1 """Git interaction library.
2 bup repositories are in Git format. This library allows us to
3 interact with the Git data structures.
6 from __future__ import absolute_import, print_function
7 import os, sys, zlib, subprocess, struct, stat, re, tempfile, glob
8 from array import array
9 from binascii import hexlify, unhexlify
10 from collections import namedtuple
11 from itertools import islice
13 from bup import _helpers, hashsplit, path, midx, bloom, xstat
14 from bup.compat import (buffer,
15 byte_int, bytes_from_byte, bytes_from_uint,
22 from bup.io import path_msg
23 from bup.helpers import (Sha1, add_error, chunkyreader, debug1, debug2,
30 mmap_read, mmap_readwrite,
32 progress, qprogress, stat_if_exists,
38 repodir = None # The default repository, once initialized
40 _typemap = {b'blob': 3, b'tree': 2, b'commit': 1, b'tag': 4}
41 _typermap = {v: k for k, v in items(_typemap)}
48 class GitError(Exception):
52 def _gitenv(repo_dir=None):
55 return merge_dict(environ, {b'GIT_DIR': os.path.abspath(repo_dir)})
57 def _git_wait(cmd, p):
60 raise GitError('%r returned %d' % (cmd, rv))
62 def _git_exo(cmd, **kwargs):
63 kwargs['check'] = False
64 result = exo(cmd, **kwargs)
66 if proc.returncode != 0:
67 raise GitError('%r returned %d' % (cmd, proc.returncode))
70 def git_config_get(option, repo_dir=None, opttype=None, cfg_file=None):
71 assert not (repo_dir and cfg_file), "repo_dir and cfg_file cannot both be used"
72 cmd = [b'git', b'config', b'--null']
74 cmd.extend([b'--file', cfg_file])
76 cmd.extend([b'--int'])
77 elif opttype == 'bool':
78 cmd.extend([b'--bool'])
80 assert opttype is None
81 cmd.extend([b'--get', option])
84 env = _gitenv(repo_dir=repo_dir)
85 p = subprocess.Popen(cmd, stdout=subprocess.PIPE, env=env,
87 # with --null, git writes out a trailing \0 after the value
88 r = p.stdout.read()[:-1]
93 elif opttype == 'bool':
94 # git converts to 'true' or 'false'
98 raise GitError('%r returned %d' % (cmd, rc))
102 def parse_tz_offset(s):
103 """UTC offset in seconds."""
104 tz_off = (int(s[1:3]) * 60 * 60) + (int(s[3:5]) * 60)
105 if bytes_from_byte(s[0]) == b'-':
109 def parse_commit_gpgsig(sig):
110 """Return the original signature bytes.
112 i.e. with the "gpgsig " header and the leading space character on
113 each continuation line removed.
118 assert sig.startswith(b'gpgsig ')
120 return sig.replace(b'\n ', b'\n')
122 # FIXME: derived from http://git.rsbx.net/Documents/Git_Data_Formats.txt
123 # Make sure that's authoritative.
126 # https://github.com/git/git/blob/master/Documentation/technical/signature-format.txt
127 # The continuation lines have only one leading space.
129 _start_end_char = br'[^ .,:;<>"\'\0\n]'
130 _content_char = br'[^\0\n<>]'
131 _safe_str_rx = br'(?:%s{1,2}|(?:%s%s*%s))' \
133 _start_end_char, _content_char, _start_end_char)
134 _tz_rx = br'[-+]\d\d[0-5]\d'
135 _parent_rx = br'(?:parent [abcdefABCDEF0123456789]{40}\n)'
136 # Assumes every following line starting with a space is part of the
137 # mergetag. Is there a formal commit blob spec?
138 _mergetag_rx = br'(?:\nmergetag object [abcdefABCDEF0123456789]{40}(?:\n [^\0\n]*)*)'
139 _commit_rx = re.compile(br'''tree (?P<tree>[abcdefABCDEF0123456789]{40})
140 (?P<parents>%s*)author (?P<author_name>%s) <(?P<author_mail>%s)> (?P<asec>\d+) (?P<atz>%s)
141 committer (?P<committer_name>%s) <(?P<committer_mail>%s)> (?P<csec>\d+) (?P<ctz>%s)(?P<mergetag>%s?)
142 (?P<gpgsig>gpgsig .*\n(?: .*\n)*)?
143 (?P<message>(?:.|\n)*)''' % (_parent_rx,
144 _safe_str_rx, _safe_str_rx, _tz_rx,
145 _safe_str_rx, _safe_str_rx, _tz_rx,
147 _parent_hash_rx = re.compile(br'\s*parent ([abcdefABCDEF0123456789]{40})\s*')
149 # Note that the author_sec and committer_sec values are (UTC) epoch
150 # seconds, and for now the mergetag is not included.
