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,
21 from bup.io import path_msg
22 from bup.helpers import (Sha1, add_error, chunkyreader, debug1, debug2,
29 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 items(_typemap)}
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:]))
377 def find_offset(self, hash):
378 """Get the offset of an object inside the index file."""
379 idx = self._idx_from_hash(hash)
381 return self._ofs_from_idx(idx)
384 def exists(self, hash, want_source=False):
385 """Return nonempty if the object exists in this index."""
386 if hash and (self._idx_from_hash(hash) != None):
387 return want_source and os.path.basename(self.name) or True
390 def _idx_from_hash(self, hash):
391 global _total_searches, _total_steps
393 assert(len(hash) == 20)
394 b1 = byte_int(hash[0])
395 start = self.fanout[b1-1] # range -1..254
396 end = self.fanout[b1] # range 0..255
398 _total_steps += 1 # lookup table is a step
401 mid = start + (end - start) // 2
402 v = self._idx_to_hash(mid)
412 class PackIdxV1(PackIdx):
413 """Object representation of a Git pack index (version 1) file."""
414 def __init__(self, filename, f):
416 self.idxnames = [self.name]
417 self.map = mmap_read(f)
418 # Min size for 'L' is 4, which is sufficient for struct's '!I'
419 self.fanout = array('L', struct.unpack('!256I', self.map))
420 self.fanout.append(0) # entry "-1"
421 self.nsha = self.fanout[255]
422 self.sha_ofs = 256 * 4
423 # Avoid slicing shatable for individual hashes (very high overhead)
424 self.shatable = buffer(self.map, self.sha_ofs, self.nsha * 24)
429 def __exit__(self, type, value, traceback):
430 with pending_raise(value, rethrow=False):
434 return int(self.nsha) # int() from long for python 2
436 def _ofs_from_idx(self, idx):
437 if idx >= self.nsha or idx < 0:
438 raise IndexError('invalid pack index index %d' % idx)
439 ofs = self.sha_ofs + idx * 24
440 return struct.unpack_from('!I', self.map, offset=ofs)[0]
442 def _idx_to_hash(self, idx):
443 if idx >= self.nsha or idx < 0:
444 raise IndexError('invalid pack index index %d' % idx)
445 ofs = self.sha_ofs + idx * 24 + 4
446 return self.map[ofs : ofs + 20]
449 start = self.sha_ofs + 4
450 for ofs in range(start, start + 24 * self.nsha, 24):
451 yield self.map[ofs : ofs + 20]
454 if self.map is not None:
460 class PackIdxV2(PackIdx):
461 """Object representation of a Git pack index (version 2) file."""
462 def __init__(self, filename, f):
464 self.idxnames = [self.name]
465 self.map = mmap_read(f)
466 assert self.map[0:8] == b'\377tOc\0\0\0\2'
467 # Min size for 'L' is 4, which is sufficient for struct's '!I'
468 self.fanout = array('L', struct.unpack_from('!256I', self.map, offset=8))
469 self.fanout.append(0)
470 self.nsha = self.fanout[255]
471 self.sha_ofs = 8 + 256*4
472 self.ofstable_ofs = self.sha_ofs + self.nsha * 20 + self.nsha * 4
473 self.ofs64table_ofs = self.ofstable_ofs + self.nsha * 4
474 # Avoid slicing this for individual hashes (very high overhead)
475 self.shatable = buffer(self.map, self.sha_ofs, self.nsha*20)
480 def __exit__(self, type, value, traceback):
481 with pending_raise(value, rethrow=False):
485 return int(self.nsha) # int() from long for python 2
487 def _ofs_from_idx(self, idx):
488 if idx >= self.nsha or idx < 0:
489 raise IndexError('invalid pack index index %d' % idx)
490 ofs_ofs = self.ofstable_ofs + idx * 4
491 ofs = struct.unpack_from('!I', self.map, offset=ofs_ofs)[0]
493 idx64 = ofs & 0x7fffffff
494 ofs64_ofs = self.ofs64table_ofs + idx64 * 8
495 ofs = struct.unpack_from('!Q', self.map, offset=ofs64_ofs)[0]
498 def _idx_to_hash(self, idx):
499 if idx >= self.nsha or idx < 0:
500 raise IndexError('invalid pack index index %d' % idx)
501 ofs = self.sha_ofs + idx * 20
502 return self.map[ofs : ofs + 20]
506 for ofs in range(start, start + 20 * self.nsha, 20):
507 yield self.map[ofs : ofs + 20]
510 if self.map is not None:
518 def __init__(self, dir, ignore_midx=False):
520 assert(_mpi_count == 0) # these things suck tons of VM; don't waste it
525 self.do_bloom = False
527 self.ignore_midx = ignore_midx
533 assert(_mpi_count == 0)
536 return iter(idxmerge(self.packs))
539 return sum(len(pack) for pack in self.packs)
541 def exists(self, hash, want_source=False):
542 """Return nonempty if the object exists in the index files."""
