECT_TRUE(JsonAndBack("{ \"vector_foo_double\": [ 9.0, 7.0, 1.0 ] }",
+ "{ \"vector_foo_double\": [ 9, 7, 1 ] }"));
EXPECT_TRUE(JsonAndBack("{ \"vector_foo_string\": [ \"bar\", \"baz\" ] }"));
EXPECT_TRUE(JsonAndBack("{ \"vector_foo_string\": [ ] }"));
@@ -336,8 +335,7 @@
/* foo */
"vector_foo_double": [ 9, 7, 1 ] /* foo */
} /* foo */)",
- "{ \"vector_foo_double\": [ 9.0, 7.0, 1.0 ] }",
- TestReflection::kNo));
+ "{ \"vector_foo_double\": [ 9, 7, 1 ] }"));
}
// Tests that C++ style comments get stripped.
@@ -346,8 +344,7 @@
// foo
"vector_foo_double": [ 9, 7, 1 ] // foo
} // foo)",
- "{ \"vector_foo_double\": [ 9.0, 7.0, 1.0 ] }",
- TestReflection::kNo));
+ "{ \"vector_foo_double\": [ 9, 7, 1 ] }"));
// Test empty comment on its own line doesn't remove the next line.
EXPECT_TRUE(JsonAndBack(R"({
@@ -355,8 +352,8 @@
"vector_foo_double": [ 9, 7, 1 ], // foo
"vector_foo_float": [ 3, 1, 4 ]
} // foo)",
- "{ \"vector_foo_float\": [ 3.0, 1.0, 4.0 ], "
- "\"vector_foo_double\": [ 9.0, 7.0, 1.0 ] }",
+ "{ \"vector_foo_float\": [ 3, 1, 4 ], "
+ "\"vector_foo_double\": [ 9, 7, 1 ] }",
TestReflection::kNo));
// Test empty comment at end of line doesn't remove the next line.
@@ -365,8 +362,8 @@
"vector_foo_double": [ 2, 7, 1 ], //
"vector_foo_float": [ 3, 1, 4 ]
} // foo)",
- "{ \"vector_foo_float\": [ 3.0, 1.0, 4.0 ], "
- "\"vector_foo_double\": [ 2.0, 7.0, 1.0 ] }",
+ "{ \"vector_foo_float\": [ 3, 1, 4 ], "
+ "\"vector_foo_double\": [ 2, 7, 1 ] }",
TestReflection::kNo));
// Test empty comment at end of document doesn't cause error.
@@ -375,8 +372,8 @@
"vector_foo_double": [ 5, 6, 7 ], // foo
"vector_foo_float": [ 7, 8, 9 ]
} //)",
- "{ \"vector_foo_float\": [ 7.0, 8.0, 9.0 ], "
- "\"vector_foo_double\": [ 5.0, 6.0, 7.0 ] }",
+ "{ \"vector_foo_float\": [ 7, 8, 9 ], "
+ "\"vector_foo_double\": [ 5, 6, 7 ] }",
TestReflection::kNo));
}
@@ -389,8 +386,8 @@
}
// foo
/* foo */)",
- "{ \"vector_foo_double\": [ 9.0, 7.0, 1.0 ] }",
- TestReflection::kNo));
+ "{ \"vector_foo_double\": [ 9, 7, 1 ] }",
+ TestReflection::kYes));
}
// Tests that multiple arrays get properly handled.
@@ -398,8 +395,6 @@
EXPECT_TRUE(
JsonAndBack("{ \"vector_foo_float\": [ 9, 7, 1 ], \"vector_foo_double\": "
"[ 9, 7, 1 ] }",
- "{ \"vector_foo_float\": [ 9.0, 7.0, 1.0 ], "
- "\"vector_foo_double\": [ 9.0, 7.0, 1.0 ] }",
TestReflection::kNo));
}