def get_packed_refs(self):
"""Get contents of the packed-refs file.
:return: Dictionary mapping ref names to SHA1s
:note: Will return an empty dictionary when no packed-refs file is
present.
"""
# TODO: invalidate the cache on repacking
if self._packed_refs is None:
# set both to empty because we want _peeled_refs to be
# None if and only if _packed_refs is also None.
self._packed_refs = {}
self._peeled_refs = {}
path = os.path.join(self.path, b'packed-refs')
try:
f = GitFile(path, 'rb')
except IOError as e:
if e.errno == errno.ENOENT:
return {}
raise
with f:
first_line = next(iter(f)).rstrip()
if (first_line.startswith(b'# pack-refs') and b' peeled' in
first_line):
for sha, name, peeled in read_packed_refs_with_peeled(f):
self._packed_refs[name] = sha
if peeled:
self._peeled_refs[name] = peeled
else:
f.seek(0)
for sha, name in read_packed_refs(f):
self._packed_refs[name] = sha
return self._packed_refs
def get_packed_refs(self):
"""Get contents of the packed-refs file.
Returns: Dictionary mapping ref names to SHA1s
Note: Will return an empty dictionary when no packed-refs file is
present.
"""
# TODO: invalidate the cache on repacking
if self._packed_refs is None:
# set both to empty because we want _peeled_refs to be
# None if and only if _packed_refs is also None.
self._packed_refs = {}
self._peeled_refs = {}
path = os.path.join(self.path, b"packed-refs")
try:
f = GitFile(path, "rb")
except FileNotFoundError:
return {}
with f:
first_line = next(iter(f)).rstrip()
if first_line.startswith(b"# pack-refs") and b" peeled" in first_line:
for sha, name, peeled in read_packed_refs_with_peeled(f):
self._packed_refs[name] = sha
if peeled:
self._peeled_refs[name] = peeled
else:
f.seek(0)
for sha, name in read_packed_refs(f):
self._packed_refs[name] = sha
return self._packed_refs
def get_packed_refs(self):
"""Get contents of the packed-refs file.
:return: Dictionary mapping ref names to SHA1s
:note: Will return an empty dictionary when no packed-refs file is
present.
"""
# TODO: invalidate the cache on repacking
if self._packed_refs is None:
# set both to empty because we want _peeled_refs to be
# None if and only if _packed_refs is also None.
self._packed_refs = {}
self._peeled_refs = {}
path = os.path.join(self.path, b'packed-refs')
try:
f = GitFile(path, 'rb')
except IOError as e:
if e.errno == errno.ENOENT:
return {}
raise
with f:
first_line = next(iter(f)).rstrip()
if (first_line.startswith(b'# pack-refs')
and b' peeled' in first_line):
for sha, name, peeled in read_packed_refs_with_peeled(f):
self._packed_refs[name] = sha
if peeled:
self._peeled_refs[name] = peeled
else:
f.seek(0)
for sha, name in read_packed_refs(f):
self._packed_refs[name] = sha
return self._packed_refs
def get_packed_refs(self):
"""Get contents of the packed-refs file.
:return: Dictionary mapping ref names to SHA1s
:note: Will return an empty dictionary when no packed-refs file is
present.
