interact with the Git data structures.
"""
+from __future__ import absolute_import
import errno, os, sys, zlib, time, subprocess, struct, stat, re, tempfile, glob
from collections import namedtuple
from itertools import islice
+from numbers import Integral
-from bup import _helpers, hashsplit, path, midx, bloom, xstat
+from bup import _helpers, compat, hashsplit, path, midx, bloom, xstat
+from bup.compat import range
from bup.helpers import (Sha1, add_error, chunkyreader, debug1, debug2,
fdatasync,
- hostname, localtime, log, merge_iter,
+ hostname, localtime, log,
+ merge_dict,
+ merge_iter,
mmap_read, mmap_readwrite,
parse_num,
- progress, qprogress, stat_if_exists,
- unlink, username, userfullname,
+ progress, qprogress, shstr, stat_if_exists,
+ unlink,
utc_offset_str)
+from bup.pwdgrp import username, userfullname
verbose = 0
ignore_midx = 0
pass
+def _gitenv(repo_dir=None):
+ if not repo_dir:
+ repo_dir = repo()
+ return merge_dict(os.environ, {'GIT_DIR': os.path.abspath(repo_dir)})
+
def _git_wait(cmd, p):
rv = p.wait()
if rv != 0:
- raise GitError('%s returned %d' % (cmd, rv))
+ raise GitError('%s returned %d' % (shstr(cmd), rv))
def _git_capture(argv):
- p = subprocess.Popen(argv, stdout=subprocess.PIPE, preexec_fn = _gitenv())
+ p = subprocess.Popen(argv, stdout=subprocess.PIPE, env=_gitenv())
r = p.stdout.read()
_git_wait(repr(argv), p)
return r
def git_config_get(option, repo_dir=None):
cmd = ('git', 'config', '--get', option)
p = subprocess.Popen(cmd, stdout=subprocess.PIPE,
- preexec_fn=_gitenv(repo_dir=repo_dir))
+ env=_gitenv(repo_dir=repo_dir))
r = p.stdout.read()
rc = p.wait()
if rc == 0:
_start_end_char, _content_char, _start_end_char)
_tz_rx = r'[-+]\d\d[0-5]\d'
_parent_rx = r'(?:parent [abcdefABCDEF0123456789]{40}\n)'
+# Assumes every following line starting with a space is part of the
+# mergetag. Is there a formal commit blob spec?
+_mergetag_rx = r'(?:\nmergetag object [abcdefABCDEF0123456789]{40}(?:\n [^\0\n]*)*)'
_commit_rx = re.compile(r'''tree (?P<tree>[abcdefABCDEF0123456789]{40})
(?P<parents>%s*)author (?P<author_name>%s) <(?P<author_mail>%s)> (?P<asec>\d+) (?P<atz>%s)
-committer (?P<committer_name>%s) <(?P<committer_mail>%s)> (?P<csec>\d+) (?P<ctz>%s)
+committer (?P<committer_name>%s) <(?P<committer_mail>%s)> (?P<csec>\d+) (?P<ctz>%s)(?P<mergetag>%s?)
(?P<message>(?:.|\n)*)''' % (_parent_rx,
_safe_str_rx, _safe_str_rx, _tz_rx,
- _safe_str_rx, _safe_str_rx, _tz_rx))
+ _safe_str_rx, _safe_str_rx, _tz_rx,
+ _mergetag_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.
+# Note that the author_sec and committer_sec values are (UTC) epoch
+# seconds, and for now the mergetag is not included.
CommitInfo = namedtuple('CommitInfo', ['tree', 'parents',
'author_name', 'author_mail',
'author_sec', 'author_offset',
message=matches['message'])
-def get_commit_items(id, cp):
- commit_it = cp.get(id)
- assert(next(commit_it) == 'commit')
- commit_content = ''.join(commit_it)
- return parse_commit(commit_content)
+def get_cat_data(cat_iterator, expected_type):
+ _, kind, _ = next(cat_iterator)
+ if kind != expected_type:
+ raise Exception('expected %r, saw %r' % (expected_type, kind))
+ return ''.join(cat_iterator)
+def get_commit_items(id, cp):
+ return parse_commit(get_cat_data(cp.get(id), 'commit'))
def _local_git_date_str(epoch_sec):
return '%d %s' % (epoch_sec, utc_offset_str(epoch_sec))
def repo(sub = '', repo_dir=None):
"""Get the path to the git repository or one of its subdirectories."""
