bup repositories are in Git format. This library allows us to
interact with the Git data structures.
"""
-import os, errno, zlib, time, subprocess, struct, stat, re, tempfile
-import heapq
+import os, sys, zlib, time, subprocess, struct, stat, re, tempfile
from bup.helpers import *
+from bup import _helpers, path
+
+MIDX_VERSION = 2
verbose = 0
ignore_midx = 0
_typemap = { 'blob':3, 'tree':2, 'commit':1, 'tag':4 }
_typermap = { 3:'blob', 2:'tree', 1:'commit', 4:'tag' }
+_total_searches = 0
+_total_steps = 0
+
class GitError(Exception):
pass
return os.path.join(repodir, sub)
+def auto_midx(objdir):
+ args = [path.exe(), 'midx', '--auto', '--dir', objdir]
+ try:
+ rv = subprocess.call(args, stdout=open('/dev/null', 'w'))
+ except OSError, 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))
+
+
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
class PackIdx:
- """Object representation of a Git pack index file."""
- def __init__(self, filename):
+ def __init__(self):
+ assert(0)
+
+ def find_offset(self, hash):
+ """Get the offset of an object inside the index file."""
+ idx = self._idx_from_hash(hash)
+ if idx != None:
+ return self._ofs_from_idx(idx)
+ return None
+
+ def exists(self, hash):
+ """Return nonempty if the object exists in this index."""
+ return hash and (self._idx_from_hash(hash) != None) and True or None
+
+ def __len__(self):
+ return int(self.fanout[255])
+
+ def _idx_from_hash(self, hash):
+ global _total_searches, _total_steps
+ _total_searches += 1
+ assert(len(hash) == 20)
+ b1 = ord(hash[0])
+ start = self.fanout[b1-1] # range -1..254
+ end = self.fanout[b1] # range 0..255
+ want = str(hash)
+ _total_steps += 1 # lookup table is a step
+ while start < end:
+ _total_steps += 1
+ mid = start + (end-start)/2
+ v = self._idx_to_hash(mid)
+ if v < want:
+ start = mid+1
+ elif v > want:
+ end = mid
+ else: # got it!
+ return mid
+ return None
+
+
+class PackIdxV1(PackIdx):
+ """Object representation of a Git pack index (version 1) file."""
+ def __init__(self, filename, f):
self.name = filename
- self.map = mmap_read(open(filename))
+ self.idxnames = [self.name]
+ self.map = mmap_read(f)
+ self.fanout = list(struct.unpack('!256I',
+ 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)
+
+ def _ofs_from_idx(self, idx):
+ return struct.unpack('!I', str(self.shatable[idx*24 : idx*24+4]))[0]
+
+ def _idx_to_hash(self, idx):
+ return str(self.shatable[idx*24+4 : idx*24+24])
+
+ def __iter__(self):
+ for i in xrange(self.fanout[255]):
+ yield buffer(self.map, 256*4 + 24*i + 4, 20)
+
+
+class PackIdxV2(PackIdx):
+ """Object representation of a Git pack index (version 2) file."""
+ def __init__(self, filename, f):
+ self.name = filename
+ self.idxnames = [self.name]
+ self.map = mmap_read(f)
assert(str(self.map[0:8]) == '\377tOc\0\0\0\2')
self.fanout = list(struct.unpack('!256I',
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.ofstable = buffer(self.map,
8 + 256*4 + nsha*20 + nsha*4,
nsha*4)
ofs = struct.unpack('!I', str(buffer(self.ofstable, idx*4, 4)))[0]
if ofs & 0x80000000:
idx64 = ofs & 0x7fffffff
- ofs = struct.unpack('!I',
+ ofs = struct.unpack('!Q',
str(buffer(self.ofs64table, idx64*8, 8)))[0]
return ofs
- def _idx_from_hash(self, hash):
- assert(len(hash) == 20)
- b1 = ord(hash[0])
- start = self.fanout[b1-1] # range -1..254
- end = self.fanout[b1] # range 0..255
- buf = buffer(self.map, 8 + 256*4, end*20)
- want = str(hash)
- while start < end:
- mid = start + (end-start)/2
- v = str(buf[mid*20:(mid+1)*20])
- if v < want:
- start = mid+1
- elif v > want:
- end = mid
- else: # got it!