"""
# TODO: invalidate the cache on repacking
if self._packed_refs is None:
self._packed_refs = {}
path = os.path.join(self.path, 'packed-refs')
try:
f = GitFile(path, 'rb')
except IOError, e:
if e.errno == errno.ENOENT:
return {}
raise
try:
first_line = iter(f).next().rstrip()
if (first_line.startswith("# pack-refs") and " peeled" in
first_line):
self._peeled_refs = {}
for sha, name, peeled in read_packed_refs_with_peeled(f):
self._packed_refs[name] = sha
if peeled:
self._peeled_refs[name] = peeled
else:
f.seek(0)
for sha, name in read_packed_refs(f):
self._packed_refs[name] = sha
finally:
f.close()
def test_readonly(self):
f = GitFile(self.path('foo'), 'rb')
self.assertTrue(isinstance(f, io.IOBase))
self.assertEqual(b'foo contents', f.read())
self.assertEqual(b'', f.read())
f.seek(4)
self.assertEqual(b'contents', f.read())
f.close()
def test_readonly(self):
f = GitFile(self.path('foo'), 'rb')
self.assertTrue(isinstance(f, io.IOBase))
self.assertEqual(b'foo contents', f.read())
self.assertEqual(b'', f.read())
f.seek(4)
self.assertEqual(b'contents', f.read())
f.close()
def test_readonly(self):
f = GitFile(self.path("foo"), "rb")
self.assertTrue(isinstance(f, io.IOBase))
self.assertEqual(b"foo contents", f.read())
self.assertEqual(b"", f.read())
f.seek(4)
self.assertEqual(b"contents", f.read())
f.close()
def test_readonly(self):
f = GitFile(self.path("foo"), "rb")
self.assertTrue(isinstance(f, io.IOBase))
self.assertEqual(b"foo contents", f.read())
self.assertEqual(b"", f.read())
f.seek(4)
self.assertEqual(b"contents", f.read())
f.close()
def test_readonly(self):
import _io
f = GitFile(self.path('foo'), 'rb')
self.assertTrue(isinstance(f, _io.BufferedReader))
self.assertEqual(b'foo contents', f.read())
self.assertEqual(b'', f.read())
f.seek(4)
self.assertEqual(b'contents', f.read())
f.close()
def test_write(self):
foo = self.path('foo')
foo_lock = '%s.lock' % foo
with open(foo, 'rb') as orig_f:
self.assertEqual(orig_f.read(), b'foo contents')
self.assertFalse(os.path.exists(foo_lock))
f = GitFile(foo, 'wb')
self.assertFalse(f.closed)
self.assertRaises(AttributeError, getattr, f, 'not_a_file_property')
self.assertTrue(os.path.exists(foo_lock))
f.write(b'new stuff')
f.seek(4)
f.write(b'contents')
f.close()
self.assertFalse(os.path.exists(foo_lock))
with open(foo, 'rb') as new_f:
self.assertEqual(b'new contents', new_f.read())
def test_write(self):
foo = self.path('foo')
foo_lock = '%s.lock' % foo
orig_f = open(foo, 'rb')
self.assertEqual(orig_f.read(), b'foo contents')
orig_f.close()
self.assertFalse(os.path.exists(foo_lock))
f = GitFile(foo, 'wb')
self.assertFalse(f.closed)
self.assertRaises(AttributeError, getattr, f, 'not_a_file_property')
self.assertTrue(os.path.exists(foo_lock))
f.write(b'new stuff')
f.seek(4)
f.write(b'contents')
f.close()
self.assertFalse(os.path.exists(foo_lock))
new_f = open(foo, 'rb')
self.assertEqual(b'new contents', new_f.read())
new_f.close()
def test_write(self):
foo = self.path("foo")
foo_lock = "%s.lock" % foo
orig_f = open(foo, "rb")
self.assertEqual(orig_f.read(), b"foo contents")
orig_f.close()
self.assertFalse(os.path.exists(foo_lock))
f = GitFile(foo, "wb")
self.assertFalse(f.closed)
self.assertRaises(AttributeError, getattr, f, "not_a_file_property")
self.assertTrue(os.path.exists(foo_lock))
f.write(b"new stuff")
f.seek(4)
f.write(b"contents")
f.close()
self.assertFalse(os.path.exists(foo_lock))
new_f = open(foo, "rb")
self.assertEqual(b"new contents", new_f.read())
new_f.close()
def test_write(self):
foo = self.path("foo")
foo_lock = "%s.lock" % foo
orig_f = open(foo, "rb")
self.assertEqual(orig_f.read(), b"foo contents")
orig_f.close()
self.assertFalse(os.path.exists(foo_lock))
f = GitFile(foo, "wb")
self.assertFalse(f.closed)
self.assertRaises(AttributeError, getattr, f, "not_a_file_property")
self.assertTrue(os.path.exists(foo_lock))
f.write(b"new stuff")
f.seek(4)
f.write(b"contents")
f.close()
self.assertFalse(os.path.exists(foo_lock))
new_f = open(foo, "rb")
self.assertEqual(b"new contents", new_f.read())
new_f.close()
class PackData(object):
"""The data contained in a packfile.
Pack files can be accessed both sequentially for exploding a pack, and
directly with the help of an index to retrieve a specific object.
The objects within are either complete or a delta aginst another.
The header is variable length. If the MSB of each byte is set then it
indicates that the subsequent byte is still part of the header.
For the first byte the next MS bits are the type, which tells you the type
of object, and whether it is a delta. The LS byte is the lowest bits of the
size. For each subsequent byte the LS 7 bits are the next MS bits of the
size, i.e. the last byte of the header contains the MS bits of the size.
For the complete objects the data is stored as zlib deflated data.
The size in the header is the uncompressed object size, so to uncompress
you need to just keep feeding data to zlib until you get an object back,
or it errors on bad data. This is done here by just giving the complete
buffer from the start of the deflated object on. This is bad, but until I
get mmap sorted out it will have to do.
Currently there are no integrity checks done. Also no attempt is made to try
and detect the delta case, or a request for an object at the wrong position.
It will all just throw a zlib or KeyError.
"""
def __init__(self, filename, file=None, size=None):
"""Create a PackData object that represents the pack in the given filename.
The file must exist and stay readable until the object is disposed of. It
must also stay the same size. It will be mapped whenever needed.
Currently there is a restriction on the size of the pack as the python
mmap implementation is flawed.
"""
self._filename = filename
self._size = size
self._header_size = 12
if file is None:
self._file = GitFile(self._filename, 'rb')
else:
self._file = file
(version, self._num_objects) = read_pack_header(self._file)
self._offset_cache = LRUSizeCache(1024*1024*20,
compute_size=_compute_object_size)
@classmethod
def from_file(cls, file, size):
return cls(str(file), file=file, size=size)
@classmethod
def from_path(cls, path):
return cls(filename=path)
def close(self):
self._file.close()
def _get_size(self):
if self._size is not None:
return self._size
self._size = os.path.getsize(self._filename)
assert self._size >= self._header_size, "%s is too small for a packfile (%d < %d)" % (self._filename, self._size, self._header_size)
return self._size
def __len__(self):
"""Returns the number of objects in this pack."""
return self._num_objects
def calculate_checksum(self):
"""Calculate the checksum for this pack.
:return: 20-byte binary SHA1 digest
"""
s = make_sha()
self._file.seek(0)
todo = self._get_size() - 20
while todo > 0:
x = self._file.read(min(todo, 1<<16))
s.update(x)
todo -= len(x)
return s.digest()
def resolve_object(self, offset, type, obj, get_ref, get_offset=None):
"""Resolve an object, possibly resolving deltas when necessary.
:return: Tuple with object type and contents.
"""
if type not in (6, 7): # Not a delta
return type, obj
if get_offset is None:
get_offset = self.get_object_at
if type == 6: # offset delta
(delta_offset, delta) = obj
assert isinstance(delta_offset, int)
assert isinstance(delta, str)
base_offset = offset-delta_offset
type, base_obj = get_offset(base_offset)
assert isinstance(type, int)
elif type == 7: # ref delta
(basename, delta) = obj
assert isinstance(basename, str) and len(basename) == 20
assert isinstance(delta, str)
type, base_obj = get_ref(basename)
assert isinstance(type, int)
# Can't be a ofs delta, as we wouldn't know the base offset
assert type != 6
base_offset = None
type, base_text = self.resolve_object(base_offset, type, base_obj, get_ref)
if base_offset is not None:
self._offset_cache[base_offset] = type, base_text
ret = (type, apply_delta(base_text, delta))
return ret
def iterobjects(self, progress=None):
class ObjectIterator(object):
def __init__(self, pack):
self.i = 0
self.offset = pack._header_size
self.num = len(pack)
self.map = pack._file
def __iter__(self):
return self
def __len__(self):
return self.num
def next(self):
if self.i == self.num:
raise StopIteration
self.map.seek(self.offset)
(type, obj, total_size, unused) = unpack_object(self.map.read)
self.map.seek(self.offset)
crc32 = zlib.crc32(self.map.read(total_size)) & 0xffffffff
ret = (self.offset, type, obj, crc32)
self.offset += total_size
if progress:
progress(self.i, self.num)
self.i+=1
return ret
return ObjectIterator(self)
def iterentries(self, ext_resolve_ref=None, progress=None):
"""Yield entries summarizing the contents of this pack.
