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,
20 from bup.io import path_msg
21 from bup.helpers import (Sha1, add_error, chunkyreader, debug1, debug2,
28 mmap_read, mmap_readwrite,
30 progress, qprogress, stat_if_exists,
36 repodir = None # The default repository, once initialized
38 _typemap = {b'blob': 3, b'tree': 2, b'commit': 1, b'tag': 4}
39 _typermap = {v: k for k, v in _typemap.items()}
46 class GitError(Exception):
50 def _gitenv(repo_dir=None):
53 return merge_dict(environ, {b'GIT_DIR': os.path.abspath(repo_dir)})
55 def _git_wait(cmd, p):
58 raise GitError('%r returned %d' % (cmd, rv))
60 def _git_exo(cmd, **kwargs):
61 kwargs['check'] = False
62 result = exo(cmd, **kwargs)
64 if proc.returncode != 0:
65 raise GitError('%r returned %d' % (cmd, proc.returncode))
68 def git_config_get(option, repo_dir=None, opttype=None, cfg_file=None):
69 assert not (repo_dir and cfg_file), "repo_dir and cfg_file cannot both be used"
70 cmd = [b'git', b'config', b'--null']
72 cmd.extend([b'--file', cfg_file])
74 cmd.extend([b'--int'])
75 elif opttype == 'bool':
76 cmd.extend([b'--bool'])
78 assert opttype is None
79 cmd.extend([b'--get', option])
82 env = _gitenv(repo_dir=repo_dir)
83 p = subprocess.Popen(cmd, stdout=subprocess.PIPE, env=env,
85 # with --null, git writes out a trailing \0 after the value
86 r = p.stdout.read()[:-1]
91 elif opttype == 'bool':
92 # git converts to 'true' or 'false'
96 raise GitError('%r returned %d' % (cmd, rc))
100 def parse_tz_offset(s):
101 """UTC offset in seconds."""
102 tz_off = (int(s[1:3]) * 60 * 60) + (int(s[3:5]) * 60)
103 if bytes_from_byte(s[0]) == b'-':
107 def parse_commit_gpgsig(sig):
108 """Return the original signature bytes.
110 i.e. with the "gpgsig " header and the leading space character on
111 each continuation line removed.
116 assert sig.startswith(b'gpgsig ')
118 return sig.replace(b'\n ', b'\n')
120 # FIXME: derived from http://git.rsbx.net/Documents/Git_Data_Formats.txt
121 # Make sure that's authoritative.
124 # https://github.com/git/git/blob/master/Documentation/technical/signature-format.txt
125 # The continuation lines have only one leading space.
127 _start_end_char = br'[^ .,:;<>"\'\0\n]'
128 _content_char = br'[^\0\n<>]'
129 _safe_str_rx = br'(?:%s{1,2}|(?:%s%s*%s))' \
131 _start_end_char, _content_char, _start_end_char)
132 _tz_rx = br'[-+]\d\d[0-5]\d'
133 _parent_rx = br'(?:parent [abcdefABCDEF0123456789]{40}\n)'
134 # Assumes every following line starting with a space is part of the
135 # mergetag. Is there a formal commit blob spec?
136 _mergetag_rx = br'(?:\nmergetag object [abcdefABCDEF0123456789]{40}(?:\n [^\0\n]*)*)'
137 _commit_rx = re.compile(br'''tree (?P<tree>[abcdefABCDEF0123456789]{40})
138 (?P<parents>%s*)author (?P<author_name>%s) <(?P<author_mail>%s)> (?P<asec>\d+) (?P<atz>%s)
139 committer (?P<committer_name>%s) <(?P<committer_mail>%s)> (?P<csec>\d+) (?P<ctz>%s)(?P<mergetag>%s?)
140 (?P<gpgsig>gpgsig .*\n(?: .*\n)*)?
141 (?P<message>(?:.|\n)*)''' % (_parent_rx,
142 _safe_str_rx, _safe_str_rx, _tz_rx,
143 _safe_str_rx, _safe_str_rx, _tz_rx,
145 _parent_hash_rx = re.compile(br'\s*parent ([abcdefABCDEF0123456789]{40})\s*')
147 # Note that the author_sec and committer_sec values are (UTC) epoch
148 # seconds, and for now the mergetag is not included.
