17
17
"""Core compression logic for compressing streams of related files."""
19
from __future__ import absolute_import
24
from .lazy_import import lazy_import
25
lazy_import(globals(), """
30
31
graph as _mod_graph,
37
from bzrlib.btree_index import BTreeBuilder
38
from bzrlib.lru_cache import LRUSizeCache
39
from bzrlib.tsort import topo_sort
40
from bzrlib.versionedfile import (
39
from breezy.repofmt import pack_repo
40
from breezy.i18n import gettext
43
from .btree_index import BTreeBuilder
44
from .lru_cache import LRUSizeCache
50
from .versionedfile import (
42
53
AbsentContentFactory,
43
54
ChunkedContentFactory,
44
55
FulltextContentFactory,
56
VersionedFilesWithFallbacks,
48
59
# Minimum number of uncompressed bytes to try fetch at once when retrieving
49
60
# groupcompress blocks.
52
_USE_LZMA = False and (pylzma is not None)
54
63
# osutils.sha_string('')
55
64
_null_sha1 = 'da39a3ee5e6b4b0d3255bfef95601890afd80709'
135
144
self._content = ''.join(self._content_chunks)
136
145
self._content_chunks = None
137
146
if self._content is None:
138
if self._z_content is None:
147
# We join self._z_content_chunks here, because if we are
148
# decompressing, then it is *very* likely that we have a single
150
if self._z_content_chunks is None:
139
151
raise AssertionError('No content to decompress')
140
if self._z_content == '':
152
z_content = ''.join(self._z_content_chunks)
141
154
self._content = ''
142
155
elif self._compressor_name == 'lzma':
143
156
# We don't do partial lzma decomp yet
144
self._content = pylzma.decompress(self._z_content)
158
self._content = pylzma.decompress(z_content)
145
159
elif self._compressor_name == 'zlib':
146
160
# Start a zlib decompressor
147
161
if num_bytes * 4 > self._content_length * 3:
148
162
# If we are requesting more that 3/4ths of the content,
149
163
# just extract the whole thing in a single pass
150
164
num_bytes = self._content_length
151
self._content = zlib.decompress(self._z_content)
165
self._content = zlib.decompress(z_content)
153
167
self._z_content_decompressor = zlib.decompressobj()
154
168
# Seed the decompressor with the uncompressed bytes, so
155
169
# that the rest of the code is simplified
156
170
self._content = self._z_content_decompressor.decompress(
157
self._z_content, num_bytes + _ZLIB_DECOMP_WINDOW)
171
z_content, num_bytes + _ZLIB_DECOMP_WINDOW)
158
172
if not self._z_content_decompressor.unconsumed_tail:
159
173
self._z_content_decompressor = None
207
221
# XXX: Define some GCCorrupt error ?
208
222
raise AssertionError('Invalid bytes: (%d) != %d + %d' %
209
223
(len(bytes), pos, self._z_content_length))
210
self._z_content = bytes[pos:]
224
self._z_content_chunks = (bytes[pos:],)
227
def _z_content(self):
228
"""Return z_content_chunks as a simple string.
230
Meant only to be used by the test suite.
232
if self._z_content_chunks is not None:
233
return ''.join(self._z_content_chunks)
213
237
def from_bytes(cls, bytes):
269
293
self._content_length = length
270
294
self._content_chunks = content_chunks
271
295
self._content = None
272
self._z_content = None
296
self._z_content_chunks = None
274
298
def set_content(self, content):
275
299
"""Set the content of this block."""
276
300
self._content_length = len(content)
277
301
self._content = content
278
self._z_content = None
280
def _create_z_content_using_lzma(self):
281
if self._content_chunks is not None:
282
self._content = ''.join(self._content_chunks)
283
self._content_chunks = None
284
if self._content is None:
285
raise AssertionError('Nothing to compress')
286
self._z_content = pylzma.compress(self._content)
287
self._z_content_length = len(self._z_content)
289
def _create_z_content_from_chunks(self):
302
self._z_content_chunks = None
304
def _create_z_content_from_chunks(self, chunks):
290
305
compressor = zlib.compressobj(zlib.Z_DEFAULT_COMPRESSION)
291
compressed_chunks = map(compressor.compress, self._content_chunks)