151 CommitInfo = namedtuple('CommitInfo', ['tree', 'parents',
152 'author_name', 'author_mail',
153 'author_sec', 'author_offset',
154 'committer_name', 'committer_mail',
155 'committer_sec', 'committer_offset',
159 def parse_commit(content):
160 commit_match = re.match(_commit_rx, content)
162 raise Exception('cannot parse commit %r' % content)
163 matches = commit_match.groupdict()
164 return CommitInfo(tree=matches['tree'],
165 parents=re.findall(_parent_hash_rx, matches['parents']),
166 author_name=matches['author_name'],
167 author_mail=matches['author_mail'],
168 author_sec=int(matches['asec']),
169 author_offset=parse_tz_offset(matches['atz']),
170 committer_name=matches['committer_name'],
171 committer_mail=matches['committer_mail'],
172 committer_sec=int(matches['csec']),
173 committer_offset=parse_tz_offset(matches['ctz']),
174 gpgsig=parse_commit_gpgsig(matches['gpgsig']),
175 message=matches['message'])
178 def get_cat_data(cat_iterator, expected_type):
179 _, kind, _ = next(cat_iterator)
180 if kind != expected_type:
181 raise Exception('expected %r, saw %r' % (expected_type, kind))
182 return b''.join(cat_iterator)
184 def get_commit_items(id, cp):
185 return parse_commit(get_cat_data(cp.get(id), b'commit'))
187 def _local_git_date_str(epoch_sec):
188 return b'%d %s' % (epoch_sec, utc_offset_str(epoch_sec))
191 def _git_date_str(epoch_sec, tz_offset_sec):
192 offs = tz_offset_sec // 60
193 return b'%d %s%02d%02d' \
195 b'+' if offs >= 0 else b'-',
200 def repo(sub = b'', repo_dir=None):
201 """Get the path to the git repository or one of its subdirectories."""
202 repo_dir = repo_dir or repodir
204 raise GitError('You should call check_repo_or_die()')
206 # If there's a .git subdirectory, then the actual repo is in there.
207 gd = os.path.join(repo_dir, b'.git')
208 if os.path.exists(gd):
211 return os.path.join(repo_dir, sub)
215 re.compile(br'([^0-9a-z]|\b)([0-9a-z]{7})[0-9a-z]{33}([^0-9a-z]|\b)')
218 return _shorten_hash_rx.sub(br'\1\2*\3', s)
222 full = os.path.abspath(path)
223 fullrepo = os.path.abspath(repo(b''))
224 if not fullrepo.endswith(b'/'):
226 if full.startswith(fullrepo):
227 path = full[len(fullrepo):]
228 if path.startswith(b'index-cache/'):
229 path = path[len(b'index-cache/'):]
230 return shorten_hash(path)
233 def auto_midx(objdir):
234 args = [path.exe(), b'midx', b'--auto', b'--dir', objdir]
236 rv = subprocess.call(args, stdout=open(os.devnull, 'w'))
238 # make sure 'args' gets printed to help with debugging
239 add_error('%r: exception: %s' % (args, e))
242 add_error('%r: returned %d' % (args, rv))
244 args = [path.exe(), b'bloom', b'--dir', objdir]
246 rv = subprocess.call(args, stdout=open(os.devnull, 'w'))
248 # make sure 'args' gets printed to help with debugging
249 add_error('%r: exception: %s' % (args, e))
252 add_error('%r: returned %d' % (args, rv))
255 def mangle_name(name, mode, gitmode):
256 """Mangle a file name to present an abstract name for segmented files.
257 Mangled file names will have the ".bup" extension added to them. If a
258 file's name already ends with ".bup", a ".bupl" extension is added to
259 disambiguate normal files from segmented ones.
261 if stat.S_ISREG(mode) and not stat.S_ISREG(gitmode):
262 assert(stat.S_ISDIR(gitmode))
263 return name + b'.bup'
264 elif name.endswith(b'.bup') or name[:-1].endswith(b'.bup'):
265 return name + b'.bupl'
270 (BUP_NORMAL, BUP_CHUNKED) = (0,1)
271 def demangle_name(name, mode):
272 """Remove name mangling from a file name, if necessary.
274 The return value is a tuple (demangled_filename,mode), where mode is one of
277 * BUP_NORMAL : files that should be read as-is from the repository
278 * BUP_CHUNKED : files that were chunked and need to be reassembled
280 For more information on the name mangling algorithm, see mangle_name()
282 if name.endswith(b'.bupl'):
283 return (name[:-5], BUP_NORMAL)
284 elif name.endswith(b'.bup'):
285 return (name[:-4], BUP_CHUNKED)
286 elif name.endswith(b'.bupm'):
288 BUP_CHUNKED if stat.S_ISDIR(mode) else BUP_NORMAL)
289 return (name, BUP_NORMAL)
292 def calc_hash(type, content):
293 """Calculate some content's hash in the Git fashion."""
294 header = b'%s %d\0' % (type, len(content))
300 def shalist_item_sort_key(ent):
301 (mode, name, id) = ent
302 assert(mode+0 == mode)
303 if stat.S_ISDIR(mode):
309 def tree_encode(shalist):
310 """Generate a git tree object from (mode,name,hash) tuples."""
311 shalist = sorted(shalist, key = shalist_item_sort_key)
313 for (mode,name,bin) in shalist:
315 assert(mode+0 == mode)
317 assert(len(bin) == 20)
318 s = b'%o %s\0%s' % (mode,name,bin)
319 assert s[0] != b'0' # 0-padded octal is not acceptable in a git tree
324 def tree_decode(buf):
325 """Generate a list of (mode,name,hash) from the git tree object in buf."""