149 CommitInfo = namedtuple('CommitInfo', ['tree', 'parents',
150 'author_name', 'author_mail',
151 'author_sec', 'author_offset',
152 'committer_name', 'committer_mail',
153 'committer_sec', 'committer_offset',
157 def parse_commit(content):
158 commit_match = re.match(_commit_rx, content)
160 raise Exception('cannot parse commit %r' % content)
161 matches = commit_match.groupdict()
162 return CommitInfo(tree=matches['tree'],
163 parents=re.findall(_parent_hash_rx, matches['parents']),
164 author_name=matches['author_name'],
165 author_mail=matches['author_mail'],
166 author_sec=int(matches['asec']),
167 author_offset=parse_tz_offset(matches['atz']),
168 committer_name=matches['committer_name'],
169 committer_mail=matches['committer_mail'],
170 committer_sec=int(matches['csec']),
171 committer_offset=parse_tz_offset(matches['ctz']),
172 gpgsig=parse_commit_gpgsig(matches['gpgsig']),
173 message=matches['message'])
176 def get_cat_data(cat_iterator, expected_type):
177 _, kind, _ = next(cat_iterator)
178 if kind != expected_type:
179 raise Exception('expected %r, saw %r' % (expected_type, kind))
180 return b''.join(cat_iterator)
182 def get_commit_items(id, cp):
183 return parse_commit(get_cat_data(cp.get(id), b'commit'))
185 def _local_git_date_str(epoch_sec):
186 return b'%d %s' % (epoch_sec, utc_offset_str(epoch_sec))
189 def _git_date_str(epoch_sec, tz_offset_sec):
190 offs = tz_offset_sec // 60
191 return b'%d %s%02d%02d' \
193 b'+' if offs >= 0 else b'-',
198 def repo(sub = b'', repo_dir=None):
199 """Get the path to the git repository or one of its subdirectories."""
200 repo_dir = repo_dir or repodir
202 raise GitError('You should call check_repo_or_die()')
204 # If there's a .git subdirectory, then the actual repo is in there.
205 gd = os.path.join(repo_dir, b'.git')
206 if os.path.exists(gd):
209 return os.path.join(repo_dir, sub)
213 re.compile(br'([^0-9a-z]|\b)([0-9a-z]{7})[0-9a-z]{33}([^0-9a-z]|\b)')
216 return _shorten_hash_rx.sub(br'\1\2*\3', s)
220 full = os.path.abspath(path)
221 fullrepo = os.path.abspath(repo(b''))
222 if not fullrepo.endswith(b'/'):
224 if full.startswith(fullrepo):
225 path = full[len(fullrepo):]
226 if path.startswith(b'index-cache/'):
227 path = path[len(b'index-cache/'):]
228 return shorten_hash(path)
231 def auto_midx(objdir):
232 args = [path.exe(), b'midx', b'--auto', b'--dir', objdir]
234 rv = subprocess.call(args, stdout=open(os.devnull, 'w'))
236 # make sure 'args' gets printed to help with debugging
237 add_error('%r: exception: %s' % (args, e))
240 add_error('%r: returned %d' % (args, rv))
242 args = [path.exe(), b'bloom', b'--dir', objdir]
244 rv = subprocess.call(args, stdout=open(os.devnull, 'w'))
246 # make sure 'args' gets printed to help with debugging
247 add_error('%r: exception: %s' % (args, e))
250 add_error('%r: returned %d' % (args, rv))
253 def mangle_name(name, mode, gitmode):
254 """Mangle a file name to present an abstract name for segmented files.
255 Mangled file names will have the ".bup" extension added to them. If a
256 file's name already ends with ".bup", a ".bupl" extension is added to
257 disambiguate normal files from segmented ones.
259 if stat.S_ISREG(mode) and not stat.S_ISREG(gitmode):
260 assert(stat.S_ISDIR(gitmode))
261 return name + b'.bup'
262 elif name.endswith(b'.bup') or name[:-1].endswith(b'.bup'):
263 return name + b'.bupl'
268 (BUP_NORMAL, BUP_CHUNKED) = (0,1)
269 def demangle_name(name, mode):
270 """Remove name mangling from a file name, if necessary.
272 The return value is a tuple (demangled_filename,mode), where mode is one of
275 * BUP_NORMAL : files that should be read as-is from the repository
276 * BUP_CHUNKED : files that were chunked and need to be reassembled
278 For more information on the name mangling algorithm, see mangle_name()
280 if name.endswith(b'.bupl'):
281 return (name[:-5], BUP_NORMAL)
282 elif name.endswith(b'.bup'):
283 return (name[:-4], BUP_CHUNKED)
284 elif name.endswith(b'.bupm'):
286 BUP_CHUNKED if stat.S_ISDIR(mode) else BUP_NORMAL)
287 return (name, BUP_NORMAL)
290 def calc_hash(type, content):
291 """Calculate some content's hash in the Git fashion."""
292 header = b'%s %d\0' % (type, len(content))
298 def shalist_item_sort_key(ent):
299 (mode, name, id) = ent
300 assert(mode+0 == mode)
301 if stat.S_ISDIR(mode):
307 def tree_encode(shalist):
308 """Generate a git tree object from (mode,name,hash) tuples."""