306
# Peak in this point is 1 fulltext, 1 compressed text, + zlib overhead
307
# (measured peak is maybe 30MB over the above...)
308
compressed_chunks = list(map(compressor.compress, chunks))
292
309
compressed_chunks.append(compressor.flush())
293
self._z_content = ''.join(compressed_chunks)
294
self._z_content_length = len(self._z_content)
310
# Ignore empty chunks
311
self._z_content_chunks = [c for c in compressed_chunks if c]
312
self._z_content_length = sum(map(len, self._z_content_chunks))
296
314
def _create_z_content(self):
297
if self._z_content is not None:
300
self._create_z_content_using_lzma()
315
if self._z_content_chunks is not None:
302
317
if self._content_chunks is not None:
303
self._create_z_content_from_chunks()
305
self._z_content = zlib.compress(self._content)
306
self._z_content_length = len(self._z_content)
318
chunks = self._content_chunks
320
chunks = (self._content,)
321
self._create_z_content_from_chunks(chunks)
324
"""Create the byte stream as a series of 'chunks'"""
325
self._create_z_content()
326
header = self.GCB_HEADER
327
chunks = ['%s%d\n%d\n'
328
% (header, self._z_content_length, self._content_length),
330
chunks.extend(self._z_content_chunks)
331
total_len = sum(map(len, chunks))
332
return total_len, chunks
308
334
def to_bytes(self):
309
335
"""Encode the information into a byte stream."""
310
self._create_z_content()
312
header = self.GCB_LZ_HEADER
314
header = self.GCB_HEADER
316
'%d\n%d\n' % (self._z_content_length, self._content_length),
336
total_len, chunks = self.to_chunks()
319
337
return ''.join(chunks)
321
339
def _dump(self, include_text=False):
435
453
# Grab and cache the raw bytes for this entry
436
454
# and break the ref-cycle with _manager since we don't need it
438
self._manager._prepare_for_extract()
457
self._manager._prepare_for_extract()
458
except zlib.error as value:
459
raise errors.DecompressCorruption("zlib: " + str(value))
439
460
block = self._manager._block
440
461
self._bytes = block.extract(self.key, self._start, self._end)
441
462
# There are code paths that first extract as fulltext, and then
460
481
_full_enough_block_size = 3*1024*1024 # size at which we won't repack
461
482
_full_enough_mixed_block_size = 2*768*1024 # 1.5MB
463
def __init__(self, block):
484
def __init__(self, block, get_compressor_settings=None):
464
485
self._block = block
465
486
# We need to preserve the ordering
466
487
self._factories = []
467
488
self._last_byte = 0
489
self._get_settings = get_compressor_settings
490
self._compressor_settings = None
492
def _get_compressor_settings(self):
493
if self._compressor_settings is not None:
494
return self._compressor_settings
496
if self._get_settings is not None:
497
settings = self._get_settings()
499
vf = GroupCompressVersionedFiles
500
settings = vf._DEFAULT_COMPRESSOR_SETTINGS
501
self._compressor_settings = settings
502
return self._compressor_settings
469
504
def add_factory(self, key, parents, start, end):
470
505
if not self._factories:
503
538
new_block.set_content(self._block._content[:last_byte])
504
539
self._block = new_block
541
def _make_group_compressor(self):
542
return GroupCompressor(self._get_compressor_settings())
506
544
def _rebuild_block(self):
507
545
"""Create a new GroupCompressBlock with only the referenced texts."""
508
compressor = GroupCompressor()
546
compressor = self._make_group_compressor()
509
547
tstart = time.time()
