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 new_packs = set(d.values())
687 if not p in new_packs:
689 new_packs = list(new_packs)
690 new_packs.sort(reverse=True, key=lambda x: len(x))
691 self.packs = new_packs
692 if self.bloom is None and os.path.exists(bfull):
693 self.bloom = bloom.ShaBloom(bfull)
695 if self.bloom and self.bloom.valid() and len(self.bloom) >= len(self):
699 self.bloom, bloom_tmp = None, self.bloom
701 except BaseException as ex:
702 with pending_raise(ex):
706 debug1('PackIdxList: using %d index%s.\n'
707 % (len(self.packs), len(self.packs)!=1 and 'es' or ''))
710 """Insert an additional object in the list."""
714 def open_idx(filename):
715 if filename.endswith(b'.idx'):
716 f = open(filename, 'rb')
718 if header[0:4] == b'\377tOc':
719 version = struct.unpack('!I', header[4:8])[0]
721 return PackIdxV2(filename, f)
723 raise GitError('%s: expected idx file version 2, got %d'
724 % (path_msg(filename), version))
725 elif len(header) == 8 and header[0:4] < b'\377tOc':
726 return PackIdxV1(filename, f)
728 raise GitError('%s: unrecognized idx file header'
729 % path_msg(filename))
730 elif filename.endswith(b'.midx'):
731 return midx.PackMidx(filename)
733 raise GitError('idx filenames must end with .idx or .midx')
736 def idxmerge(idxlist, final_progress=True):
737 """Generate a list of all the objects reachable in a PackIdxList."""
738 def pfunc(count, total):
739 qprogress('Reading indexes: %.2f%% (%d/%d)\r'
740 % (count*100.0/total, count, total))
741 def pfinal(count, total):
743 progress('Reading indexes: %.2f%% (%d/%d), done.\n'
744 % (100, total, total))
745 return merge_iter(idxlist, 10024, pfunc, pfinal)
748 def create_commit_blob(tree, parent,
749 author, adate_sec, adate_tz,
750 committer, cdate_sec, cdate_tz,
752 if adate_tz is not None:
753 adate_str = _git_date_str(adate_sec, adate_tz)
755 adate_str = _local_git_date_str(adate_sec)
756 if cdate_tz is not None:
757 cdate_str = _git_date_str(cdate_sec, cdate_tz)
759 cdate_str = _local_git_date_str(cdate_sec)
761 if tree: l.append(b'tree %s' % hexlify(tree))
762 if parent: l.append(b'parent %s' % hexlify(parent))
763 if author: l.append(b'author %s %s' % (author, adate_str))
764 if committer: l.append(b'committer %s %s' % (committer, cdate_str))
769 def _make_objcache():
770 return PackIdxList(repo(b'objects/pack'))
772 # bup-gc assumes that it can disable all PackWriter activities
773 # (bloom/midx/cache) via the constructor and close() arguments.
776 """Writes Git objects inside a pack file."""
777 def __init__(self, objcache_maker=_make_objcache, compression_level=1,
778 run_midx=True, on_pack_finish=None,
779 max_pack_size=None, max_pack_objects=None, repo_dir=None):
781 self.repo_dir = repo_dir or repo()
788 self.objcache_maker = objcache_maker
790 self.compression_level = compression_level
791 self.run_midx=run_midx
792 self.on_pack_finish = on_pack_finish
793 if not max_pack_size:
794 max_pack_size = git_config_get(b'pack.packSizeLimit',
795 repo_dir=self.repo_dir,
797 if not max_pack_size:
798 # larger packs slow down pruning
799 max_pack_size = 1000 * 1000 * 1000
800 self.max_pack_size = max_pack_size
801 # cache memory usage is about 83 bytes per object
802 self.max_pack_objects = max_pack_objects if max_pack_objects \
803 else max(1, self.max_pack_size // 5000)
808 def __exit__(self, type, value, traceback):
809 with pending_raise(value, rethrow=False):
814 objdir = dir = os.path.join(self.repo_dir, b'objects')
815 fd, name = tempfile.mkstemp(suffix=b'.pack', dir=objdir)
817 self.file = os.fdopen(fd, 'w+b')
822 self.parentfd = os.open(objdir, os.O_RDONLY)
828 assert name.endswith(b'.pack')
829 self.filename = name[:-5]
830 self.file.write(b'PACK\0\0\0\2\0\0\0\0')
831 self.idx = PackIdxV2Writer()
833 def _raw_write(self, datalist, sha):
836 # in case we get interrupted (eg. KeyboardInterrupt), it's best if
837 # the file never has a *partial* blob. So let's make sure it's
838 # all-or-nothing. (The blob shouldn't be very big anyway, thanks
839 # to our hashsplit algorithm.) f.write() does its own buffering,
840 # but that's okay because we'll flush it in _end().