309 shalist = sorted(shalist, key = shalist_item_sort_key)
311 for (mode,name,bin) in shalist:
313 assert(mode+0 == mode)
315 assert(len(bin) == 20)
316 s = b'%o %s\0%s' % (mode,name,bin)
317 assert s[0] != b'0' # 0-padded octal is not acceptable in a git tree
322 def tree_decode(buf):
323 """Generate a list of (mode,name,hash) from the git tree object in buf."""
325 while ofs < len(buf):
326 z = buf.find(b'\0', ofs)
328 spl = buf[ofs:z].split(b' ', 1)
329 assert(len(spl) == 2)
331 sha = buf[z+1:z+1+20]
333 yield (int(mode, 8), name, sha)
336 def _encode_packobj(type, content, compression_level=1):
337 if compression_level not in (0, 1, 2, 3, 4, 5, 6, 7, 8, 9):
338 raise ValueError('invalid compression level %s' % compression_level)
341 szbits = (sz & 0x0f) | (_typemap[type]<<4)
344 if sz: szbits |= 0x80
345 szout += bytes_from_uint(szbits)
350 z = zlib.compressobj(compression_level)
352 yield z.compress(content)
356 def _decode_packobj(buf):
359 type = _typermap[(c & 0x70) >> 4]
366 sz |= (c & 0x7f) << shift
370 return (type, zlib.decompress(buf[i+1:]))
373 class PackIdx(object):
374 def find_offset(self, hash):
375 """Get the offset of an object inside the index file."""
376 idx = self._idx_from_hash(hash)
378 return self._ofs_from_idx(idx)
381 def exists(self, hash, want_source=False):
382 """Return nonempty if the object exists in this index."""
383 if hash and (self._idx_from_hash(hash) != None):
384 return want_source and os.path.basename(self.name) or True
387 def _idx_from_hash(self, hash):
388 global _total_searches, _total_steps
390 assert(len(hash) == 20)
391 b1 = byte_int(hash[0])
392 start = self.fanout[b1-1] # range -1..254
393 end = self.fanout[b1] # range 0..255
395 _total_steps += 1 # lookup table is a step
398 mid = start + (end - start) // 2
399 v = self._idx_to_hash(mid)
409 class PackIdxV1(PackIdx):
410 """Object representation of a Git pack index (version 1) file."""
411 def __init__(self, filename, f):
412 super(PackIdxV1, self).__init__()
415 self.idxnames = [self.name]
416 self.map = mmap_read(f)
417 # Min size for 'L' is 4, which is sufficient for struct's '!I'
418 self.fanout = array('L', struct.unpack('!256I', self.map))
419 self.fanout.append(0) # entry "-1"
420 self.nsha = self.fanout[255]
421 self.sha_ofs = 256 * 4
422 # Avoid slicing shatable for individual hashes (very high overhead)
423 self.shatable = buffer(self.map, self.sha_ofs, self.nsha * 24)
428 def __exit__(self, type, value, traceback):
429 with pending_raise(value, rethrow=False):
433 return int(self.nsha) # int() from long for python 2
435 def _ofs_from_idx(self, idx):
436 if idx >= self.nsha or idx < 0:
437 raise IndexError('invalid pack index index %d' % idx)
438 ofs = self.sha_ofs + idx * 24
439 return struct.unpack_from('!I', self.map, offset=ofs)[0]
441 def _idx_to_hash(self, idx):
442 if idx >= self.nsha or idx < 0:
443 raise IndexError('invalid pack index index %d' % idx)
444 ofs = self.sha_ofs + idx * 24 + 4
445 return self.map[ofs : ofs + 20]
448 start = self.sha_ofs + 4
449 for ofs in range(start, start + 24 * self.nsha, 24):
450 yield self.map[ofs : ofs + 20]
454 if self.map is not None:
463 class PackIdxV2(PackIdx):
464 """Object representation of a Git pack index (version 2) file."""
465 def __init__(self, filename, f):
466 super(PackIdxV2, self).__init__()
469 self.idxnames = [self.name]
470 self.map = mmap_read(f)
471 assert self.map[0:8] == b'\377tOc\0\0\0\2'
472 # Min size for 'L' is 4, which is sufficient for struct's '!I'
473 self.fanout = array('L', struct.unpack_from('!256I', self.map, offset=8))
474 self.fanout.append(0)
475 self.nsha = self.fanout[255]
476 self.sha_ofs = 8 + 256*4
477 self.ofstable_ofs = self.sha_ofs + self.nsha * 20 + self.nsha * 4
478 self.ofs64table_ofs = self.ofstable_ofs + self.nsha * 4
479 # Avoid slicing this for individual hashes (very high overhead)
480 self.shatable = buffer(self.map, self.sha_ofs, self.nsha*20)
485 def __exit__(self, type, value, traceback):
486 with pending_raise(value, rethrow=False):
490 return int(self.nsha) # int() from long for python 2
492 def _ofs_from_idx(self, idx):
493 if idx >= self.nsha or idx < 0:
494 raise IndexError('invalid pack index index %d' % idx)
495 ofs_ofs = self.ofstable_ofs + idx * 4
496 ofs = struct.unpack_from('!I', self.map, offset=ofs_ofs)[0]
498 idx64 = ofs & 0x7fffffff
499 ofs64_ofs = self.ofs64table_ofs + idx64 * 8
500 ofs = struct.unpack_from('!Q', self.map, offset=ofs64_ofs)[0]
503 def _idx_to_hash(self, idx):
504 if idx >= self.nsha or idx < 0:
505 raise IndexError('invalid pack index index %d' % idx)
506 ofs = self.sha_ofs + idx * 20
507 return self.map[ofs : ofs + 20]
511 for ofs in range(start, start + 20 * self.nsha, 20):
512 yield self.map[ofs : ofs + 20]
516 if self.map is not None:
527 def __init__(self, dir, ignore_midx=False):
529 # Q: was this also intended to prevent opening multiple repos?
530 assert(_mpi_count == 0) # these things suck tons of VM; don't waste it
536 self.do_bloom = False
538 self.ignore_midx = ignore_midx
541 except BaseException as ex:
542 with pending_raise(ex):
548 assert _mpi_count == 0
551 assert _mpi_count == 0
553 self.bloom, bloom = None, self.bloom
554 self.packs, packs = None, self.packs
556 with ExitStack() as stack:
558 stack.enter_context(pack)
565 def __exit__(self, type, value, traceback):
566 with pending_raise(value, rethrow=False):
573 return iter(idxmerge(self.packs))
576 return sum(len(pack) for pack in self.packs)
578 def exists(self, hash, want_source=False):
579 """Return nonempty if the object exists in the index files."""
580 global _total_searches
582 if hash in self.also:
584 if self.do_bloom and self.bloom:
585 if self.bloom.exists(hash):
586 self.do_bloom = False
588 _total_searches -= 1 # was counted by bloom
590 for i in range(len(self.packs)):
592 _total_searches -= 1 # will be incremented by sub-pack
593 ix = p.exists(hash, want_source=want_source)
595 # reorder so most recently used packs are searched first
596 self.packs = [p] + self.packs[:i] + self.packs[i+1:]
601 def refresh(self, skip_midx = False):
602 """Refresh the index list.
603 This method verifies if .midx files were superseded (e.g. all of its
604 contents are in another, bigger .midx file) and removes the superseded
607 If skip_midx is True, all work on .midx files will be skipped and .midx
608 files will be removed from the list.
610 The instance variable 'ignore_midx' can force this function to
611 always act as if skip_midx was True.
613 if self.bloom is not None:
615 self.bloom = None # Always reopen the bloom as it may have been relaced
616 self.do_bloom = False
617 skip_midx = skip_midx or self.ignore_midx
618 d = dict((p.name, p) for p in self.packs
619 if not skip_midx or not isinstance(p, midx.PackMidx))
620 if os.path.exists(self.dir):
623 midxes = set(glob.glob(os.path.join(self.dir, b'*.midx')))
624 # remove any *.midx files from our list that no longer exist
625 for ix in list(d.values()):
626 if not isinstance(ix, midx.PackMidx):
628 if ix.name in midxes:
633 self.packs.remove(ix)
634 for ix in self.packs:
635 if isinstance(ix, midx.PackMidx):
636 for name in ix.idxnames:
637 d[os.path.join(self.dir, name)] = ix
640 mx = midx.PackMidx(full)
641 (mxd, mxf) = os.path.split(mx.name)
643 for n in mx.idxnames:
644 if not os.path.exists(os.path.join(mxd, n)):
645 log(('warning: index %s missing\n'
647 % (path_msg(n), path_msg(mxf)))
654 midxl.sort(key=lambda ix:
655 (-len(ix), -xstat.stat(ix.name).st_mtime))
658 for sub in ix.idxnames:
659 found = d.get(os.path.join(self.dir, sub))
660 if not found or isinstance(found, PackIdx):
661 # doesn't exist, or exists but not in a midx
666 for name in ix.idxnames:
667 d[os.path.join(self.dir, name)] = ix
668 elif not ix.force_keep:
669 debug1('midx: removing redundant: %s\n'
670 % path_msg(os.path.basename(ix.name)))
673 for full in glob.glob(os.path.join(self.dir, b'*.idx')):
677 except GitError as e:
681 bfull = os.path.join(self.dir, b'bup.bloom')
682 new_packs = set(d.values())
684 if not p in new_packs:
686 new_packs = list(new_packs)
687 new_packs.sort(reverse=True, key=lambda x: len(x))
688 self.packs = new_packs
689 if self.bloom is None and os.path.exists(bfull):
690 self.bloom = bloom.ShaBloom(bfull)
692 if self.bloom and self.bloom.valid() and len(self.bloom) >= len(self):
696 self.bloom, bloom_tmp = None, self.bloom
698 except BaseException as ex:
699 with pending_raise(ex):
703 debug1('PackIdxList: using %d index%s.\n'
704 % (len(self.packs), len(self.packs)!=1 and 'es' or ''))
707 """Insert an additional object in the list."""
711 def open_idx(filename):
712 if filename.endswith(b'.idx'):
713 f = open(filename, 'rb')
715 if header[0:4] == b'\377tOc':
716 version = struct.unpack('!I', header[4:8])[0]
718 return PackIdxV2(filename, f)
720 raise GitError('%s: expected idx file version 2, got %d'
721 % (path_msg(filename), version))
722 elif len(header) == 8 and header[0:4] < b'\377tOc':
723 return PackIdxV1(filename, f)
725 raise GitError('%s: unrecognized idx file header'
726 % path_msg(filename))
727 elif filename.endswith(b'.midx'):
728 return midx.PackMidx(filename)
730 raise GitError('idx filenames must end with .idx or .midx')
733 def idxmerge(idxlist, final_progress=True):
734 """Generate a list of all the objects reachable in a PackIdxList."""
735 def pfunc(count, total):
736 qprogress('Reading indexes: %.2f%% (%d/%d)\r'
737 % (count*100.0/total, count, total))
738 def pfinal(count, total):
740 progress('Reading indexes: %.2f%% (%d/%d), done.\n'
741 % (100, total, total))
742 return merge_iter(idxlist, 10024, pfunc, pfinal)
745 def create_commit_blob(tree, parent,
746 author, adate_sec, adate_tz,
747 committer, cdate_sec, cdate_tz,
749 if adate_tz is not None:
750 adate_str = _git_date_str(adate_sec, adate_tz)
752 adate_str = _local_git_date_str(adate_sec)
753 if cdate_tz is not None:
754 cdate_str = _git_date_str(cdate_sec, cdate_tz)
756 cdate_str = _local_git_date_str(cdate_sec)
758 if tree: l.append(b'tree %s' % hexlify(tree))
759 if parent: l.append(b'parent %s' % hexlify(parent))
760 if author: l.append(b'author %s %s' % (author, adate_str))
761 if committer: l.append(b'committer %s %s' % (committer, cdate_str))
766 def _make_objcache():
767 return PackIdxList(repo(b'objects/pack'))
769 # bup-gc assumes that it can disable all PackWriter activities
770 # (bloom/midx/cache) via the constructor and close() arguments.
772 class PackWriter(object):
773 """Writes Git objects inside a pack file."""
774 def __init__(self, objcache_maker=_make_objcache, compression_level=1,
775 run_midx=True, on_pack_finish=None,
776 max_pack_size=None, max_pack_objects=None, repo_dir=None):
778 self.repo_dir = repo_dir or repo()
785 self.objcache_maker = objcache_maker
787 self.compression_level = compression_level
788 self.run_midx=run_midx
789 self.on_pack_finish = on_pack_finish
790 if not max_pack_size:
791 max_pack_size = git_config_get(b'pack.packSizeLimit',
792 repo_dir=self.repo_dir,
794 if not max_pack_size:
795 # larger packs slow down pruning
796 max_pack_size = 1000 * 1000 * 1000
797 self.max_pack_size = max_pack_size
798 # cache memory usage is about 83 bytes per object
799 self.max_pack_objects = max_pack_objects if max_pack_objects \
800 else max(1, self.max_pack_size // 5000)
805 def __exit__(self, type, value, traceback):
806 with pending_raise(value, rethrow=False):
811 objdir = dir = os.path.join(self.repo_dir, b'objects')
812 fd, name = tempfile.mkstemp(suffix=b'.pack', dir=objdir)
814 self.file = os.fdopen(fd, 'w+b')
819 self.parentfd = os.open(objdir, os.O_RDONLY)
825 assert name.endswith(b'.pack')
826 self.filename = name[:-5]
827 self.file.write(b'PACK\0\0\0\2\0\0\0\0')
828 self.idx = PackIdxV2Writer()
830 def _raw_write(self, datalist, sha):
833 # in case we get interrupted (eg. KeyboardInterrupt), it's best if
834 # the file never has a *partial* blob. So let's make sure it's
835 # all-or-nothing. (The blob shouldn't be very big anyway, thanks
836 # to our hashsplit algorithm.) f.write() does its own buffering,
837 # but that's okay because we'll flush it in _end().