510
548
old_length = self._block._content_length
523
561
# block? It seems hard to come up with a method that it would
524
562
# expand, since we do full compression again. Perhaps based on a
525
563
# request that ends up poorly ordered?
564
# TODO: If the content would have expanded, then we would want to
565
# handle a case where we need to split the block.
566
# Now that we have a user-tweakable option
567
# (max_bytes_to_index), it is possible that one person set it
568
# to a very low value, causing poor compression.
526
569
delta = time.time() - tstart
527
570
self._block = new_block
528
571
trace.mutter('creating new compressed block on-the-fly in %.3fs'
679
722
z_header_bytes = zlib.compress(header_bytes)
681
724
z_header_bytes_len = len(z_header_bytes)
682
block_bytes = self._block.to_bytes()
725
block_bytes_len, block_chunks = self._block.to_chunks()
683
726
lines.append('%d\n%d\n%d\n' % (z_header_bytes_len, header_bytes_len,
685
728
lines.append(z_header_bytes)
686
lines.append(block_bytes)
687
del z_header_bytes, block_bytes
729
lines.extend(block_chunks)
730
del z_header_bytes, block_chunks
731
# TODO: This is a point where we will double the memory consumption. To
732
# avoid this, we probably have to switch to a 'chunked' api
688
733
return ''.join(lines)
691
736
def from_bytes(cls, bytes):
692
737
# TODO: This does extra string copying, probably better to do it a
738
# different way. At a minimum this creates 2 copies of the
694
740
(storage_kind, z_header_len, header_len,
695
741
block_len, rest) = bytes.split('\n', 4)
855
905
After calling this, the compressor should no longer be used
857
# TODO: this causes us to 'bloat' to 2x the size of content in the
858
# group. This has an impact for 'commit' of large objects.
859
# One possibility is to use self._content_chunks, and be lazy and
860
# only fill out self._content as a full string when we actually
861
# need it. That would at least drop the peak memory consumption
862
# for 'commit' down to ~1x the size of the largest file, at a
863
# cost of increased complexity within this code. 2x is still <<
864
# 3x the size of the largest file, so we are doing ok.
865
907
self._block.set_chunked_content(self.chunks, self.endpoint)
866
908
self.chunks = None
867
909
self._delta_index = None
886
928
class PythonGroupCompressor(_CommonGroupCompressor):
930
def __init__(self, settings=None):
889
931
"""Create a GroupCompressor.
891
933
Used only if the pyrex version is not available.
893
super(PythonGroupCompressor, self).__init__()
935
super(PythonGroupCompressor, self).__init__(settings)
894
936
self._delta_index = LinesDeltaIndex([])
895
937
# The actual content is managed by LinesDeltaIndex
896
938
self.chunks = self._delta_index.lines
934
976
It contains code very similar to SequenceMatcher because of having a similar
935
977
task. However some key differences apply:
936
- there is no junk, we want a minimal edit not a human readable diff.
937
- we don't filter very common lines (because we don't know where a good
938
range will start, and after the first text we want to be emitting minmal
940
- we chain the left side, not the right side
941
- we incrementally update the adjacency matrix as new lines are provided.
942
- we look for matches in all of the left side, so the routine which does
943
the analagous task of find_longest_match does not need to filter on the
979
* there is no junk, we want a minimal edit not a human readable diff.
980
* we don't filter very common lines (because we don't know where a good
981
range will start, and after the first text we want to be emitting minmal
983
* we chain the left side, not the right side
984
* we incrementally update the adjacency matrix as new lines are provided.
985
* we look for matches in all of the left side, so the routine which does
986
the analagous task of find_longest_match does not need to filter on the
948
super(PyrexGroupCompressor, self).__init__()
949
self._delta_index = DeltaIndex()
990
def __init__(self, settings=None):
991
super(PyrexGroupCompressor, self).__init__(settings)
992
max_bytes_to_index = self._settings.get('max_bytes_to_index', 0)
993
self._delta_index = DeltaIndex(max_bytes_to_index=max_bytes_to_index)
951
995
def _compress(self, key, bytes, max_delta_size, soft=False):
952
996
"""see _CommonGroupCompressor._compress"""
1027
1071
index = _GCGraphIndex(graph_index, lambda:True, parents=parents,
1028
1072
add_callback=graph_index.add_nodes,
1029
1073
inconsistency_fatal=inconsistency_fatal)
1030
access = knit._DirectPackAccess({})
1074
access = pack_repo._DirectPackAccess({})
1031
1075
access.set_writer(writer, graph_index, (transport, 'newpack'))
1032
1076
result = GroupCompressVersionedFiles(index, access, delta)
1033
1077
result.stream = stream
1044
1088
class _BatchingBlockFetcher(object):
1045
1089
"""Fetch group compress blocks in batches.
1047
1091
:ivar total_bytes: int of expected number of bytes needed to fetch the
1048
1092
currently pending batch.
1051
def __init__(self, gcvf, locations):
1095
def __init__(self, gcvf, locations, get_compressor_settings=None):
1052
1096
self.gcvf = gcvf
1053
1097
self.locations = locations
1102
1147
def yield_factories(self, full_flush=False):
1103
1148
"""Yield factories for keys added since the last yield. They will be
1104
1149
returned in the order they were added via add_key.
1106
1151
:param full_flush: by default, some results may not be returned in case
1107
1152
they can be part of the next batch. If full_flush is True, then
1108
1153
all results are returned.
