def validate_tree(src_id, dest_id):
rmrf(b'restore-src')
rmrf(b'restore-dest')
mkdir(b'restore-src')
mkdir(b'restore-dest')
commit_env = merge_dict(environ, {b'GIT_COMMITTER_DATE': b'2014-01-01 01:01'})
# Create a commit so the archive contents will have matching timestamps.
src_c = exo((b'git', b'--git-dir', b'get-src',
b'commit-tree', b'-m', b'foo', src_id),
env=commit_env).out.strip()
dest_c = exo((b'git', b'--git-dir', b'get-dest',
b'commit-tree', b'-m', b'foo', dest_id),
env=commit_env).out.strip()
exr = verify_rcz(b'git --git-dir get-src archive %s | tar xvf - -C restore-src'
% bquote(src_c),
shell=True)
if exr.rc != 0: return False
exr = verify_rcz(b'git --git-dir get-dest archive %s | tar xvf - -C restore-dest'
% bquote(dest_c),
shell=True)
if exr.rc != 0: return False
# git archive doesn't include an entry for ./.
unlink(b'restore-src/pax_global_header')
unlink(b'restore-dest/pax_global_header')
ex((b'touch', b'-r', b'restore-src', b'restore-dest'))
verify_trees_match(b'restore-src/', b'restore-dest/')
rmrf(b'restore-src')
rmrf(b'restore-dest')
def do_midx_dir(path, outfilename):
already = {}
sizes = {}
if opt.force and not opt.auto:
midxs = [] # don't use existing midx files
else:
midxs = glob.glob('%s/*.midx' % path)
contents = {}
for mname in midxs:
m = git.open_idx(mname)
contents[mname] = [('%s/%s' % (path,i)) for i in m.idxnames]
sizes[mname] = len(m)
# sort the biggest+newest midxes first, so that we can eliminate
# smaller (or older) redundant ones that come later in the list
midxs.sort(key=lambda ix: (-sizes[ix], -xstat.stat(ix).st_mtime))
for mname in midxs:
any = 0
for iname in contents[mname]:
if not already.get(iname):
already[iname] = 1
any = 1
if not any:
debug1('%r is redundant\n' % mname)
unlink(mname)
already[mname] = 1
midxs = [k for k in midxs if not already.get(k)]
idxs = [k for k in glob.glob('%s/*.idx' % path) if not already.get(k)]
for iname in idxs:
i = git.open_idx(iname)
sizes[iname] = len(i)
all = [(sizes[n],n) for n in (midxs + idxs)]
# FIXME: what are the optimal values? Does this make sense?
DESIRED_HWM = opt.force and 1 or 5
DESIRED_LWM = opt.force and 1 or 2
existed = dict((name,1) for sz,name in all)
debug1('midx: %d indexes; want no more than %d.\n'
% (len(all), DESIRED_HWM))
if len(all) <= DESIRED_HWM:
debug1('midx: nothing to do.\n')
while len(all) > DESIRED_HWM:
all.sort()
part1 = [name for sz,name in all[:len(all)-DESIRED_LWM+1]]
part2 = all[len(all)-DESIRED_LWM+1:]
all = list(do_midx_group(path, outfilename, part1)) + part2
if len(all) > DESIRED_HWM:
debug1('\nStill too many indexes (%d > %d). Merging again.\n'
% (len(all), DESIRED_HWM))
if opt['print']:
for sz,name in all:
if not existed.get(name):
print name
def do_midx_dir(path, outfilename):
already = {}
sizes = {}
if opt.force and not opt.auto:
midxs = [] # don't use existing midx files
else:
midxs = glob.glob('%s/*.midx' % path)
contents = {}
for mname in midxs:
m = git.open_idx(mname)
contents[mname] = [('%s/%s' % (path, i)) for i in m.idxnames]
sizes[mname] = len(m)
# sort the biggest+newest midxes first, so that we can eliminate
# smaller (or older) redundant ones that come later in the list
midxs.sort(key=lambda ix: (-sizes[ix], -xstat.stat(ix).st_mtime))
for mname in midxs:
any = 0
for iname in contents[mname]:
if not already.get(iname):
already[iname] = 1
any = 1
if not any:
debug1('%r is redundant\n' % mname)
unlink(mname)
already[mname] = 1
midxs = [k for k in midxs if not already.get(k)]
idxs = [k for k in glob.glob('%s/*.idx' % path) if not already.get(k)]
for iname in idxs:
i = git.open_idx(iname)
sizes[iname] = len(i)
all = [(sizes[n], n) for n in (midxs + idxs)]
