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. Commits with no parents have
   8   generation number 1; commits with parents have generation number
   9   one more than the maximum generation number of its parents. We
  10   reserve zero as special, and can be used to mark a generation
  11   number invalid or as "not computed".
  12
  13 - The root tree OID.
  14
  15 - The commit date.
  16
  17 - The parents of the commit, stored using positional references within
  18   the graph file.
  19
  20 - The Bloom filter of the commit carrying the paths that were changed between
  21   the commit and its first parent, if requested.
  22
  23 These positional references are stored as unsigned 32-bit integers
  24 corresponding to the array position within the list of commit OIDs. Due
  25 to some special constants we use to track parents, we can store at most
  26 (1 << 30) + (1 << 29) + (1 << 28) - 1 (around 1.8 billion) commits.
  27
  28 == Commit graph files have the following format:
  29
  30 In order to allow extensions that add extra data to the graph, we organize
  31 the body into "chunks" and provide a binary lookup table at the beginning
  32 of the body. The header includes certain values, such as number of chunks
  33 and hash type.
  34
  35 All multi-byte numbers are in network byte order.
  36
  37 HEADER:
  38
  39   4-byte signature:
  40       The signature is: {'C', 'G', 'P', 'H'}
  41
  42   1-byte version number:
  43       Currently, the only valid version is 1.
  44
  45   1-byte Hash Version
  46       We infer the hash length (H) from this value:
  47         1 => SHA-1
  48         2 => SHA-256
  49       If the hash type does not match the repository's hash algorithm, the
  50       commit-graph file should be ignored with a warning presented to the
  51       user.
  52
  53   1-byte number (C) of "chunks"
  54
  55   1-byte number (B) of base commit-graphs
  56       We infer the length (H*B) of the Base Graphs chunk
  57       from this value.
  58
  59 CHUNK LOOKUP:
  60
  61   (C + 1) * 12 bytes listing the table of contents for the chunks:
  62       First 4 bytes describe the chunk id. Value 0 is a terminating label.
  63       Other 8 bytes provide the byte-offset in current file for chunk to
  64       start. (Chunks are ordered contiguously in the file, so you can infer
  65       the length using the next chunk position if necessary.) Each chunk
  66       ID appears at most once.
  67
  68   The remaining data in the body is described one chunk at a time, and
  69   these chunks may be given in any order. Chunks are required unless
  70   otherwise specified.
  71
  72 CHUNK DATA:
  73
  74   OID Fanout (ID: {'O', 'I', 'D', 'F'}) (256 * 4 bytes)
  75       The ith entry, F[i], stores the number of OIDs with first
  76       byte at most i. Thus F[255] stores the total
  77       number of commits (N).
  78
  79   OID Lookup (ID: {'O', 'I', 'D', 'L'}) (N * H bytes)
  80       The OIDs for all commits in the graph, sorted in ascending order.
  81
  82   Commit Data (ID: {'C', 'D', 'A', 'T' }) (N * (H + 16) bytes)
  83     * The first H bytes are for the OID of the root tree.
  84     * The next 8 bytes are for the positions of the first two parents
  85       of the ith commit. Stores value 0x7000000 if no parent in that
  86       position. If there are more than two parents, the second value
  87       has its most-significant bit on and the other bits store an array
  88       position into the Extra Edge List chunk.
  89     * The next 8 bytes store the generation number 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   Extra Edge List (ID: {'E', 'D', 'G', 'E'}) [Optional]
  97       This list of 4-byte values store the second through nth parents for
  98       all octopus merges. The second parent value in the commit data stores
  99       an array position within this list along with the most-significant bit
 100       on. Starting at that array position, iterate through this list of commit
 101       positions for the parents until reaching a value with the most-significant
 102       bit on. The other bits correspond to the position of the last parent.
 103
 104   Bloom Filter Index (ID: {'B', 'I', 'D', 'X'}) (N * 4 bytes) [Optional]
 105     * The ith entry, BIDX[i], stores the number of bytes in all Bloom filters
 106       from commit 0 to commit i (inclusive) in lexicographic order. The Bloom
 107       filter for the i-th commit spans from BIDX[i-1] to BIDX[i] (plus header
 108       length), where BIDX[-1] is 0.
 109     * The BIDX chunk is ignored if the BDAT chunk is not present.
 110
 111   Bloom Filter Data (ID: {'B', 'D', 'A', 'T'}) [Optional]
 112     * It starts with header consisting of three unsigned 32-bit integers:
 113       - Version of the hash algorithm being used. We currently only support
 114         value 1 which corresponds to the 32-bit version of the murmur3 hash
 115         implemented exactly as described in
 116         https://en.wikipedia.org/wiki/MurmurHash#Algorithm and the double
 117         hashing technique using seed values 0x293ae76f and 0x7e646e2 as
 118         described in https://doi.org/10.1007/978-3-540-30494-4_26 "Bloom Filters
 119         in Probabilistic Verification"
 120       - The number of times a path is hashed and hence the number of bit positions
 121               that cumulatively determine whether a file is present in the commit.
 122       - The minimum number of bits 'b' per entry in the Bloom filter. If the filter
 123               contains 'n' entries, then the filter size is the minimum number of 64-bit
 124               words that contain n*b bits.
 125     * The rest of the chunk is the concatenation of all the computed Bloom
 126       filters for the commits in lexicographic order.
 127     * Note: Commits with no changes or more than 512 changes have Bloom filters
 128       of length zero.
 129     * The BDAT chunk is present if and only if BIDX is present.
 130
 131   Base Graphs List (ID: {'B', 'A', 'S', 'E'}) [Optional]
 132       This list of H-byte hashes describe a set of B commit-graph files that
 133       form a commit-graph chain. The graph position for the ith commit in this
 134       file's OID Lookup chunk is equal to i plus the number of commits in all
 135       base graphs.  If B is non-zero, this chunk must exist.
 136
 137 TRAILER:
 138
 139         H-byte HASH-checksum of all of the above.