X-Git-Url: https://arthur.barton.de/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=lib%2Fbup%2Fgit.py;h=31ae34216e6dde4b6a6c375d4f02db883ce6d7b1;hb=53db6ee308661b43ee738443e0be23b1485c99be;hp=fc48e51ee8eee4122aa3ae175bc433827de188e5;hpb=9bb94e0f684e311f7a8c71a912b6d0e812a04e5f;p=bup.git diff --git a/lib/bup/git.py b/lib/bup/git.py index fc48e51..31ae342 100644 --- a/lib/bup/git.py +++ b/lib/bup/git.py @@ -2,16 +2,27 @@ bup repositories are in Git format. This library allows us to interact with the Git data structures. """ -import os, sys, zlib, time, subprocess, struct, stat, re, tempfile, heapq -from bup.helpers import * -from bup import _helpers, path -MIDX_VERSION = 2 +import errno, os, sys, zlib, time, subprocess, struct, stat, re, tempfile, glob +from collections import namedtuple +from itertools import islice + +from bup import _helpers, hashsplit, path, midx, bloom, xstat +from bup.helpers import (Sha1, add_error, chunkyreader, debug1, debug2, + fdatasync, + hostname, localtime, log, merge_iter, + mmap_read, mmap_readwrite, + progress, qprogress, stat_if_exists, + unlink, username, userfullname, + utc_offset_str) + + +max_pack_size = 1000*1000*1000 # larger packs will slow down pruning +max_pack_objects = 200*1000 # cache memory usage is about 83 bytes per object verbose = 0 ignore_midx = 0 -home_repodir = os.path.expanduser('~/.bup') -repodir = None +repodir = None # The default repository, once initialized _typemap = { 'blob':3, 'tree':2, 'commit':1, 'tag':4 } _typermap = { 3:'blob', 2:'tree', 1:'commit', 4:'tag' } @@ -24,25 +35,132 @@ class GitError(Exception): pass -def repo(sub = ''): +def parse_tz_offset(s): + """UTC offset in seconds.""" + tz_off = (int(s[1:3]) * 60 * 60) + (int(s[3:5]) * 60) + if s[0] == '-': + return - tz_off + return tz_off + + +# FIXME: derived from http://git.rsbx.net/Documents/Git_Data_Formats.txt +# Make sure that's authoritative. +_start_end_char = r'[^ .,:;<>"\'\0\n]' +_content_char = r'[^\0\n<>]' +_safe_str_rx = '(?:%s{1,2}|(?:%s%s*%s))' \ + % (_start_end_char, + _start_end_char, _content_char, _start_end_char) +_tz_rx = r'[-+]\d\d[0-5]\d' +_parent_rx = r'(?:parent [abcdefABCDEF0123456789]{40}\n)' +_commit_rx = re.compile(r'''tree (?P[abcdefABCDEF0123456789]{40}) +(?P%s*)author (?P%s) <(?P%s)> (?P\d+) (?P%s) +committer (?P%s) <(?P%s)> (?P\d+) (?P%s) + +(?P(?:.|\n)*)''' % (_parent_rx, + _safe_str_rx, _safe_str_rx, _tz_rx, + _safe_str_rx, _safe_str_rx, _tz_rx)) +_parent_hash_rx = re.compile(r'\s*parent ([abcdefABCDEF0123456789]{40})\s*') + + +# Note that the author_sec and committer_sec values are (UTC) epoch seconds. +CommitInfo = namedtuple('CommitInfo', ['tree', 'parents', + 'author_name', 'author_mail', + 'author_sec', 'author_offset', + 'committer_name', 'committer_mail', + 'committer_sec', 'committer_offset', + 'message']) + +def parse_commit(content): + commit_match = re.match(_commit_rx, content) + if not commit_match: + raise Exception('cannot parse commit %r' % content) + matches = commit_match.groupdict() + return CommitInfo(tree=matches['tree'], + parents=re.findall(_parent_hash_rx, matches['parents']), + author_name=matches['author_name'], + author_mail=matches['author_mail'], + author_sec=int(matches['asec']), + author_offset=parse_tz_offset(matches['atz']), + committer_name=matches['committer_name'], + committer_mail=matches['committer_mail'], + committer_sec=int(matches['csec']), + committer_offset=parse_tz_offset(matches['ctz']), + message=matches['message']) + + +def get_commit_items(id, cp): + commit_it = cp.get(id) + assert(commit_it.next() == 'commit') + commit_content = ''.join(commit_it) + return parse_commit(commit_content) + + +def _local_git_date_str(epoch_sec): + return '%d %s' % (epoch_sec, utc_offset_str(epoch_sec)) + + +def _git_date_str(epoch_sec, tz_offset_sec): + offs = tz_offset_sec // 60 + return '%d %s%02d%02d' \ + % (epoch_sec, + '+' if offs >= 0 else '-', + abs(offs) // 60, + abs(offs) % 60) + + +def repo(sub = '', repo_dir=None): """Get the path to the git repository or one of its subdirectories.""" global repodir - if not repodir: + repo_dir = repo_dir or repodir + if not repo_dir: raise GitError('You should call check_repo_or_die()') # If there's a .git subdirectory, then the actual repo is in there. - gd = os.path.join(repodir, '.git') + gd = os.path.join(repo_dir, '.git') if os.path.exists(gd): repodir = gd - return os.path.join(repodir, sub) + return os.path.join(repo_dir, sub) + + +def shorten_hash(s): + return re.sub(r'([^0-9a-z]|\b)([0-9a-z]{7})[0-9a-z]{33}([^0-9a-z]|\b)', + r'\1\2*\3', s) + + +def repo_rel(path): + full = os.path.abspath(path) + fullrepo = os.path.abspath(repo('')) + if not fullrepo.endswith('/'): + fullrepo += '/' + if full.startswith(fullrepo): + path = full[len(fullrepo):] + if path.startswith('index-cache/'): + path = path[len('index-cache/'):] + return shorten_hash(path) + + +def all_packdirs(): + paths = [repo('objects/pack')] + paths += glob.glob(repo('index-cache/*/.')) + return paths def auto_midx(objdir): args = [path.exe(), 'midx', '--auto', '--dir', objdir] try: rv = subprocess.call(args, stdout=open('/dev/null', 'w')) - except OSError, e: + except