1 # pack.py -- For dealing wih packed git objects.
2 # Copyright (C) 2007 James Westby <jw+debian@jameswestby.net>
3 # The code is loosely based on that in the sha1_file.c file from git itself,
4 # which is Copyright (C) Linus Torvalds, 2005 and distributed under the
7 # This program is free software; you can redistribute it and/or
8 # modify it under the terms of the GNU General Public License
9 # as published by the Free Software Foundation; version 2
12 # This program is distributed in the hope that it will be useful,
13 # but WITHOUT ANY WARRANTY; without even the implied warranty of
14 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 # GNU General Public License for more details.
17 # You should have received a copy of the GNU General Public License
18 # along with this program; if not, write to the Free Software
19 # Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston,
22 """Classes for dealing with packed git objects.
24 A pack is a compact representation of a bunch of objects, stored
25 using deltas where possible.
27 They have two parts, the pack file, which stores the data, and an index
28 that tells you where the data is.
30 To find an object you look in all of the index files 'til you find a
31 match for the object name. You then use the pointer got from this as
32 a pointer in to the corresponding packfile.
38 from objects import (ShaFile,
42 hex_to_sha = lambda hex: int(hex, 16)
44 def multi_ord(map, start, count):
46 for i in range(count):
47 value = value * 256 + ord(map[start+i])
50 max_size = 256 * 1024 * 1024
52 class PackIndex(object):
53 """An index in to a packfile.
55 Given a sha id of an object a pack index can tell you the location in the
56 packfile of that object if it has it.
58 To do the looup it opens the file, and indexes first 256 4 byte groups
59 with the first byte of the sha id. The value in the four byte group indexed
60 is the end of the group that shares the same starting byte. Subtract one
61 from the starting byte and index again to find the start of the group.
62 The values are sorted by sha id within the group, so do the math to find
63 the start and end offset and then bisect in to find if the value is present.
66 header_record_size = 4
67 header_size = 256 * header_record_size
70 record_size = sha_bytes + index_size
72 def __init__(self, filename):
73 """Create a pack index object.
75 Provide it with the name of the index file to consider, and it will map
78 self._filename = filename
79 assert os.path.exists(filename), "%s is not a pack index" % filename
80 # Take the size now, so it can be checked each time we map the file to
81 # ensure that it hasn't changed.
82 self._size = os.path.getsize(filename)
83 assert self._size > self.header_size, "%s is too small to be a packfile" % \
85 assert self._size < max_size, "%s is larger than 256 meg, and it " \
86 "might not be a good idea to mmap it. If you want to go ahead " \
87 "delete this check, or get python to support mmap offsets so that " \
88 "I can map the files sensibly"
90 def object_index(self, sha):
91 """Return the index in to the corresponding packfile for the object.
93 Given the name of an object it will return the offset that object lives
94 at within the corresponding pack file. If the pack file doesn't have the
95 object then None will be returned.
97 size = os.path.getsize(self._filename)
98 assert size == self._size, "Pack index %s has changed size, I don't " \
99 "like that" % self._filename
100 f = open(self._filename, 'rb')
102 map = mmap.mmap(f.fileno(), size, access=mmap.ACCESS_READ)
103 return self._object_index(map, sha)
107 def _object_index(self, map, hexsha):
108 """See object_index"""
109 first_byte = hex_to_sha(hexsha[:2])
110 header_offset = self.header_record_size * first_byte
111 start = multi_ord(map, header_offset-self.header_record_size, self.header_record_size)
112 end = multi_ord(map, header_offset, self.header_record_size)
113 sha = hex_to_sha(hexsha)
116 offset = self.header_size + (i * self.record_size)
117 file_sha = multi_ord(map, offset + self.index_size, self.sha_bytes)
119 return multi_ord(map, offset, self.index_size)
127 class PackData(object):
128 """The data contained in a packfile.
130 Pack files can be accessed both sequentially for exploding a pack, and
131 directly with the help of an index to retrieve a specific object.
133 The objects within are either complete or a delta aginst another.
135 The header is variable length. If the MSB of each byte is set then it
136 indicates that the subsequent byte is still part of the header.
137 For the first byte the next MS bits are the type, which tells you the type
138 of object, and whether it is a delta. The LS byte is the lowest bits of the
139 size. For each subsequent byte the LS 7 bits are the next MS bits of the
140 size, i.e. the last byte of the header contains the MS bits of the size.
142 For the complete objects the data is stored as zlib deflated data.
143 The size in the header is the uncompressed object size, so to uncompress
144 you need to just keep feeding data to zlib until you get an object back,
145 or it errors on bad data. This is done here by just giving the complete
146 buffer from the start of the deflated object on. This is bad, but until I
147 get mmap sorted out it will have to do.
149 Currently there are no integrity checks done. Also no attempt is made to try
150 and detect the delta case, or a request for an object at the wrong position.
151 It will all just throw a zlib or KeyError.
154 def __init__(self, filename):
155 """Create a PackData object that represents the pack in the given filename.
157 The file must exist and stay readable until the object is disposed of. It
158 must also stay the same size. It will be mapped whenever needed.
160 Currently there is a restriction on the size of the pack as the python
161 mmap implementation is flawed.
163 self._filename = filename
164 assert os.path.exists(filename), "%s is not a packfile" % filename
165 self._size = os.path.getsize(filename)
166 assert self._size < max_size, "%s is larger than 256 meg, and it " \
167 "might not be a good idea to mmap it. If you want to go ahead " \
168 "delete this check, or get python to support mmap offsets so that " \
169 "I can map the files sensibly"
171 def get_object_at(self, offset):
172 """Given an offset in to the packfile return the object that is there.
174 Using the associated index the location of an object can be looked up, and
175 then the packfile can be asked directly for that object using this
178 Currently only non-delta objects are supported.
180 size = os.path.getsize(self._filename)
181 assert size == self._size, "Pack data %s has changed size, I don't " \
182 "like that" % self._filename
183 f = open(self._filename, 'rb')
185 map = mmap.mmap(f.fileno(), size, access=mmap.ACCESS_READ)
186 return self._get_object_at(map, offset)
190 def _get_object_at(self, map, offset):
191 first_byte = ord(map[offset])
192 sign_extend = first_byte & 0x80
193 type = (first_byte >> 4) & 0x07
194 size = first_byte & 0x0f
196 while sign_extend > 0:
197 byte = ord(map[offset+cur_offset+1])
198 sign_extend = byte & 0x80
199 size_part = byte & 0x7f
200 size += size_part << ((cur_offset * 7) + 4)
202 raw_base = offset+cur_offset+1
203 # The size is the inflated size, so we have no idea what the deflated size
204 # is, so for now give it as much as we have. It should really iterate
205 # feeding it more data if it doesn't decompress, but as we have the whole
206 # thing then just use it.
208 uncomp = _decompress(raw)
209 obj = ShaFile.from_raw_string(type, uncomp)