543 global _total_searches
545 if hash in self.also:
547 if self.do_bloom and self.bloom:
548 if self.bloom.exists(hash):
549 self.do_bloom = False
551 _total_searches -= 1 # was counted by bloom
553 for i in range(len(self.packs)):
555 _total_searches -= 1 # will be incremented by sub-pack
556 ix = p.exists(hash, want_source=want_source)
558 # reorder so most recently used packs are searched first
559 self.packs = [p] + self.packs[:i] + self.packs[i+1:]
564 def refresh(self, skip_midx = False):
565 """Refresh the index list.
566 This method verifies if .midx files were superseded (e.g. all of its
567 contents are in another, bigger .midx file) and removes the superseded
570 If skip_midx is True, all work on .midx files will be skipped and .midx
571 files will be removed from the list.
573 The instance variable 'ignore_midx' can force this function to
574 always act as if skip_midx was True.
576 if self.bloom is not None:
578 self.bloom = None # Always reopen the bloom as it may have been relaced
579 self.do_bloom = False
580 skip_midx = skip_midx or self.ignore_midx
581 d = dict((p.name, p) for p in self.packs
582 if not skip_midx or not isinstance(p, midx.PackMidx))
583 if os.path.exists(self.dir):
586 midxes = set(glob.glob(os.path.join(self.dir, b'*.midx')))
587 # remove any *.midx files from our list that no longer exist
588 for ix in list(d.values()):
589 if not isinstance(ix, midx.PackMidx):
591 if ix.name in midxes:
596 self.packs.remove(ix)
597 for ix in self.packs:
598 if isinstance(ix, midx.PackMidx):
599 for name in ix.idxnames:
600 d[os.path.join(self.dir, name)] = ix
603 mx = midx.PackMidx(full)
604 (mxd, mxf) = os.path.split(mx.name)
606 for n in mx.idxnames:
607 if not os.path.exists(os.path.join(mxd, n)):
608 log(('warning: index %s missing\n'
610 % (path_msg(n), path_msg(mxf)))
618 midxl.sort(key=lambda ix:
619 (-len(ix), -xstat.stat(ix.name).st_mtime))
622 for sub in ix.idxnames:
623 found = d.get(os.path.join(self.dir, sub))
624 if not found or isinstance(found, PackIdx):
625 # doesn't exist, or exists but not in a midx
630 for name in ix.idxnames:
631 d[os.path.join(self.dir, name)] = ix
632 elif not ix.force_keep:
633 debug1('midx: removing redundant: %s\n'
634 % path_msg(os.path.basename(ix.name)))
637 for full in glob.glob(os.path.join(self.dir, b'*.idx')):
641 except GitError as e:
645 bfull = os.path.join(self.dir, b'bup.bloom')
646 if self.bloom is None and os.path.exists(bfull):
647 self.bloom = bloom.ShaBloom(bfull)
648 self.packs = list(set(d.values()))
649 self.packs.sort(reverse=True, key=lambda x: len(x))
650 if self.bloom and self.bloom.valid() and len(self.bloom) >= len(self):
654 debug1('PackIdxList: using %d index%s.\n'
655 % (len(self.packs), len(self.packs)!=1 and 'es' or ''))
658 """Insert an additional object in the list."""