841 oneblob = b''.join(datalist)
847 crc = zlib.crc32(oneblob) & 0xffffffff
848 self._update_idx(sha, crc, nw)
853 def _update_idx(self, sha, crc, size):
856 self.idx.add(sha, crc, self.file.tell() - size)
858 def _write(self, sha, type, content):
862 sha = calc_hash(type, content)
863 size, crc = self._raw_write(_encode_packobj(type, content,
864 self.compression_level),
866 if self.outbytes >= self.max_pack_size \
867 or self.count >= self.max_pack_objects:
871 def _require_objcache(self):
872 if self.objcache is None and self.objcache_maker:
873 self.objcache = self.objcache_maker()
874 if self.objcache is None:
876 "PackWriter not opened or can't check exists w/o objcache")
878 def exists(self, id, want_source=False):
879 """Return non-empty if an object is found in the object cache."""
880 self._require_objcache()
881 return self.objcache.exists(id, want_source=want_source)
883 def just_write(self, sha, type, content):
884 """Write an object to the pack file without checking for duplication."""
885 self._write(sha, type, content)
886 # If nothing else, gc doesn't have/want an objcache
887 if self.objcache is not None:
888 self.objcache.add(sha)
890 def maybe_write(self, type, content):
891 """Write an object to the pack file if not present and return its id."""
892 sha = calc_hash(type, content)
893 if not self.exists(sha):
894 self._require_objcache()
895 self.just_write(sha, type, content)
898 def new_blob(self, blob):
899 """Create a blob object in the pack with the supplied content."""
900 return self.maybe_write(b'blob', blob)
902 def new_tree(self, shalist):
903 """Create a tree object in the pack."""
904 content = tree_encode(shalist)
905 return self.maybe_write(b'tree', content)
907 def new_commit(self, tree, parent,
908 author, adate_sec, adate_tz,
909 committer, cdate_sec, cdate_tz,
911 """Create a commit object in the pack. The date_sec values must be
912 epoch-seconds, and if a tz is None, the local timezone is assumed."""
913 content = create_commit_blob(tree, parent,
914 author, adate_sec, adate_tz,
915 committer, cdate_sec, cdate_tz,
917 return self.maybe_write(b'commit', content)
919 def _end(self, run_midx=True, abort=False):
920 # Ignores run_midx during abort
923 self.file, f = None, self.file
924 self.idx, idx = None, self.idx
925 self.parentfd, pfd, = None, self.parentfd
928 with nullcontext_if_not(self.objcache), \
929 finalized(pfd, lambda x: x is not None and os.close(x)), \
933 os.unlink(self.filename + b'.pack')
936 # update object count
938 cp = struct.pack('!i', self.count)
942 # calculate the pack sha1sum
945 for b in chunkyreader(f):
947 packbin = sum.digest()
950 fdatasync(f.fileno())
953 idx.write(self.filename + b'.idx', packbin)
954 nameprefix = os.path.join(self.repo_dir,
955 b'objects/pack/pack-' + hexlify(packbin))
956 if os.path.exists(self.filename + b'.map'):
957 os.unlink(self.filename + b'.map')
958 os.rename(self.filename + b'.pack', nameprefix + b'.pack')
959 os.rename(self.filename + b'.idx', nameprefix + b'.idx')
962 auto_midx(os.path.join(self.repo_dir, b'objects/pack'))
963 if self.on_pack_finish:
964 self.on_pack_finish(nameprefix)
967 # Must be last -- some of the code above depends on it
971 """Remove the pack file from disk."""