- return mid
- return None
-
- def find_offset(self, hash):
- """Get the offset of an object inside the index file."""
- idx = self._idx_from_hash(hash)
- if idx != None:
- return self._ofs_from_idx(idx)
- return None
-
- def exists(self, hash):
- """Return nonempty if the object exists in this index."""
- return hash and (self._idx_from_hash(hash) != None) and True or None
+ def _idx_to_hash(self, idx):
+ return str(self.shatable[idx*20:(idx+1)*20])
def __iter__(self):
for i in xrange(self.fanout[255]):
yield buffer(self.map, 8 + 256*4 + 20*i, 20)
- def __len__(self):
- return int(self.fanout[255])
-
-def extract_bits(buf, nbits):
- """Take the first 'nbits' bits from 'buf' and return them as an integer."""
- mask = (1<<nbits) - 1
- v = struct.unpack('!I', buf[0:4])[0]
- v = (v >> (32-nbits)) & mask
- return v
+extract_bits = _helpers.extract_bits
class PackMidx:
"""
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[0:8]) == 'MIDX\0\0\0\1':
- log('Warning: ignoring old-style midx %r\n' % filename)
- self.bits = 0
- self.entries = 1
- self.fanout = buffer('\0\0\0\0')
- self.shalist = buffer('\0'*20)
- self.idxnames = []
- else:
- assert(str(self.map[0:8]) == 'MIDX\0\0\0\2')
- self.bits = struct.unpack('!I', self.map[8:12])[0]
- 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')
+ 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 struct.unpack('!I', s)[0]
-
+ 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:
- mid = start + (end-start)/2
- v = str(self.shalist[mid*20:(mid+1)*20])
+ _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))
def __iter__(self):
return iter(idxmerge(self.packs))
+ def __len__(self):
+ return sum(len(pack) for pack in self.packs)
+
def exists(self, hash):
"""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)):
p = self.packs[i]
+ _total_searches -= 1 # will be incremented by sub-pack
if p.exists(hash):
# reorder so most recently used packs are searched first
self.packs = [p] + self.packs[:i] + self.packs[i+1:]
log(('warning: index %s missing\n' +
' used by %s\n') % (n, mxf))
broken += 1
- if not broken:
+ if broken:
+ del mx
+ unlink(full)
+ else:
midxl.append(mx)
midxl.sort(lambda x,y: -cmp(len(x),len(y)))
for ix in midxl:
d[os.path.join(self.dir, name)] = ix
any += 1
break
- if not any:
- log('midx: removing redundant: %s\n'
- % os.path.basename(ix.name))
+ if not any and not ix.force_keep:
+ debug1('midx: removing redundant: %s\n'
+ % os.path.basename(ix.name))
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):
- ix = PackIdx(full)
+ try:
+ ix = open_idx(full)
+ except GitError, e:
+ add_error(e)
+ continue
d[full] = ix
self.packs = list(set(d.values()))
- log('PackIdxList: using %d index%s.\n'
+ 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
return name
-def idxmerge(idxlist):
- """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))
+def open_idx(filename):
+ if filename.endswith('.idx'):
+ f = open(filename, 'rb')
+ header = f.read(8)
+ if header[0:4] == '\377tOc':
+ version = struct.unpack('!I', header[4:8])[0]
+ if version == 2:
+ return PackIdxV2(filename, f)
+ else:
+ raise GitError('%s: expected idx file version 2, got %d'
+ % (filename, version))
+ elif len(header) == 8 and header[0:4] < '\377tOc':
+ return PackIdxV1(filename, f)
else:
- heapq.heappop(heap)
- log('Reading indexes: %.2f%% (%d/%d), done.\n' % (100, total, total))
+ raise GitError('%s: unrecognized idx file header' % filename)
+ elif filename.endswith('.midx'):
+ return 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."""