1136
1181
memos_to_get_stack.pop()
1138
1183
block = self.batch_memos[read_memo]
1139
self.manager = _LazyGroupContentManager(block)
1184
self.manager = _LazyGroupContentManager(block,
1185
get_compressor_settings=self._get_compressor_settings)
1140
1186
self.last_read_memo = read_memo
1141
1187
start, end = index_memo[3:5]
1142
1188
self.manager.add_factory(key, parents, start, end)
1149
1195
self.total_bytes = 0
1152
class GroupCompressVersionedFiles(VersionedFiles):
1198
class GroupCompressVersionedFiles(VersionedFilesWithFallbacks):
1153
1199
"""A group-compress based VersionedFiles implementation."""
1155
def __init__(self, index, access, delta=True, _unadded_refs=None):
1201
# This controls how the GroupCompress DeltaIndex works. Basically, we
1202
# compute hash pointers into the source blocks (so hash(text) => text).
1203
# However each of these references costs some memory in trade against a
1204
# more accurate match result. For very large files, they either are
1205
# pre-compressed and change in bulk whenever they change, or change in just
1206
# local blocks. Either way, 'improved resolution' is not very helpful,
1207
# versus running out of memory trying to track everything. The default max
1208
# gives 100% sampling of a 1MB file.
1209
_DEFAULT_MAX_BYTES_TO_INDEX = 1024 * 1024
1210
_DEFAULT_COMPRESSOR_SETTINGS = {'max_bytes_to_index':
1211
_DEFAULT_MAX_BYTES_TO_INDEX}
1213
def __init__(self, index, access, delta=True, _unadded_refs=None,
1156
1215
"""Create a GroupCompressVersionedFiles object.
1158
1217
:param index: The index object storing access and graph data.
1159
1218
:param access: The access object storing raw data.
1160
1219
:param delta: Whether to delta compress or just entropy compress.
1161
1220
:param _unadded_refs: private parameter, don't use.
1221
:param _group_cache: private parameter, don't use.
1163
1223
self._index = index
1164
1224
self._access = access
1166
1226
if _unadded_refs is None:
1167
1227
_unadded_refs = {}
1168
1228
self._unadded_refs = _unadded_refs
1169
self._group_cache = LRUSizeCache(max_size=50*1024*1024)
1170
self._fallback_vfs = []
1229
if _group_cache is None:
1230
_group_cache = LRUSizeCache(max_size=50*1024*1024)
1231
self._group_cache = _group_cache
1232
self._immediate_fallback_vfs = []
1233
self._max_bytes_to_index = None
1172
1235
def without_fallbacks(self):
1173
1236
"""Return a clone of this object without any fallbacks configured."""
1174
1237
return GroupCompressVersionedFiles(self._index, self._access,
1175
self._delta, _unadded_refs=dict(self._unadded_refs))
1238
self._delta, _unadded_refs=dict(self._unadded_refs),
1239
_group_cache=self._group_cache)
1177
1241
def add_lines(self, key, parents, lines, parent_texts=None,
1178
1242
left_matching_blocks=None, nostore_sha=None, random_id=False,
1182
1246
:param key: The key tuple of the text to add.
1183
1247
:param parents: The parents key tuples of the text to add.
1184
1248
:param lines: A list of lines. Each line must be a bytestring. And all
1185
of them except the last must be terminated with \n and contain no
1186
other \n's. The last line may either contain no \n's or a single
1187
terminating \n. If the lines list does meet this constraint the add
1188
routine may error or may succeed - but you will be unable to read
1189
the data back accurately. (Checking the lines have been split
1249
of them except the last must be terminated with \\n and contain no
1250
other \\n's. The last line may either contain no \\n's or a single
1251
terminating \\n. If the lines list does meet this constraint the
1252
add routine may error or may succeed - but you will be unable to
1253
read the data back accurately. (Checking the lines have been split
1190
1254
correctly is expensive and extremely unlikely to catch bugs so it
1191
1255
is not done at runtime unless check_content is True.)
1192
1256
:param parent_texts: An optional dictionary containing the opaque
1287
1351
self._check_lines_not_unicode(lines)
1288
1352
self._check_lines_are_lines(lines)
1290
def get_known_graph_ancestry(self, keys):
1291
"""Get a KnownGraph instance with the ancestry of keys."""