973 self._end(abort=True)
975 def breakpoint(self):
976 """Clear byte and object counts and return the last processed id."""
977 id = self._end(self.run_midx)
978 self.outbytes = self.count = 0
981 def close(self, run_midx=True):
982 """Close the pack file and move it to its definitive path."""
984 return self._end(run_midx=run_midx)
990 class PackIdxV2Writer:
992 self.idx = list(list() for i in range(256))
995 def add(self, sha, crc, offs):
998 self.idx[byte_int(sha[0])].append((sha, crc, offs))
1000 def write(self, filename, packbin):
1002 for section in self.idx:
1003 for entry in section:
1004 if entry[2] >= 2**31:
1007 # Length: header + fan-out + shas-and-crcs + overflow-offsets
1008 index_len = 8 + (4 * 256) + (28 * self.count) + (8 * ofs64_count)
1010 idx_f = open(filename, 'w+b')
1012 idx_f.truncate(index_len)
1013 fdatasync(idx_f.fileno())
1014 idx_map = mmap_readwrite(idx_f, close=False)
1016 count = _helpers.write_idx(filename, idx_map, self.idx,
1018 assert(count == self.count)
1025 idx_f = open(filename, 'a+b')
1027 idx_f.write(packbin)
1030 b = idx_f.read(8 + 4*256)
1033 for b in chunkyreader(idx_f, 20 * self.count):
1036 for b in chunkyreader(idx_f):
1038 idx_f.write(idx_sum.digest())
1039 fdatasync(idx_f.fileno())
1044 def list_refs(patterns=None, repo_dir=None,
1045 limit_to_heads=False, limit_to_tags=False):
1046 """Yield (refname, hash) tuples for all repository refs unless
1047 patterns are specified. In that case, only include tuples for
1048 refs matching those patterns (cf. git-show-ref(1)). The limits
1049 restrict the result items to refs/heads or refs/tags. If both
1050 limits are specified, items from both sources will be included.
1053 argv = [b'git', b'show-ref']
1055 argv.append(b'--heads')
1057 argv.append(b'--tags')
1060 argv.extend(patterns)
1061 p = subprocess.Popen(argv, env=_gitenv(repo_dir), stdout=subprocess.PIPE,
1063 out = p.stdout.read().strip()
1064 rv = p.wait() # not fatal
1068 for d in out.split(b'\n'):
1069 sha, name = d.split(b' ', 1)
1070 yield name, unhexlify(sha)
1073 def read_ref(refname, repo_dir = None):
1074 """Get the commit id of the most recent commit made on a given ref."""
1075 refs = list_refs(patterns=[refname], repo_dir=repo_dir, limit_to_heads=True)
1076 l = tuple(islice(refs, 2))
1084 def rev_list_invocation(ref_or_refs, format=None):
1085 if isinstance(ref_or_refs, bytes):
1086 refs = (ref_or_refs,)
1089 argv = [b'git', b'rev-list']
1092 argv.append(b'--pretty=format:' + format)
1094 assert not ref.startswith(b'-')
1100 def rev_list(ref_or_refs, parse=None, format=None, repo_dir=None):
1101 """Yield information about commits as per "git rev-list". If a format
1102 is not provided, yield one hex hash at a time. If a format is
1103 provided, pass it to rev-list and call parse(git_stdout) for each
1104 commit with the stream positioned just after the rev-list "commit
1105 HASH" header line. When a format is provided yield (oidx,
1106 parse(git_stdout)) for each commit.