+ def pfunc(count, total):
+ progress('Reading indexes: %.2f%% (%d/%d)\r'
+ % (count*100.0/total, count, total))
+ def pfinal(count, total):
+ if final_progress:
+ log('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'))
-
class PackWriter:
"""Writes Git objects insid a pack file."""
- def __init__(self, objcache_maker=None):
+ def __init__(self, objcache_maker=_make_objcache):
self.count = 0
self.outbytes = 0
self.filename = None
self.file = None
+ self.idx = None
self.objcache_maker = objcache_maker
self.objcache = None
def __del__(self):
self.close()
- def _make_objcache(self):
- if not self.objcache:
- if self.objcache_maker:
- self.objcache = self.objcache_maker()
- else:
- self.objcache = PackIdxList(repo('objects/pack'))
-
def _open(self):
if not self.file:
- self._make_objcache()
(fd,name) = tempfile.mkstemp(suffix='.pack', dir=repo('objects'))
self.file = os.fdopen(fd, 'w+b')
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))
- def _raw_write(self, datalist):
+ # 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
# in case we get interrupted (eg. KeyboardInterrupt), it's best if
# to our hashsplit algorithm.) f.write() does its own buffering,
# but that's okay because we'll flush it in _end().
oneblob = ''.join(datalist)
- f.write(oneblob)
- self.outbytes += len(oneblob)
+ try:
+ f.write(oneblob)
+ except IOError, e:
+ raise GitError, e, sys.exc_info()[2]
+ nw = len(oneblob)
+ crc = zlib.crc32(oneblob) & 0xffffffff
+ self._update_idx(sha, crc, nw)
+ self.outbytes += nw
self.count += 1
+ return nw, crc
- def _write(self, bin, type, content):
+ def _update_idx(self, sha, crc, size):
+ assert(sha)
+ if self.idx:
+ self.idx[ord(sha[0])].append((sha, crc, self.file.tell() - size))
+
+ def _write(self, sha, type, content):
if verbose:
log('>')
- self._raw_write(_encode_packobj(type, content))
- return bin
+ if not sha:
+ sha = calc_hash(type, content)
+ size, crc = self._raw_write(_encode_packobj(type, content), sha=sha)
+ return sha
def breakpoint(self):
"""Clear byte and object counts and return the last processed id."""
"""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()
+ if self.objcache is None:
+ raise GitError(
+ "PackWriter not opened or can't check exists w/o objcache")
+
def exists(self, id):
"""Return non-empty if an object is found in the object cache."""
- if not self.objcache:
- self._make_objcache()
+ self._require_objcache()
return self.objcache.exists(id)
def maybe_write(self, type, content):
"""Write an object to the pack file if not present and return its id."""
- bin = calc_hash(type, content)
- if not self.exists(bin):
- self._write(bin, type, content)
- self.objcache.add(bin)
- return bin
+ self._require_objcache()
+ sha = calc_hash(type, content)
+ if not self.exists(sha):
+ self._write(sha, type, content)
+ self.objcache.add(sha)
+ return sha
def new_blob(self, blob):
"""Create a blob object in the pack with the supplied content."""
l.append(msg)
return self.maybe_write('commit', '\n'.join(l))
- def new_commit(self, parent, tree, msg):
+ def new_commit(self, parent, tree, date, msg):
"""Create a commit object in the pack."""
- now = time.time()
userline = '%s <%s@%s>' % (userfullname(), username(), hostname())
commit = self._new_commit(tree, parent,
- userline, now, userline, now,
+ userline, date, userline, date,
msg)
return commit
"""Remove the pack file from disk."""