# FIXME: what are the optimal values? Does this make sense?
DESIRED_HWM = opt.force and 1 or 5
DESIRED_LWM = opt.force and 1 or 2
existed = dict((name, 1) for sz, name in all)
debug1('midx: %d indexes; want no more than %d.\n' %
(len(all), DESIRED_HWM))
if len(all) <= DESIRED_HWM:
debug1('midx: nothing to do.\n')
while len(all) > DESIRED_HWM:
all.sort()
part1 = [name for sz, name in all[:len(all) - DESIRED_LWM + 1]]
part2 = all[len(all) - DESIRED_LWM + 1:]
all = list(do_midx_group(path, outfilename, part1)) + part2
if len(all) > DESIRED_HWM:
debug1('\nStill too many indexes (%d > %d). Merging again.\n' %
(len(all), DESIRED_HWM))
if opt['print']:
for sz, name in all:
if not existed.get(name):
print name
def create_path(n, fullname, meta):
if meta:
meta.create_path(fullname)
else:
# These fallbacks are important -- meta could be null if, for
# example, save created a "fake" item, i.e. a new strip/graft
# path element, etc. You can find cases like that by
# searching for "Metadata()".
unlink(fullname)
if stat.S_ISDIR(n.mode):
mkdirp(fullname)
elif stat.S_ISLNK(n.mode):
os.symlink(n.readlink(), fullname)
def prepare_save(self):
""" Commit all of the relevant data to disk. Do as much work
as possible without actually making the changes visible."""
if self._pending_save:
raise Error('save of %r already in progress' % self._filename)
with self._cleanup:
if self._node_paths:
dir, name = os.path.split(self._filename)
self._pending_save = atomically_replaced_file(self._filename,
mode='wb',
buffering=65536)
with self._cleanup.enter_context(self._pending_save) as f:
pickle.dump(self._node_paths, f, 2)
else: # No data
self._cleanup.callback(lambda: unlink(self._filename))
self._cleanup = self._cleanup.pop_all()
def refresh(self, skip_midx=False):
"""Refresh the index list.
This method verifies if .midx files were superseded (e.g. all of its
contents are in another, bigger .midx file) and removes the superseded
files.
If skip_midx is True, all work on .midx files will be skipped and .midx
files will be removed from the list.
The instance variable 'ignore_midx' can force this function to
always act as if skip_midx was True.
"""
if self.bloom is not None:
self.bloom.close()
self.bloom = None # Always reopen the bloom as it may have been relaced
self.do_bloom = False
skip_midx = skip_midx or self.ignore_midx
d = dict((p.name, p) for p in self.packs
if not skip_midx or not isinstance(p, midx.PackMidx))
if os.path.exists(self.dir):
if not skip_midx:
midxl = []
midxes = set(glob.glob(os.path.join(self.dir, b'*.midx')))
# remove any *.midx files from our list that no longer exist
for ix in list(d.values()):
if not isinstance(ix, midx.PackMidx):
continue
if ix.name in midxes:
continue
# remove the midx
del d[ix.name]
ix.close()
self.packs.remove(ix)
for ix in self.packs:
if isinstance(ix, midx.PackMidx):
for name in ix.idxnames:
d[os.path.join(self.dir, name)] = ix
for full in midxes:
if not d.get(full):
mx = midx.PackMidx(full)
(mxd, mxf) = os.path.split(mx.name)
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') %
(path_msg(n), path_msg(mxf)))
broken = True
if broken:
mx.close()
del mx
unlink(full)
else:
midxl.append(mx)
midxl.sort(
key=lambda ix: (-len(ix), -xstat.stat(ix.name).st_mtime))
for ix in midxl:
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
any_needed = True
break
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' %
path_msg(os.path.basename(ix.name)))
ix.close()
unlink(ix.name)
for full in glob.glob(os.path.join(self.dir, b'*.idx')):
if not d.get(full):
try:
ix = open_idx(full)
except GitError as e:
add_error(e)
continue
d[full] = ix
bfull = os.path.join(self.dir, b'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(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:
self.bloom = None
debug1('PackIdxList: using %d index%s.\n' %
(len(self.packs), len(self.packs) != 1 and 'es' or ''))
def _do_midx(outdir, outfilename, infilenames, prefixstr,
auto=False, force=False):
global _first
if not outfilename:
assert(outdir)
sum = hexlify(Sha1(b'\0'.join(infilenames)).digest())
outfilename = b'%s/midx-%s.midx' % (outdir, sum)
inp = []
total = 0
allfilenames = []
midxs = []
try:
for name in infilenames:
ix = git.open_idx(name)
midxs.append(ix)
inp.append((
ix.map,
len(ix),
ix.sha_ofs,
isinstance(ix, midx.PackMidx) and ix.which_ofs or 0,
len(allfilenames),
))
for n in ix.idxnames:
allfilenames.append(os.path.basename(n))
total += len(ix)
inp.sort(reverse=True, key=lambda x: x[0][x[2] : x[2] + 20])
if not _first: _first = outdir
dirprefix = (_first != outdir) and git.repo_rel(outdir) + b': ' or b''
debug1('midx: %s%screating from %d files (%d objects).\n'
% (dirprefix, prefixstr, len(infilenames), total))
if (auto and (total < 1024 and len(infilenames) < 3)) \
or ((auto or force) and len(infilenames) < 2) \
or (force and not total):
debug1('midx: nothing to do.\n')
return
pages = int(total/SHA_PER_PAGE) or 1
bits = int(math.ceil(math.log(pages, 2)))
entries = 2**bits
debug1('midx: table size: %d (%d bits)\n' % (entries*4, bits))
unlink(outfilename)
with atomically_replaced_file(outfilename, 'wb') as f:
f.write(b'MIDX')
f.write(struct.pack('!II', midx.MIDX_VERSION, bits))
assert(f.tell() == 12)
f.truncate(12 + 4*entries + 20*total + 4*total)
f.flush()
fdatasync(f.fileno())
fmap = mmap_readwrite(f, close=False)
count = merge_into(fmap, bits, total, inp)
del fmap # Assume this calls msync() now.