1292
# Note that this is identical to
1293
# KnitVersionedFiles.get_known_graph_ancestry, but they don't share
1295
parent_map, missing_keys = self._index.find_ancestry(keys)
1296
for fallback in self._fallback_vfs:
1297
if not missing_keys:
1299
(f_parent_map, f_missing_keys) = fallback._index.find_ancestry(
1301
parent_map.update(f_parent_map)
1302
missing_keys = f_missing_keys
1303
kg = _mod_graph.KnownGraph(parent_map)
1306
1354
def get_parent_map(self, keys):
1307
1355
"""Get a map of the graph parents of keys.
1366
1414
yield read_memo, cached[read_memo]
1367
1415
except KeyError:
1368
1416
# Read the block, and cache it.
1369
zdata = raw_records.next()
1417
zdata = next(raw_records)
1370
1418
block = GroupCompressBlock.from_bytes(zdata)
1371
1419
self._group_cache[read_memo] = block
1372
1420
cached[read_memo] = block
1447
1495
The returned objects should be in the order defined by 'ordering',
1448
1496
which can weave between different sources.
1449
1498
:param ordering: Must be one of 'topological' or 'groupcompress'
1450
1499
:return: List of [(source, [keys])] tuples, such that all keys are in
1451
1500
the defined order, regardless of source.
1453
1502
if ordering == 'topological':
1454
present_keys = topo_sort(parent_map)
1503
present_keys = tsort.topo_sort(parent_map)
1456
1505
# ordering == 'groupcompress'
1457
1506
# XXX: This only optimizes for the target ordering. We may need
1493
1542
# This is the group the bytes are stored in, followed by the
1494
1543
# location in the group
1495
1544
return locations[key][0]
1496
present_keys = sorted(locations.iterkeys(), key=get_group)
1497
1545
# We don't have an ordering for keys in the in-memory object, but
1498
1546
# lets process the in-memory ones first.
1499
present_keys = list(unadded_keys) + present_keys
1547
present_keys = list(unadded_keys)
1548
present_keys.extend(sorted(locations, key=get_group))
1500
1549
# Now grab all of the ones from other sources
1501
1550
source_keys = [(self, present_keys)]
1502
1551
source_keys.extend(source_result)
1546
1595
# - we encounter an unadded ref, or
1547
1596
# - we run out of keys, or
1548
1597
# - the total bytes to retrieve for this batch > BATCH_SIZE
1549
batcher = _BatchingBlockFetcher(self, locations)
1598
batcher = _BatchingBlockFetcher(self, locations,
1599
get_compressor_settings=self._get_compressor_settings)
1550
1600
for source, keys in source_keys:
1551
1601
if source is self:
1552
1602
for key in keys:
1598
1648
for _ in self._insert_record_stream(stream, random_id=False):
1651
def _get_compressor_settings(self):
1652
if self._max_bytes_to_index is None:
1653
# TODO: VersionedFiles don't know about their containing
1654
# repository, so they don't have much of an idea about their
1655
# location. So for now, this is only a global option.
1656
c = config.GlobalConfig()
1657
val = c.get_user_option('bzr.groupcompress.max_bytes_to_index')
1661
except ValueError as e:
1662
trace.warning('Value for '
1663
'"bzr.groupcompress.max_bytes_to_index"'
1664
' %r is not an integer'
1668
val = self._DEFAULT_MAX_BYTES_TO_INDEX
1669
self._max_bytes_to_index = val
1670
return {'max_bytes_to_index': self._max_bytes_to_index}
1672
def _make_group_compressor(self):
1673
return GroupCompressor(self._get_compressor_settings())
1601
1675
def _insert_record_stream(self, stream, random_id=False, nostore_sha=None,
1602
1676
reuse_blocks=True):
1603
1677
"""Internal core to insert a record stream into this container.