838 oneblob = b''.join(datalist)
844 crc = zlib.crc32(oneblob) & 0xffffffff
845 self._update_idx(sha, crc, nw)
850 def _update_idx(self, sha, crc, size):
853 self.idx.add(sha, crc, self.file.tell() - size)
855 def _write(self, sha, type, content):
859 size, crc = self._raw_write(_encode_packobj(type, content,
860 self.compression_level),
862 if self.outbytes >= self.max_pack_size \
863 or self.count >= self.max_pack_objects:
867 def _require_objcache(self):
868 if self.objcache is None and self.objcache_maker:
869 self.objcache = self.objcache_maker()
870 if self.objcache is None:
872 "PackWriter not opened or can't check exists w/o objcache")
874 def exists(self, id, want_source=False):
875 """Return non-empty if an object is found in the object cache."""
876 self._require_objcache()
877 return self.objcache.exists(id, want_source=want_source)
879 def just_write(self, sha, type, content):
880 """Write an object to the pack file without checking for duplication."""
881 self._write(sha, type, content)
882 # If nothing else, gc doesn't have/want an objcache
883 if self.objcache is not None:
884 self.objcache.add(sha)
886 def maybe_write(self, type, content):
887 """Write an object to the pack file if not present and return its id."""
888 sha = calc_hash(type, content)
889 if not self.exists(sha):
890 self._require_objcache()
891 self.just_write(sha, type, content)
894 def new_blob(self, blob):
895 """Create a blob object in the pack with the supplied content."""
896 return self.maybe_write(b'blob', blob)
898 def new_tree(self, shalist):
899 """Create a tree object in the pack."""
900 content = tree_encode(shalist)
901 return self.maybe_write(b'tree', content)
903 def new_commit(self, tree, parent,
904 author, adate_sec, adate_tz,
905 committer, cdate_sec, cdate_tz,
907 """Create a commit object in the pack. The date_sec values must be
908 epoch-seconds, and if a tz is None, the local timezone is assumed."""
909 content = create_commit_blob(tree, parent,
910 author, adate_sec, adate_tz,
911 committer, cdate_sec, cdate_tz,
913 return self.maybe_write(b'commit', content)
915 def _end(self, run_midx=True, abort=False):
916 # Ignores run_midx during abort
917 self.parentfd, pfd, = None, self.parentfd
918 self.file, f = None, self.file
919 self.idx, idx = None, self.idx
921 with nullcontext_if_not(self.objcache), \
922 finalized(pfd, lambda x: x is not None and os.close(x)), \
923 nullcontext_if_not(f):
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, force=False):
1156 """Update a repository reference.
1158 With force=True, don't care about the previous ref (oldval);
1159 with force=False oldval must be either a sha1 or None (for an
1160 entirely new branch)
1163 assert oldval is None
1168 oldarg = [hexlify(oldval)]
1169 assert refname.startswith(b'refs/heads/') \
1170 or refname.startswith(b'refs/tags/')
1171 p = subprocess.Popen([b'git', b'update-ref', refname,
1172 hexlify(newval)] + oldarg,
1173 env=_gitenv(repo_dir),
1175 _git_wait(b'git update-ref', p)
1178 def delete_ref(refname, oldvalue=None):
1179 """Delete a repository reference (see git update-ref(1))."""
1180 assert refname.startswith(b'refs/')
1181 oldvalue = [] if not oldvalue else [oldvalue]
1182 p = subprocess.Popen([b'git', b'update-ref', b'-d', refname] + oldvalue,
1185 _git_wait('git update-ref', p)
1189 """Return the global repodir or BUP_DIR when either is set, or ~/.bup.
1190 Usually, if you are interacting with a bup repository, you would
1191 not be calling this function but using check_repo_or_die().
1196 repo = environ.get(b'BUP_DIR')
1198 repo = os.path.expanduser(b'~/.bup')
1202 def init_repo(path=None):
1203 """Create the Git bare repository for bup in a given path."""
1205 repodir = path or guess_repo()
1206 d = repo() # appends a / to the path
1207 parent = os.path.dirname(os.path.dirname(d))
1208 if parent and not os.path.exists(parent):
1209 raise GitError('parent directory "%s" does not exist\n'
1211 if os.path.exists(d) and not os.path.isdir(os.path.join(d, b'.')):
1212 raise GitError('"%s" exists but is not a directory\n' % path_msg(d))
1213 p = subprocess.Popen([b'git', b'--bare', b'init'], stdout=sys.stderr,
1216 _git_wait('git init', p)
1217 # Force the index version configuration in order to ensure bup works
1218 # regardless of the version of the installed Git binary.