327 while ofs < len(buf):
328 z = buf.find(b'\0', ofs)
330 spl = buf[ofs:z].split(b' ', 1)
331 assert(len(spl) == 2)
333 sha = buf[z+1:z+1+20]
335 yield (int(mode, 8), name, sha)
338 def _encode_packobj(type, content, compression_level=1):
339 if compression_level not in (0, 1, 2, 3, 4, 5, 6, 7, 8, 9):
340 raise ValueError('invalid compression level %s' % compression_level)
343 szbits = (sz & 0x0f) | (_typemap[type]<<4)
346 if sz: szbits |= 0x80
347 szout += bytes_from_uint(szbits)
352 z = zlib.compressobj(compression_level)
354 yield z.compress(content)
358 def _decode_packobj(buf):
361 type = _typermap[(c & 0x70) >> 4]
368 sz |= (c & 0x7f) << shift
372 return (type, zlib.decompress(buf[i+1:]))
379 def find_offset(self, hash):
380 """Get the offset of an object inside the index file."""
381 idx = self._idx_from_hash(hash)
383 return self._ofs_from_idx(idx)
386 def exists(self, hash, want_source=False):
387 """Return nonempty if the object exists in this index."""
388 if hash and (self._idx_from_hash(hash) != None):
389 return want_source and os.path.basename(self.name) or True
392 def _idx_from_hash(self, hash):
393 global _total_searches, _total_steps
395 assert(len(hash) == 20)
396 b1 = byte_int(hash[0])
397 start = self.fanout[b1-1] # range -1..254
398 end = self.fanout[b1] # range 0..255
400 _total_steps += 1 # lookup table is a step
403 mid = start + (end - start) // 2
404 v = self._idx_to_hash(mid)
414 class PackIdxV1(PackIdx):
415 """Object representation of a Git pack index (version 1) file."""
416 def __init__(self, filename, f):
419 self.idxnames = [self.name]
420 self.map = mmap_read(f)
421 # Min size for 'L' is 4, which is sufficient for struct's '!I'
422 self.fanout = array('L', struct.unpack('!256I', self.map))
423 self.fanout.append(0) # entry "-1"
424 self.nsha = self.fanout[255]
425 self.sha_ofs = 256 * 4
426 # Avoid slicing shatable for individual hashes (very high overhead)
427 self.shatable = buffer(self.map, self.sha_ofs, self.nsha * 24)
432 def __exit__(self, type, value, traceback):
433 with pending_raise(value, rethrow=False):
437 return int(self.nsha) # int() from long for python 2
439 def _ofs_from_idx(self, idx):
440 if idx >= self.nsha or idx < 0:
441 raise IndexError('invalid pack index index %d' % idx)
442 ofs = self.sha_ofs + idx * 24
443 return struct.unpack_from('!I', self.map, offset=ofs)[0]
445 def _idx_to_hash(self, idx):
446 if idx >= self.nsha or idx < 0:
447 raise IndexError('invalid pack index index %d' % idx)
448 ofs = self.sha_ofs + idx * 24 + 4
449 return self.map[ofs : ofs + 20]
452 start = self.sha_ofs + 4
453 for ofs in range(start, start + 24 * self.nsha, 24):
454 yield self.map[ofs : ofs + 20]
458 if self.map is not None:
467 class PackIdxV2(PackIdx):
468 """Object representation of a Git pack index (version 2) file."""
469 def __init__(self, filename, f):
472 self.idxnames = [self.name]
473 self.map = mmap_read(f)
474 assert self.map[0:8] == b'\377tOc\0\0\0\2'
475 # Min size for 'L' is 4, which is sufficient for struct's '!I'
476 self.fanout = array('L', struct.unpack_from('!256I', self.map, offset=8))
477 self.fanout.append(0)
478 self.nsha = self.fanout[255]
479 self.sha_ofs = 8 + 256*4
480 self.ofstable_ofs = self.sha_ofs + self.nsha * 20 + self.nsha * 4
481 self.ofs64table_ofs = self.ofstable_ofs + self.nsha * 4
482 # Avoid slicing this for individual hashes (very high overhead)
483 self.shatable = buffer(self.map, self.sha_ofs, self.nsha*20)
488 def __exit__(self, type, value, traceback):
489 with pending_raise(value, rethrow=False):
493 return int(self.nsha) # int() from long for python 2
495 def _ofs_from_idx(self, idx):
496 if idx >= self.nsha or idx < 0:
497 raise IndexError('invalid pack index index %d' % idx)
498 ofs_ofs = self.ofstable_ofs + idx * 4
499 ofs = struct.unpack_from('!I', self.map, offset=ofs_ofs)[0]
501 idx64 = ofs & 0x7fffffff
502 ofs64_ofs = self.ofs64table_ofs + idx64 * 8
503 ofs = struct.unpack_from('!Q', self.map, offset=ofs64_ofs)[0]
506 def _idx_to_hash(self, idx):
507 if idx >= self.nsha or idx < 0:
508 raise IndexError('invalid pack index index %d' % idx)
509 ofs = self.sha_ofs + idx * 20
510 return self.map[ofs : ofs + 20]
514 for ofs in range(start, start + 20 * self.nsha, 20):
515 yield self.map[ofs : ofs + 20]
519 if self.map is not None:
530 def __init__(self, dir, ignore_midx=False):
532 # Q: was this also intended to prevent opening multiple repos?
533 assert(_mpi_count == 0) # these things suck tons of VM; don't waste it
539 self.do_bloom = False
541 self.ignore_midx = ignore_midx
544 except BaseException as ex:
545 with pending_raise(ex):
551 assert _mpi_count == 0
554 assert _mpi_count == 0
556 self.bloom, bloom = None, self.bloom
557 self.packs, packs = None, self.packs
559 with ExitStack() as stack:
561 stack.enter_context(pack)
568 def __exit__(self, type, value, traceback):
569 with pending_raise(value, rethrow=False):
576 return iter(idxmerge(self.packs))
579 return sum(len(pack) for pack in self.packs)
581 def exists(self, hash, want_source=False):
582 """Return nonempty if the object exists in the index files."""