- global 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(repo_dir, '.git')
if os.path.exists(gd):
- repodir = gd
+ repo_dir = gd
return os.path.join(repo_dir, sub)
self.name = 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 = list(struct.unpack('!256I', buffer(self.map, 0, 256 * 4)))
self.fanout.append(0) # entry "-1"
nsha = self.fanout[255]
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]
+ ofs = idx * 24
+ return struct.unpack('!I', self.shatable[ofs : ofs + 4])[0]
def _idx_to_hash(self, idx):
- return str(self.shatable[idx*24+4 : idx*24+24])
+ ofs = idx * 24 + 4
+ return self.shatable[ofs : ofs + 20]
def __iter__(self):
- for i in xrange(self.fanout[255]):
- yield buffer(self.map, 256*4 + 24*i + 4, 20)
+ count = self.fanout[255]
+ start = 256 * 4 + 4
+ for ofs in range(start, start + (24 * count), 24):
+ yield self.map[ofs : ofs + 20]
class PackIdxV2(PackIdx):
self.name = filename
self.idxnames = [self.name]
self.map = mmap_read(f)
- assert(str(self.map[0:8]) == '\377tOc\0\0\0\2')
+ assert self.map[0:8] == b'\377tOc\0\0\0\2'
self.fanout = list(struct.unpack('!256I',
- str(buffer(self.map, 8, 256*4))))
+ buffer(self.map[8 : 8 + 256 * 4])))
self.fanout.append(0) # entry "-1"
nsha = self.fanout[255]
self.sha_ofs = 8 + 256*4
8 + 256*4 + nsha*20 + nsha*4 + nsha*4)
def _ofs_from_idx(self, idx):
- ofs = struct.unpack('!I', str(buffer(self.ofstable, idx*4, 4)))[0]
+ i = idx * 4
+ ofs = struct.unpack('!I', self.ofstable[i : i + 4])[0]
if ofs & 0x80000000:
idx64 = ofs & 0x7fffffff
- ofs = struct.unpack('!Q',
- str(buffer(self.ofs64table, idx64*8, 8)))[0]
+ idx64_i = idx64 * 8
+ ofs = struct.unpack('!Q', self.ofs64table[idx64_i : idx64_i + 8])[0]
return ofs
def _idx_to_hash(self, idx):
- return str(self.shatable[idx*20:(idx+1)*20])
+ return 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)
+ count = self.fanout[255]
+ start = 8 + 256 * 4
+ for ofs in range(start, start + (20 * count), 20):
+ yield self.map[ofs : ofs + 20]
_mpi_count = 0
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)))
+ self.packs.sort(reverse=True, key=lambda x: len(x))
if self.bloom and self.bloom.valid() and len(self.bloom) >= len(self):
self.do_bloom = True
else:
"""Writes Git objects inside a pack file."""
def __init__(self, objcache_maker=_make_objcache, compression_level=1,
run_midx=True, on_pack_finish=None,
- max_pack_size=None, max_pack_objects=None):
- self.repo_dir = repo()
+ max_pack_size=None, max_pack_objects=None, repo_dir=None):
+ self.repo_dir = repo_dir or repo()
self.file = None
self.parentfd = None
self.count = 0
def __del__(self):
self.close()
+ def __enter__(self):
+ return self
+
+ def __exit__(self, type, value, traceback):
+ self.close()
+
def _open(self):
if not self.file:
objdir = dir = os.path.join(self.repo_dir, 'objects')
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()."""
+ """Write an object to the pack file without checking for duplication."""
self._write(sha, type, content)
+ # If nothing else, gc doesn't have/want an objcache
+ if self.objcache is not None:
+ self.objcache.add(sha)
def maybe_write(self, type, content):
"""Write an object to the pack file if not present and return its id."""
sha = calc_hash(type, content)
if not self.exists(sha):
- self.just_write(sha, type, content)
self._require_objcache()
- self.objcache.add(sha)
+ self.just_write(sha, type, content)
return sha
def new_blob(self, blob):
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(patterns=None, repo_dir=None,
limit_to_heads=False, limit_to_tags=False):
"""Yield (refname, hash) tuples for all repository refs unless
argv.append('--')
if patterns:
argv.extend(patterns)
- p = subprocess.Popen(argv,
- preexec_fn = _gitenv(repo_dir),
- stdout = subprocess.PIPE)
+ p = subprocess.Popen(argv, env=_gitenv(repo_dir), stdout=subprocess.PIPE)
out = p.stdout.read().strip()
rv = p.wait() # not fatal
if rv:
return None
-def rev_list(ref, count=None, repo_dir=None):
- """Generate a list of reachable commits in reverse chronological order.
