128
128
# _z_content because of this.
129
129
if num_bytes is None:
130
130
num_bytes = self._content_length
131
if self._content_length is not None:
132
assert num_bytes <= self._content_length
133
if self._content is None:
134
assert self._z_content is not None
131
elif (self._content_length is not None
132
and num_bytes > self._content_length):
133
raise AssertionError(
134
'requested num_bytes (%d) > content length (%d)'
135
% (num_bytes, self._content_length))
136
# Expand the content if required
137
if self._content is None:
138
if self._content_chunks is not None:
139
self._content = ''.join(self._content_chunks)
140
self._content_chunks = None
141
if self._content is None:
142
if self._z_content is None:
143
raise AssertionError('No content to decompress')
135
144
if self._z_content == '':
136
145
self._content = ''
137
146
elif self._compressor_name == 'lzma':
138
147
# We don't do partial lzma decomp yet
139
148
self._content = pylzma.decompress(self._z_content)
149
elif self._compressor_name == 'zlib':
141
150
# Start a zlib decompressor
142
assert self._compressor_name == 'zlib'
143
151
if num_bytes is None:
144
152
self._content = zlib.decompress(self._z_content)
157
169
# We must have already decompressed everything
159
171
# If we got this far, and don't have a decompressor, something is wrong
160
assert self._z_content_decompressor is not None
172
if self._z_content_decompressor is None:
173
raise AssertionError(
174
'No decompressor to decompress %d bytes' % num_bytes)
161
175
remaining_decomp = self._z_content_decompressor.unconsumed_tail
162
176
if num_bytes is None:
163
177
if remaining_decomp:
164
178
# We don't know how much is left, but we'll decompress it all
165
179
self._content += self._z_content_decompressor.decompress(
166
180
remaining_decomp)
167
# Note: There what I consider a bug in zlib.decompressobj
181
# Note: There's what I consider a bug in zlib.decompressobj
168
182
# If you pass back in the entire unconsumed_tail, only
169
183
# this time you don't pass a max-size, it doesn't
170
184
# change the unconsumed_tail back to None/''.
171
185
# However, we know we are done with the whole stream
172
186
self._z_content_decompressor = None
187
# XXX: Why is this the only place in this routine we set this?
173
188
self._content_length = len(self._content)
175
# If we have nothing left to decomp, we ran out of decomp bytes
176
assert remaining_decomp
190
if not remaining_decomp:
191
raise AssertionError('Nothing left to decompress')
177
192
needed_bytes = num_bytes - len(self._content)
178
193
# We always set max_size to 32kB over the minimum needed, so that
179
194
# zlib will give us as much as we really want.
202
219
pos2 = bytes.index('\n', pos, pos + 14)
203
220
self._content_length = int(bytes[pos:pos2])
205
assert len(bytes) == (pos + self._z_content_length)
222
if len(bytes) != (pos + self._z_content_length):
223
# XXX: Define some GCCorrupt error ?
224
raise AssertionError('Invalid bytes: (%d) != %d + %d' %
225
(len(bytes), pos, self._z_content_length))
206
226
self._z_content = bytes[pos:]
207
assert len(self._z_content) == self._z_content_length
210
229
def from_bytes(cls, bytes):
212
if bytes[:6] not in (cls.GCB_HEADER, cls.GCB_LZ_HEADER):
213
raise ValueError('bytes did not start with %r' % (cls.GCB_HEADER,))
231
if bytes[:6] not in cls.GCB_KNOWN_HEADERS:
232
raise ValueError('bytes did not start with any of %r'
233
% (cls.GCB_KNOWN_HEADERS,))
234
# XXX: why not testing the whole header ?
214
235
if bytes[4] == 'z':
215
236
out._compressor_name = 'zlib'
216
237
elif bytes[4] == 'l':
254
275
bytes = apply_delta_to_source(self._content, content_start, end)
278
def set_chunked_content(self, content_chunks, length):
279
"""Set the content of this block to the given chunks."""