662 def open_idx(filename):
663 if filename.endswith(b'.idx'):
664 f = open(filename, 'rb')
666 if header[0:4] == b'\377tOc':
667 version = struct.unpack('!I', header[4:8])[0]
669 return PackIdxV2(filename, f)
671 raise GitError('%s: expected idx file version 2, got %d'
672 % (path_msg(filename), version))
673 elif len(header) == 8 and header[0:4] < b'\377tOc':
674 return PackIdxV1(filename, f)
676 raise GitError('%s: unrecognized idx file header'
677 % path_msg(filename))
678 elif filename.endswith(b'.midx'):
679 return midx.PackMidx(filename)
681 raise GitError('idx filenames must end with .idx or .midx')
684 def idxmerge(idxlist, final_progress=True):
685 """Generate a list of all the objects reachable in a PackIdxList."""
686 def pfunc(count, total):
687 qprogress('Reading indexes: %.2f%% (%d/%d)\r'
688 % (count*100.0/total, count, total))
689 def pfinal(count, total):
691 progress('Reading indexes: %.2f%% (%d/%d), done.\n'
692 % (100, total, total))
693 return merge_iter(idxlist, 10024, pfunc, pfinal)
696 def create_commit_blob(tree, parent,
697 author, adate_sec, adate_tz,
698 committer, cdate_sec, cdate_tz,
700 if adate_tz is not None:
701 adate_str = _git_date_str(adate_sec, adate_tz)
703 adate_str = _local_git_date_str(adate_sec)
704 if cdate_tz is not None:
705 cdate_str = _git_date_str(cdate_sec, cdate_tz)
707 cdate_str = _local_git_date_str(cdate_sec)
709 if tree: l.append(b'tree %s' % hexlify(tree))
710 if parent: l.append(b'parent %s' % hexlify(parent))
711 if author: l.append(b'author %s %s' % (author, adate_str))
712 if committer: l.append(b'committer %s %s' % (committer, cdate_str))
718 def _make_objcache():
719 return PackIdxList(repo(b'objects/pack'))
721 # bup-gc assumes that it can disable all PackWriter activities
722 # (bloom/midx/cache) via the constructor and close() arguments.
725 """Writes Git objects inside a pack file."""
726 def __init__(self, objcache_maker=_make_objcache, compression_level=1,
727 run_midx=True, on_pack_finish=None,
728 max_pack_size=None, max_pack_objects=None, repo_dir=None):
729 self.repo_dir = repo_dir or repo()
736 self.objcache_maker = objcache_maker
738 self.compression_level = compression_level
739 self.run_midx=run_midx
740 self.on_pack_finish = on_pack_finish
741 if not max_pack_size:
742 max_pack_size = git_config_get(b'pack.packSizeLimit',
743 repo_dir=self.repo_dir,
745 if not max_pack_size:
746 # larger packs slow down pruning
747 max_pack_size = 1000 * 1000 * 1000
748 self.max_pack_size = max_pack_size
749 # cache memory usage is about 83 bytes per object
750 self.max_pack_objects = max_pack_objects if max_pack_objects \
751 else max(1, self.max_pack_size // 5000)
759 def __exit__(self, type, value, traceback):
760 with pending_raise(value, rethrow=False):
765 objdir = dir = os.path.join(self.repo_dir, b'objects')
766 fd, name = tempfile.mkstemp(suffix=b'.pack', dir=objdir)
768 self.file = os.fdopen(fd, 'w+b')
773 self.parentfd = os.open(objdir, os.O_RDONLY)
779 assert name.endswith(b'.pack')
780 self.filename = name[:-5]
781 self.file.write(b'PACK\0\0\0\2\0\0\0\0')
782 self.idx = PackIdxV2Writer()
784 def _raw_write(self, datalist, sha):
787 # in case we get interrupted (eg. KeyboardInterrupt), it's best if
788 # the file never has a *partial* blob. So let's make sure it's
789 # all-or-nothing. (The blob shouldn't be very big anyway, thanks
790 # to our hashsplit algorithm.) f.write() does its own buffering,
791 # but that's okay because we'll flush it in _end().