583 global _total_searches
585 if hash in self.also:
587 if self.do_bloom and self.bloom:
588 if self.bloom.exists(hash):
589 self.do_bloom = False
591 _total_searches -= 1 # was counted by bloom
593 for i in range(len(self.packs)):
595 _total_searches -= 1 # will be incremented by sub-pack
596 ix = p.exists(hash, want_source=want_source)
598 # reorder so most recently used packs are searched first
599 self.packs = [p] + self.packs[:i] + self.packs[i+1:]
604 def refresh(self, skip_midx = False):
605 """Refresh the index list.
606 This method verifies if .midx files were superseded (e.g. all of its
607 contents are in another, bigger .midx file) and removes the superseded
610 If skip_midx is True, all work on .midx files will be skipped and .midx
611 files will be removed from the list.
613 The instance variable 'ignore_midx' can force this function to
614 always act as if skip_midx was True.
616 if self.bloom is not None:
618 self.bloom = None # Always reopen the bloom as it may have been relaced
619 self.do_bloom = False
620 skip_midx = skip_midx or self.ignore_midx
621 d = dict((p.name, p) for p in self.packs
622 if not skip_midx or not isinstance(p, midx.PackMidx))
623 if os.path.exists(self.dir):
626 midxes = set(glob.glob(os.path.join(self.dir, b'*.midx')))
627 # remove any *.midx files from our list that no longer exist
628 for ix in list(d.values()):
629 if not isinstance(ix, midx.PackMidx):
631 if ix.name in midxes:
636 self.packs.remove(ix)
637 for ix in self.packs:
638 if isinstance(ix, midx.PackMidx):
639 for name in ix.idxnames:
640 d[os.path.join(self.dir, name)] = ix
643 mx = midx.PackMidx(full)
644 (mxd, mxf) = os.path.split(mx.name)
646 for n in mx.idxnames:
647 if not os.path.exists(os.path.join(mxd, n)):
648 log(('warning: index %s missing\n'
650 % (path_msg(n), path_msg(mxf)))
657 midxl.sort(key=lambda ix:
658 (-len(ix), -xstat.stat(ix.name).st_mtime))
661 for sub in ix.idxnames:
662 found = d.get(os.path.join(self.dir, sub))
663 if not found or isinstance(found, PackIdx):
664 # doesn't exist, or exists but not in a midx
669 for name in ix.idxnames:
670 d[os.path.join(self.dir, name)] = ix
671 elif not ix.force_keep:
672 debug1('midx: removing redundant: %s\n'
673 % path_msg(os.path.basename(ix.name)))
676 for full in glob.glob(os.path.join(self.dir, b'*.idx')):
680 except GitError as e:
684 bfull = os.path.join(self.dir, b'bup.bloom')
685 self.packs = list(set(d.values()))
686 self.packs.sort(reverse=True, key=lambda x: len(x))
687 if self.bloom is None and os.path.exists(bfull):
688 self.bloom = bloom.ShaBloom(bfull)
690 if self.bloom and self.bloom.valid() and len(self.bloom) >= len(self):
694 self.bloom, bloom_tmp = None, self.bloom
696 except BaseException as ex:
697 with pending_raise(ex):
701 debug1('PackIdxList: using %d index%s.\n'
702 % (len(self.packs), len(self.packs)!=1 and 'es' or ''))
705 """Insert an additional object in the list."""
709 def open_idx(filename):
710 if filename.endswith(b'.idx'):
711 f = open(filename, 'rb')
713 if header[0:4] == b'\377tOc':
714 version = struct.unpack('!I', header[4:8])[0]
716 return PackIdxV2(filename, f)
718 raise GitError('%s: expected idx file version 2, got %d'
719 % (path_msg(filename), version))
720 elif len(header) == 8 and header[0:4] < b'\377tOc':
721 return PackIdxV1(filename, f)
723 raise GitError('%s: unrecognized idx file header'
724 % path_msg(filename))
725 elif filename.endswith(b'.midx'):
726 return midx.PackMidx(filename)
728 raise GitError('idx filenames must end with .idx or .midx')
731 def idxmerge(idxlist, final_progress=True):
732 """Generate a list of all the objects reachable in a PackIdxList."""
733 def pfunc(count, total):
734 qprogress('Reading indexes: %.2f%% (%d/%d)\r'
735 % (count*100.0/total, count, total))
736 def pfinal(count, total):
738 progress('Reading indexes: %.2f%% (%d/%d), done.\n'
739 % (100, total, total))
740 return merge_iter(idxlist, 10024, pfunc, pfinal)
743 def create_commit_blob(tree, parent,
744 author, adate_sec, adate_tz,
745 committer, cdate_sec, cdate_tz,
747 if adate_tz is not None:
748 adate_str = _git_date_str(adate_sec, adate_tz)
750 adate_str = _local_git_date_str(adate_sec)
751 if cdate_tz is not None:
752 cdate_str = _git_date_str(cdate_sec, cdate_tz)
754 cdate_str = _local_git_date_str(cdate_sec)
756 if tree: l.append(b'tree %s' % hexlify(tree))
757 if parent: l.append(b'parent %s' % hexlify(parent))
758 if author: l.append(b'author %s %s' % (author, adate_str))
759 if committer: l.append(b'committer %s %s' % (committer, cdate_str))
764 def _make_objcache():
765 return PackIdxList(repo(b'objects/pack'))
767 # bup-gc assumes that it can disable all PackWriter activities
768 # (bloom/midx/cache) via the constructor and close() arguments.
