4 == Checksums and object IDs
6 In a repository using the traditional SHA-1, pack checksums, index checksums,
7 and object IDs (object names) mentioned below are all computed using SHA-1.
8 Similarly, in SHA-256 repositories, these values are computed using SHA-256.
10 == pack-*.pack files have the following format:
12 - A header appears at the beginning and consists of the following:
15 The signature is: {'P', 'A', 'C', 'K'}
17 4-byte version number (network byte order):
18 Git currently accepts version number 2 or 3 but
19 generates version 2 only.
21 4-byte number of objects contained in the pack (network byte order)
23 Observation: we cannot have more than 4G versions ;-) and
24 more than 4G objects in a pack.
26 - The header is followed by number of object entries, each of
27 which looks like this:
29 (undeltified representation)
30 n-byte type and length (3-bit type, (n-1)*7+4-bit length)
33 (deltified representation)
34 n-byte type and length (3-bit type, (n-1)*7+4-bit length)
35 base object name if OBJ_REF_DELTA or a negative relative
36 offset from the delta object's position in the pack if this
37 is an OBJ_OFS_DELTA object
40 Observation: length of each object is encoded in a variable
41 length format and is not constrained to 32-bit or anything.
43 - The trailer records a pack checksum of all of the above.
47 Valid object types are:
56 Type 5 is reserved for future expansion. Type 0 is invalid.
58 === Deltified representation
60 Conceptually there are only four object types: commit, tree, tag and
61 blob. However to save space, an object could be stored as a "delta" of
62 another "base" object. These representations are assigned new types
63 ofs-delta and ref-delta, which is only valid in a pack file.
65 Both ofs-delta and ref-delta store the "delta" to be applied to
66 another object (called 'base object') to reconstruct the object. The
67 difference between them is, ref-delta directly encodes base object
68 name. If the base object is in the same pack, ofs-delta encodes
69 the offset of the base object in the pack instead.
71 The base object could also be deltified if it's in the same pack.
72 Ref-delta can also refer to an object outside the pack (i.e. the
73 so-called "thin pack"). When stored on disk however, the pack should
74 be self contained to avoid cyclic dependency.
76 The delta data is a sequence of instructions to reconstruct an object
77 from the base object. If the base object is deltified, it must be
78 converted to canonical form first. Each instruction appends more and
79 more data to the target object until it's complete. There are two
80 supported instructions so far: one for copy a byte range from the
81 source object and one for inserting new data embedded in the
84 Each instruction has variable length. Instruction type is determined
85 by the seventh bit of the first octet. The following diagrams follow
86 the convention in RFC 1951 (Deflate compressed data format).
88 ==== Instruction to copy from base object
90 +----------+---------+---------+---------+---------+-------+-------+-------+
91 | 1xxxxxxx | offset1 | offset2 | offset3 | offset4 | size1 | size2 | size3 |
92 +----------+---------+---------+---------+---------+-------+-------+-------+
94 This is the instruction format to copy a byte range from the source
95 object. It encodes the offset to copy from and the number of bytes to
96 copy. Offset and size are in little-endian order.
98 All offset and size bytes are optional. This is to reduce the
99 instruction size when encoding small offsets or sizes. The first seven
100 bits in the first octet determines which of the next seven octets is
101 present. If bit zero is set, offset1 is present. If bit one is set
102 offset2 is present and so on.
104 Note that a more compact instruction does not change offset and size
105 encoding. For example, if only offset2 is omitted like below, offset3
106 still contains bits 16-23. It does not become offset2 and contains
107 bits 8-15 even if it's right next to offset1.
109 +----------+---------+---------+
110 | 10000101 | offset1 | offset3 |
111 +----------+---------+---------+
113 In its most compact form, this instruction only takes up one byte
114 (0x80) with both offset and size omitted, which will have default
115 values zero. There is another exception: size zero is automatically
116 converted to 0x10000.
118 ==== Instruction to add new data
120 +----------+============+
122 +----------+============+
124 This is the instruction to construct target object without the base
125 object. The following data is appended to the target object. The first
126 seven bits of the first octet determines the size of data in
127 bytes. The size must be non-zero.
129 ==== Reserved instruction
131 +----------+============
133 +----------+============
135 This is the instruction reserved for future expansion.
137 == Original (version 1) pack-*.idx files have the following format:
139 - The header consists of 256 4-byte network byte order
140 integers. N-th entry of this table records the number of
141 objects in the corresponding pack, the first byte of whose
142 object name is less than or equal to N. This is called the
143 'first-level fan-out' table.
145 - The header is followed by sorted 24-byte entries, one entry
146 per object in the pack. Each entry is:
148 4-byte network byte order integer, recording where the
149 object is stored in the packfile as the offset from the
152 one object name of the appropriate size.
154 - The file is concluded with a trailer:
156 A copy of the pack checksum at the end of the corresponding
159 Index checksum of all of the above.
163 -- +--------------------------------+
164 fanout | fanout[0] = 2 (for example) |-.
165 table +--------------------------------+ |
167 +--------------------------------+ |
169 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
170 | fanout[255] = total objects |---.
171 -- +--------------------------------+ | |
173 index | object name 00XXXXXXXXXXXXXXXX | | |
174 table +--------------------------------+ | |
176 | object name 00XXXXXXXXXXXXXXXX | | |
177 +--------------------------------+<+ |
179 | | object name 01XXXXXXXXXXXXXXXX | |
180 | +--------------------------------+ |
182 | | object name 01XXXXXXXXXXXXXXXX | |
183 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
185 | | object name FFXXXXXXXXXXXXXXXX | |
186 --| +--------------------------------+<--+
187 trailer | | packfile checksum |
188 | +--------------------------------+
189 | | idxfile checksum |
190 | +--------------------------------+
195 packed object header:
196 1-byte size extension bit (MSB)
199 n-byte sizeN (as long as MSB is set, each 7-bit)
200 size0..sizeN form 4+7+7+..+7 bit integer, size0
201 is the least significant part, and sizeN is the
202 most significant part.
204 If it is not DELTA, then deflated bytes (the size above
205 is the size before compression).
206 If it is REF_DELTA, then
207 base object name (the size above is the
208 size of the delta data that follows).
209 delta data, deflated.
210 If it is OFS_DELTA, then
211 n-byte offset (see below) interpreted as a negative
212 offset from the type-byte of the header of the
213 ofs-delta entry (the size above is the size of
214 the delta data that follows).
215 delta data, deflated.
218 n bytes with MSB set in all but the last one.
219 The offset is then the number constructed by
220 concatenating the lower 7 bit of each byte, and
221 for n >= 2 adding 2^7 + 2^14 + ... + 2^(7*(n-1))
226 == Version 2 pack-*.idx files support packs larger than 4 GiB, and
227 have some other reorganizations. They have the format:
229 - A 4-byte magic number '\377tOc' which is an unreasonable
232 - A 4-byte version number (= 2)
234 - A 256-entry fan-out table just like v1.
236 - A table of sorted object names. These are packed together
237 without offset values to reduce the cache footprint of the
238 binary search for a specific object name.
240 - A table of 4-byte CRC32 values of the packed object data.
241 This is new in v2 so compressed data can be copied directly
242 from pack to pack during repacking without undetected
245 - A table of 4-byte offset values (in network byte order).
246 These are usually 31-bit pack file offsets, but large
247 offsets are encoded as an index into the next table with
250 - A table of 8-byte offset entries (empty for pack files less
251 than 2 GiB). Pack files are organized with heavily used
252 objects toward the front, so most object references should
253 not need to refer to this table.
255 - The same trailer as a v1 pack file:
257 A copy of the pack checksum at the end of
258 corresponding packfile.
260 Index checksum of all of the above.
262 == multi-pack-index (MIDX) files have the following format:
264 The multi-pack-index files refer to multiple pack-files and loose objects.
266 In order to allow extensions that add extra data to the MIDX, we organize
267 the body into "chunks" and provide a lookup table at the beginning of the
268 body. The header includes certain length values, such as the number of packs,
269 the number of base MIDX files, hash lengths and types.
271 All 4-byte numbers are in network order.
276 The signature is: {'M', 'I', 'D', 'X'}
278 1-byte version number:
279 Git only writes or recognizes version 1.
281 1-byte Object Id Version
282 We infer the length of object IDs (OIDs) from this value:
285 If the hash type does not match the repository's hash algorithm,
286 the multi-pack-index file should be ignored with a warning
287 presented to the user.
289 1-byte number of "chunks"
291 1-byte number of base multi-pack-index files:
292 This value is currently always zero.
294 4-byte number of pack files
298 (C + 1) * 12 bytes providing the chunk offsets:
299 First 4 bytes describe chunk id. Value 0 is a terminating label.
300 Other 8 bytes provide offset in current file for chunk to start.
301 (Chunks are provided in file-order, so you can infer the length
302 using the next chunk position if necessary.)
304 The CHUNK LOOKUP matches the table of contents from
305 link:technical/chunk-format.html[the chunk-based file format].
307 The remaining data in the body is described one chunk at a time, and
308 these chunks may be given in any order. Chunks are required unless
313 Packfile Names (ID: {'P', 'N', 'A', 'M'})
314 Stores the packfile names as concatenated, null-terminated strings.
315 Packfiles must be listed in lexicographic order for fast lookups by
316 name. This is the only chunk not guaranteed to be a multiple of four
317 bytes in length, so should be the last chunk for alignment reasons.
319 OID Fanout (ID: {'O', 'I', 'D', 'F'})
320 The ith entry, F[i], stores the number of OIDs with first
321 byte at most i. Thus F[255] stores the total
324 OID Lookup (ID: {'O', 'I', 'D', 'L'})
325 The OIDs for all objects in the MIDX are stored in lexicographic
328 Object Offsets (ID: {'O', 'O', 'F', 'F'})
329 Stores two 4-byte values for every object.
330 1: The pack-int-id for the pack storing this object.
331 2: The offset within the pack.
332 If all offsets are less than 2^32, then the large offset chunk
333 will not exist and offsets are stored as in IDX v1.
334 If there is at least one offset value larger than 2^32-1, then
335 the large offset chunk must exist, and offsets larger than
336 2^31-1 must be stored in it instead. If the large offset chunk
337 exists and the 31st bit is on, then removing that bit reveals
338 the row in the large offsets containing the 8-byte offset of
341 [Optional] Object Large Offsets (ID: {'L', 'O', 'F', 'F'})
342 8-byte offsets into large packfiles.
346 Index checksum of the above contents.