280
# If we have lots of short lines, it is may be more efficient to join
281
# the content ahead of time. If the content is <10MiB, we don't really
282
# care about the extra memory consumption, so we can just pack it and
283
# be done. However, timing showed 18s => 17.9s for repacking 1k revs of
284
# mysql, which is below the noise margin
285
self._content_length = length
286
self._content_chunks = content_chunks
288
self._z_content = None
257
290
def set_content(self, content):
258
291
"""Set the content of this block."""
259
292
self._content_length = len(content)
260
293
self._content = content
261
294
self._z_content = None
296
def _create_z_content_using_lzma(self):
297
if self._content_chunks is not None:
298
self._content = ''.join(self._content_chunks)
299
self._content_chunks = None
300
if self._content is None:
301
raise AssertionError('Nothing to compress')
302
self._z_content = pylzma.compress(self._content)
303
self._z_content_length = len(self._z_content)
305
def _create_z_content_from_chunks(self):
306
compressor = zlib.compressobj(zlib.Z_DEFAULT_COMPRESSION)
307
compressed_chunks = map(compressor.compress, self._content_chunks)
308
compressed_chunks.append(compressor.flush())
309
self._z_content = ''.join(compressed_chunks)
310
self._z_content_length = len(self._z_content)
312
def _create_z_content(self):
313
if self._z_content is not None:
316
self._create_z_content_using_lzma()
318
if self._content_chunks is not None:
319
self._create_z_content_from_chunks()
321
self._z_content = zlib.compress(self._content)
322
self._z_content_length = len(self._z_content)
263
324
def to_bytes(self):
264
325
"""Encode the information into a byte stream."""
265
compress = zlib.compress
267
compress = pylzma.compress
268
if self._z_content is None:
269
assert self._content is not None
270
self._z_content = compress(self._content)
271
self._z_content_length = len(self._z_content)
326
self._create_z_content()
273
328
header = self.GCB_LZ_HEADER
280
335
return ''.join(chunks)
337
def _dump(self, include_text=False):
338
"""Take this block, and spit out a human-readable structure.
340
:param include_text: Inserts also include text bits, chose whether you
341
want this displayed in the dump or not.
342
:return: A dump of the given block. The layout is something like:
343
[('f', length), ('d', delta_length, text_length, [delta_info])]
344
delta_info := [('i', num_bytes, text), ('c', offset, num_bytes),
347
self._ensure_content()
350
while pos < self._content_length:
351
kind = self._content[pos]
353
if kind not in ('f', 'd'):
354
raise ValueError('invalid kind character: %r' % (kind,))
355
content_len, len_len = decode_base128_int(
356
self._content[pos:pos + 5])
358
if content_len + pos > self._content_length:
359
raise ValueError('invalid content_len %d for record @ pos %d'
360
% (content_len, pos - len_len - 1))
361
if kind == 'f': # Fulltext
363
text = self._content[pos:pos+content_len]
364
result.append(('f', content_len, text))
366
result.append(('f', content_len))
367
elif kind == 'd': # Delta
368
delta_content = self._content[pos:pos+content_len]
370
# The first entry in a delta is the decompressed length
371
decomp_len, delta_pos = decode_base128_int(delta_content)
372
result.append(('d', content_len, decomp_len, delta_info))
374
while delta_pos < content_len:
375
c = ord(delta_content[delta_pos])
379
delta_pos) = decode_copy_instruction(delta_content, c,
382
text = self._content[offset:offset+length]
383
delta_info.append(('c', offset, length, text))
385
delta_info.append(('c', offset, length))
386
measured_len += length
389
txt = delta_content[delta_pos:delta_pos+c]
392
delta_info.append(('i', c, txt))
395
if delta_pos != content_len:
396
raise ValueError('Delta consumed a bad number of bytes:'
397
' %d != %d' % (delta_pos, content_len))
398
if measured_len != decomp_len:
399
raise ValueError('Delta claimed fulltext was %d bytes, but'
400
' extraction resulted in %d bytes'
401
% (decomp_len, measured_len))
283
406
class _LazyGroupCompressFactory(object):
284
407
"""Yield content from a GroupCompressBlock on demand."""