1109 assert bool(parse) == bool(format)
1110 p = subprocess.Popen(rev_list_invocation(ref_or_refs,
1112 env=_gitenv(repo_dir),
1113 stdout = subprocess.PIPE,
1116 for line in p.stdout:
1119 line = p.stdout.readline()
1122 if not s.startswith(b'commit '):
1123 raise Exception('unexpected line ' + repr(s))
1126 yield s, parse(p.stdout)
1127 line = p.stdout.readline()
1129 rv = p.wait() # not fatal
1131 raise GitError('git rev-list returned error %d' % rv)
1134 def rev_parse(committish, repo_dir=None):
1135 """Resolve the full hash for 'committish', if it exists.
1137 Should be roughly equivalent to 'git rev-parse'.
1139 Returns the hex value of the hash if it is found, None if 'committish' does
1140 not correspond to anything.
1142 head = read_ref(committish, repo_dir=repo_dir)
1144 debug2("resolved from ref: commit = %s\n" % hexlify(head))
1147 if len(committish) == 40:
1149 hash = unhexlify(committish)
1153 with PackIdxList(repo(b'objects/pack', repo_dir=repo_dir)) as pL:
1160 def update_ref(refname, newval, oldval, repo_dir=None):
1161 """Update a repository reference."""
1164 assert refname.startswith(b'refs/heads/') \
1165 or refname.startswith(b'refs/tags/')
1166 p = subprocess.Popen([b'git', b'update-ref', refname,
1167 hexlify(newval), hexlify(oldval)],
1168 env=_gitenv(repo_dir),
1170 _git_wait(b'git update-ref', p)
1173 def delete_ref(refname, oldvalue=None):
1174 """Delete a repository reference (see git update-ref(1))."""
1175 assert refname.startswith(b'refs/')
1176 oldvalue = [] if not oldvalue else [oldvalue]
1177 p = subprocess.Popen([b'git', b'update-ref', b'-d', refname] + oldvalue,
1180 _git_wait('git update-ref', p)
1183 def guess_repo(path=None):
1184 """Set the path value in the global variable "repodir".
1185 This makes bup look for an existing bup repository, but not fail if a
1186 repository doesn't exist. Usually, if you are interacting with a bup
1187 repository, you would not be calling this function but using
1188 check_repo_or_die().
1194 repodir = environ.get(b'BUP_DIR')
1196 repodir = os.path.expanduser(b'~/.bup')
1199 def init_repo(path=None):
1200 """Create the Git bare repository for bup in a given path."""
1202 d = repo() # appends a / to the path
1203 parent = os.path.dirname(os.path.dirname(d))
1204 if parent and not os.path.exists(parent):
1205 raise GitError('parent directory "%s" does not exist\n'
1207 if os.path.exists(d) and not os.path.isdir(os.path.join(d, b'.')):
1208 raise GitError('"%s" exists but is not a directory\n' % path_msg(d))
1209 p = subprocess.Popen([b'git', b'--bare', b'init'], stdout=sys.stderr,
1212 _git_wait('git init', p)
1213 # Force the index version configuration in order to ensure bup works
1214 # regardless of the version of the installed Git binary.
1215 p = subprocess.Popen([b'git', b'config', b'pack.indexVersion', '2'],
1216 stdout=sys.stderr, env=_gitenv(), close_fds=True)
1217 _git_wait('git config', p)
1219 p = subprocess.Popen([b'git', b'config', b'core.logAllRefUpdates', b'true'],
1220 stdout=sys.stderr, env=_gitenv(), close_fds=True)
1221 _git_wait('git config', p)
1224 def check_repo_or_die(path=None):
1225 """Check to see if a bup repository probably exists, and abort if not."""
1228 pst = stat_if_exists(top + b'/objects/pack')
1229 if pst and stat.S_ISDIR(pst.st_mode):
1232 top_st = stat_if_exists(top)
1234 log('error: repository %r does not exist (see "bup help init")\n'
1237 log('error: %s is not a repository\n' % path_msg(top))
1241 def is_suitable_git(ver_str):
1242 if not ver_str.startswith(b'git version '):
1243 return 'unrecognized'
1244 ver_str = ver_str[len(b'git version '):]
1245 if ver_str.startswith(b'0.'):
1246 return 'insufficient'
1247 if ver_str.startswith(b'1.'):
1248 if re.match(br'1\.[012345]rc', ver_str):
1249 return 'insufficient'
1250 if re.match(br'1\.[01234]\.', ver_str):
1251 return 'insufficient'
1252 if re.match(br'1\.5\.[012345]($|\.)', ver_str):
1253 return 'insufficient'
1254 if re.match(br'1\.5\.6-rc', ver_str):
1255 return 'insufficient'
1257 if re.match(br'[0-9]+(\.|$)?', ver_str):
1263 def require_suitable_git(ver_str=None):
1264 """Raise GitError if the version of git isn't suitable.