1627
1701
# This will go up to fulltexts for gc to gc fetching, which isn't
1629
self._compressor = GroupCompressor()
1703
self._compressor = self._make_group_compressor()
1630
1704
self._unadded_refs = {}
1631
1705
keys_to_add = []
1633
bytes = self._compressor.flush().to_bytes()
1634
self._compressor = GroupCompressor()
1707
bytes_len, chunks = self._compressor.flush().to_chunks()
1708
self._compressor = self._make_group_compressor()
1709
# Note: At this point we still have 1 copy of the fulltext (in
1710
# record and the var 'bytes'), and this generates 2 copies of
1711
# the compressed text (one for bytes, one in chunks)
1712
# TODO: Push 'chunks' down into the _access api, so that we don't
1713
# have to double compressed memory here
1714
# TODO: Figure out how to indicate that we would be happy to free
1715
# the fulltext content at this point. Note that sometimes we
1716
# will want it later (streaming CHK pages), but most of the
1717
# time we won't (everything else)
1718
bytes = ''.join(chunks)
1635
1720
index, start, length = self._access.add_raw_records(
1636
1721
[(None, len(bytes))], bytes)[0]
1656
1741
raise errors.RevisionNotPresent(record.key, self)
1658
1743
if record.key in inserted_keys:
1659
trace.note('Insert claimed random_id=True,'
1660
' but then inserted %r two times', record.key)
1744
trace.note(gettext('Insert claimed random_id=True,'
1745
' but then inserted %r two times'), record.key)
1662
1747
inserted_keys.add(record.key)
1663
1748
if reuse_blocks:
1802
1887
"""See VersionedFiles.keys."""
1803
1888
if 'evil' in debug.debug_flags:
1804
1889
trace.mutter_callsite(2, "keys scales with size of history")
1805
sources = [self._index] + self._fallback_vfs
1890
sources = [self._index] + self._immediate_fallback_vfs
1807
1892
for source in sources:
1808
1893
result.update(source.keys())
1897
class _GCBuildDetails(object):
1898
"""A blob of data about the build details.
1900
This stores the minimal data, which then allows compatibility with the old
1901
api, without taking as much memory.
1904
__slots__ = ('_index', '_group_start', '_group_end', '_basis_end',
1905
'_delta_end', '_parents')
1908
compression_parent = None
1910
def __init__(self, parents, position_info):
1911
self._parents = parents
1912
(self._index, self._group_start, self._group_end, self._basis_end,
1913
self._delta_end) = position_info
1916
return '%s(%s, %s)' % (self.__class__.__name__,
1917
self.index_memo, self._parents)
1920
def index_memo(self):
1921
return (self._index, self._group_start, self._group_end,
1922
self._basis_end, self._delta_end)
1925
def record_details(self):
1926
return static_tuple.StaticTuple(self.method, None)
1928
def __getitem__(self, offset):
1929
"""Compatibility thunk to act like a tuple."""
1931
return self.index_memo
1933
return self.compression_parent # Always None
1935
return self._parents
1937
return self.record_details
1939
raise IndexError('offset out of range')
1812
1945
class _GCGraphIndex(object):
1813
1946
"""Mapper from GroupCompressVersionedFiles needs into GraphIndex storage."""
1817
1950
inconsistency_fatal=True, track_new_keys=False):
1818
1951
"""Construct a _GCGraphIndex on a graph_index.
1820
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1953
:param graph_index: An implementation of breezy.index.GraphIndex.
1821
1954
:param is_locked: A callback, returns True if the index is locked and
1823
1956
:param parents: If True, record knits parents, if not do not record
1902
2035
if self._parents:
1903
for key, (value, node_refs) in keys.iteritems():
2036
for key, (value, node_refs) in viewitems(keys):
1904
2037
result.append((key, value, node_refs))
1906
for key, (value, node_refs) in keys.iteritems():
2039
for key, (value, node_refs) in viewitems(keys):
1907
2040
result.append((key, value))
1908
2041
records = result
1909
2042
key_dependencies = self._key_dependencies
1989
2122
:param keys: An iterable of keys.
1990
2123
:return: A dict of key:
1991
2124
(index_memo, compression_parent, parents, record_details).
1993
opaque structure to pass to read_records to extract the raw
1996
Content that this record is built upon, may be None
1998
Logical parents of this node
2000
extra information about the content which needs to be passed to
2001
Factory.parse_record
2126
* index_memo: opaque structure to pass to read_records to extract
2128
* compression_parent: Content that this record is built upon, may
2130
* parents: Logical parents of this node
2131
* record_details: extra information about the content which needs
2132
to be passed to Factory.parse_record
2003
2134
self._check_read()
2033
2163
# each, or about 7MB. Note that it might be even more when you consider
2034
2164
# how PyInt is allocated in separate slabs. And you can't return a slab
2035
2165
# to the OS if even 1 int on it is in use. Note though that Python uses
2036
# a LIFO when re-using PyInt slots, which probably causes more
2166
# a LIFO when re-using PyInt slots, which might cause more
2037
2167
# fragmentation.
2038
2168
start = int(bits[0])
2039
2169
start = self._int_cache.setdefault(start, start)
added
removed
breezy/groupcompress.py