432
553
# time (self._block._content) is a little expensive.
433
554
self._block._ensure_content(self._last_byte)
435
def _check_rebuild_block(self):
556
def _check_rebuild_action(self):
436
557
"""Check to see if our block should be repacked."""
437
558
total_bytes_used = 0
438
559
last_byte_used = 0
439
560
for factory in self._factories:
440
561
total_bytes_used += factory._end - factory._start
441
last_byte_used = max(last_byte_used, factory._end)
442
# If we are using most of the bytes from the block, we have nothing
443
# else to check (currently more that 1/2)
562
if last_byte_used < factory._end:
563
last_byte_used = factory._end
564
# If we are using more than half of the bytes from the block, we have
565
# nothing else to check
444
566
if total_bytes_used * 2 >= self._block._content_length:
446
# Can we just strip off the trailing bytes? If we are going to be
447
# transmitting more than 50% of the front of the content, go ahead
567
return None, last_byte_used, total_bytes_used
568
# We are using less than 50% of the content. Is the content we are
569
# using at the beginning of the block? If so, we can just trim the
570
# tail, rather than rebuilding from scratch.
448
571
if total_bytes_used * 2 > last_byte_used:
449
self._trim_block(last_byte_used)
572
return 'trim', last_byte_used, total_bytes_used
452
574
# We are using a small amount of the data, and it isn't just packed
453
575
# nicely at the front, so rebuild the content.
460
582
# expanding many deltas into fulltexts, as well.
461
583
# If we build a cheap enough 'strip', then we could try a strip,
462
584
# if that expands the content, we then rebuild.
463
self._rebuild_block()
585
return 'rebuild', last_byte_used, total_bytes_used
587
def check_is_well_utilized(self):
588
"""Is the current block considered 'well utilized'?
590
This heuristic asks if the current block considers itself to be a fully
591
developed group, rather than just a loose collection of data.
593
if len(self._factories) == 1:
594
# A block of length 1 could be improved by combining with other
595
# groups - don't look deeper. Even larger than max size groups
596
# could compress well with adjacent versions of the same thing.
598
action, last_byte_used, total_bytes_used = self._check_rebuild_action()
599
block_size = self._block._content_length
600
if total_bytes_used < block_size * self._max_cut_fraction:
601
# This block wants to trim itself small enough that we want to
602
# consider it under-utilized.
604
# TODO: This code is meant to be the twin of _insert_record_stream's
605
# 'start_new_block' logic. It would probably be better to factor
606
# out that logic into a shared location, so that it stays
608
# We currently assume a block is properly utilized whenever it is >75%
609
# of the size of a 'full' block. In normal operation, a block is
610
# considered full when it hits 4MB of same-file content. So any block
611
# >3MB is 'full enough'.
612
# The only time this isn't true is when a given block has large-object
613
# content. (a single file >4MB, etc.)
614
# Under these circumstances, we allow a block to grow to
615
# 2 x largest_content. Which means that if a given block had a large
616
# object, it may actually be under-utilized. However, given that this
617
# is 'pack-on-the-fly' it is probably reasonable to not repack large
618
# content blobs on-the-fly. Note that because we return False for all
619
# 1-item blobs, we will repack them; we may wish to reevaluate our
620
# treatment of large object blobs in the future.
621
if block_size >= self._full_enough_block_size:
623
# If a block is <3MB, it still may be considered 'full' if it contains
624
# mixed content. The current rule is 2MB of mixed content is considered
625
# full. So check to see if this block contains mixed content, and
626
# set the threshold appropriately.
628
for factory in self._factories:
629
prefix = factory.key[:-1]
630
if common_prefix is None:
631
common_prefix = prefix
632
elif prefix != common_prefix:
633
# Mixed content, check the size appropriately
634
if block_size >= self._full_enough_mixed_block_size:
637
# The content failed both the mixed check and the single-content check
638
# so obviously it is not fully utilized
639
# TODO: there is one other constraint that isn't being checked
640
# namely, that the entries in the block are in the appropriate
641
# order. For example, you could insert the entries in exactly
642
# reverse groupcompress order, and we would think that is ok.
643
# (all the right objects are in one group, and it is fully
644
# utilized, etc.) For now, we assume that case is rare,
645
# especially since we should always fetch in 'groupcompress'
649
def _check_rebuild_block(self):
650
action, last_byte_used, total_bytes_used = self._check_rebuild_action()
654
self._trim_block(last_byte_used)
655
elif action == 'rebuild':
656
self._rebuild_block()
658
raise ValueError('unknown rebuild action: %r' % (action,))
465
660
def _wire_bytes(self):
466
661
"""Return a byte stream suitable for transmitting over the wire."""
593
790
:param soft: Do a 'soft' compression. This means that we require larger
594
791
ranges to match to be considered for a copy command.
596
:return: The sha1 of lines, the start and end offsets in the delta, the
597
type ('fulltext' or 'delta') and the number of bytes accumulated in
598
the group output so far.
793
:return: The sha1 of lines, the start and end offsets in the delta, and
794
the type ('fulltext' or 'delta').
600
796
:seealso VersionedFiles.add_lines:
602
798
if not bytes: # empty, like a dir entry, etc
603
799
if nostore_sha == _null_sha1:
604
800
raise errors.ExistingContent()
605
return _null_sha1, 0, 0, 'fulltext', 0
801
return _null_sha1, 0, 0, 'fulltext'
606
802
# we assume someone knew what they were doing when they passed it in
607
803
if expected_sha is not None:
608
804
sha1 = expected_sha
614
810
if key[-1] is None:
615
811
key = key[:-1] + ('sha1:' + sha1,)
617
return self._compress(key, bytes, sha1, len(bytes) / 2, soft)
813
start, end, type = self._compress(key, bytes, len(bytes) / 2, soft)
814
return sha1, start, end, type
619
def _compress(self, key, bytes, sha1, max_delta_size, soft=False):
816
def _compress(self, key, bytes, max_delta_size, soft=False):
620
817
"""Compress lines with label key.
622
819
:param key: A key tuple. It is stored in the output for identification
703
904
def __init__(self):
704
905
"""Create a GroupCompressor.
706
:param delta: If False, do not compress records.
907
Used only if the pyrex version is not available.
708
909
super(PythonGroupCompressor, self).__init__()
709
910
self._delta_index = LinesDeltaIndex([])
710
911
# The actual content is managed by LinesDeltaIndex
711
912
self.chunks = self._delta_index.lines
713
def _compress(self, key, bytes, sha1, max_delta_size, soft=False):
914
def _compress(self, key, bytes, max_delta_size, soft=False):
714
915
"""see _CommonGroupCompressor._compress"""
715
bytes_length = len(bytes)
916
input_len = len(bytes)
716
917
new_lines = osutils.split_lines(bytes)
717
out_lines, index_lines = self._delta_index.make_delta(new_lines,
718
bytes_length=bytes_length, soft=soft)
918
out_lines, index_lines = self._delta_index.make_delta(
919
new_lines, bytes_length=input_len, soft=soft)
719
920
delta_length = sum(map(len, out_lines))
720
921
if delta_length > max_delta_size:
721
922
# The delta is longer than the fulltext, insert a fulltext
722
923
type = 'fulltext'
723
out_lines = ['f', encode_base128_int(bytes_length)]
924
out_lines = ['f', encode_base128_int(input_len)]
724
925
out_lines.extend(new_lines)
725
926
index_lines = [False, False]
726
927
index_lines.extend([True] * len(new_lines))
727
out_length = len(out_lines[1]) + bytes_length + 1
729
929
# this is a worthy delta, output it
731
931
out_lines[0] = 'd'
732
932
# Update the delta_length to include those two encoded integers
733
933
out_lines[1] = encode_base128_int(delta_length)
734
out_length = len(out_lines[3]) + 1 + delta_length
735
start = self.endpoint # Before insertion
736
chunk_start = len(self._delta_index.lines)
935
start = self.endpoint
936
chunk_start = len(self.chunks)
937
self._last = (chunk_start, self.endpoint)
737
938
self._delta_index.extend_lines(out_lines, index_lines)
738
939
self.endpoint = self._delta_index.endpoint
739
self.input_bytes += bytes_length
740
chunk_end = len(self._delta_index.lines)
940
self.input_bytes += input_len
941
chunk_end = len(self.chunks)
741
942
self.labels_deltas[key] = (start, chunk_start,
742
943
self.endpoint, chunk_end)
743
return sha1, start, self.endpoint, type, out_length
944
return start, self.endpoint, type
746
947
class PyrexGroupCompressor(_CommonGroupCompressor):
857
1057
versioned_files.stream.close()
1060
class _BatchingBlockFetcher(object):
1061
"""Fetch group compress blocks in batches.