1266 Rely on ver_str when provided, rather than invoking the git in the
1271 if _git_great is not None:
1273 if environ.get(b'BUP_GIT_VERSION_IS_FINE', b'').lower() \
1274 in (b'yes', b'true', b'1'):
1278 ver_str, _, _ = _git_exo([b'git', b'--version'])
1279 status = is_suitable_git(ver_str)
1280 if status == 'unrecognized':
1281 raise GitError('Unexpected git --version output: %r' % ver_str)
1282 if status == 'insufficient':
1283 log('error: git version must be at least 1.5.6\n')
1285 if status == 'suitable':
1292 """Link to 'git cat-file' that is used to retrieve blob data."""
1293 def __init__(self, repo_dir = None):
1294 require_suitable_git()
1295 self.repo_dir = repo_dir
1296 self.p = self.inprogress = None
1298 def close(self, wait=False):
1299 self.p, p = None, self.p
1300 self.inprogress = None
1305 # This will handle pending exceptions correctly once
1315 self.p = subprocess.Popen([b'git', b'cat-file', b'--batch'],
1316 stdin=subprocess.PIPE,
1317 stdout=subprocess.PIPE,
1320 env=_gitenv(self.repo_dir))
1323 """Yield (oidx, type, size), followed by the data referred to by ref.
1324 If ref does not exist, only yield (None, None, None).
1327 if not self.p or self.p.poll() != None:
1330 poll_result = self.p.poll()
1331 assert(poll_result == None)
1333 log('get: opening %r while %r is open\n' % (ref, self.inprogress))
1334 assert(not self.inprogress)
1335 assert ref.find(b'\n') < 0
1336 assert ref.find(b'\r') < 0
1337 assert not ref.startswith(b'-')
1338 self.inprogress = ref
1339 self.p.stdin.write(ref + b'\n')
1340 self.p.stdin.flush()
1341 hdr = self.p.stdout.readline()
1343 raise GitError('unexpected cat-file EOF (last request: %r, exit: %s)'
1344 % (ref, self.p.poll() or 'none'))
1345 if hdr.endswith(b' missing\n'):
1346 self.inprogress = None
1347 yield None, None, None
1349 info = hdr.split(b' ')
1350 if len(info) != 3 or len(info[0]) != 40:
1351 raise GitError('expected object (id, type, size), got %r' % info)
1352 oidx, typ, size = info
1355 it = chunkyreader(self.p.stdout, size)
1356 yield oidx, typ, size
1357 for blob in chunkyreader(self.p.stdout, size):
1359 readline_result = self.p.stdout.readline()
1360 assert readline_result == b'\n'
1361 self.inprogress = None
1362 except Exception as ex:
1363 with pending_raise(ex):
1366 def _join(self, it):
1367 _, typ, _ = next(it)
1371 elif typ == b'tree':
1372 treefile = b''.join(it)
1373 for (mode, name, sha) in tree_decode(treefile):
1374 for blob in self.join(hexlify(sha)):
1376 elif typ == b'commit':
1377 treeline = b''.join(it).split(b'\n')[0]
1378 assert treeline.startswith(b'tree ')
1379 for blob in self.join(treeline[5:]):
1382 raise GitError('invalid object type %r: expected blob/tree/commit'
1386 """Generate a list of the content of all blobs that can be reached
1387 from an object. The hash given in 'id' must point to a blob, a tree
1388 or a commit. The content of all blobs that can be seen from trees or
1389 commits will be added to the list.