+def rev_list_invocation(ref_or_refs, count=None, format=None):
+ if isinstance(ref_or_refs, compat.str_type):
+ refs = (ref_or_refs,)
+ else:
+ refs = ref_or_refs
+ argv = ['git', 'rev-list']
+ if isinstance(count, Integral):
+ argv.extend(['-n', str(count)])
+ elif count:
+ raise ValueError('unexpected count argument %r' % count)
+
+ if format:
+ argv.append('--pretty=format:' + format)
+ for ref in refs:
+ assert not ref.startswith('-')
+ argv.append(ref)
+ argv.append('--')
+ return argv
+
- This generator walks through commits, from child to parent, that are
- reachable via the specified ref and yields a series of tuples of the form
- (date,hash).
+def rev_list(ref_or_refs, count=None, parse=None, format=None, repo_dir=None):
+ """Yield information about commits as per "git rev-list". If a format
+ is not provided, yield one hex hash at a time. If a format is
+ provided, pass it to rev-list and call parse(git_stdout) for each
+ commit with the stream positioned just after the rev-list "commit
+ HASH" header line. When a format is provided yield (oidx,
+ parse(git_stdout)) for each commit.
- If count is a non-zero integer, limit the number of commits to "count"
- objects.
"""
- assert(not ref.startswith('-'))
- opts = []
- if count:
- opts += ['-n', str(atoi(count))]
- argv = ['git', 'rev-list', '--pretty=format:%at'] + opts + [ref, '--']
- p = subprocess.Popen(argv,
- preexec_fn = _gitenv(repo_dir),
+ assert bool(parse) == bool(format)
+ p = subprocess.Popen(rev_list_invocation(ref_or_refs, count=count,
+ format=format),
+ env=_gitenv(repo_dir),
stdout = subprocess.PIPE)
- commit = None
- for row in p.stdout:
- s = row.strip()
- if s.startswith('commit '):
- commit = s[7:].decode('hex')
- else:
- date = int(s)
- yield (date, commit)
+ if not format:
+ for line in p.stdout:
+ yield line.strip()
+ else:
+ line = p.stdout.readline()
+ while line:
+ s = line.strip()
+ if not s.startswith('commit '):
+ raise Exception('unexpected line ' + s)
+ s = s[7:]
+ assert len(s) == 40
+ yield s, parse(p.stdout)
+ line = p.stdout.readline()
+
rv = p.wait() # not fatal
if rv:
raise GitError, 'git rev-list returned error %d' % rv
or refname.startswith('refs/tags/'))
p = subprocess.Popen(['git', 'update-ref', refname,
newval.encode('hex'), oldval.encode('hex')],
- preexec_fn = _gitenv(repo_dir))
+ env=_gitenv(repo_dir))
_git_wait('git update-ref', p)
assert(refname.startswith('refs/'))
oldvalue = [] if not oldvalue else [oldvalue]
p = subprocess.Popen(['git', 'update-ref', '-d', refname] + oldvalue,
- preexec_fn = _gitenv())
+ env=_gitenv())
_git_wait('git update-ref', p)
if os.path.exists(d) and not os.path.isdir(os.path.join(d, '.')):
raise GitError('"%s" exists but is not a directory\n' % d)
p = subprocess.Popen(['git', '--bare', 'init'], stdout=sys.stderr,
- preexec_fn = _gitenv())
+ env=_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, env=_gitenv())
_git_wait('git config', p)
# Enable the reflog
p = subprocess.Popen(['git', 'config', 'core.logAllRefUpdates', 'true'],
- stdout=sys.stderr, preexec_fn = _gitenv())
+ stdout=sys.stderr, env=_gitenv())
_git_wait('git config', p)
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."""
stdout=subprocess.PIPE,
close_fds = True,
bufsize = 4096,
- preexec_fn = _gitenv(self.repo_dir))
+ env=_gitenv(self.repo_dir))
- def get(self, id, size=False):
- """Yield the object type, and then an iterator over the data referred
- to by the id ref. If size is true, yield (obj_type, obj_size)
- instead of just the type.
+ def get(self, ref):
+ """Yield (oidx, type, size), followed by the data referred to by ref.
+ If ref does not exist, only yield (None, None, None).