OSError as e: + # make sure 'args' gets printed to help with debugging + add_error('%r: exception: %s' % (args, e)) + raise + if rv: + add_error('%r: returned %d' % (args, rv)) + + args = [path.exe(), 'bloom', '--dir', objdir] + try: + rv = subprocess.call(args, stdout=open('/dev/null', 'w')) + except OSError as e: # make sure 'args' gets printed to help with debugging add_error('%r: exception: %s' % (args, e)) raise @@ -54,9 +172,10 @@ def mangle_name(name, mode, gitmode): """Mangle a file name to present an abstract name for segmented files. Mangled file names will have the ".bup" extension added to them. If a file's name already ends with ".bup", a ".bupl" extension is added to - disambiguate normal files from semgmented ones. + disambiguate normal files from segmented ones. """ if stat.S_ISREG(mode) and not stat.S_ISREG(gitmode): + assert(stat.S_ISDIR(gitmode)) return name + '.bup' elif name.endswith('.bup') or name[:-1].endswith('.bup'): return name + '.bupl' @@ -65,26 +184,77 @@ def mangle_name(name, mode, gitmode): (BUP_NORMAL, BUP_CHUNKED) = (0,1) -def demangle_name(name): +def demangle_name(name, mode): """Remove name mangling from a file name, if necessary. The return value is a tuple (demangled_filename,mode), where mode is one of the following: * BUP_NORMAL : files that should be read as-is from the repository - * BUP_CHUNKED : files that were chunked and need to be assembled + * BUP_CHUNKED : files that were chunked and need to be reassembled - For more information on the name mangling algorythm, see mangle_name() + For more information on the name mangling algorithm, see mangle_name() """ if name.endswith('.bupl'): return (name[:-5], BUP_NORMAL) elif name.endswith('.bup'): return (name[:-4], BUP_CHUNKED) + elif name.endswith('.bupm'): + return (name[:-5], + BUP_CHUNKED if stat.S_ISDIR(mode) else BUP_NORMAL) else: return (name, BUP_NORMAL) -def _encode_packobj(type, content): +def calc_hash(type, content): + """Calculate some content's hash in the Git fashion.""" + header = '%s %d\0' % (type, len(content)) + sum = Sha1(header) + sum.update(content) + return sum.digest() + + +def shalist_item_sort_key(ent): + (mode, name, id) = ent + assert(mode+0 == mode) + if stat.S_ISDIR(mode): + return name + '/' + else: + return name + + +def tree_encode(shalist): + """Generate a git tree object from (mode,name,hash) tuples.""" + shalist = sorted(shalist, key = shalist_item_sort_key) + l = [] + for (mode,name,bin) in shalist: + assert(mode) + assert(mode+0 == mode) + assert(name) + assert(len(bin) == 20) + s = '%o %s\0%s' % (mode,name,bin) + assert(s[0] != '0') # 0-padded octal is not acceptable in a git tree + l.append(s) + return ''.join(l) + + +def tree_decode(buf): + """Generate a list of (mode,name,hash) from the git tree object in buf.""" + ofs = 0 + while ofs < len(buf): + z = buf.find('\0', ofs) + assert(z > ofs) + spl = buf[ofs:z].split(' ', 1) + assert(len(spl) == 2) + mode,name = spl + sha = buf[z+1:z+1+20] + ofs = z+1+20 + yield (int(mode, 8), name, sha) + + +def _encode_packobj(type, content, compression_level=1): + if compression_level not in (0, 1, 2, 3, 4, 5, 6, 7, 8, 9): + raise ValueError('invalid compression level %s' % compression_level) szout = '' sz = len(content) szbits = (sz & 0x0f) | (_typemap[type]<<4) @@ -96,14 +266,14 @@ def _encode_packobj(type, content): break szbits = sz & 0x7f sz >>= 7 - z = zlib.compressobj(1) + z = zlib.compressobj(compression_level) yield szout yield z.compress(content) yield z.flush() -def _encode_looseobj(type, content): - z = zlib.compressobj(1) +def _encode_looseobj(type, content, compression_level=1): + z = zlib.compressobj(compression_level) yield z.compress('%s %d\0' % (type, len(content))) yield z.compress(content) yield z.flush() @@ -151,9 +321,11 @@ class PackIdx: return self._ofs_from_idx(idx) return None - def exists(self, hash): + def exists(self, hash, want_source=False): """Return nonempty if the object exists in this index.""" - return hash and (self._idx_from_hash(hash) != None) and True or None + if hash and (self._idx_from_hash(hash) != None): + return want_source and os.path.basename(self.name) or True + return None def __len__(self): return int(self.fanout[255]) @@ -190,7 +362,8 @@ class PackIdxV1(PackIdx): str(buffer(self.map, 0, 256*4)))) self.fanout.append(0) # entry "-1" nsha = self.fanout[255] - self.shatable = buffer(self.map, 256*4, nsha*24) + self.sha_ofs = 256*4 + self.shatable = buffer(self.map, self.sha_ofs, nsha*24) def _ofs_from_idx(self, idx): return struct.unpack('!I', str(self.shatable[idx*24 : idx*24+4]))[0] @@ -214,9 +387,10 @@ class PackIdxV2(PackIdx): str(buffer(self.map, 8, 256*4)))) self.fanout.append(0) # entry "-1" nsha = self.fanout[255] - self.shatable = buffer(self.map, 8 + 256*4, nsha*20) + self.sha_ofs = 8 + 256*4 + self.shatable = buffer(self.map, self.sha_ofs, nsha*20) self.ofstable = buffer(self.map, - 8 + 256*4 + nsha*20 + nsha*4, + self.sha_ofs + nsha*20 + nsha*4, nsha*4) self.ofs64table = buffer(self.map, 8 + 256*4 + nsha*20 + nsha*4 + nsha*4) @@ -237,101 +411,6 @@ class PackIdxV2(PackIdx): yield buffer(self.map, 8 + 256*4 + 20*i, 20) -extract_bits = _helpers.extract_bits - - -class PackMidx: - """Wrapper which contains data from multiple index files. - Multiple index (.midx) files constitute a wrapper around index (.idx) files - and make it possible for bup to expand Git's indexing capabilities to vast - amounts of files. - """ - def __init__(self, filename): - self.name = filename - self.force_keep = False - assert(filename.endswith('.midx')) - self.map = mmap_read(open(filename)) - if str(self.map[0:4]) != 'MIDX': - log('Warning: skipping: invalid MIDX header in %r\n' % filename) - self.force_keep = True - return self._init_failed() - ver = struct.unpack('!I', self.map[4:8])[0] - if ver < MIDX_VERSION: - log('Warning: ignoring old-style (v%d) midx %r\n' - % (ver, filename)) - self.force_keep = False # old stuff is boring - return self._init_failed() - if ver > MIDX_VERSION: - log('Warning: ignoring too-new (v%d) midx %r\n' - % (ver, filename)) - self.force_keep = True # new stuff is exciting - return self._init_failed() - - self.bits = _helpers.firstword(self.map[8:12]) - self.entries = 2**self.bits - self.fanout = buffer(self.map, 12, self.entries*4) - shaofs = 12 + self.entries*4 - nsha = self._fanget(self.entries-1) - self.shalist = buffer(self.map, shaofs, nsha*20) - self.idxnames = str(self.map[shaofs + 20*nsha:]).split('\0') - - def _init_failed(self): - self.bits = 0 - self.entries = 1 - self.fanout = buffer('\0\0\0\0') - self.shalist = buffer('\0'*20) - self.idxnames = [] - - def _fanget(self, i): - start = i*4 - s = self.fanout[start:start+4] - return _helpers.firstword(s) - - def _get(self, i): - return str(self.shalist[i*20:(i+1)*20]) - - def exists(self, hash): - """Return nonempty if the object exists in the index files.""" - global _total_searches, _total_steps - _total_searches += 1 - want = str(hash) - el = extract_bits(want, self.bits) - if el: - start = self._fanget(el-1) - startv = el << (32-self.bits) - else: - start = 0 - startv = 0 - end = self._fanget(el) - endv = (el+1) << (32-self.bits) - _total_steps += 1 # lookup table is a step - hashv = _helpers.firstword(hash) - #print '(%08x) %08x %08x %08x' % (extract_bits(want, 32), startv, hashv, endv) - while start < end: - _total_steps += 1 - #print '! %08x %08x %08x %d - %d' % (startv, hashv, endv, start, end) - mid = start + (hashv-startv)*(end-start-1)/(endv-startv) - #print ' %08x %08x %08x %d %d %d' % (startv, hashv, endv, start, mid, end) - v = self._get(mid) - #print ' %08x' % self._num(v) - if v < want: - start = mid+1 - startv = _helpers.firstword(v) - elif v > want: - end = mid - endv = _helpers.firstword(v) - else: # got it! - return True - return None - - def __iter__(self): - for i in xrange(self._fanget(self.entries-1)): - yield buffer(self.shalist, i*20, 20) - - def __len__(self): - return int(self._fanget(self.entries-1)) - - _mpi_count = 0 class PackIdxList: def __init__(self, dir): @@ -339,8 +418,10 @@ class PackIdxList: assert(_mpi_count == 0) # these things suck tons of VM; don't waste it _mpi_count += 1 self.dir = dir - self.also = {} + self.also = set() self.packs = [] + self.do_bloom = False + self.bloom = None self.refresh() def __del__(self): @@ -354,19 +435,27 @@ class PackIdxList: def __len__(self): return sum(len(pack) for pack in self.packs) - def exists(self, hash): + def exists(self, hash, want_source=False): """Return nonempty if the object exists in the index files.""" global _total_searches _total_searches += 1 if hash in self.also: return True - for i in range(len(self.packs)): + if self.do_bloom and self.bloom: + if self.bloom.exists(hash): + self.do_bloom = False + else: + _total_searches -= 1 # was counted by bloom + return None + for i in xrange(len(self.packs)): p = self.packs[i] _total_searches -= 1 # will be incremented by sub-pack - if p.exists(hash): + ix = p.exists(hash, want_source=want_source) + if ix: # reorder so most recently used packs are searched first self.packs = [p] + self.packs[:i] + self.packs[i+1:] - return p.name + return ix + self.do_bloom = True return None def refresh(self, skip_midx = False): @@ -381,100 +470,76 @@ class PackIdxList: The module-global variable 'ignore_midx' can force this function to always act as if skip_midx was True. """ + self.bloom = None # Always reopen the bloom as it may have been relaced + self.do_bloom = False skip_midx = skip_midx or ignore_midx d = dict((p.name, p) for p in self.packs - if not skip_midx or not isinstance(p, PackMidx)) + if not skip_midx or not isinstance(p, midx.PackMidx)) if os.path.exists(self.dir): if not skip_midx: midxl = [] for ix in self.packs: - if isinstance(ix, PackMidx): + if isinstance(ix, midx.PackMidx): for name in ix.idxnames: d[os.path.join(self.dir, name)] = ix - for f in os.listdir(self.dir): - full = os.path.join(self.dir, f) - if f.endswith('.midx') and not d.get(full): - mx = PackMidx(full) + for full in glob.glob(os.path.join(self.dir,'*.midx')): + if not d.get(full): + mx = midx.PackMidx(full) (mxd, mxf) = os.path.split(mx.name) - broken = 0 + broken = False for n in mx.idxnames: if not os.path.exists(os.path.join(mxd, n)): log(('warning: index %s missing\n' + ' used by %s\n') % (n, mxf)) - broken += 1 + broken = True if broken: + mx.close() del mx unlink(full) else: midxl.append(mx) - midxl.sort(lambda x,y: -cmp(len(x),len(y))) + midxl.sort(key=lambda ix: + (-len(ix), -xstat.stat(ix.name).st_mtime)) for ix in midxl: - any = 0 + any_needed = False for sub in ix.idxnames: found = d.get(os.path.join(self.dir, sub)) if not found or isinstance(found, PackIdx): # doesn't exist, or exists but not in a midx - d[ix.name] = ix - for name in ix.idxnames: - d[os.path.join(self.dir, name)] = ix - any += 1 + any_needed = True break - if not any and not ix.force_keep: + if any_needed: + d[ix.name] = ix + for name in ix.idxnames: + d[os.path.join(self.dir, name)] = ix + elif not ix.force_keep: debug1('midx: removing redundant: %s\n' % os.path.basename(ix.name)) + ix.close() unlink(ix.name) - for f in os.listdir(self.dir): - full = os.path.join(self.dir, f) - if f.endswith('.idx') and not d.get(full): + for full in glob.glob(os.path.join(self.dir,'*.idx')): + if not d.get(full): try: ix = open_idx(full) - except GitError, e: + except GitError as e: add_error(e) continue d[full] = ix + bfull = os.path.join(self.dir, 'bup.bloom') + if self.bloom is None and os.path.exists(bfull): + self.bloom = bloom.ShaBloom(bfull) self.packs = list(set(d.values())) + self.packs.sort(lambda x,y: -cmp(len(x),len(y))) + if self.bloom and self.bloom.valid() and len(self.bloom) >= len(self): + self.do_bloom = True + else: + self.bloom = None debug1('PackIdxList: using %d index%s.\n' % (len(self.packs), len(self.packs)!=1 and 'es' or '')) - def packname_containing(self, hash): - # figure out which pack contains a given hash. - # FIXME: if the midx file format would just *store* this information, - # we could calculate it a lot more efficiently. But it's not needed - # often, so let's do it like this. - for f in os.listdir(self.dir): - if f.endswith('.idx'): - full = os.path.join(self.dir, f) - try: - ix = open_idx(full) - except GitError, e: - add_error(e) - continue - if ix.exists(hash): - return full - def add(self, hash): """Insert an additional object in the list.""" - self.also[hash] = 1 - - def zap_also(self): - """Remove all additional objects from the list.""" - self.also = {} - - -def calc_hash(type, content): - """Calculate some content's hash in the Git fashion.""" - header = '%s %d\0' % (type, len(content)) - sum = Sha1(header) - sum.update(content) - return sum.digest() - - -def _shalist_sort_key(ent): - (mode, name, id) = ent - if stat.S_ISDIR(int(mode, 8)): - return name + '/' - else: - return name + self.also.add(hash) def open_idx(filename): @@ -493,65 +558,69 @@ def open_idx(filename): else: raise GitError('%s: unrecognized idx file header' % filename) elif filename.endswith('.midx'): - return PackMidx(filename) + return midx.PackMidx(filename) else: raise GitError('idx filenames must end with .idx or .midx') def idxmerge(idxlist, final_progress=True): """Generate a list of all the objects reachable in a PackIdxList.""" - total = sum(len(i) for i in idxlist) - iters = (iter(i) for i in idxlist) - heap = [(next(it), it) for it in iters] - heapq.heapify(heap) - count = 0 - last = None - while heap: - if (count % 10024) == 0: - progress('Reading indexes: %.2f%% (%d/%d)\r' - % (count*100.0/total, count, total)) - (e, it) = heap[0] - if e != last: - yield e - last = e - count += 1 - e = next(it) - if e: - heapq.heapreplace(heap, (e, it)) - else: - heapq.heappop(heap) - if final_progress: - log('Reading indexes: %.2f%% (%d/%d), done.\n' % (100, total, total)) + def pfunc(count, total): + qprogress('Reading indexes: %.2f%% (%d/%d)\r' + % (count*100.0/total, count, total)) + def pfinal(count, total): + if final_progress: + progress('Reading indexes: %.2f%% (%d/%d), done.\n' + % (100, total, total)) + return merge_iter(idxlist, 10024, pfunc, pfinal) def _make_objcache(): return PackIdxList(repo('objects/pack')) +# bup-gc assumes that it can disable all PackWriter activities +# (bloom/midx/cache) via the constructor and close() arguments. + class PackWriter: - """Writes Git objects insid a pack file.""" - def __init__(self, objcache_maker=_make_objcache): + """Writes Git objects inside a pack file.""" + def __init__(self, objcache_maker=_make_objcache, compression_level=1, + run_midx=True, on_pack_finish=None): + self.file = None + self.parentfd = None self.count = 0 self.outbytes = 0 self.filename = None - self.file = None self.idx = None self.objcache_maker = objcache_maker self.objcache = None + self.compression_level = compression_level + self.run_midx=run_midx + self.on_pack_finish = on_pack_finish def __del__(self): self.close() def _open(self): if not self.file: - (fd,name) = tempfile.mkstemp(suffix='.pack', dir=repo('objects')) - self.file = os.fdopen(fd, 'w+b') + objdir = dir=repo('objects') + fd, name = tempfile.mkstemp(suffix='.pack', dir=objdir) + try: + self.file = os.fdopen(fd, 'w+b') + except: + os.close(fd) + raise + try: + self.parentfd = os.open(objdir, os.O_RDONLY) + except: + f = self.file + self.file = None + f.close() + raise assert(name.endswith('.pack')) self.filename = name[:-5] self.file.write('PACK\0\0\0\2\0\0\0\0') self.idx = list(list() for i in xrange(256)) - # the 'sha' parameter is used in client.py's _raw_write(), but not needed - # in this basic version. def _raw_write(self, datalist, sha): self._open() f = self.file @@ -563,7 +632,7 @@ class PackWriter: oneblob = ''.join(datalist) try: f.write(oneblob) - except IOError, e: + except IOError as e: raise GitError, e, sys.exc_info()[2] nw = len(oneblob) crc = zlib.crc32(oneblob) & 0xffffffff @@ -582,19 +651,19 @@ class PackWriter: log('>') if not sha: sha = calc_hash(type, content) - size, crc = self._raw_write(_encode_packobj(type, content), sha=sha) + size, crc = self._raw_write(_encode_packobj(type, content, + self.compression_level), + sha=sha) + if self.outbytes >= max_pack_size or self.count >= max_pack_objects: + self.breakpoint() return sha def breakpoint(self): """Clear byte and object counts and return the last processed id.""" - id = self._end() + id = self._end(self.run_midx) self.outbytes = self.count = 0 return id - def write(self, type, content): - """Write an object in this pack file.""" - return self._write(calc_hash(type, content), type, content) - def _require_objcache(self): if self.objcache is None and self.objcache_maker: self.objcache = self.objcache_maker() @@ -602,17 +671,22 @@ class PackWriter: raise GitError( "PackWriter not opened or can't check exists w/o objcache") - def exists(self, id): + def exists(self, id, want_source=False): """Return non-empty if an object is found in the object cache.""" self._require_objcache() - return self.objcache.exists(id) + return self.objcache.exists(id, want_source=want_source) + + def just_write(self, sha, type, content): + """Write an object to the pack file, bypassing the objcache. Fails if + sha exists().""" + self._write(sha, type, content) def maybe_write(self, type, content): """Write an object to the pack file if not present and return its id.""" - self._require_objcache() sha = calc_hash(type, content) if not self.exists(sha): - self._write(sha, type, content) + self.just_write(sha, type, content) + self._require_objcache() self.objcache.add(sha) return sha @@ -622,140 +696,174 @@ class PackWriter: def new_tree(self, shalist): """Create a tree object in the pack.""" - shalist = sorted(shalist, key = _shalist_sort_key) - l = [] - for (mode,name,bin) in shalist: - assert(mode) - assert(mode != '0') - assert(mode[0] != '0') - assert(name) - assert(len(bin) == 20) - l.append('%s %s\0%s' % (mode,name,bin)) - return self.maybe_write('tree', ''.join(l)) - - def _new_commit(self, tree, parent, author, adate, committer, cdate, msg): + content = tree_encode(shalist) + return self.maybe_write('tree', content) + + def new_commit(self, tree, parent, + author, adate_sec, adate_tz, + committer, cdate_sec, cdate_tz, + msg): + """Create a commit object in the pack. The date_sec values must be + epoch-seconds, and if a tz is None, the local timezone is assumed.""" + if adate_tz: + adate_str = _git_date_str(adate_sec, adate_tz) + else: + adate_str = _local_git_date_str(adate_sec) + if cdate_tz: + cdate_str = _git_date_str(cdate_sec, cdate_tz) + else: + cdate_str = _local_git_date_str(cdate_sec) l = [] if tree: l.append('tree %s' % tree.encode('hex')) if parent: l.append('parent %s' % parent.encode('hex')) - if author: l.append('author %s %s' % (author, _git_date(adate))) - if committer: l.append('committer %s %s' % (committer, _git_date(cdate))) + if author: l.append('author %s %s' % (author, adate_str)) + if committer: l.append('committer %s %s' % (committer, cdate_str)) l.append('') l.append(msg) return self.maybe_write('commit', '\n'.join(l)) - def new_commit(self, parent, tree, date, msg): - """Create a commit object in the pack.""" - userline = '%s <%s@%s>' % (userfullname(), username(), hostname()) - commit = self._new_commit(tree, parent, - userline, date, userline, date, - msg) - return commit - def abort(self): """Remove the pack file from disk.""" f = self.file if f: - self.idx = None + pfd = self.parentfd self.file = None - f.close() - os.unlink(self.filename + '.pack') + self.parentfd = None + self.idx = None + try: + try: + os.unlink(self.filename + '.pack') + finally: + f.close() + finally: + if pfd is not None: + os.close(pfd) def _end(self, run_midx=True): f = self.file if not f: return None self.file = None - self.objcache = None - idx = self.idx - self.idx = None + try: + self.objcache = None + idx = self.idx + self.idx = None + + # update object count + f.seek(8) + cp = struct.pack('!i', self.count) + assert(len(cp) == 4) + f.write(cp) + + # calculate the pack sha1sum + f.seek(0) + sum = Sha1() + for b in chunkyreader(f): + sum.update(b) + packbin = sum.digest() + f.write(packbin) + fdatasync(f.fileno()) + finally: + f.close() - # update object count - f.seek(8) - cp = struct.pack('!i', self.count) - assert(len(cp) == 4) - f.write(cp) - - # calculate the pack sha1sum - f.seek(0) - sum = Sha1() - for b in chunkyreader(f): - sum.update(b) - packbin = sum.digest() - f.write(packbin) - f.close() - - idx_f = open(self.filename + '.idx', 'wb') - obj_list_sha = self._write_pack_idx_v2(idx_f, idx, packbin) - idx_f.close() + obj_list_sha = self._write_pack_idx_v2(self.filename + '.idx', idx, packbin) nameprefix = repo('objects/pack/pack-%s' % obj_list_sha) if os.path.exists(self.filename + '.map'): os.unlink(self.filename + '.map') os.rename(self.filename + '.pack', nameprefix + '.pack') os.rename(self.filename + '.idx', nameprefix + '.idx') + try: + os.fsync(self.parentfd) + finally: + os.close(self.parentfd) if run_midx: auto_midx(repo('objects/pack')) + + if self.on_pack_finish: + self.on_pack_finish(nameprefix) + return nameprefix def close(self, run_midx=True): """Close the pack file and move it to its definitive path.""" return self._end(run_midx=run_midx) - def _write_pack_idx_v2(self, file, idx, packbin): - sum = Sha1() - - def write(data): - file.write(data) - sum.update(data) - - write('\377tOc\0\0\0\2') - - n = 0 - for part in idx: - n += len(part) - write(struct.pack('!i', n)) - part.sort(key=lambda x: x[0]) - - obj_list_sum = Sha1() - for part in idx: - for entry in part: - write(entry[0]) - obj_list_sum.update(entry[0]) - for part in idx: - for entry in part: - write(struct.pack('!I', entry[1])) - ofs64_list = [] - for part in idx: - for entry in part: - if entry[2] & 0x80000000: - write(struct.pack('!I', 0x80000000 | len(ofs64_list))) - ofs64_list.append(struct.pack('!Q', entry[2])) - else: - write(struct.pack('!i', entry[2])) - for ofs64 in ofs64_list: - write(ofs64) - - write(packbin) - file.write(sum.digest()) - return obj_list_sum.hexdigest() - - -def _git_date(date): - return '%d %s' % (date, time.strftime('%z', time.localtime(date))) - - -def _gitenv(): - os.environ['GIT_DIR'] = os.path.abspath(repo()) - + def _write_pack_idx_v2(self, filename, idx, packbin): + ofs64_count = 0 + for section in idx: + for entry in section: + if entry[2] >= 2**31: + ofs64_count += 1 + + # Length: header + fan-out + shas-and-crcs + overflow-offsets + index_len = 8 + (4 * 256) + (28 * self.count) + (8 * ofs64_count) + idx_map = None + idx_f = open(filename, 'w+b') + try: + idx_f.truncate(index_len) + fdatasync(idx_f.fileno()) + idx_map = mmap_readwrite(idx_f, close=False) + try: + count = _helpers.write_idx(filename, idx_map, idx, self.count) + assert(count == self.count) + idx_map.flush() + finally: + idx_map.close() + finally: + idx_f.close() + + idx_f = open(filename, 'a+b') + try: + idx_f.write(packbin) + idx_f.seek(0) + idx_sum = Sha1() + b = idx_f.read(8 + 4*256) + idx_sum.update(b) + + obj_list_sum = Sha1() + for b in chunkyreader(idx_f, 20*self.count): + idx_sum.update(b) + obj_list_sum.update(b) + namebase = obj_list_sum.hexdigest() + + for b in chunkyreader(idx_f): + idx_sum.update(b) + idx_f.write(idx_sum.digest()) + fdatasync(idx_f.fileno()) + return namebase + finally: + idx_f.close() + + +def _gitenv(repo_dir = None): + if not repo_dir: + repo_dir = repo() + def env(): + os.environ['GIT_DIR'] = os.path.abspath(repo_dir) + return env + + +def list_refs(refnames=None, repo_dir=None, + limit_to_heads=False, limit_to_tags=False): + """Yield (refname, hash) tuples for all repository refs unless + refnames are specified. In that case, only include tuples for + those refs. The limits restrict the result items to refs/heads or + refs/tags. If both limits are specified, items from both sources + will be included. -def list_refs(refname = None): - """Generate a list of tuples in the form (refname,hash). - If a ref name is specified, list only this particular ref. """ - argv = ['git', 'show-ref', '--'] - if refname: - argv += [refname] - p = subprocess.Popen(argv, preexec_fn = _gitenv, stdout = subprocess.PIPE) + argv = ['git', 'show-ref'] + if limit_to_heads: + argv.append('--heads') + if limit_to_tags: + argv.append('--tags') + argv.append('--') + if refnames: + argv += refnames + p = subprocess.Popen(argv, + preexec_fn = _gitenv(repo_dir), + stdout = subprocess.PIPE) out = p.stdout.read().strip() rv = p.wait() # not fatal if rv: @@ -766,9 +874,10 @@ def list_refs(refname = None): yield (name, sha.decode('hex')) -def read_ref(refname): +def read_ref(refname, repo_dir = None): """Get the commit id of the most recent commit made on a given ref.""" - l = list(list_refs(refname)) + refs = list_refs(refnames=[refname], repo_dir=repo_dir, limit_to_heads=True) + l = tuple(islice(refs, 2)) if l: assert(len(l) == 1) return l[0][1] @@ -776,7 +885,7 @@ def read_ref(refname): return None -def rev_list(ref, count=None): +def rev_list(ref, count=None, repo_dir=None): """Generate a list of reachable commits in reverse chronological order. This generator walks through commits, from child to parent, that are @@ -790,8 +899,10 @@ def rev_list(ref, count=None): opts = [] if count: opts += ['-n', str(atoi(count))] - argv = ['git', 'rev-list', '--pretty=format:%ct'] + opts + [ref, '--'] - p = subprocess.Popen(argv, preexec_fn = _gitenv, stdout = subprocess.PIPE) + argv = ['git', 'rev-list', '--pretty=format:%at'] + opts + [ref, '--'] + p = subprocess.Popen(argv, + preexec_fn = _gitenv(repo_dir), + stdout = subprocess.PIPE) commit = None for row in p.stdout: s = row.strip() @@ -805,14 +916,18 @@ def rev_list(ref, count=None): raise GitError, 'git rev-list returned error %d' % rv -def rev_get_date(ref): - """Get the date of the latest commit on the specified ref.""" - for (date, commit) in rev_list(ref, count=1): - return date - raise GitError, 'no such commit %r' % ref +def get_commit_dates(refs, repo_dir=None): + """Get the dates for the specified commit refs. For now, every unique + string in refs must resolve to a different commit or this + function will fail.""" + result = [] + for ref in refs: + commit = get_commit_items(ref, cp(repo_dir)) + result.append(commit.author_sec) + return result -def rev_parse(committish): +def rev_parse(committish, repo_dir=None): """Resolve the full hash for 'committish', if it exists. Should be roughly equivalent to 'git rev-parse'. @@ -820,12 +935,12 @@ def rev_parse(committish): Returns the hex value of the hash if it is found, None if 'committish' does not correspond to anything. """ - head = read_ref(committish) + head = read_ref(committish, repo_dir=repo_dir) if head: debug2("resolved from ref: commit = %s\n" % head.encode('hex')) return head - pL = PackIdxList(repo('objects/pack')) + pL = PackIdxList(repo('objects/pack', repo_dir=repo_dir)) if len(committish) == 40: try: @@ -839,14 +954,24 @@ def rev_parse(committish): return None -def update_ref(refname, newval, oldval): - """Change the commit pointed to by a branch.""" +def update_ref(refname, newval, oldval, repo_dir=None): + """Update a repository reference.""" if not oldval: oldval = '' - assert(refname.startswith('refs/heads/')) + assert(refname.startswith('refs/heads/') \ + or refname.startswith('refs/tags/')) p = subprocess.Popen(['git', 'update-ref', refname, newval.encode('hex'), oldval.encode('hex')], - preexec_fn = _gitenv) + preexec_fn = _gitenv(repo_dir)) + _git_wait('git update-ref', p) + + +def delete_ref(refname, oldvalue=None): + """Delete a repository reference (see git update-ref(1)).""" + assert(refname.startswith('refs/')) + oldvalue = [] if not oldvalue else [oldvalue] + p = subprocess.Popen(['git', 'update-ref', '-d', refname] + oldvalue, + preexec_fn = _gitenv()) _git_wait('git update-ref', p) @@ -874,42 +999,36 @@ def init_repo(path=None): if parent and not os.path.exists(parent): raise GitError('parent directory "%s" does not exist\n' % parent) if os.path.exists(d) and not os.path.isdir(os.path.join(d, '.')): - raise GitError('"%d" exists but is not a directory\n' % d) + raise GitError('"%s" exists but is not a directory\n' % d) p = subprocess.Popen(['git', '--bare', 'init'], stdout=sys.stderr, - preexec_fn = _gitenv) + preexec_fn = _gitenv()) _git_wait('git init', p) # Force the index version configuration in order to ensure bup works # regardless of the version of the installed Git binary. p = subprocess.Popen(['git', 'config', 'pack.indexVersion', '2'], - stdout=sys.stderr, preexec_fn = _gitenv) + stdout=sys.stderr, preexec_fn = _gitenv()) + _git_wait('git config', p) + # Enable the reflog + p = subprocess.Popen(['git', 'config', 'core.logAllRefUpdates', 'true'], + stdout=sys.stderr, preexec_fn = _gitenv()) _git_wait('git config', p) def check_repo_or_die(path=None): - """Make sure a bup repository exists, and abort if not. - If the path to a particular repository was not specified, this function - initializes the default repository automatically. - """ + """Check to see if a bup repository probably exists, and abort if not.""" guess_repo(path) - if not os.path.isdir(repo('objects/pack/.')): - if repodir == home_repodir: - init_repo() - else: - log('error: %r is not a bup/git repository\n' % repo()) + top = repo() + pst = stat_if_exists(top + '/objects/pack') + if pst and stat.S_ISDIR(pst.st_mode): + return + if not pst: + top_st = stat_if_exists(top) + if not top_st: + log('error: repository %r does not exist (see "bup help init")\n' + % top) sys.exit(15) - - -def treeparse(buf): - """Generate a list of (mode, name, hash) tuples of objects from 'buf'.""" - ofs = 0 - while ofs < len(buf): - z = buf[ofs:].find('\0') - assert(z > 0) - spl = buf[ofs:ofs+z].split(' ', 1) - assert(len(spl) == 2) - sha = buf[ofs+z+1:ofs+z+1+20] - ofs += z+1+20 - yield (spl[0], spl[1], sha) + log('error: %r is not a repository\n' % top) + sys.exit(14) _ver = None @@ -946,7 +1065,7 @@ def _git_wait(cmd, p): def _git_capture(argv): - p = subprocess.Popen(argv, stdout=subprocess.PIPE, preexec_fn = _gitenv) + p = subprocess.Popen(argv, stdout=subprocess.PIPE, preexec_fn = _gitenv()) r = p.stdout.read() _git_wait(repr(argv), p) return r @@ -964,7 +1083,7 @@ class _AbortableIter: def next(self): try: return self.it.next() - except StopIteration, e: + except StopIteration as e: self.done = True raise except: @@ -982,11 +1101,18 @@ class _AbortableIter: self.abort() +class MissingObject(KeyError): + def __init__(self, id): + self.id = id + KeyError.