1219 p = subprocess.Popen([b'git', b'config', b'pack.indexVersion', '2'],
1220 stdout=sys.stderr, env=_gitenv(), close_fds=True)
1221 _git_wait('git config', p)
1223 p = subprocess.Popen([b'git', b'config', b'core.logAllRefUpdates', b'true'],
1224 stdout=sys.stderr, env=_gitenv(), close_fds=True)
1225 _git_wait('git config', p)
1228 def check_repo_or_die(path=None):
1229 """Check to see if a bup repository probably exists, and abort if not."""
1231 repodir = path or guess_repo()
1233 pst = stat_if_exists(top + b'/objects/pack')
1234 if pst and stat.S_ISDIR(pst.st_mode):
1237 top_st = stat_if_exists(top)
1239 log('error: repository %r does not exist (see "bup help init")\n'
1242 log('error: %s is not a repository\n' % path_msg(top))
1246 def is_suitable_git(ver_str):
1247 if not ver_str.startswith(b'git version '):
1248 return 'unrecognized'
1249 ver_str = ver_str[len(b'git version '):]
1250 if ver_str.startswith(b'0.'):
1251 return 'insufficient'
1252 if ver_str.startswith(b'1.'):
1253 if re.match(br'1\.[012345]rc', ver_str):
1254 return 'insufficient'
1255 if re.match(br'1\.[01234]\.', ver_str):
1256 return 'insufficient'
1257 if re.match(br'1\.5\.[012345]($|\.)', ver_str):
1258 return 'insufficient'
1259 if re.match(br'1\.5\.6-rc', ver_str):
1260 return 'insufficient'
1262 if re.match(br'[0-9]+(\.|$)?', ver_str):
1268 def require_suitable_git(ver_str=None):
1269 """Raise GitError if the version of git isn't suitable.
1271 Rely on ver_str when provided, rather than invoking the git in the
1276 if _git_great is not None:
1278 if environ.get(b'BUP_GIT_VERSION_IS_FINE', b'').lower() \
1279 in (b'yes', b'true', b'1'):
1283 ver_str, _, _ = _git_exo([b'git', b'--version'])
1284 status = is_suitable_git(ver_str)
1285 if status == 'unrecognized':
1286 raise GitError('Unexpected git --version output: %r' % ver_str)
1287 if status == 'insufficient':
1288 log('error: git version must be at least 1.5.6\n')
1290 if status == 'suitable':
1297 """Link to 'git cat-file' that is used to retrieve blob data."""
1298 def __init__(self, repo_dir = None):
1299 require_suitable_git()
1300 self.repo_dir = repo_dir
1301 self.p = self.inprogress = None
1303 def close(self, wait=False):
1304 self.p, p = None, self.p
1305 self.inprogress = None
1310 # This will handle pending exceptions correctly once
1320 self.p = subprocess.Popen([b'git', b'cat-file', b'--batch'],
1321 stdin=subprocess.PIPE,
1322 stdout=subprocess.PIPE,
1325 env=_gitenv(self.repo_dir))
1328 """Yield (oidx, type, size), followed by the data referred to by ref.
1329 If ref does not exist, only yield (None, None, None).
1332 if not self.p or self.p.poll() != None:
1335 poll_result = self.p.poll()
1336 assert(poll_result == None)
1338 log('get: opening %r while %r is open\n' % (ref, self.inprogress))
1339 assert(not self.inprogress)
1340 assert ref.find(b'\n') < 0
1341 assert ref.find(b'\r') < 0
1342 assert not ref.startswith(b'-')
1343 self.inprogress = ref
1344 self.p.stdin.write(ref + b'\n')
1345 self.p.stdin.flush()
1346 hdr = self.p.stdout.readline()
1348 raise GitError('unexpected cat-file EOF (last request: %r, exit: %s)'
1349 % (ref, self.p.poll() or 'none'))
1350 if hdr.endswith(b' missing\n'):
1351 self.inprogress = None
1352 yield None, None, None
1354 info = hdr.split(b' ')
1355 if len(info) != 3 or len(info[0]) != 40:
1356 raise GitError('expected object (id, type, size), got %r' % info)
1357 oidx, typ, size = info
1360 it = chunkyreader(self.p.stdout, size)
1361 yield oidx, typ, size
1362 for blob in chunkyreader(self.p.stdout, size):
1364 readline_result = self.p.stdout.readline()
1365 assert readline_result == b'\n'
1366 self.inprogress = None
1367 except Exception as ex:
1368 with pending_raise(ex):
1371 def _join(self, it):
1372 _, typ, _ = next(it)
1376 elif typ == b'tree':
1377 treefile = b''.join(it)
1378 for (mode, name, sha) in tree_decode(treefile):
1379 for blob in self.join(hexlify(sha)):
1381 elif typ == b'commit':
1382 treeline = b''.join(it).split(b'\n')[0]
1383 assert treeline.startswith(b'tree ')
1384 for blob in self.join(treeline[5:]):
1387 raise GitError('invalid object type %r: expected blob/tree/commit'
1391 """Generate a list of the content of all blobs that can be reached
1392 from an object. The hash given in 'id' must point to a blob, a tree
1393 or a commit. The content of all blobs that can be seen from trees or
1394 commits will be added to the list.