"""
if not self.p or self.p.poll() != None:
poll_result = self.p.poll()
assert(poll_result == None)
if self.inprogress:
- log('get: opening %r while %r is open\n' % (id, self.inprogress))
+ log('get: opening %r while %r is open\n' % (ref, self.inprogress))
assert(not self.inprogress)
- assert(id.find('\n') < 0)
- assert(id.find('\r') < 0)
- assert(not id.startswith('-'))
- self.inprogress = id
- self.p.stdin.write('%s\n' % id)
+ assert(ref.find('\n') < 0)
+ assert(ref.find('\r') < 0)
+ assert(not ref.startswith('-'))
+ self.inprogress = ref
+ self.p.stdin.write('%s\n' % ref)
self.p.stdin.flush()
hdr = self.p.stdout.readline()
if hdr.endswith(' missing\n'):
self.inprogress = None
- raise MissingObject(id.decode('hex'))
- spl = hdr.split(' ')
- if len(spl) != 3 or len(spl[0]) != 40:
- raise GitError('expected blob, got %r' % spl)
- hex, typ, sz = spl
- sz = int(sz)
- it = _AbortableIter(chunkyreader(self.p.stdout, sz),
+ yield None, None, None
+ return
+ info = hdr.split(' ')
+ if len(info) != 3 or len(info[0]) != 40:
+ raise GitError('expected object (id, type, size), got %r' % info)
+ oidx, typ, size = info
+ size = int(size)
+ it = _AbortableIter(chunkyreader(self.p.stdout, size),
onabort=self._abort)
try:
- if size:
- yield typ, sz
- else:
- yield typ
+ yield oidx, typ, size
for blob in it:
yield blob
readline_result = self.p.stdout.readline()
raise
def _join(self, it):
- type = next(it)
- if type == 'blob':
+ _, typ, _ = next(it)
+ if typ == 'blob':
for blob in it:
yield blob
- elif type == 'tree':
+ elif typ == 'tree':
treefile = ''.join(it)
for (mode, name, sha) in tree_decode(treefile):
for blob in self.join(sha.encode('hex')):
yield blob
- elif type == 'commit':
+ elif typ == 'commit':
treeline = ''.join(it).split('\n')[0]
assert(treeline.startswith('tree '))
for blob in self.join(treeline[5:]):
yield blob
else:
raise GitError('invalid object type %r: expected blob/tree/commit'
- % type)
+ % typ)
def join(self, id):
"""Generate a list of the content of all blobs that can be reached
return tags
-WalkItem = namedtuple('WalkItem', ['id', 'type', 'mode',
+class MissingObject(KeyError):
+ def __init__(self, oid):
+ self.oid = oid
+ KeyError.__init__(self, 'object %r is missing' % oid.encode('hex'))
+
+
+WalkItem = namedtuple('WalkItem', ['oid', '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
# ...
-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.
+def walk_object(get_ref, oidx, stop_at=None, include_data=None):
+ """Yield everything reachable from oidx via get_ref (which must behave
+ like CatPipe get) as a WalkItem, stopping whenever stop_at(oidx)
+ 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)]
+ pending = [(oidx, [], [], None)]
while len(pending):
- id, parent_path, chunk_path, mode = pending.pop()
- if stop_at and stop_at(id):
+ oidx, parent_path, chunk_path, mode = pending.pop()
+ oid = oidx.decode('hex')
+ if stop_at and stop_at(oidx):
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',
+ yield WalkItem(oid=oid, type='blob',
chunk_path=chunk_path, path=parent_path,
mode=mode,
data=None)
continue
- item_it = cat_pipe.get(id)
- type = next(item_it)
- if type not in ('blob', 'commit', 'tree'):
- raise Exception('unexpected repository object type %r' % type)
+ item_it = get_ref(oidx)
+ get_oidx, typ, _ = next(item_it)
+ if not get_oidx:
+ raise MissingObject(oidx.decode('hex'))
+ if typ not in ('blob', 'commit', 'tree'):
+ raise Exception('unexpected repository object type %r' % typ)
# FIXME: set the mode based on the type when the mode is None
- if type == 'blob' and not include_data:
+ if typ == 'blob' and not include_data:
# Dump data until we can ask cat_pipe not to fetch it
for ignored in item_it:
pass
else:
data = ''.join(item_it)
- yield WalkItem(id=id, type=type,
+ yield WalkItem(oid=oid, type=typ,
chunk_path=chunk_path, path=parent_path,
mode=mode,
data=(data if include_data else None))
- if type == 'commit':
+ if typ == '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':
+ elif typ == 'tree':
for mode, name, ent_id in tree_decode(data):
demangled, bup_type = demangle_name(name, mode)
if chunk_path: