# Copyright (C) 2010, 2011 Sebastian Thiel (byronimo@gmail.com) and contributors
#
# This module is part of GitDB and is released under
# the New BSD License: https://opensource.org/license/bsd-3-clause/
"""Contains basic c-functions which usually contain performance critical code
Keeping this code separate from the beginning makes it easier to out-source
it into c later, if required"""
import zlib
from gitdb.util import byte_ord
decompressobj = zlib.decompressobj
import mmap
from itertools import islice
from functools import reduce
from gitdb.const import NULL_BYTE, BYTE_SPACE
from gitdb.utils.encoding import force_text
from gitdb.typ import (
str_blob_type,
str_commit_type,
str_tree_type,
str_tag_type,
)
from io import StringIO
# INVARIANTS
OFS_DELTA = 6
REF_DELTA = 7
delta_types = (OFS_DELTA, REF_DELTA)
type_id_to_type_map = {
0: b'', # EXT 1
1: str_commit_type,
2: str_tree_type,
3: str_blob_type,
4: str_tag_type,
5: b'', # EXT 2
OFS_DELTA: "OFS_DELTA", # OFFSET DELTA
REF_DELTA: "REF_DELTA" # REFERENCE DELTA
}
type_to_type_id_map = {
str_commit_type: 1,
str_tree_type: 2,
str_blob_type: 3,
str_tag_type: 4,
"OFS_DELTA": OFS_DELTA,
"REF_DELTA": REF_DELTA,
}
# used when dealing with larger streams
chunk_size = 1000 * mmap.PAGESIZE
__all__ = ('is_loose_object', 'loose_object_header_info', 'msb_size', 'pack_object_header_info',
'write_object', 'loose_object_header', 'stream_copy', 'apply_delta_data',
'is_equal_canonical_sha', 'connect_deltas', 'DeltaChunkList', 'create_pack_object_header')
#{ Structures
def _set_delta_rbound(d, size):
"""Truncate the given delta to the given size
:param size: size relative to our target offset, may not be 0, must be smaller or equal
to our size
:return: d"""
d.ts = size
# NOTE: data is truncated automatically when applying the delta
# MUST NOT DO THIS HERE
return d
def _move_delta_lbound(d, bytes):
"""Move the delta by the given amount of bytes, reducing its size so that its
right bound stays static
:param bytes: amount of bytes to move, must be smaller than delta size
:return: d"""
if bytes == 0:
return
d.to += bytes
d.so += bytes
d.ts -= bytes
if d.data is not None:
d.data = d.data[bytes:]
# END handle data
return d
def delta_duplicate(src):
return DeltaChunk(src.to, src.ts, src.so, src.data)
def delta_chunk_apply(dc, bbuf, write):
"""Apply own data to the target buffer
:param bbuf: buffer providing source bytes for copy operations
:param write: write method to call with data to write"""
if dc.data is None:
# COPY DATA FROM SOURCE
write(bbuf[dc.so:dc.so + dc.ts])
else:
# APPEND DATA
# what's faster: if + 4 function calls or just a write with a slice ?
# Considering data can be larger than 127 bytes now, it should be worth it
if dc.ts < len(dc.data):
write(dc.data[:dc.ts])
else:
write(dc.data)
# END handle truncation
# END handle chunk mode
class DeltaChunk:
"""Represents a piece of a delta, it can either add new data, or copy existing
one from a source buffer"""
__slots__ = (
'to', # start offset in the target buffer in bytes
'ts', # size of this chunk in the target buffer in bytes
'so', # start offset in the source buffer in bytes or None
'data', # chunk of bytes to be added to the target buffer,
# DeltaChunkList to use as base, or None
)
def __init__(self, to, ts, so, data):
self.to = to
self.ts = ts
self.so = so
self.data = data
def __repr__(self):
return "DeltaChunk(%i, %i, %s, %s)" % (self.to, self.ts, self.so, self.data or "")
#{ Interface
def rbound(self):
return self.to + self.ts
def has_data(self):
""":return: True if the instance has data to add to the target stream"""
return self.data is not None
#} END interface
def _closest_index(dcl, absofs):
""":return: index at which the given absofs should be inserted. The index points
to the DeltaChunk with a target buffer absofs that equals or is greater than
absofs.
**Note:** global method for performance only, it belongs to DeltaChunkList"""
lo = 0
hi = len(dcl)
while lo < hi:
mid = (lo + hi) / 2
dc = dcl[mid]
if dc.to > absofs:
hi = mid
elif dc.rbound() > absofs or dc.to == absofs:
return mid
else:
lo = mid + 1
# END handle bound
# END for each delta absofs
return len(dcl) - 1
def delta_list_apply(dcl, bbuf, write):
"""Apply the chain's changes and write the final result using the passed
write function.
:param bbuf: base buffer containing the base of all deltas contained in this
list. It will only be used if the chunk in question does not have a base
chain.
:param write: function taking a string of bytes to write to the output"""
for dc in dcl:
delta_chunk_apply(dc, bbuf, write)
# END for each dc
def delta_list_slice(dcl, absofs, size, ndcl):
""":return: Subsection of this list at the given absolute offset, with the given
size in bytes.
:return: None"""
cdi = _closest_index(dcl, absofs) # delta start index
cd = dcl[cdi]
slen = len(dcl)
lappend = ndcl.append
if cd.to != absofs:
tcd = DeltaChunk(cd.to, cd.ts, cd.so, cd.data)
_move_delta_lbound(tcd, absofs - cd.to)
tcd.ts = min(tcd.ts, size)
lappend(tcd)
size -= tcd.ts
cdi += 1
# END lbound overlap handling
while cdi < slen and size:
# are we larger than the current block
cd = dcl[cdi]
if cd.ts 1:
# if first_data_index is not None:
nd = StringIO() # new data
so = self[first_data_index].to # start offset in target buffer
for x in range(first_data_index, i - 1):
xdc = self[x]
nd.write(xdc.data[:xdc.ts])
# END collect data
del(self[first_data_index:i - 1])
buf = nd.getvalue()
self.insert(first_data_index, DeltaChunk(so, len(buf), 0, buf))
slen = len(self)
i = first_data_index + 1
# END concatenate data
first_data_index = None
continue
# END skip non-data chunks
if first_data_index is None:
first_data_index = i - 1
# END iterate list
# if slen_orig != len(self):
# print "INFO: Reduced delta list len to %f %% of former size" % ((float(len(self)) / slen_orig) * 100)
return self
def check_integrity(self, target_size=-1):
"""Verify the list has non-overlapping chunks only, and the total size matches
target_size
:param target_size: if not -1, the total size of the chain must be target_size
:raise AssertionError: if the size doesn't match"""
if target_size > -1:
assert self[-1].rbound() == target_size
assert reduce(lambda x, y: x + y, (d.ts for d in self), 0) == target_size
# END target size verification
if len(self) < 2:
return
# check data
for dc in self:
assert dc.ts > 0
if dc.has_data():
assert len(dc.data) >= dc.ts
# END for each dc
left = islice(self, 0, len(self) - 1)
right = iter(self)
right.next()
# this is very pythonic - we might have just use index based access here,
# but this could actually be faster
for lft, rgt in zip(left, right):
assert lft.rbound() == rgt.to
assert lft.to + lft.ts == rgt.to
# END for each pair
class TopdownDeltaChunkList(DeltaChunkList):
"""Represents a list which is generated by feeding its ancestor streams one by
one"""
__slots__ = tuple()
def connect_with_next_base(self, bdcl):
"""Connect this chain with the next level of our base delta chunklist.
The goal in this game is to mark as many of our chunks rigid, hence they
cannot be changed by any of the upcoming bases anymore. Once all our
chunks are marked like that, we can stop all processing
:param bdcl: data chunk list being one of our bases. They must be fed in
consecutively and in order, towards the earliest ancestor delta
:return: True if processing was done. Use it to abort processing of
remaining streams if False is returned"""
nfc = 0 # number of frozen chunks
dci = 0 # delta chunk index
slen = len(self) # len of self
ccl = list() # temporary list
while dci < slen:
dc = self[dci]
dci += 1
# all add-chunks which are already topmost don't need additional processing
if dc.data is not None:
nfc += 1
continue
# END skip add chunks
# copy chunks
# integrate the portion of the base list into ourselves. Lists
# dont support efficient insertion ( just one at a time ), but for now
# we live with it. Internally, its all just a 32/64bit pointer, and
# the portions of moved memory should be smallish. Maybe we just rebuild
# ourselves in order to reduce the amount of insertions ...
del(ccl[:])
delta_list_slice(bdcl, dc.so, dc.ts, ccl)
# move the target bounds into place to match with our chunk
ofs = dc.to - dc.so
for cdc in ccl:
cdc.to += ofs
# END update target bounds
if len(ccl) == 1:
self[dci - 1] = ccl[0]
else:
# maybe try to compute the expenses here, and pick the right algorithm
# It would normally be faster than copying everything physically though
# TODO: Use a deque here, and decide by the index whether to extend
# or extend left !
post_dci = self[dci:]
del(self[dci - 1:]) # include deletion of dc
self.extend(ccl)
self.extend(post_dci)
slen = len(self)
dci += len(ccl) - 1 # deleted dc, added rest
# END handle chunk replacement
# END for each chunk
if nfc == slen:
return False
# END handle completeness
return True
#} END structures
#{ Routines
def is_loose_object(m):
"""
:return: True the file contained in memory map m appears to be a loose object.
Only the first two bytes are needed"""
b0, b1 = map(ord, m[:2])
word = (b0 > 4) & 7 # numeric type
size = c & 15 # starting size
s = 4 # starting bit-shift size
while c & 0x80:
c = byte_ord(data[i])
i += 1
size += (c & 0x7f) = 4
while obj_size:
hdr.append(c | 0x80)
c = obj_size & 0x7f
obj_size >>= 7
# END until size is consumed
hdr.append(c)
# end handle interpreter
return hdr
def msb_size(data, offset=0):
"""
:return: tuple(read_bytes, size) read the msb size from the given random
access data starting at the given byte offset"""
size = 0
i = 0
l = len(data)
hit_msb = False
while i < l:
c = data[i + offset]
size |= (c & 0x7f)