James Kuszmaul | 48dd4c8 | 2021-10-27 20:04:08 -0700 | [diff] [blame^] | 1 | Snappy compressed format description |
| 2 | Last revised: 2011-10-05 |
| 3 | |
| 4 | |
| 5 | This is not a formal specification, but should suffice to explain most |
| 6 | relevant parts of how the Snappy format works. It is originally based on |
| 7 | text by Zeev Tarantov. |
| 8 | |
| 9 | Snappy is a LZ77-type compressor with a fixed, byte-oriented encoding. |
| 10 | There is no entropy encoder backend nor framing layer -- the latter is |
| 11 | assumed to be handled by other parts of the system. |
| 12 | |
| 13 | This document only describes the format, not how the Snappy compressor nor |
| 14 | decompressor actually works. The correctness of the decompressor should not |
| 15 | depend on implementation details of the compressor, and vice versa. |
| 16 | |
| 17 | |
| 18 | 1. Preamble |
| 19 | |
| 20 | The stream starts with the uncompressed length (up to a maximum of 2^32 - 1), |
| 21 | stored as a little-endian varint. Varints consist of a series of bytes, |
| 22 | where the lower 7 bits are data and the upper bit is set iff there are |
| 23 | more bytes to be read. In other words, an uncompressed length of 64 would |
| 24 | be stored as 0x40, and an uncompressed length of 2097150 (0x1FFFFE) |
| 25 | would be stored as 0xFE 0xFF 0x7F. |
| 26 | |
| 27 | |
| 28 | 2. The compressed stream itself |
| 29 | |
| 30 | There are two types of elements in a Snappy stream: Literals and |
| 31 | copies (backreferences). There is no restriction on the order of elements, |
| 32 | except that the stream naturally cannot start with a copy. (Having |
| 33 | two literals in a row is never optimal from a compression point of |
| 34 | view, but nevertheless fully permitted.) Each element starts with a tag byte, |
| 35 | and the lower two bits of this tag byte signal what type of element will |
| 36 | follow: |
| 37 | |
| 38 | 00: Literal |
| 39 | 01: Copy with 1-byte offset |
| 40 | 10: Copy with 2-byte offset |
| 41 | 11: Copy with 4-byte offset |
| 42 | |
| 43 | The interpretation of the upper six bits are element-dependent. |
| 44 | |
| 45 | |
| 46 | 2.1. Literals (00) |
| 47 | |
| 48 | Literals are uncompressed data stored directly in the byte stream. |
| 49 | The literal length is stored differently depending on the length |
| 50 | of the literal: |
| 51 | |
| 52 | - For literals up to and including 60 bytes in length, the upper |
| 53 | six bits of the tag byte contain (len-1). The literal follows |
| 54 | immediately thereafter in the bytestream. |
| 55 | - For longer literals, the (len-1) value is stored after the tag byte, |
| 56 | little-endian. The upper six bits of the tag byte describe how |
| 57 | many bytes are used for the length; 60, 61, 62 or 63 for |
| 58 | 1-4 bytes, respectively. The literal itself follows after the |
| 59 | length. |
| 60 | |
| 61 | |
| 62 | 2.2. Copies |
| 63 | |
| 64 | Copies are references back into previous decompressed data, telling |
| 65 | the decompressor to reuse data it has previously decoded. |
| 66 | They encode two values: The _offset_, saying how many bytes back |
| 67 | from the current position to read, and the _length_, how many bytes |
| 68 | to copy. Offsets of zero can be encoded, but are not legal; |
| 69 | similarly, it is possible to encode backreferences that would |
| 70 | go past the end of the block (offset > current decompressed position), |
| 71 | which is also nonsensical and thus not allowed. |
| 72 | |
| 73 | As in most LZ77-based compressors, the length can be larger than the offset, |
| 74 | yielding a form of run-length encoding (RLE). For instance, |
| 75 | "xababab" could be encoded as |
| 76 | |
| 77 | <literal: "xab"> <copy: offset=2 length=4> |
| 78 | |
| 79 | Note that since the current Snappy compressor works in 32 kB |
| 80 | blocks and does not do matching across blocks, it will never produce |
| 81 | a bitstream with offsets larger than about 32768. However, the |
| 82 | decompressor should not rely on this, as it may change in the future. |
| 83 | |
| 84 | There are several different kinds of copy elements, depending on |
| 85 | the amount of bytes to be copied (length), and how far back the |
| 86 | data to be copied is (offset). |
| 87 | |
| 88 | |
| 89 | 2.2.1. Copy with 1-byte offset (01) |
| 90 | |
| 91 | These elements can encode lengths between [4..11] bytes and offsets |
| 92 | between [0..2047] bytes. (len-4) occupies three bits and is stored |
| 93 | in bits [2..4] of the tag byte. The offset occupies 11 bits, of which the |
| 94 | upper three are stored in the upper three bits ([5..7]) of the tag byte, |
| 95 | and the lower eight are stored in a byte following the tag byte. |
| 96 | |
| 97 | |
| 98 | 2.2.2. Copy with 2-byte offset (10) |
| 99 | |
| 100 | These elements can encode lengths between [1..64] and offsets from |
| 101 | [0..65535]. (len-1) occupies six bits and is stored in the upper |
| 102 | six bits ([2..7]) of the tag byte. The offset is stored as a |
| 103 | little-endian 16-bit integer in the two bytes following the tag byte. |
| 104 | |
| 105 | |
| 106 | 2.2.3. Copy with 4-byte offset (11) |
| 107 | |
| 108 | These are like the copies with 2-byte offsets (see previous subsection), |
| 109 | except that the offset is stored as a 32-bit integer instead of a |
| 110 | 16-bit integer (and thus will occupy four bytes). |