interact with the Git data structures.
"""
import os, sys, zlib, time, subprocess, struct, stat, re, tempfile, glob
+from collections import namedtuple
+
from bup.helpers import *
-from bup import _helpers, path, midx, bloom
+from bup import _helpers, path, midx, bloom, xstat
-SEEK_END=2 # os.SEEK_END is not defined in python 2.4
+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
_typemap = { 'blob':3, 'tree':2, 'commit':1, 'tag':4 }
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<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)
+
+(?P<message>(?:.|\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 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):
"""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'
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)
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):
szout = ''
sz = len(content)
szbits = (sz & 0x0f) | (_typemap[type]<<4)
break
szbits = sz & 0x7f
sz >>= 7
- z = zlib.compressobj(1)
+ if compression_level > 9:
+ compression_level = 9
+ elif compression_level < 0:
+ compression_level = 0
+ 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()
' used by %s\n') % (n, mxf))
broken = True
if broken:
+ mx.close()
del mx
unlink(full)
else:
midxl.append(mx)
midxl.sort(key=lambda ix:
- (-len(ix), -os.stat(ix.name).st_mtime))
+ (-len(ix), -xstat.stat(ix.name).st_mtime))
for ix in midxl:
any_needed = False
for sub in ix.idxnames:
elif not ix.force_keep:
debug1('midx: removing redundant: %s\n'
% os.path.basename(ix.name))
+ ix.close()
unlink(ix.name)
for full in glob.glob(os.path.join(self.dir,'*.idx')):
if not d.get(full):
self.also.add(hash)
-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
-
-
def open_idx(filename):
if filename.endswith('.idx'):
f = open(filename, 'rb')
class PackWriter:
"""Writes Git objects inside a pack file."""
- def __init__(self, objcache_maker=_make_objcache):
+ def __init__(self, objcache_maker=_make_objcache, compression_level=1):
self.count = 0
self.outbytes = 0
self.filename = None
self.idx = None
self.objcache_maker = objcache_maker
self.objcache = None
+ self.compression_level = compression_level
def __del__(self):
self.close()
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):
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._require_objcache()
self.objcache.add(sha)
return sha
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))
+ content = tree_encode(shalist)
+ return self.maybe_write('tree', content)
def _new_commit(self, tree, parent, author, adate, committer, cdate, msg):
l = []
return self._end(run_midx=run_midx)
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')
- idx_f.write('\377tOc\0\0\0\2')
-
- ofs64_ofs = 8 + 4*256 + 28*self.count
- idx_f.truncate(ofs64_ofs)
- idx_f.seek(0)
- idx_map = mmap_readwrite(idx_f, close=False)
- idx_f.seek(0, SEEK_END)
- count = _helpers.write_idx(idx_f, idx_map, idx, self.count)
- assert(count == self.count)
- idx_map.close()
- 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):
+ try:
+ idx_f.truncate(index_len)
+ idx_map = mmap_readwrite(idx_f, close=False)
+ count = _helpers.write_idx(filename, idx_map, idx, self.count)
+ assert(count == self.count)
+ finally:
+ if idx_map: idx_map.close()
+ 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.update(b)
- namebase = obj_list_sum.hexdigest()
- for b in chunkyreader(idx_f):
- idx_sum.update(b)
- idx_f.write(idx_sum.digest())
- idx_f.close()
+ 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()
- return namebase
+ for b in chunkyreader(idx_f):
+ idx_sum.update(b)
+ idx_f.write(idx_sum.digest())
+ return namebase
+ finally:
+ idx_f.close()
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 _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(refname = None):
+def list_refs(refname = None, repo_dir = 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)
+ p = subprocess.Popen(argv,
+ preexec_fn = _gitenv(repo_dir),
+ stdout = subprocess.PIPE)
out = p.stdout.read().strip()
rv = p.wait() # not fatal
if rv:
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))
+ l = list(list_refs(refname, repo_dir))
if l:
assert(len(l) == 1)
return l[0][1]
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
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()
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'.
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:
return None
-def update_ref(refname, newval, oldval):
+def update_ref(refname, newval, oldval, repo_dir=None):
"""Change the commit pointed to by a branch."""
if not oldval:
oldval = ''
assert(refname.startswith('refs/heads/'))
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)
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)
initializes the default repository automatically.
"""
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())
+ try:
+ os.stat(repo('objects/pack/.'))
+ except OSError, e:
+ if e.errno == errno.ENOENT:
+ log('error: %r is not a bup repository; run "bup init"\n'
+ % repo())
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)
+ else:
+ log('error: %s\n' % e)
+ sys.exit(14)
_ver = None
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
_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:
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()
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)
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:
it.abort()
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)
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':
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
+ if not repo_dir:
+ repo_dir = 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():
+ for (n,c) in list_refs(repo_dir = repo_dir):
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