792 oneblob = b''.join(datalist)
798 crc = zlib.crc32(oneblob) & 0xffffffff
799 self._update_idx(sha, crc, nw)
804 def _update_idx(self, sha, crc, size):
807 self.idx.add(sha, crc, self.file.tell() - size)
809 def _write(self, sha, type, content):
813 sha = calc_hash(type, content)
814 size, crc = self._raw_write(_encode_packobj(type, content,
815 self.compression_level),
817 if self.outbytes >= self.max_pack_size \
818 or self.count >= self.max_pack_objects:
822 def _require_objcache(self):
823 if self.objcache is None and self.objcache_maker:
824 self.objcache = self.objcache_maker()
825 if self.objcache is None:
827 "PackWriter not opened or can't check exists w/o objcache")
829 def exists(self, id, want_source=False):
830 """Return non-empty if an object is found in the object cache."""
831 self._require_objcache()
832 return self.objcache.exists(id, want_source=want_source)
834 def just_write(self, sha, type, content):
835 """Write an object to the pack file without checking for duplication."""
836 self._write(sha, type, content)
837 # If nothing else, gc doesn't have/want an objcache
838 if self.objcache is not None:
839 self.objcache.add(sha)
841 def maybe_write(self, type, content):
842 """Write an object to the pack file if not present and return its id."""
843 sha = calc_hash(type, content)
844 if not self.exists(sha):
845 self._require_objcache()
846 self.just_write(sha, type, content)
849 def new_blob(self, blob):
850 """Create a blob object in the pack with the supplied content."""
851 return self.maybe_write(b'blob', blob)
853 def new_tree(self, shalist):
854 """Create a tree object in the pack."""
855 content = tree_encode(shalist)
856 return self.maybe_write(b'tree', content)
858 def new_commit(self, tree, parent,
859 author, adate_sec, adate_tz,
860 committer, cdate_sec, cdate_tz,
862 """Create a commit object in the pack. The date_sec values must be
863 epoch-seconds, and if a tz is None, the local timezone is assumed."""
864 content = create_commit_blob(tree, parent,
865 author, adate_sec, adate_tz,
866 committer, cdate_sec, cdate_tz,
868 return self.maybe_write(b'commit', content)
870 def _end(self, run_midx=True, abort=False):
871 # Ignores run_midx during abort
874 self.file, f = None, self.file
875 self.idx, idx = None, self.idx
876 self.parentfd, pfd, = None, self.parentfd
879 with finalized(pfd, lambda x: x is not None and os.close(x)), \
883 os.unlink(self.filename + b'.pack')
886 # update object count
888 cp = struct.pack('!i', self.count)
892 # calculate the pack sha1sum
895 for b in chunkyreader(f):
897 packbin = sum.digest()
900 fdatasync(f.fileno())
903 idx.write(self.filename + b'.idx', packbin)
904 nameprefix = os.path.join(self.repo_dir,
905 b'objects/pack/pack-' + hexlify(packbin))
906 if os.path.exists(self.filename + b'.map'):
907 os.unlink(self.filename + b'.map')
908 os.rename(self.filename + b'.pack', nameprefix + b'.pack')
909 os.rename(self.filename + b'.idx', nameprefix + b'.idx')
912 auto_midx(os.path.join(self.repo_dir, b'objects/pack'))
913 if self.on_pack_finish:
914 self.on_pack_finish(nameprefix)
918 """Remove the pack file from disk."""
919 self._end(abort=True)
921 def breakpoint(self):
922 """Clear byte and object counts and return the last processed id."""
923 id = self._end(self.run_midx)
924 self.outbytes = self.count = 0
927 def close(self, run_midx=True):
928 """Close the pack file and move it to its definitive path."""
929 return self._end(run_midx=run_midx)
932 class PackIdxV2Writer:
934 self.idx = list(list() for i in range(256))
937 def add(self, sha, crc, offs):
940 self.idx[byte_int(sha[0])].append((sha, crc, offs))
942 def write(self, filename, packbin):
944 for section in self.idx:
945 for entry in section:
946 if entry[2] >= 2**31:
949 # Length: header + fan-out + shas-and-crcs + overflow-offsets
950 index_len = 8 + (4 * 256) + (28 * self.count) + (8 * ofs64_count)
952 idx_f = open(filename, 'w+b')
954 idx_f.truncate(index_len)
955 fdatasync(idx_f.fileno())
956 idx_map = mmap_readwrite(idx_f, close=False)
958 count = _helpers.write_idx(filename, idx_map, self.idx,
960 assert(count == self.count)
967 idx_f = open(filename, 'a+b')
972 b = idx_f.read(8 + 4*256)
975 for b in chunkyreader(idx_f, 20 * self.count):
978 for b in chunkyreader(idx_f):
980 idx_f.write(idx_sum.digest())
981 fdatasync(idx_f.fileno())
986 def list_refs(patterns=None, repo_dir=None,
987 limit_to_heads=False, limit_to_tags=False):
988 """Yield (refname, hash) tuples for all repository refs unless
989 patterns are specified. In that case, only include tuples for
990 refs matching those patterns (cf. git-show-ref(1)). The limits
991 restrict the result items to refs/heads or refs/tags. If both
992 limits are specified, items from both sources will be included.
995 argv = [b'git', b'show-ref']
997 argv.append(b'--heads')
999 argv.append(b'--tags')
1002 argv.extend(patterns)
1003 p = subprocess.Popen(argv, env=_gitenv(repo_dir), stdout=subprocess.PIPE,
1005 out = p.stdout.read().strip()
1006 rv = p.wait() # not fatal
1010 for d in out.split(b'\n'):
1011 sha, name = d.split(b' ', 1)
1012 yield name, unhexlify(sha)
1015 def read_ref(refname, repo_dir = None):
1016 """Get the commit id of the most recent commit made on a given ref."""
1017 refs = list_refs(patterns=[refname], repo_dir=repo_dir, limit_to_heads=True)
1018 l = tuple(islice(refs, 2))
1026 def rev_list_invocation(ref_or_refs, format=None):
1027 if isinstance(ref_or_refs, bytes):
1028 refs = (ref_or_refs,)
1031 argv = [b'git', b'rev-list']
1034 argv.append(b'--pretty=format:' + format)
1036 assert not ref.startswith(b'-')
1042 def rev_list(ref_or_refs, parse=None, format=None, repo_dir=None):
1043 """Yield information about commits as per "git rev-list". If a format
1044 is not provided, yield one hex hash at a time. If a format is
1045 provided, pass it to rev-list and call parse(git_stdout) for each
1046 commit with the stream positioned just after the rev-list "commit
1047 HASH" header line. When a format is provided yield (oidx,
1048 parse(git_stdout)) for each commit.
1051 assert bool(parse) == bool(format)
1052 p = subprocess.Popen(rev_list_invocation(ref_or_refs,
1054 env=_gitenv(repo_dir),
1055 stdout = subprocess.PIPE,
1058 for line in p.stdout:
1061 line = p.stdout.readline()
1064 if not s.startswith(b'commit '):
1065 raise Exception('unexpected line ' + repr(s))
1068 yield s, parse(p.stdout)
1069 line = p.stdout.readline()
1071 rv = p.wait() # not fatal
1073 raise GitError('git rev-list returned error %d' % rv)
1076 def rev_parse(committish, repo_dir=None):
1077 """Resolve the full hash for 'committish', if it exists.
1079 Should be roughly equivalent to 'git rev-parse'.
1081 Returns the hex value of the hash if it is found, None if 'committish' does
1082 not correspond to anything.
1084 head = read_ref(committish, repo_dir=repo_dir)
1086 debug2("resolved from ref: commit = %s\n" % hexlify(head))
1089 pL = PackIdxList(repo(b'objects/pack', repo_dir=repo_dir))
1091 if len(committish) == 40:
1093 hash = unhexlify(committish)
1103 def update_ref(refname, newval, oldval, repo_dir=None):
1104 """Update a repository reference."""
1107 assert refname.startswith(b'refs/heads/') \
1108 or refname.startswith(b'refs/tags/')
1109 p = subprocess.Popen([b'git', b'update-ref', refname,
1110 hexlify(newval), hexlify(oldval)],
1111 env=_gitenv(repo_dir),
1113 _git_wait(b'git update-ref', p)
1116 def delete_ref(refname, oldvalue=None):
1117 """Delete a repository reference (see git update-ref(1))."""
1118 assert refname.startswith(b'refs/')
1119 oldvalue = [] if not oldvalue else [oldvalue]
1120 p = subprocess.Popen([b'git', b'update-ref', b'-d', refname] + oldvalue,
1123 _git_wait('git update-ref', p)
1126 def guess_repo(path=None):
1127 """Set the path value in the global variable "repodir".
1128 This makes bup look for an existing bup repository, but not fail if a
1129 repository doesn't exist. Usually, if you are interacting with a bup
1130 repository, you would not be calling this function but using
1131 check_repo_or_die().
1137 repodir = environ.get(b'BUP_DIR')
1139 repodir = os.path.expanduser(b'~/.bup')
1142 def init_repo(path=None):
1143 """Create the Git bare repository for bup in a given path."""
1145 d = repo() # appends a / to the path
1146 parent = os.path.dirname(os.path.dirname(d))
1147 if parent and not os.path.exists(parent):
1148 raise GitError('parent directory "%s" does not exist\n'
1150 if os.path.exists(d) and not os.path.isdir(os.path.join(d, b'.')):
1151 raise GitError('"%s" exists but is not a directory\n' % path_msg(d))
1152 p = subprocess.Popen([b'git', b'--bare', b'init'], stdout=sys.stderr,
1155 _git_wait('git init', p)
1156 # Force the index version configuration in order to ensure bup works
1157 # regardless of the version of the installed Git binary.
1158 p = subprocess.Popen([b'git', b'config', b'pack.indexVersion', '2'],
1159 stdout=sys.stderr, env=_gitenv(), close_fds=True)
1160 _git_wait('git config', p)
1162 p = subprocess.Popen([b'git', b'config', b'core.logAllRefUpdates', b'true'],
1163 stdout=sys.stderr, env=_gitenv(), close_fds=True)
1164 _git_wait('git config', p)
1167 def check_repo_or_die(path=None):
1168 """Check to see if a bup repository probably exists, and abort if not."""
1171 pst = stat_if_exists(top + b'/objects/pack')
1172 if pst and stat.S_ISDIR(pst.st_mode):
1175 top_st = stat_if_exists(top)
1177 log('error: repository %r does not exist (see "bup help init")\n'
1180 log('error: %s is not a repository\n' % path_msg(top))
1184 def is_suitable_git(ver_str):
1185 if not ver_str.startswith(b'git version '):
1186 return 'unrecognized'
1187 ver_str = ver_str[len(b'git version '):]
1188 if ver_str.startswith(b'0.'):
1189 return 'insufficient'
1190 if ver_str.startswith(b'1.'):
1191 if re.match(br'1\.[012345]rc', ver_str):
1192 return 'insufficient'
1193 if re.match(br'1\.[01234]\.', ver_str):
1194 return 'insufficient'
1195 if re.match(br'1\.5\.[012345]($|\.)', ver_str):
1196 return 'insufficient'
1197 if re.match(br'1\.5\.6-rc', ver_str):
1198 return 'insufficient'
1200 if re.match(br'[0-9]+(\.|$)?', ver_str):
1206 def require_suitable_git(ver_str=None):
1207 """Raise GitError if the version of git isn't suitable.
1209 Rely on ver_str when provided, rather than invoking the git in the
1214 if _git_great is not None:
1216 if environ.get(b'BUP_GIT_VERSION_IS_FINE', b'').lower() \
1217 in (b'yes', b'true', b'1'):
1221 ver_str, _, _ = _git_exo([b'git', b'--version'])
1222 status = is_suitable_git(ver_str)
1223 if status == 'unrecognized':
1224 raise GitError('Unexpected git --version output: %r' % ver_str)
1225 if status == 'insufficient':
1226 log('error: git version must be at least 1.5.6\n')
1228 if status == 'suitable':
1234 class _AbortableIter:
1235 def __init__(self, it, onabort = None):
1237 self.onabort = onabort
1245 return next(self.it)
1246 except StopIteration as e:
1256 """Abort iteration and call the abortion callback, if needed."""
1267 """Link to 'git cat-file' that is used to retrieve blob data."""
1268 def __init__(self, repo_dir = None):
1269 require_suitable_git()
1270 self.repo_dir = repo_dir
1271 self.p = self.inprogress = None
1273 def close(self, wait=False):
1279 self.inprogress = None
1287 self.p = subprocess.Popen([b'git', b'cat-file', b'--batch'],
1288 stdin=subprocess.PIPE,
1289 stdout=subprocess.PIPE,
1292 env=_gitenv(self.repo_dir))
1295 """Yield (oidx, type, size), followed by the data referred to by ref.
1296 If ref does not exist, only yield (None, None, None).
1299 if not self.p or self.p.poll() != None:
1302 poll_result = self.p.poll()
1303 assert(poll_result == None)
1305 log('get: opening %r while %r is open\n' % (ref, self.inprogress))
1306 assert(not self.inprogress)
1307 assert ref.find(b'\n') < 0
1308 assert ref.find(b'\r') < 0
1309 assert not ref.startswith(b'-')
1310 self.inprogress = ref
1311 self.p.stdin.write(ref + b'\n')
1312 self.p.stdin.flush()
1313 hdr = self.p.stdout.readline()
1315 raise GitError('unexpected cat-file EOF (last request: %r, exit: %s)'
1316 % (ref, self.p.poll() or 'none'))
1317 if hdr.endswith(b' missing\n'):
1318 self.inprogress = None
1319 yield None, None, None
1321 info = hdr.split(b' ')
1322 if len(info) != 3 or len(info[0]) != 40:
1323 raise GitError('expected object (id, type, size), got %r' % info)
1324 oidx, typ, size = info
1326 it = _AbortableIter(chunkyreader(self.p.stdout, size),
1329 yield oidx, typ, size
1332 readline_result = self.p.stdout.readline()
1333 assert readline_result == b'\n'
1334 self.inprogress = None
1335 except Exception as e:
1339 def _join(self, it):
1340 _, typ, _ = next(it)
1344 elif typ == b'tree':
1345 treefile = b''.join(it)
1346 for (mode, name, sha) in tree_decode(treefile):
1347 for blob in self.join(hexlify(sha)):
1349 elif typ == b'commit':
1350 treeline = b''.join(it).split(b'\n')[0]
1351 assert treeline.startswith(b'tree ')
1352 for blob in self.join(treeline[5:]):
1355 raise GitError('invalid object type %r: expected blob/tree/commit'
1359 """Generate a list of the content of all blobs that can be reached
1360 from an object. The hash given in 'id' must point to a blob, a tree
1361 or a commit. The content of all blobs that can be seen from trees or
1362 commits will be added to the list.
1364 for d in self._join(self.get(id)):
1370 def cp(repo_dir=None):
1371 """Create a CatPipe object or reuse the already existing one."""
1374 repo_dir = repodir or repo()
1375 repo_dir = os.path.abspath(repo_dir)
1376 cp = _cp.get(repo_dir)
1378 cp = CatPipe(repo_dir)
1383 def close_catpipes():
1384 # FIXME: chain exceptions
1386 _, cp = _cp.popitem()
1390 def tags(repo_dir = None):
1391 """Return a dictionary of all tags in the form {hash: [tag_names, ...]}."""
1393 for n, c in list_refs(repo_dir = repo_dir, limit_to_tags=True):
1394 assert n.startswith(b'refs/tags/')
1398 tags[c].append(name) # more than one tag can point at 'c'
1402 class MissingObject(KeyError):
1403 def __init__(self, oid):
1405 KeyError.__init__(self, 'object %r is missing' % hexlify(oid))
1408 WalkItem = namedtuple('WalkItem', ['oid', 'type', 'mode',
1409 'path', 'chunk_path', 'data'])
1410 # The path is the mangled path, and if an item represents a fragment
1411 # of a chunked file, the chunk_path will be the chunked subtree path
1412 # for the chunk, i.e. ['', '2d3115e', ...]. The top-level path for a
1413 # chunked file will have a chunk_path of ['']. So some chunk subtree
1414 # of the file '/foo/bar/baz' might look like this:
1416 # item.path = ['foo', 'bar', 'baz.bup']
1417 # item.chunk_path = ['', '2d3115e', '016b097']
1418 # item.type = 'tree'
1422 def walk_object(get_ref, oidx, stop_at=None, include_data=None):
1423 """Yield everything reachable from oidx via get_ref (which must behave
1424 like CatPipe get) as a WalkItem, stopping whenever stop_at(oidx)
1425 returns true. Throw MissingObject if a hash encountered is
1426 missing from the repository, and don't read or return blob content
1427 in the data field unless include_data is set.
1430 # Maintain the pending stack on the heap to avoid stack overflow
1431 pending = [(oidx, [], [], None)]
1433 oidx, parent_path, chunk_path, mode = pending.pop()
1434 oid = unhexlify(oidx)
1435 if stop_at and stop_at(oidx):
1438 if (not include_data) and mode and stat.S_ISREG(mode):
1439 # If the object is a "regular file", then it's a leaf in
1440 # the graph, so we can skip reading the data if the caller
1441 # hasn't requested it.
1442 yield WalkItem(oid=oid, type=b'blob',
1443 chunk_path=chunk_path, path=parent_path,
1448 item_it = get_ref(oidx)
1449 get_oidx, typ, _ = next(item_it)
1451 raise MissingObject(unhexlify(oidx))
1452 if typ not in (b'blob', b'commit', b'tree'):
1453 raise Exception('unexpected repository object type %r' % typ)
1455 # FIXME: set the mode based on the type when the mode is None
1456 if typ == b'blob' and not include_data:
1457 # Dump data until we can ask cat_pipe not to fetch it
1458 for ignored in item_it:
1462 data = b''.join(item_it)
1464 yield WalkItem(oid=oid, type=typ,
1465 chunk_path=chunk_path, path=parent_path,
1467 data=(data if include_data else None))
1469 if typ == b'commit':
1470 commit_items = parse_commit(data)
1471 for pid in commit_items.parents:
1472 pending.append((pid, parent_path, chunk_path, mode))
1473 pending.append((commit_items.tree, parent_path, chunk_path,
1474 hashsplit.GIT_MODE_TREE))
1475 elif typ == b'tree':
1476 for mode, name, ent_id in tree_decode(data):
1477 demangled, bup_type = demangle_name(name, mode)
1479 sub_path = parent_path
1480 sub_chunk_path = chunk_path + [name]
1482 sub_path = parent_path + [name]
1483 if bup_type == BUP_CHUNKED:
1484 sub_chunk_path = [b'']
1486 sub_chunk_path = chunk_path
1487 pending.append((hexlify(ent_id), sub_path, sub_chunk_path,