Documentation/technical/commit-graph-format.txt

   1 Git commit graph format
   2 =======================
   3
   4 The Git commit graph stores a list of commit OIDs and some associated
   5 metadata, including:
   6
   7 - The generation number of the commit.
   8
   9 - The root tree OID.
  10
  11 - The commit date.
  12
  13 - The parents of the commit, stored using positional references within
  14   the graph file.
  15
  16 - The Bloom filter of the commit carrying the paths that were changed between
  17   the commit and its first parent, if requested.
  18
  19 These positional references are stored as unsigned 32-bit integers
  20 corresponding to the array position within the list of commit OIDs. Due
  21 to some special constants we use to track parents, we can store at most
  22 (1 << 30) + (1 << 29) + (1 << 28) - 1 (around 1.8 billion) commits.
  23
  24 == Commit graph files have the following format:
  25
  26 In order to allow extensions that add extra data to the graph, we organize
  27 the body into "chunks" and provide a binary lookup table at the beginning
  28 of the body. The header includes certain values, such as number of chunks
  29 and hash type.
  30
  31 All multi-byte numbers are in network byte order.
  32
  33 HEADER:
  34
  35   4-byte signature:
  36       The signature is: {'C', 'G', 'P', 'H'}
  37
  38   1-byte version number:
  39       Currently, the only valid version is 1.
  40
  41   1-byte Hash Version
  42       We infer the hash length (H) from this value:
  43         1 => SHA-1
  44         2 => SHA-256
  45       If the hash type does not match the repository's hash algorithm, the
  46       commit-graph file should be ignored with a warning presented to the
  47       user.
  48
  49   1-byte number (C) of "chunks"
  50
  51   1-byte number (B) of base commit-graphs
  52       We infer the length (H*B) of the Base Graphs chunk
  53       from this value.
  54
  55 CHUNK LOOKUP:
  56
  57   (C + 1) * 12 bytes listing the table of contents for the chunks:
  58       First 4 bytes describe the chunk id. Value 0 is a terminating label.
  59       Other 8 bytes provide the byte-offset in current file for chunk to
  60       start. (Chunks are ordered contiguously in the file, so you can infer
  61       the length using the next chunk position if necessary.) Each chunk
  62       ID appears at most once.
  63
  64   The CHUNK LOOKUP matches the table of contents from
  65   link:technical/chunk-format.html[the chunk-based file format].
  66
  67   The remaining data in the body is described one chunk at a time, and
  68   these chunks may be given in any order. Chunks are required unless
  69   otherwise specified.
  70
  71 CHUNK DATA:
  72
  73   OID Fanout (ID: {'O', 'I', 'D', 'F'}) (256 * 4 bytes)
  74       The ith entry, F[i], stores the number of OIDs with first
  75       byte at most i. Thus F[255] stores the total
  76       number of commits (N).
  77
  78   OID Lookup (ID: {'O', 'I', 'D', 'L'}) (N * H bytes)
  79       The OIDs for all commits in the graph, sorted in ascending order.
  80
  81   Commit Data (ID: {'C', 'D', 'A', 'T' }) (N * (H + 16) bytes)
  82     * The first H bytes are for the OID of the root tree.
  83     * The next 8 bytes are for the positions of the first two parents
  84       of the ith commit. Stores value 0x70000000 if no parent in that
  85       position. If there are more than two parents, the second value
  86       has its most-significant bit on and the other bits store an array
  87       position into the Extra Edge List chunk.
  88     * The next 8 bytes store the topological level (generation number v1)
  89       of the commit and
  90       the commit time in seconds since EPOCH. The generation number
  91       uses the higher 30 bits of the first 4 bytes, while the commit
  92       time uses the 32 bits of the second 4 bytes, along with the lowest
  93       2 bits of the lowest byte, storing the 33rd and 34th bit of the
  94       commit time.
  95
  96   Generation Data (ID: {'G', 'D', 'A', '2' }) (N * 4 bytes) [Optional]
  97     * This list of 4-byte values store corrected commit date offsets for the
  98       commits, arranged in the same order as commit data chunk.