1063
:ivar total_bytes: int of expected number of bytes needed to fetch the
1064
currently pending batch.
1067
def __init__(self, gcvf, locations):
1069
self.locations = locations
1071
self.batch_memos = {}
1072
self.memos_to_get = []
1073
self.total_bytes = 0
1074
self.last_read_memo = None
1077
def add_key(self, key):
1078
"""Add another to key to fetch.
1080
:return: The estimated number of bytes needed to fetch the batch so
1083
self.keys.append(key)
1084
index_memo, _, _, _ = self.locations[key]
1085
read_memo = index_memo[0:3]
1086
# Three possibilities for this read_memo:
1087
# - it's already part of this batch; or
1088
# - it's not yet part of this batch, but is already cached; or
1089
# - it's not yet part of this batch and will need to be fetched.
1090
if read_memo in self.batch_memos:
1091
# This read memo is already in this batch.
1092
return self.total_bytes
1094
cached_block = self.gcvf._group_cache[read_memo]
1096
# This read memo is new to this batch, and the data isn't cached
1098
self.batch_memos[read_memo] = None
1099
self.memos_to_get.append(read_memo)
1100
byte_length = read_memo[2]
1101
self.total_bytes += byte_length
1103
# This read memo is new to this batch, but cached.
1104
# Keep a reference to the cached block in batch_memos because it's
1105
# certain that we'll use it when this batch is processed, but
1106
# there's a risk that it would fall out of _group_cache between now
1108
self.batch_memos[read_memo] = cached_block
1109
return self.total_bytes
1111
def _flush_manager(self):
1112
if self.manager is not None:
1113
for factory in self.manager.get_record_stream():
1116
self.last_read_memo = None
1118
def yield_factories(self, full_flush=False):
1119
"""Yield factories for keys added since the last yield. They will be
1120
returned in the order they were added via add_key.
1122
:param full_flush: by default, some results may not be returned in case
1123
they can be part of the next batch. If full_flush is True, then
1124
all results are returned.
1126
if self.manager is None and not self.keys:
1128
# Fetch all memos in this batch.
1129
blocks = self.gcvf._get_blocks(self.memos_to_get)
1130
# Turn blocks into factories and yield them.
1131
memos_to_get_stack = list(self.memos_to_get)
1132
memos_to_get_stack.reverse()
1133
for key in self.keys:
1134
index_memo, _, parents, _ = self.locations[key]
1135
read_memo = index_memo[:3]
1136
if self.last_read_memo != read_memo:
1137
# We are starting a new block. If we have a
1138
# manager, we have found everything that fits for
1139
# now, so yield records
1140
for factory in self._flush_manager():
1142
# Now start a new manager.
1143
if memos_to_get_stack and memos_to_get_stack[-1] == read_memo:
1144
# The next block from _get_blocks will be the block we
1146
block_read_memo, block = blocks.next()
1147
if block_read_memo != read_memo:
1148
raise AssertionError(
1149
"block_read_memo out of sync with read_memo"
1150
"(%r != %r)" % (block_read_memo, read_memo))
1151
self.batch_memos[read_memo] = block
1152
memos_to_get_stack.pop()
1154
block = self.batch_memos[read_memo]
1155
self.manager = _LazyGroupContentManager(block)
1156
self.last_read_memo = read_memo
1157
start, end = index_memo[3:5]
1158
self.manager.add_factory(key, parents, start, end)
1160
for factory in self._flush_manager():
1163
self.batch_memos.clear()
1164
del self.memos_to_get[:]
1165
self.total_bytes = 0
860
1168
class GroupCompressVersionedFiles(VersionedFiles):
861
1169
"""A group-compress based VersionedFiles implementation."""