1396 for d in self._join(self.get(id)):
1402 def cp(repo_dir=None):
1403 """Create a CatPipe object or reuse the already existing one."""
1406 repo_dir = repodir or repo()
1407 repo_dir = os.path.abspath(repo_dir)
1408 cp = _cp.get(repo_dir)
1410 cp = CatPipe(repo_dir)
1415 def close_catpipes():
1416 # FIXME: chain exceptions
1418 _, cp = _cp.popitem()
1422 def tags(repo_dir = None):
1423 """Return a dictionary of all tags in the form {hash: [tag_names, ...]}."""
1425 for n, c in list_refs(repo_dir = repo_dir, limit_to_tags=True):
1426 assert n.startswith(b'refs/tags/')
1430 tags[c].append(name) # more than one tag can point at 'c'
1434 class MissingObject(KeyError):
1435 def __init__(self, oid):
1437 KeyError.__init__(self, 'object %r is missing' % hexlify(oid))
1440 WalkItem = namedtuple('WalkItem', ['oid', 'type', 'mode',
1441 'path', 'chunk_path', 'data'])
1442 # The path is the mangled path, and if an item represents a fragment
1443 # of a chunked file, the chunk_path will be the chunked subtree path
1444 # for the chunk, i.e. ['', '2d3115e', ...]. The top-level path for a
1445 # chunked file will have a chunk_path of ['']. So some chunk subtree
1446 # of the file '/foo/bar/baz' might look like this:
1448 # item.path = ['foo', 'bar', 'baz.bup']
1449 # item.chunk_path = ['', '2d3115e', '016b097']
1450 # item.type = 'tree'
1454 def walk_object(get_ref, oidx, stop_at=None, include_data=None):
1455 """Yield everything reachable from oidx via get_ref (which must behave
1456 like CatPipe get) as a WalkItem, stopping whenever stop_at(oidx)
1457 returns true. Throw MissingObject if a hash encountered is
1458 missing from the repository, and don't read or return blob content
1459 in the data field unless include_data is set.
1462 # Maintain the pending stack on the heap to avoid stack overflow
1463 pending = [(oidx, [], [], None)]
1465 oidx, parent_path, chunk_path, mode = pending.pop()
1466 oid = unhexlify(oidx)
1467 if stop_at and stop_at(oidx):
1470 if (not include_data) and mode and stat.S_ISREG(mode):
1471 # If the object is a "regular file", then it's a leaf in
1472 # the graph, so we can skip reading the data if the caller
1473 # hasn't requested it.
1474 yield WalkItem(oid=oid, type=b'blob',
1475 chunk_path=chunk_path, path=parent_path,
1480 item_it = get_ref(oidx)
1481 get_oidx, typ, _ = next(item_it)
1483 raise MissingObject(unhexlify(oidx))
1484 if typ not in (b'blob', b'commit', b'tree'):
1485 raise Exception('unexpected repository object type %r' % typ)
1487 # FIXME: set the mode based on the type when the mode is None
1488 if typ == b'blob' and not include_data:
1489 # Dump data until we can ask cat_pipe not to fetch it
1490 for ignored in item_it:
1494 data = b''.join(item_it)
1496 yield WalkItem(oid=oid, type=typ,
1497 chunk_path=chunk_path, path=parent_path,
1499 data=(data if include_data else None))
1501 if typ == b'commit':
1502 commit_items = parse_commit(data)
1503 for pid in commit_items.parents:
1504 pending.append((pid, parent_path, chunk_path, mode))
1505 pending.append((commit_items.tree, parent_path, chunk_path,
1506 hashsplit.GIT_MODE_TREE))
1507 elif typ == b'tree':
1508 for mode, name, ent_id in tree_decode(data):
1509 demangled, bup_type = demangle_name(name, mode)
1511 sub_path = parent_path
1512 sub_chunk_path = chunk_path + [name]
1514 sub_path = parent_path + [name]
1515 if bup_type == BUP_CHUNKED:
1516 sub_chunk_path = [b'']
1518 sub_chunk_path = chunk_path
1519 pending.append((hexlify(ent_id), sub_path, sub_chunk_path,