1391 for d in self._join(self.get(id)):
1397 def cp(repo_dir=None):
1398 """Create a CatPipe object or reuse the already existing one."""
1401 repo_dir = repodir or repo()
1402 repo_dir = os.path.abspath(repo_dir)
1403 cp = _cp.get(repo_dir)
1405 cp = CatPipe(repo_dir)
1410 def close_catpipes():
1411 # FIXME: chain exceptions
1413 _, cp = _cp.popitem()
1417 def tags(repo_dir = None):
1418 """Return a dictionary of all tags in the form {hash: [tag_names, ...]}."""
1420 for n, c in list_refs(repo_dir = repo_dir, limit_to_tags=True):
1421 assert n.startswith(b'refs/tags/')
1425 tags[c].append(name) # more than one tag can point at 'c'
1429 class MissingObject(KeyError):
1430 def __init__(self, oid):
1432 KeyError.__init__(self, 'object %r is missing' % hexlify(oid))
1435 WalkItem = namedtuple('WalkItem', ['oid', 'type', 'mode',
1436 'path', 'chunk_path', 'data'])
1437 # The path is the mangled path, and if an item represents a fragment
1438 # of a chunked file, the chunk_path will be the chunked subtree path
1439 # for the chunk, i.e. ['', '2d3115e', ...]. The top-level path for a
1440 # chunked file will have a chunk_path of ['']. So some chunk subtree
1441 # of the file '/foo/bar/baz' might look like this:
1443 # item.path = ['foo', 'bar', 'baz.bup']
1444 # item.chunk_path = ['', '2d3115e', '016b097']
1445 # item.type = 'tree'
1449 def walk_object(get_ref, oidx, stop_at=None, include_data=None):
1450 """Yield everything reachable from oidx via get_ref (which must behave
1451 like CatPipe get) as a WalkItem, stopping whenever stop_at(oidx)
1452 returns true. Throw MissingObject if a hash encountered is
1453 missing from the repository, and don't read or return blob content
1454 in the data field unless include_data is set.
1457 # Maintain the pending stack on the heap to avoid stack overflow
1458 pending = [(oidx, [], [], None)]
1460 oidx, parent_path, chunk_path, mode = pending.pop()
1461 oid = unhexlify(oidx)
1462 if stop_at and stop_at(oidx):
1465 if (not include_data) and mode and stat.S_ISREG(mode):
1466 # If the object is a "regular file", then it's a leaf in
1467 # the graph, so we can skip reading the data if the caller
1468 # hasn't requested it.
1469 yield WalkItem(oid=oid, type=b'blob',
1470 chunk_path=chunk_path, path=parent_path,
1475 item_it = get_ref(oidx)
1476 get_oidx, typ, _ = next(item_it)
1478 raise MissingObject(unhexlify(oidx))
1479 if typ not in (b'blob', b'commit', b'tree'):
1480 raise Exception('unexpected repository object type %r' % typ)
1482 # FIXME: set the mode based on the type when the mode is None
1483 if typ == b'blob' and not include_data:
1484 # Dump data until we can ask cat_pipe not to fetch it
1485 for ignored in item_it:
1489 data = b''.join(item_it)
1491 yield WalkItem(oid=oid, type=typ,
1492 chunk_path=chunk_path, path=parent_path,
1494 data=(data if include_data else None))
1496 if typ == b'commit':
1497 commit_items = parse_commit(data)
1498 for pid in commit_items.parents:
1499 pending.append((pid, parent_path, chunk_path, mode))
1500 pending.append((commit_items.tree, parent_path, chunk_path,
1501 hashsplit.GIT_MODE_TREE))
1502 elif typ == b'tree':
1503 for mode, name, ent_id in tree_decode(data):
1504 demangled, bup_type = demangle_name(name, mode)
1506 sub_path = parent_path
1507 sub_chunk_path = chunk_path + [name]
1509 sub_path = parent_path + [name]
1510 if bup_type == BUP_CHUNKED:
1511 sub_chunk_path = [b'']
1513 sub_chunk_path = chunk_path
1514 pending.append((hexlify(ent_id), sub_path, sub_chunk_path,