f.seek(0, os.SEEK_END)
f.write(b'\0'.join(allfilenames))
finally:
for ix in midxs:
if isinstance(ix, midx.PackMidx):
ix.close()
midxs = None
inp = None
# This is just for testing (if you enable this, don't clear inp above)
if 0:
p = midx.PackMidx(outfilename)
assert(len(p.idxnames) == len(infilenames))
log(repr(p.idxnames) + '\n')
assert(len(p) == total)
for pe, e in p, git.idxmerge(inp, final_progress=False):
pin = next(pi)
assert(i == pin)
assert(p.exists(i))
return total, outfilename
def refresh(self, skip_midx = False):
"""Refresh the index list.
This method verifies if .midx files were superseded (e.g. all of its
contents are in another, bigger .midx file) and removes the superseded
files.
If skip_midx is True, all work on .midx files will be skipped and .midx
files will be removed from the list.
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, midx.PackMidx))
if os.path.exists(self.dir):
if not skip_midx:
midxl = []
for ix in self.packs:
if isinstance(ix, midx.PackMidx):
for name in ix.idxnames:
d[os.path.join(self.dir, name)] = ix
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 = 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 = True
if broken:
mx.close()
del mx
unlink(full)
else:
midxl.append(mx)
midxl.sort(key=lambda ix:
(-len(ix), -xstat.stat(ix.name).st_mtime))
for ix in midxl:
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
any_needed = True
break
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 full in glob.glob(os.path.join(self.dir,'*.idx')):
if not d.get(full):
try:
ix = open_idx(full)
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 clear_bloom(dir):
unlink(os.path.join(dir, 'bup.bloom'))
def cleanup_testfs():
subprocess.call(['umount', 'testfs'])
helpers.unlink('testfs.img')
def clear_bloom(dir):
unlink(os.path.join(dir, 'bup.bloom'))
def cleanup_testfs():
subprocess.call(['umount', 'testfs'])
helpers.unlink('testfs.img')
def _do_midx(outdir, outfilename, infilenames, prefixstr):
global _first
if not outfilename:
assert(outdir)
sum = Sha1('\0'.join(infilenames)).hexdigest()
outfilename = '%s/midx-%s.midx' % (outdir, sum)
inp = []
total = 0
allfilenames = []
midxs = []
try:
for name in infilenames:
ix = git.open_idx(name)
midxs.append(ix)
inp.append((
ix.map,
len(ix),
ix.sha_ofs,
isinstance(ix, midx.PackMidx) and ix.which_ofs or 0,
len(allfilenames),
))
for n in ix.idxnames:
allfilenames.append(os.path.basename(n))
total += len(ix)
inp.sort(lambda x,y: cmp(str(y[0][y[2]:y[2]+20]),str(x[0][x[2]:x[2]+20])))
if not _first: _first = outdir
dirprefix = (_first != outdir) and git.repo_rel(outdir)+': ' or ''
debug1('midx: %s%screating from %d files (%d objects).\n'
% (dirprefix, prefixstr, len(infilenames), total))
if (opt.auto and (total < 1024 and len(infilenames) < 3)) \
or ((opt.auto or opt.force) and len(infilenames) < 2) \
or (opt.force and not total):
debug1('midx: nothing to do.\n')
return
pages = int(total/SHA_PER_PAGE) or 1
bits = int(math.ceil(math.log(pages, 2)))
entries = 2**bits
debug1('midx: table size: %d (%d bits)\n' % (entries*4, bits))
unlink(outfilename)
with atomically_replaced_file(outfilename, 'wb') as f:
f.write('MIDX')
f.write(struct.pack('!II', midx.MIDX_VERSION, bits))
assert(f.tell() == 12)
f.truncate(12 + 4*entries + 20*total + 4*total)
f.flush()
fdatasync(f.fileno())
fmap = mmap_readwrite(f, close=False)
count = merge_into(fmap, bits, total, inp)
del fmap # Assume this calls msync() now.
f.seek(0, os.SEEK_END)
f.write('\0'.join(allfilenames))
finally:
for ix in midxs:
if isinstance(ix, midx.PackMidx):
ix.close()
midxs = None
inp = None
# This is just for testing (if you enable this, don't clear inp above)
if 0:
p = midx.PackMidx(outfilename)
assert(len(p.idxnames) == len(infilenames))
print p.idxnames
assert(len(p) == total)
for pe, e in p, git.idxmerge(inp, final_progress=False):
pin = pi.next()
assert(i == pin)
assert(p.exists(i))
return total, outfilename
def cleanup_testfs():
subprocess.call(["umount", "testfs"])
helpers.unlink("testfs.img")