771 """Writes Git objects inside a pack file."""
772 def __init__(self, objcache_maker=_make_objcache, compression_level=1,
773 run_midx=True, on_pack_finish=None,
774 max_pack_size=None, max_pack_objects=None, repo_dir=None):
776 self.repo_dir = repo_dir or repo()
783 self.objcache_maker = objcache_maker
785 self.compression_level = compression_level
786 self.run_midx=run_midx
787 self.on_pack_finish = on_pack_finish
788 if not max_pack_size:
789 max_pack_size = git_config_get(b'pack.packSizeLimit',
790 repo_dir=self.repo_dir,
792 if not max_pack_size:
793 # larger packs slow down pruning
794 max_pack_size = 1000 * 1000 * 1000
795 self.max_pack_size = max_pack_size
796 # cache memory usage is about 83 bytes per object
797 self.max_pack_objects = max_pack_objects if max_pack_objects \
798 else max(1, self.max_pack_size // 5000)
803 def __exit__(self, type, value, traceback):
804 with pending_raise(value, rethrow=False):
809 objdir = dir = os.path.join(self.repo_dir, b'objects')
810 fd, name = tempfile.mkstemp(suffix=b'.pack', dir=objdir)
812 self.file = os.fdopen(fd, 'w+b')
817 self.parentfd = os.open(objdir, os.O_RDONLY)
823 assert name.endswith(b'.pack')
824 self.filename = name[:-5]
825 self.file.write(b'PACK\0\0\0\2\0\0\0\0')
826 self.idx = PackIdxV2Writer()
828 def _raw_write(self, datalist, sha):
831 # in case we get interrupted (eg. KeyboardInterrupt), it's best if
832 # the file never has a *partial* blob. So let's make sure it's
833 # all-or-nothing. (The blob shouldn't be very big anyway, thanks
834 # to our hashsplit algorithm.) f.write() does its own buffering,
835 # but that's okay because we'll flush it in _end().
836 oneblob = b''.join(datalist)
842 crc = zlib.crc32(oneblob) & 0xffffffff
843 self._update_idx(sha, crc, nw)
848 def _update_idx(self, sha, crc, size):
851 self.idx.add(sha, crc, self.file.tell() - size)
853 def _write(self, sha, type, content):
857 sha = calc_hash(type, content)
858 size, crc = self._raw_write(_encode_packobj(type, content,
859 self.compression_level),
861 if self.outbytes >= self.max_pack_size \
862 or self.count >= self.max_pack_objects:
866 def _require_objcache(self):
867 if self.objcache is None and self.objcache_maker:
868 self.objcache = self.objcache_maker()
869 if self.objcache is None:
871 "PackWriter not opened or can't check exists w/o objcache")
873 def exists(self, id, want_source=False):
874 """Return non-empty if an object is found in the object cache."""
875 self._require_objcache()
876 return self.objcache.exists(id, want_source=want_source)
878 def just_write(self, sha, type, content):
879 """Write an object to the pack file without checking for duplication."""
880 self._write(sha, type, content)
881 # If nothing else, gc doesn't have/want an objcache
882 if self.objcache is not None:
883 self.objcache.add(sha)
885 def maybe_write(self, type, content):
886 """Write an object to the pack file if not present and return its id."""
887 sha = calc_hash(type, content)
888 if not self.exists(sha):
889 self._require_objcache()
890 self.just_write(sha, type, content)
893 def new_blob(self, blob):
894 """Create a blob object in the pack with the supplied content."""
895 return self.maybe_write(b'blob', blob)
897 def new_tree(self, shalist):
898 """Create a tree object in the pack."""
899 content = tree_encode(shalist)
900 return self.maybe_write(b'tree', content)
902 def new_commit(self, tree, parent,
903 author, adate_sec, adate_tz,
904 committer, cdate_sec, cdate_tz,
906 """Create a commit object in the pack. The date_sec values must be
907 epoch-seconds, and if a tz is None, the local timezone is assumed."""
908 content = create_commit_blob(tree, parent,
909 author, adate_sec, adate_tz,
910 committer, cdate_sec, cdate_tz,
912 return self.maybe_write(b'commit', content)
914 def _end(self, run_midx=True, abort=False):
915 # Ignores run_midx during abort
918 self.file, f = None, self.file
919 self.idx, idx = None, self.idx
920 self.parentfd, pfd, = None, self.parentfd
923 with nullcontext_if_not(self.objcache), \
924 finalized(pfd, lambda x: x is not None and os.close(x)), \
928 os.unlink(self.filename + b'.pack')
931 # update object count
933 cp = struct.pack('!i', self.count)
937 # calculate the pack sha1sum
940 for b in chunkyreader(f):
942 packbin = sum.digest()
945 fdatasync(f.fileno())
948 idx.write(self.filename + b'.idx', packbin)
949 nameprefix = os.path.join(self.repo_dir,
950 b'objects/pack/pack-' + hexlify(packbin))
951 if os.path.exists(self.filename + b'.map'):
952 os.unlink(self.filename + b'.map')
953 os.rename(self.filename + b'.pack', nameprefix + b'.pack')
954 os.rename(self.filename + b'.idx', nameprefix + b'.idx')
957 auto_midx(os.path.join(self.repo_dir, b'objects/pack'))
958 if self.on_pack_finish:
959 self.on_pack_finish(nameprefix)
962 # Must be last -- some of the code above depends on it
966 """Remove the pack file from disk."""
968 self._end(abort=True)
970 def breakpoint(self):
971 """Clear byte and object counts and return the last processed id."""
972 id = self._end(self.run_midx)
973 self.outbytes = self.count = 0
976 def close(self, run_midx=True):
977 """Close the pack file and move it to its definitive path."""
979 return self._end(run_midx=run_midx)
985 class PackIdxV2Writer:
987 self.idx = list(list() for i in range(256))
990 def add(self, sha, crc, offs):
993 self.idx[byte_int(sha[0])].append((sha, crc, offs))
995 def write(self, filename, packbin):
997 for section in self.idx:
998 for entry in section:
999 if entry[2] >= 2**31:
1002 # Length: header + fan-out + shas-and-crcs + overflow-offsets
1003 index_len = 8 + (4 * 256) + (28 * self.count) + (8 * ofs64_count)
1005 idx_f = open(filename, 'w+b')
1007 idx_f.truncate(index_len)
1008 fdatasync(idx_f.fileno())
1009 idx_map = mmap_readwrite(idx_f, close=False)
1011 count = _helpers.write_idx(filename, idx_map, self.idx,
1013 assert(count == self.count)
1020 idx_f = open(filename, 'a+b')
1022 idx_f.write(packbin)
1025 b = idx_f.read(8 + 4*256)
1028 for b in chunkyreader(idx_f, 20 * self.count):
1031 for b in chunkyreader(idx_f):
1033 idx_f.write(idx_sum.digest())
1034 fdatasync(idx_f.fileno())
1039 def list_refs(patterns=None, repo_dir=None,
1040 limit_to_heads=False, limit_to_tags=False):
1041 """Yield (refname, hash) tuples for all repository refs unless
1042 patterns are specified. In that case, only include tuples for
1043 refs matching those patterns (cf. git-show-ref(1)). The limits
1044 restrict the result items to refs/heads or refs/tags. If both
1045 limits are specified, items from both sources will be included.
1048 argv = [b'git', b'show-ref']
1050 argv.append(b'--heads')
1052 argv.append(b'--tags')
1055 argv.extend(patterns)
1056 p = subprocess.Popen(argv, env=_gitenv(repo_dir), stdout=subprocess.PIPE,
1058 out = p.stdout.read().strip()
1059 rv = p.wait() # not fatal
1063 for d in out.split(b'\n'):
1064 sha, name = d.split(b' ', 1)
1065 yield name, unhexlify(sha)
1068 def read_ref(refname, repo_dir = None):
1069 """Get the commit id of the most recent commit made on a given ref."""
1070 refs = list_refs(patterns=[refname], repo_dir=repo_dir, limit_to_heads=True)
1071 l = tuple(islice(refs, 2))
1079 def rev_list_invocation(ref_or_refs, format=None):
1080 if isinstance(ref_or_refs, bytes):
1081 refs = (ref_or_refs,)
1084 argv = [b'git', b'rev-list']
1087 argv.append(b'--pretty=format:' + format)
1089 assert not ref.startswith(b'-')
1095 def rev_list(ref_or_refs, parse=None, format=None, repo_dir=None):
1096 """Yield information about commits as per "git rev-list". If a format
1097 is not provided, yield one hex hash at a time. If a format is
1098 provided, pass it to rev-list and call parse(git_stdout) for each
1099 commit with the stream positioned just after the rev-list "commit
1100 HASH" header line. When a format is provided yield (oidx,
1101 parse(git_stdout)) for each commit.
1104 assert bool(parse) == bool(format)
1105 p = subprocess.Popen(rev_list_invocation(ref_or_refs,
1107 env=_gitenv(repo_dir),
1108 stdout = subprocess.PIPE,
1111 for line in p.stdout:
1114 line = p.stdout.readline()
1117 if not s.startswith(b'commit '):
1118 raise Exception('unexpected line ' + repr(s))
1121 yield s, parse(p.stdout)
1122 line = p.stdout.readline()
1124 rv = p.wait() # not fatal
1126 raise GitError('git rev-list returned error %d' % rv)
1129 def rev_parse(committish, repo_dir=None):
1130 """Resolve the full hash for 'committish', if it exists.
1132 Should be roughly equivalent to 'git rev-parse'.
1134 Returns the hex value of the hash if it is found, None if 'committish' does
1135 not correspond to anything.
1137 head = read_ref(committish, repo_dir=repo_dir)
1139 debug2("resolved from ref: commit = %s\n" % hexlify(head))
1142 if len(committish) == 40:
1144 hash = unhexlify(committish)
1148 with PackIdxList(repo(b'objects/pack', repo_dir=repo_dir)) as pL:
1155 def update_ref(refname, newval, oldval, repo_dir=None):
1156 """Update a repository reference."""
1159 assert refname.startswith(b'refs/heads/') \
1160 or refname.startswith(b'refs/tags/')
1161 p = subprocess.Popen([b'git', b'update-ref', refname,
1162 hexlify(newval), hexlify(oldval)],
1163 env=_gitenv(repo_dir),
1165 _git_wait(b'git update-ref', p)
1168 def delete_ref(refname, oldvalue=None):
1169 """Delete a repository reference (see git update-ref(1))."""