:param ext_resolve_ref: Optional function to resolve base
objects (in case this is a thin pack)
:param progress: Progress function, called with current and
total object count.
This will yield tuples with (sha, offset, crc32)
"""
found = {}
postponed = defaultdict(list)
class Postpone(Exception):
"""Raised to postpone delta resolving."""
def get_ref_text(sha):
assert len(sha) == 20
if sha in found:
return self.get_object_at(found[sha])
if ext_resolve_ref:
try:
return ext_resolve_ref(sha)
except KeyError:
pass
raise Postpone, (sha, )
extra = []
todo = chain(self.iterobjects(progress=progress), extra)
for (offset, type, obj, crc32) in todo:
assert isinstance(offset, int)
assert isinstance(type, int)
assert isinstance(obj, tuple) or isinstance(obj, str)
try:
type, obj = self.resolve_object(offset, type, obj, get_ref_text)
except Postpone, (sha, ):
postponed[sha].append((offset, type, obj))
else:
shafile = ShaFile.from_raw_string(type, obj)
sha = shafile.sha().digest()
found[sha] = offset
yield sha, offset, crc32
extra.extend(postponed.get(sha, []))
if postponed:
raise KeyError([sha_to_hex(h) for h in postponed.keys()])
class PackIndex(object):
"""An index in to a packfile.
Given a sha id of an object a pack index can tell you the location in the
packfile of that object if it has it.
To do the loop it opens the file, and indexes first 256 4 byte groups
with the first byte of the sha id. The value in the four byte group indexed
is the end of the group that shares the same starting byte. Subtract one
from the starting byte and index again to find the start of the group.
The values are sorted by sha id within the group, so do the math to find
the start and end offset and then bisect in to find if the value is present.
"""
def __init__(self, filename, file=None, size=None):
"""Create a pack index object.
Provide it with the name of the index file to consider, and it will map
it whenever required.
"""
self._filename = filename
# Take the size now, so it can be checked each time we map the file to
# ensure that it hasn't changed.
if file is None:
self._file = GitFile(filename, 'rb')
else:
self._file = file
fileno = getattr(self._file, 'fileno', None)
if fileno is not None:
fd = self._file.fileno()
if size is None:
self._size = os.fstat(fd).st_size
else:
self._size = size
self._contents = mmap.mmap(fd, self._size,
access=mmap.ACCESS_READ)
else:
self._file.seek(0)
self._contents = self._file.read()
self._size = len(self._contents)
def __eq__(self, other):
if not isinstance(other, PackIndex):
return False
if self._fan_out_table != other._fan_out_table:
return False
for (name1, _, _), (name2, _, _) in izip(self.iterentries(), other.iterentries()):
if name1 != name2:
return False
return True
def __ne__(self, other):
return not self.__eq__(other)
def close(self):
self._file.close()
def __len__(self):
"""Return the number of entries in this pack index."""
return self._fan_out_table[-1]
def _unpack_entry(self, i):
"""Unpack the i-th entry in the index file.
:return: Tuple with object name (SHA), offset in pack file and
CRC32 checksum (if known)."""
raise NotImplementedError(self._unpack_entry)
def _unpack_name(self, i):
"""Unpack the i-th name from the index file."""
raise NotImplementedError(self._unpack_name)
def _unpack_offset(self, i):
"""Unpack the i-th object offset from the index file."""
raise NotImplementedError(self._unpack_offset)
def _unpack_crc32_checksum(self, i):
"""Unpack the crc32 checksum for the i-th object from the index file."""
raise NotImplementedError(self._unpack_crc32_checksum)
def __iter__(self):
"""Iterate over the SHAs in this pack."""
return imap(sha_to_hex, self._itersha())
def _itersha(self):
for i in range(len(self)):
yield self._unpack_name(i)
def objects_sha1(self):
"""Return the hex SHA1 over all the shas of all objects in this pack.
:note: This is used for the filename of the pack.
"""
return iter_sha1(self._itersha())
def iterentries(self):
"""Iterate over the entries in this pack index.
Will yield tuples with object name, offset in packfile and crc32 checksum.
"""
for i in range(len(self)):
yield self._unpack_entry(i)
def _read_fan_out_table(self, start_offset):
ret = []
for i in range(0x100):
ret.append(struct.unpack(">L", self._contents[start_offset+i*4:start_offset+(i+1)*4])[0])
return ret
def check(self):
"""Check that the stored checksum matches the actual checksum."""
# TODO: Check pack contents, too
return self.calculate_checksum() == self.get_stored_checksum()
def calculate_checksum(self):
"""Calculate the SHA1 checksum over this pack index.
:return: This is a 20-byte binary digest
"""
return make_sha(self._contents[:-20]).digest()
def get_pack_checksum(self):
"""Return the SHA1 checksum stored for the corresponding packfile.
:return: 20-byte binary digest
"""
return str(self._contents[-40:-20])
def get_stored_checksum(self):
"""Return the SHA1 checksum stored for this index.
:return: 20-byte binary digest
"""
return str(self._contents[-20:])
def object_index(self, sha):
"""Return the index in to the corresponding packfile for the object.
Given the name of an object it will return the offset that object lives
at within the corresponding pack file. If the pack file doesn't have the
object then None will be returned.
"""
if len(sha) == 40:
sha = hex_to_sha(sha)
return self._object_index(sha)
def _object_index(self, sha):
"""See object_index.
:param sha: A *binary* SHA string. (20 characters long)_
"""
assert len(sha) == 20
idx = ord(sha[0])
if idx == 0:
start = 0
else:
start = self._fan_out_table[idx-1]
end = self._fan_out_table[idx]
i = bisect_find_sha(start, end, sha, self._unpack_name)
if i is None:
raise KeyError(sha)
return self._unpack_offset(i)
class PackData(object):
"""The data contained in a packfile.
Pack files can be accessed both sequentially for exploding a pack, and
directly with the help of an index to retrieve a specific object.
The objects within are either complete or a delta aginst another.
The header is variable length. If the MSB of each byte is set then it
indicates that the subsequent byte is still part of the header.
For the first byte the next MS bits are the type, which tells you the type
of object, and whether it is a delta. The LS byte is the lowest bits of the
size. For each subsequent byte the LS 7 bits are the next MS bits of the
size, i.e. the last byte of the header contains the MS bits of the size.
For the complete objects the data is stored as zlib deflated data.
The size in the header is the uncompressed object size, so to uncompress
you need to just keep feeding data to zlib until you get an object back,
or it errors on bad data. This is done here by just giving the complete
buffer from the start of the deflated object on. This is bad, but until I
get mmap sorted out it will have to do.
Currently there are no integrity checks done. Also no attempt is made to
try and detect the delta case, or a request for an object at the wrong
position. It will all just throw a zlib or KeyError.
"""
def __init__(self, filename, file=None, size=None):
"""Create a PackData object representing the pack in the given filename.
The file must exist and stay readable until the object is disposed of. It
must also stay the same size. It will be mapped whenever needed.
Currently there is a restriction on the size of the pack as the python
mmap implementation is flawed.
"""
self._filename = filename
self._size = size
self._header_size = 12
if file is None:
self._file = GitFile(self._filename, 'rb')
else:
self._file = file
(version, self._num_objects) = read_pack_header(self._file.read)
self._offset_cache = LRUSizeCache(1024*1024*20,
compute_size=_compute_object_size)
self.pack = None
@classmethod
def from_file(cls, file, size):
return cls(str(file), file=file, size=size)
@classmethod
def from_path(cls, path):
return cls(filename=path)
def close(self):
self._file.close()
def _get_size(self):
if self._size is not None:
return self._size
self._size = os.path.getsize(self._filename)
if self._size < self._header_size:
errmsg = ("%s is too small for a packfile (%d < %d)" %
(self._filename, self._size, self._header_size))
raise AssertionError(errmsg)
return self._size
def __len__(self):
"""Returns the number of objects in this pack."""
return self._num_objects
def calculate_checksum(self):
"""Calculate the checksum for this pack.
:return: 20-byte binary SHA1 digest
"""
s = make_sha()
self._file.seek(0)
todo = self._get_size() - 20
while todo > 0:
x = self._file.read(min(todo, 1<<16))
s.update(x)
todo -= len(x)
return s.digest()
def get_ref(self, sha):
"""Get the object for a ref SHA, only looking in this pack."""
# TODO: cache these results
if self.pack is None:
raise KeyError(sha)
offset = self.pack.index.object_index(sha)
if not offset:
raise KeyError(sha)
type, obj = self.get_object_at(offset)
return offset, type, obj
def resolve_object(self, offset, type, obj, get_ref=None):
"""Resolve an object, possibly resolving deltas when necessary.
:return: Tuple with object type and contents.
"""
if type not in DELTA_TYPES:
return type, obj
if get_ref is None:
get_ref = self.get_ref
if type == OFS_DELTA:
(delta_offset, delta) = obj
# TODO: clean up asserts and replace with nicer error messages
assert isinstance(offset, int)
assert isinstance(delta_offset, int)
base_offset = offset-delta_offset
type, base_obj = self.get_object_at(base_offset)
assert isinstance(type, int)
elif type == REF_DELTA:
(basename, delta) = obj
assert isinstance(basename, str) and len(basename) == 20
base_offset, type, base_obj = get_ref(basename)
assert isinstance(type, int)
type, base_chunks = self.resolve_object(base_offset, type, base_obj)
chunks = apply_delta(base_chunks, delta)
# TODO(dborowitz): This can result in poor performance if large base
# objects are separated from deltas in the pack. We should reorganize
# so that we apply deltas to all objects in a chain one after the other
# to optimize cache performance.
if offset is not None:
self._offset_cache[offset] = type, chunks
return type, chunks
def iterobjects(self, progress=None):
return PackObjectIterator(self, progress)
def iterentries(self, progress=None):
"""Yield entries summarizing the contents of this pack.
:param progress: Progress function, called with current and total
object count.
:return: iterator of tuples with (sha, offset, crc32)
"""
for offset, type, obj, crc32 in self.iterobjects(progress=progress):
assert isinstance(offset, int)
assert isinstance(type, int)
assert isinstance(obj, list) or isinstance(obj, tuple)
type, obj = self.resolve_object(offset, type, obj)
yield obj_sha(type, obj), offset, crc32
def sorted_entries(self, progress=None):
"""Return entries in this pack, sorted by SHA.
:param progress: Progress function, called with current and total
object count
:return: List of tuples with (sha, offset, crc32)
"""
ret = list(self.iterentries(progress=progress))
ret.sort()
return ret
def create_index_v1(self, filename, progress=None):
"""Create a version 1 file for this data file.
:param filename: Index filename.
:param progress: Progress report function
:return: Checksum of index file
"""
entries = self.sorted_entries(progress=progress)
f = GitFile(filename, 'wb')
try:
return write_pack_index_v1(f, entries, self.calculate_checksum())
finally:
f.close()
def create_index_v2(self, filename, progress=None):
"""Create a version 2 index file for this data file.
:param filename: Index filename.
:param progress: Progress report function
:return: Checksum of index file
"""
entries = self.sorted_entries(progress=progress)
f = GitFile(filename, 'wb')
try:
return write_pack_index_v2(f, entries, self.calculate_checksum())
finally:
f.close()
def create_index(self, filename, progress=None,
version=2):
"""Create an index file for this data file.
:param filename: Index filename.
:param progress: Progress report function
:return: Checksum of index file
"""
if version == 1:
return self.create_index_v1(filename, progress)
elif version == 2:
return self.create_index_v2(filename, progress)
else:
raise ValueError("unknown index format %d" % version)
def get_stored_checksum(self):
"""Return the expected checksum stored in this pack."""
self._file.seek(self._get_size()-20)
return self._file.read(20)
def check(self):
"""Check the consistency of this pack."""
actual = self.calculate_checksum()
stored = self.get_stored_checksum()
if actual != stored:
raise ChecksumMismatch(stored, actual)
def get_object_at(self, offset):
"""Given an offset in to the packfile return the object that is there.
Using the associated index the location of an object can be looked up,
and then the packfile can be asked directly for that object using this
function.
"""
if offset in self._offset_cache:
return self._offset_cache[offset]
assert isinstance(offset, long) or isinstance(offset, int),\
"offset was %r" % offset
assert offset >= self._header_size
self._file.seek(offset)
return unpack_object(self._file.read)[:2]