  99     * If the corrected commit date offset cannot be stored within 31 bits,
 100       the value has its most-significant bit on and the other bits store
 101       the position of corrected commit date into the Generation Data Overflow
 102       chunk.
 103     * Generation Data chunk is present only when commit-graph file is written
 104       by compatible versions of Git and in case of split commit-graph chains,
 105       the topmost layer also has Generation Data chunk.
 106
 107   Generation Data Overflow (ID: {'G', 'D', 'O', '2' }) [Optional]
 108     * This list of 8-byte values stores the corrected commit date offsets
 109       for commits with corrected commit date offsets that cannot be
 110       stored within 31 bits.
 111     * Generation Data Overflow chunk is present only when Generation Data
 112       chunk is present and atleast one corrected commit date offset cannot
 113       be stored within 31 bits.
 114
 115   Extra Edge List (ID: {'E', 'D', 'G', 'E'}) [Optional]
 116       This list of 4-byte values store the second through nth parents for
 117       all octopus merges. The second parent value in the commit data stores
 118       an array position within this list along with the most-significant bit
 119       on. Starting at that array position, iterate through this list of commit
 120       positions for the parents until reaching a value with the most-significant
 121       bit on. The other bits correspond to the position of the last parent.
 122
 123   Bloom Filter Index (ID: {'B', 'I', 'D', 'X'}) (N * 4 bytes) [Optional]
 124     * The ith entry, BIDX[i], stores the number of bytes in all Bloom filters
 125       from commit 0 to commit i (inclusive) in lexicographic order. The Bloom
 126       filter for the i-th commit spans from BIDX[i-1] to BIDX[i] (plus header
 127       length), where BIDX[-1] is 0.
 128     * The BIDX chunk is ignored if the BDAT chunk is not present.
 129
 130   Bloom Filter Data (ID: {'B', 'D', 'A', 'T'}) [Optional]
 131     * It starts with header consisting of three unsigned 32-bit integers:
 132       - Version of the hash algorithm being used. We currently only support
 133         value 1 which corresponds to the 32-bit version of the murmur3 hash
 134         implemented exactly as described in
 135         https://en.wikipedia.org/wiki/MurmurHash#Algorithm and the double
 136         hashing technique using seed values 0x293ae76f and 0x7e646e2 as
 137         described in https://doi.org/10.1007/978-3-540-30494-4_26 "Bloom Filters
 138         in Probabilistic Verification"
 139       - The number of times a path is hashed and hence the number of bit positions
 140               that cumulatively determine whether a file is present in the commit.
 141       - The minimum number of bits 'b' per entry in the Bloom filter. If the filter
 142               contains 'n' entries, then the filter size is the minimum number of 64-bit
 143               words that contain n*b bits.
 144     * The rest of the chunk is the concatenation of all the computed Bloom
 145       filters for the commits in lexicographic order.
 146     * Note: Commits with no changes or more than 512 changes have Bloom filters
 147       of length one, with either all bits set to zero or one respectively.
 148     * The BDAT chunk is present if and only if BIDX is present.
 149
 150   Base Graphs List (ID: {'B', 'A', 'S', 'E'}) [Optional]
 151       This list of H-byte hashes describe a set of B commit-graph files that
 152       form a commit-graph chain. The graph position for the ith commit in this
 153       file's OID Lookup chunk is equal to i plus the number of commits in all
 154       base graphs.  If B is non-zero, this chunk must exist.
 155
 156 TRAILER:
 157
 158         H-byte HASH-checksum of all of the above.
 159
 160 == Historical Notes:
 161
 162 The Generation Data (GDA2) and Generation Data Overflow (GDO2) chunks have
 163 the number '2' in their chunk IDs because a previous version of Git wrote
 164 possibly erroneous data in these chunks with the IDs "GDAT" and "GDOV". By
 165 changing the IDs, newer versions of Git will silently ignore those older
 166 chunks and write the new information without trusting the incorrect data.