863
def __init__(self, index, access, delta=True):
1171
def __init__(self, index, access, delta=True, _unadded_refs=None):
864
1172
"""Create a GroupCompressVersionedFiles object.
866
1174
:param index: The index object storing access and graph data.
867
1175
:param access: The access object storing raw data.
868
1176
:param delta: Whether to delta compress or just entropy compress.
1177
:param _unadded_refs: private parameter, don't use.
870
1179
self._index = index
871
1180
self._access = access
872
1181
self._delta = delta
873
self._unadded_refs = {}
1182
if _unadded_refs is None:
1184
self._unadded_refs = _unadded_refs
874
1185
self._group_cache = LRUSizeCache(max_size=50*1024*1024)
875
1186
self._fallback_vfs = []
1188
def without_fallbacks(self):
1189
"""Return a clone of this object without any fallbacks configured."""
1190
return GroupCompressVersionedFiles(self._index, self._access,
1191
self._delta, _unadded_refs=dict(self._unadded_refs))
877
1193
def add_lines(self, key, parents, lines, parent_texts=None,
878
1194
left_matching_blocks=None, nostore_sha=None, random_id=False,
879
1195
check_content=True):
924
1240
nostore_sha=nostore_sha))[0]
925
1241
return sha1, length, None
1243
def _add_text(self, key, parents, text, nostore_sha=None, random_id=False):
1244
"""See VersionedFiles._add_text()."""
1245
self._index._check_write_ok()
1246
self._check_add(key, None, random_id, check_content=False)
1247
if text.__class__ is not str:
1248
raise errors.BzrBadParameterUnicode("text")
1250
# The caller might pass None if there is no graph data, but kndx
1251
# indexes can't directly store that, so we give them
1252
# an empty tuple instead.
1254
# double handling for now. Make it work until then.
1256
record = FulltextContentFactory(key, parents, None, text)
1257
sha1 = list(self._insert_record_stream([record], random_id=random_id,
1258
nostore_sha=nostore_sha))[0]
1259
return sha1, length, None
927
1261
def add_fallback_versioned_files(self, a_versioned_files):
928
1262
"""Add a source of texts for texts not present in this knit.
934
1268
def annotate(self, key):
935
1269
"""See VersionedFiles.annotate."""
937
parent_map = self.get_parent_map([key])
939
raise errors.RevisionNotPresent(key, self)
940
if parent_map[key] is not None:
941
search = graph._make_breadth_first_searcher([key])
945
present, ghosts = search.next_with_ghosts()
946
except StopIteration:
949
parent_map = self.get_parent_map(keys)
952
parent_map = {key:()}
953
head_cache = _mod_graph.FrozenHeadsCache(graph)
955
reannotate = annotate.reannotate
956
for record in self.get_record_stream(keys, 'topological', True):
958
chunks = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
959
parent_lines = [parent_cache[parent] for parent in parent_map[key]]
960
parent_cache[key] = list(
961
reannotate(parent_lines, chunks, key, None, head_cache))
962
return parent_cache[key]
964
def check(self, progress_bar=None):
1270
ann = annotate.Annotator(self)
1271
return ann.annotate_flat(key)
1273
def get_annotator(self):
1274
return annotate.Annotator(self)
1276
def check(self, progress_bar=None, keys=None):
965
1277
"""See VersionedFiles.check()."""
967
for record in self.get_record_stream(keys, 'unordered', True):
968
record.get_bytes_as('fulltext')
1280
for record in self.get_record_stream(keys, 'unordered', True):
1281
record.get_bytes_as('fulltext')
1283
return self.get_record_stream(keys, 'unordered', True)
970
1285
def _check_add(self, key, lines, random_id, check_content):
971
1286
"""check that version_id and lines are safe to add."""