1170 assert refname.startswith(b'refs/')
1171 oldvalue = [] if not oldvalue else [oldvalue]
1172 p = subprocess.Popen([b'git', b'update-ref', b'-d', refname] + oldvalue,
1175 _git_wait('git update-ref', p)
1178 def guess_repo(path=None):
1179 """Set the path value in the global variable "repodir".
1180 This makes bup look for an existing bup repository, but not fail if a
1181 repository doesn't exist. Usually, if you are interacting with a bup
1182 repository, you would not be calling this function but using
1183 check_repo_or_die().
1189 repodir = environ.get(b'BUP_DIR')
1191 repodir = os.path.expanduser(b'~/.bup')
1194 def init_repo(path=None):
1195 """Create the Git bare repository for bup in a given path."""
1197 d = repo() # appends a / to the path
1198 parent = os.path.dirname(os.path.dirname(d))
1199 if parent and not os.path.exists(parent):
1200 raise GitError('parent directory "%s" does not exist\n'
1202 if os.path.exists(d) and not os.path.isdir(os.path.join(d, b'.')):
1203 raise GitError('"%s" exists but is not a directory\n' % path_msg(d))
1204 p = subprocess.Popen([b'git', b'--bare', b'init'], stdout=sys.stderr,
1207 _git_wait('git init', p)
1208 # Force the index version configuration in order to ensure bup works
1209 # regardless of the version of the installed Git binary.
1210 p = subprocess.Popen([b'git', b'config', b'pack.indexVersion', '2'],
1211 stdout=sys.stderr, env=_gitenv(), close_fds=True)
1212 _git_wait('git config', p)
1214 p = subprocess.Popen([b'git', b'config', b'core.logAllRefUpdates', b'true'],
1215 stdout=sys.stderr, env=_gitenv(), close_fds=True)
1216 _git_wait('git config', p)
1219 def check_repo_or_die(path=None):
1220 """Check to see if a bup repository probably exists, and abort if not."""
1223 pst = stat_if_exists(top + b'/objects/pack')
1224 if pst and stat.S_ISDIR(pst.st_mode):
1227 top_st = stat_if_exists(top)
1229 log('error: repository %r does not exist (see "bup help init")\n'
1232 log('error: %s is not a repository\n' % path_msg(top))
1236 def is_suitable_git(ver_str):
1237 if not ver_str.startswith(b'git version '):
1238 return 'unrecognized'
1239 ver_str = ver_str[len(b'git version '):]
1240 if ver_str.startswith(b'0.'):
1241 return 'insufficient'
1242 if ver_str.startswith(b'1.'):
1243 if re.match(br'1\.[012345]rc', ver_str):
1244 return 'insufficient'
1245 if re.match(br'1\.[01234]\.', ver_str):
1246 return 'insufficient'
1247 if re.match(br'1\.5\.[012345]($|\.)', ver_str):
1248 return 'insufficient'
1249 if re.match(br'1\.5\.6-rc', ver_str):
1250 return 'insufficient'
1252 if re.match(br'[0-9]+(\.|$)?', ver_str):
1258 def require_suitable_git(ver_str=None):
1259 """Raise GitError if the version of git isn't suitable.
1261 Rely on ver_str when provided, rather than invoking the git in the
1266 if _git_great is not None:
1268 if environ.get(b'BUP_GIT_VERSION_IS_FINE', b'').lower() \
1269 in (b'yes', b'true', b'1'):
1273 ver_str, _, _ = _git_exo([b'git', b'--version'])
1274 status = is_suitable_git(ver_str)
1275 if status == 'unrecognized':
1276 raise GitError('Unexpected git --version output: %r' % ver_str)
1277 if status == 'insufficient':
1278 log('error: git version must be at least 1.5.6\n')
1280 if status == 'suitable':
1287 """Link to 'git cat-file' that is used to retrieve blob data."""
1288 def __init__(self, repo_dir = None):
1289 require_suitable_git()
1290 self.repo_dir = repo_dir
1291 self.p = self.inprogress = None
1293 def close(self, wait=False):
1294 self.p, p = None, self.p
1295 self.inprogress = None
1300 # This will handle pending exceptions correctly once
1310 self.p = subprocess.Popen([b'git', b'cat-file', b'--batch'],
1311 stdin=subprocess.PIPE,
1312 stdout=subprocess.PIPE,
1315 env=_gitenv(self.repo_dir))
1318 """Yield (oidx, type, size), followed by the data referred to by ref.
1319 If ref does not exist, only yield (None, None, None).
1322 if not self.p or self.p.poll() != None:
1325 poll_result = self.p.poll()
1326 assert(poll_result == None)
1328 log('get: opening %r while %r is open\n' % (ref, self.inprogress))
1329 assert(not self.inprogress)
1330 assert ref.find(b'\n') < 0
1331 assert ref.find(b'\r') < 0
1332 assert not ref.startswith(b'-')
1333 self.inprogress = ref
1334 self.p.stdin.write(ref + b'\n')
1335 self.p.stdin.flush()
1336 hdr = self.p.stdout.readline()
1338 raise GitError('unexpected cat-file EOF (last request: %r, exit: %s)'
1339 % (ref, self.p.poll() or 'none'))
1340 if hdr.endswith(b' missing\n'):
1341 self.inprogress = None
1342 yield None, None, None
1344 info = hdr.split(b' ')
1345 if len(info) != 3 or len(info[0]) != 40:
1346 raise GitError('expected object (id, type, size), got %r' % info)
1347 oidx, typ, size = info
1350 it = chunkyreader(self.p.stdout, size)
1351 yield oidx, typ, size
1352 for blob in chunkyreader(self.p.stdout, size):
1354 readline_result = self.p.stdout.readline()
1355 assert readline_result == b'\n'
1356 self.inprogress = None
1357 except Exception as ex:
1358 with pending_raise(ex):
1361 def _join(self, it):
1362 _, typ, _ = next(it)
1366 elif typ == b'tree':
1367 treefile = b''.join(it)
1368 for (mode, name, sha) in tree_decode(treefile):
1369 for blob in self.join(hexlify(sha)):
1371 elif typ == b'commit':
1372 treeline = b''.join(it).split(b'\n')[0]
1373 assert treeline.startswith(b'tree ')
1374 for blob in self.join(treeline[5:]):
1377 raise GitError('invalid object type %r: expected blob/tree/commit'
1381 """Generate a list of the content of all blobs that can be reached
1382 from an object. The hash given in 'id' must point to a blob, a tree
1383 or a commit. The content of all blobs that can be seen from trees or
1384 commits will be added to the list.
1386 for d in self._join(self.get(id)):
1392 def cp(repo_dir=None):
1393 """Create a CatPipe object or reuse the already existing one."""
1396 repo_dir = repodir or repo()
1397 repo_dir = os.path.abspath(repo_dir)
1398 cp = _cp.get(repo_dir)
1400 cp = CatPipe(repo_dir)
1405 def close_catpipes():
1406 # FIXME: chain exceptions
1408 _, cp = _cp.popitem()
1412 def tags(repo_dir = None):
1413 """Return a dictionary of all tags in the form {hash: [tag_names, ...]}."""
1415 for n, c in list_refs(repo_dir = repo_dir, limit_to_tags=True):
1416 assert n.startswith(b'refs/tags/')
1420 tags[c].append(name) # more than one tag can point at 'c'
1424 class MissingObject(KeyError):
1425 def __init__(self, oid):
1427 KeyError.__init__(self, 'object %r is missing' % hexlify(oid))
1430 WalkItem = namedtuple('WalkItem', ['oid', 'type', 'mode',
1431 'path', 'chunk_path', 'data'])
1432 # The path is the mangled path, and if an item represents a fragment
1433 # of a chunked file, the chunk_path will be the chunked subtree path
1434 # for the chunk, i.e. ['', '2d3115e', ...]. The top-level path for a
1435 # chunked file will have a chunk_path of ['']. So some chunk subtree
1436 # of the file '/foo/bar/baz' might look like this:
1438 # item.path = ['foo', 'bar', 'baz.bup']
1439 # item.chunk_path = ['', '2d3115e', '016b097']
1440 # item.type = 'tree'
1444 def walk_object(get_ref, oidx, stop_at=None, include_data=None):
1445 """Yield everything reachable from oidx via get_ref (which must behave
1446 like CatPipe get) as a WalkItem, stopping whenever stop_at(oidx)
1447 returns true. Throw MissingObject if a hash encountered is
1448 missing from the repository, and don't read or return blob content
1449 in the data field unless include_data is set.
1452 # Maintain the pending stack on the heap to avoid stack overflow
1453 pending = [(oidx, [], [], None)]
1455 oidx, parent_path, chunk_path, mode = pending.pop()
1456 oid = unhexlify(oidx)
1457 if stop_at and stop_at(oidx):
1460 if (not include_data) and mode and stat.S_ISREG(mode):
1461 # If the object is a "regular file", then it's a leaf in
1462 # the graph, so we can skip reading the data if the caller
1463 # hasn't requested it.
1464 yield WalkItem(oid=oid, type=b'blob',
1465 chunk_path=chunk_path, path=parent_path,
1470 item_it = get_ref(oidx)
1471 get_oidx, typ, _ = next(item_it)
1473 raise MissingObject(unhexlify(oidx))
1474 if typ not in (b'blob', b'commit', b'tree'):
1475 raise Exception('unexpected repository object type %r' % typ)
1477 # FIXME: set the mode based on the type when the mode is None
1478 if typ == b'blob' and not include_data:
1479 # Dump data until we can ask cat_pipe not to fetch it
1480 for ignored in item_it:
1484 data = b''.join(item_it)
1486 yield WalkItem(oid=oid, type=typ,
1487 chunk_path=chunk_path, path=parent_path,
1489 data=(data if include_data else None))
1491 if typ == b'commit':
1492 commit_items = parse_commit(data)
1493 for pid in commit_items.parents:
1494 pending.append((pid, parent_path, chunk_path, mode))
1495 pending.append((commit_items.tree, parent_path, chunk_path,
1496 hashsplit.GIT_MODE_TREE))
1497 elif typ == b'tree':
1498 for mode, name, ent_id in tree_decode(data):
1499 demangled, bup_type = demangle_name(name, mode)
1501 sub_path = parent_path
1502 sub_chunk_path = chunk_path + [name]
1504 sub_path = parent_path + [name]
1505 if bup_type == BUP_CHUNKED:
1506 sub_chunk_path = [b'']
1508 sub_chunk_path = chunk_path
1509 pending.append((hexlify(ent_id), sub_path, sub_chunk_path,