__init__(self, 'object %r is missing' % id.encode('hex')) + + _ver_warned = 0 class CatPipe: """Link to 'git cat-file' that is used to retrieve blob data.""" - def __init__(self): + def __init__(self, repo_dir = None): global _ver_warned + self.repo_dir = repo_dir wanted = ('1','5','6') if ver() < wanted: if not _ver_warned: @@ -1005,22 +1131,23 @@ class CatPipe: self.p = None self.inprogress = None - def _restart(self): + def restart(self): self._abort() self.p = subprocess.Popen(['git', 'cat-file', '--batch'], stdin=subprocess.PIPE, stdout=subprocess.PIPE, close_fds = True, bufsize = 4096, - preexec_fn = _gitenv) + preexec_fn = _gitenv(self.repo_dir)) def _fast_get(self, id): if not self.p or self.p.poll() != None: - self._restart() + self.restart() assert(self.p) - assert(self.p.poll() == None) + poll_result = self.p.poll() + assert(poll_result == None) if self.inprogress: - log('_fast_get: opening %r while %r is open' + log('_fast_get: opening %r while %r is open\n' % (id, self.inprogress)) assert(not self.inprogress) assert(id.find('\n') < 0) @@ -1032,7 +1159,7 @@ class CatPipe: hdr = self.p.stdout.readline() if hdr.endswith(' missing\n'): self.inprogress = None - raise KeyError('blob %r is missing' % id) + raise MissingObject(id.decode('hex')) spl = hdr.split(' ') if len(spl) != 3 or len(spl[0]) != 40: raise GitError('expected blob, got %r' % spl) @@ -1044,9 +1171,10 @@ class CatPipe: yield type for blob in it: yield blob - assert(self.p.stdout.readline() == '\n') + readline_result = self.p.stdout.readline() + assert(readline_result == '\n') self.inprogress = None - except Exception, e: + except Exception as e: it.abort() raise @@ -1059,7 +1187,7 @@ class CatPipe: p = subprocess.Popen(['git', 'cat-file', type, id], stdout=subprocess.PIPE, - preexec_fn = _gitenv) + preexec_fn = _gitenv(self.repo_dir)) for blob in chunkyreader(p.stdout): yield blob _git_wait('git cat-file', p) @@ -1071,7 +1199,7 @@ class CatPipe: yield blob elif type == 'tree': treefile = ''.join(it) - for (mode, name, sha) in treeparse(treefile): + for (mode, name, sha) in tree_decode(treefile): for blob in self.join(sha.encode('hex')): yield blob elif type == 'commit': @@ -1095,15 +1223,110 @@ class CatPipe: except StopIteration: log('booger!\n') -def tags(): + +_cp = {} + +def cp(repo_dir=None): + """Create a CatPipe object or reuse the already existing one.""" + global _cp, repodir + if not repo_dir: + repo_dir = repodir or repo() + repo_dir = os.path.abspath(repo_dir) + cp = _cp.get(repo_dir) + if not cp: + cp = CatPipe(repo_dir) + _cp[repo_dir] = cp + return cp + + +def tags(repo_dir = None): """Return a dictionary of all tags in the form {hash: [tag_names, ...]}.""" tags = {} - for (n,c) in list_refs(): - if n.startswith('refs/tags/'): - name = n[10:] - if not c in tags: - tags[c] = [] + for n, c in list_refs(repo_dir = repo_dir, limit_to_tags=True): + assert(n.startswith('refs/tags/')) + name = n[10:] + if not c in tags: + tags[c] = [] + tags[c].append(name) # more than one tag can point at 'c' + return tags - tags[c].append(name) # more than one tag can point at 'c' - return tags +WalkItem = namedtuple('WalkItem', ['id', 'type', 'mode', + 'path', 'chunk_path', 'data']) +# The path is the mangled path, and if an item represents a fragment +# of a chunked file, the chunk_path will be the chunked subtree path +# for the chunk, i.e. ['', '2d3115e', ...]. The top-level path for a +# chunked file will have a chunk_path of ['']. So some chunk subtree +# of the file '/foo/bar/baz' might look like this: +# +# item.path = ['foo', 'bar', 'baz.bup'] +# item.chunk_path = ['', '2d3115e', '016b097'] +# item.type = 'tree' +# ... + + +def walk_object(cat_pipe, id, + stop_at=None, + include_data=None): + """Yield everything reachable from id via cat_pipe as a WalkItem, + stopping whenever stop_at(id) returns true. Throw MissingObject + if a hash encountered is missing from the repository, and don't + read or return blob content in the data field unless include_data + is set. + """ + # Maintain the pending stack on the heap to avoid stack overflow + pending = [(id, [], [], None)] + while len(pending): + id, parent_path, chunk_path, mode = pending.pop() + if stop_at and stop_at(id): + continue + + if (not include_data) and mode and stat.S_ISREG(mode): + # If the object is a "regular file", then it's a leaf in + # the graph, so we can skip reading the data if the caller + # hasn't requested it. + yield WalkItem(id=id, type='blob', + chunk_path=chunk_path, path=parent_path, + mode=mode, + data=None) + continue + + item_it = cat_pipe.get(id) + type = item_it.next() + if type not in ('blob', 'commit', 'tree'): + raise Exception('unexpected repository object type %r' % type) + + # FIXME: set the mode based on the type when the mode is None + if type == 'blob' and not include_data: + # Dump data until we can ask cat_pipe not to fetch it + for ignored in item_it: + pass + data = None + else: + data = ''.join(item_it) + + yield WalkItem(id=id, type=type, + chunk_path=chunk_path, path=parent_path, + mode=mode, + data=(data if include_data else None)) + + if type == 'commit': + commit_items = parse_commit(data) + for pid in commit_items.parents: + pending.append((pid, parent_path, chunk_path, mode)) + pending.append((commit_items.tree, parent_path, chunk_path, + hashsplit.GIT_MODE_TREE)) + elif type == 'tree': + for mode, name, ent_id in tree_decode(data): + demangled, bup_type = demangle_name(name, mode) + if chunk_path: + sub_path = parent_path + sub_chunk_path = chunk_path + [name] + else: + sub_path = parent_path + [name] + if bup_type == BUP_CHUNKED: + sub_chunk_path = [''] + else: + sub_chunk_path = chunk_path + pending.append((ent_id.encode('hex'), sub_path, sub_chunk_path, + mode))