f = self.file
if f:
+ self.idx = None
self.file = None
f.close()
os.unlink(self.filename + '.pack')
- def _end(self):
+ 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
# update object count
f.seek(8)
# calculate the pack sha1sum
f.seek(0)
sum = Sha1()
- while 1:
- b = f.read(65536)
+ for b in chunkyreader(f):
sum.update(b)
- if not b: break
- f.write(sum.digest())
-
+ packbin = sum.digest()
+ f.write(packbin)
f.close()
- p = subprocess.Popen(['git', 'index-pack', '-v',
- '--index-version=2',
- self.filename + '.pack'],
- preexec_fn = _gitenv,
- stdout = subprocess.PIPE)
- out = p.stdout.read().strip()
- _git_wait('git index-pack', p)
- if not out:
- raise GitError('git index-pack produced no output')
- nameprefix = repo('objects/pack/%s' % out)
+ idx_f = open(self.filename + '.idx', 'wb')
+ obj_list_sha = self._write_pack_idx_v2(idx_f, idx, packbin)
+ idx_f.close()
+
+ 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')
+
+ if run_midx:
+ auto_midx(repo('objects/pack'))
return nameprefix
- def close(self):
+ def close(self, run_midx=True):
"""Close the pack file and move it to its definitive path."""
- return self._end()
+ 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):
raise GitError, 'no such commit %r' % ref
+def rev_parse(committish):
+ """Resolve the full hash for 'committish', if it exists.
+
+ Should be roughly equivalent to 'git rev-parse'.
+
+ Returns the hex value of the hash if it is found, None if 'committish' does
+ not correspond to anything.
+ """
+ head = read_ref(committish)
+ if head:
+ debug2("resolved from ref: commit = %s\n" % head.encode('hex'))
+ return head
+
+ pL = PackIdxList(repo('objects/pack'))
+
+ if len(committish) == 40:
+ try:
+ hash = committish.decode('hex')
+ except TypeError:
+ return None
+
+ if pL.exists(hash):
+ return hash
+
+ return None
+
+
def update_ref(refname, newval, oldval):
"""Change the commit pointed to by a branch."""
if not oldval:
def init_repo(path=None):
"""Create the Git bare repository for bup in a given path."""
guess_repo(path)
- d = repo()
+ d = repo() # appends a / to the path
+ parent = os.path.dirname(os.path.dirname(d))
+ 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)
p = subprocess.Popen(['git', '--bare', 'init'], stdout=sys.stderr,
sys.exit(15)
-def _treeparse(buf):
+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')
def __iter__(self):
return self
-
+
def next(self):
try:
return self.it.next()
self.done = True
if self.onabort:
self.onabort()
-
+
def __del__(self):
self.abort()
def _restart(self):
self._abort()
self.p = subprocess.Popen(['git', 'cat-file', '--batch'],
- stdin=subprocess.PIPE,
+ stdin=subprocess.PIPE,
stdout=subprocess.PIPE,
+ close_fds = True,
+ bufsize = 4096,
preexec_fn = _gitenv)
def _fast_get(self, id):
assert(self.p)
assert(self.p.poll() == None)
if self.inprogress:
- log('_fast_get: opening %r while %r is open'
+ log('_fast_get: opening %r while %r is open'
% (id, self.inprogress))
assert(not self.inprogress)
assert(id.find('\n') < 0)
assert(id.find('\r') < 0)
- assert(id[0] != '-')
+ assert(not id.startswith('-'))
self.inprogress = id
self.p.stdin.write('%s\n' % id)
+ self.p.stdin.flush()
hdr = self.p.stdout.readline()
if hdr.endswith(' missing\n'):
+ self.inprogress = None
raise KeyError('blob %r is missing' % id)
spl = hdr.split(' ')
if len(spl) != 3 or len(spl[0]) != 40:
yield blob
elif type == 'tree':
treefile = ''.join(it)
- for (mode, name, sha) in _treeparse(treefile):
+ for (mode, name, sha) in treeparse(treefile):
for blob in self.join(sha.encode('hex')):
yield blob
elif type == 'commit':
yield d
except StopIteration:
log('booger!\n')
-
-def cat(id):
- c = CatPipe()
- for d in c.join(id):
- yield d
+def tags():
+ """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] = []
+
+ tags[c].append(name) # more than one tag can point at 'c'
+
+ return tags
projects / bup.git / blobdiff