1014
1345
missing.difference_update(set(new_result))
1015
1346
return result, source_results
1017
def _get_block(self, index_memo):
1018
read_memo = index_memo[0:3]
1021
block = self._group_cache[read_memo]
1024
zdata = self._access.get_raw_records([read_memo]).next()
1025
# decompress - whole thing - this is not a bug, as it
1026
# permits caching. We might want to store the partially
1027
# decompresed group and decompress object, so that recent
1028
# texts are not penalised by big groups.
1029
block = GroupCompressBlock.from_bytes(zdata)
1030
self._group_cache[read_memo] = block
1032
# print len(zdata), len(plain)
1033
# parse - requires split_lines, better to have byte offsets
1034
# here (but not by much - we only split the region for the
1035
# recipe, and we often want to end up with lines anyway.
1348
def _get_blocks(self, read_memos):
1349
"""Get GroupCompressBlocks for the given read_memos.
1351
:returns: a series of (read_memo, block) pairs, in the order they were
1355
for read_memo in read_memos:
1357
block = self._group_cache[read_memo]
1361
cached[read_memo] = block
1363
not_cached_seen = set()
1364
for read_memo in read_memos:
1365
if read_memo in cached:
1366
# Don't fetch what we already have
1368
if read_memo in not_cached_seen:
1369
# Don't try to fetch the same data twice
1371
not_cached.append(read_memo)
1372
not_cached_seen.add(read_memo)
1373
raw_records = self._access.get_raw_records(not_cached)
1374
for read_memo in read_memos:
1376
yield read_memo, cached[read_memo]
1378
# Read the block, and cache it.
1379
zdata = raw_records.next()
1380
block = GroupCompressBlock.from_bytes(zdata)
1381
self._group_cache[read_memo] = block
1382
cached[read_memo] = block
1383
yield read_memo, block
1038
1385
def get_missing_compression_parent_keys(self):
1039
1386
"""Return the keys of missing compression parents.
1205
1552
unadded_keys, source_result)
1206
1553
for key in missing:
1207
1554
yield AbsentContentFactory(key)
1209
last_read_memo = None
1210
# TODO: This works fairly well at batching up existing groups into a
1211
# streamable format, and possibly allowing for taking one big
1212
# group and splitting it when it isn't fully utilized.
1213
# However, it doesn't allow us to find under-utilized groups and
1214
# combine them into a bigger group on the fly.
1215
# (Consider the issue with how chk_map inserts texts
1216
# one-at-a-time.) This could be done at insert_record_stream()
1217
# time, but it probably would decrease the number of
1218
# bytes-on-the-wire for fetch.
1555
# Batch up as many keys as we can until either:
1556
# - we encounter an unadded ref, or
1557
# - we run out of keys, or
1558
# - the total bytes to retrieve for this batch > BATCH_SIZE
1559
batcher = _BatchingBlockFetcher(self, locations)
1219
1560
for source, keys in source_keys:
1220
1561
if source is self:
1221
1562
for key in keys:
1222
1563
if key in self._unadded_refs:
1223
if manager is not None:
1224
for factory in manager.get_record_stream():
1226
last_read_memo = manager = None
1564
# Flush batch, then yield unadded ref from
1566
for factory in batcher.yield_factories(full_flush=True):
1227
1568
bytes, sha1 = self._compressor.extract(key)
1228
1569
parents = self._unadded_refs[key]
1229
1570
yield FulltextContentFactory(key, parents, sha1, bytes)
1231
index_memo, _, parents, (method, _) = locations[key]
1232
read_memo = index_memo[0:3]
1233
if last_read_memo != read_memo:
1234
# We are starting a new block. If we have a
1235
# manager, we have found everything that fits for
1236
# now, so yield records
1237
if manager is not None:
1238
for factory in manager.get_record_stream():
1240
# Now start a new manager
1241
block = self._get_block(index_memo)
1242
manager = _LazyGroupContentManager(block)
1243
last_read_memo = read_memo
1244
start, end = index_memo[3:5]
1245
manager.add_factory(key, parents, start, end)
1572
if batcher.add_key(key) > BATCH_SIZE:
1573
# Ok, this batch is big enough. Yield some results.
1574
for factory in batcher.yield_factories():
1247
if manager is not None:
1248
for factory in manager.get_record_stream():
1250
last_read_memo = manager = None
1577
for factory in batcher.yield_factories(full_flush=True):
1251
1579
for record in source.get_record_stream(keys, ordering,
1252
1580
include_delta_closure):
1254
if manager is not None:
1255
for factory in manager.get_record_stream():
1582
for factory in batcher.yield_factories(full_flush=True):
1258
1585
def get_sha1s(self, keys):
1259
1586
"""See VersionedFiles.get_sha1s()."""
1321
1652
self._compressor = GroupCompressor()
1323
1654
last_prefix = None
1324
last_fulltext_len = None
1325
1655
max_fulltext_len = 0
1326
1656
max_fulltext_prefix = None
1327
1657
insert_manager = None
1328
1658
block_start = None
1329
1659
block_length = None
1660
# XXX: TODO: remove this, it is just for safety checking for now
1661
inserted_keys = set()
1662
reuse_this_block = reuse_blocks
1330
1663
for record in stream:
1331
1664
# Raise an error when a record is missing.
1332
1665
if record.storage_kind == 'absent':
1333
1666
raise errors.RevisionNotPresent(record.key, self)
1668
if record.key in inserted_keys:
1669
trace.note('Insert claimed random_id=True,'
1670
' but then inserted %r two times', record.key)
1672
inserted_keys.add(record.key)
1334
1673
if reuse_blocks:
1335
1674
# If the reuse_blocks flag is set, check to see if we can just
1336
1675
# copy a groupcompress block as-is.
1676
# We only check on the first record (groupcompress-block) not
1677
# on all of the (groupcompress-block-ref) entries.
1678
# The reuse_this_block flag is then kept for as long as
1679
if record.storage_kind == 'groupcompress-block':
1680
# Check to see if we really want to re-use this block
1681
insert_manager = record._manager
1682
reuse_this_block = insert_manager.check_is_well_utilized()
1684
reuse_this_block = False
1685
if reuse_this_block:
1686
# We still want to reuse this block
1337
1687
if record.storage_kind == 'groupcompress-block':
1338
1688
# Insert the raw block into the target repo
1339
1689
insert_manager = record._manager
1340
insert_manager._check_rebuild_block()
1341
1690
bytes = record._manager._block.to_bytes()
1342
1691
_, start, length = self._access.add_raw_records(
1343
1692
[(None, len(bytes))], bytes)[0]
1371
1725
if max_fulltext_len < len(bytes):
1372
1726
max_fulltext_len = len(bytes)
1373
1727
max_fulltext_prefix = prefix
1374
(found_sha1, start_point, end_point, type,
1375
length) = self._compressor.compress(record.key,
1376
bytes, record.sha1, soft=soft,
1377
nostore_sha=nostore_sha)
1378
# delta_ratio = float(len(bytes)) / length
1728
(found_sha1, start_point, end_point,
1729
type) = self._compressor.compress(record.key,
1730
bytes, record.sha1, soft=soft,
1731
nostore_sha=nostore_sha)
1732
# delta_ratio = float(len(bytes)) / (end_point - start_point)
1379
1733
# Check if we want to continue to include that text
1380
1734
if (prefix == max_fulltext_prefix
1381
1735
and end_point < 2 * max_fulltext_len):
1476
1830
:param add_callback: If not None, allow additions to the index and call
1477
1831
this callback with a list of added GraphIndex nodes:
1478
1832
[(node, value, node_refs), ...]
1833
:param track_external_parent_refs: As keys are added, keep track of the
1834
keys they reference, so that we can query get_missing_parents(),
1836
:param inconsistency_fatal: When asked to add records that are already
1837
present, and the details are inconsistent with the existing
1838
record, raise an exception instead of warning (and skipping the
1480
1841
self._add_callback = add_callback
1481
1842
self._graph_index = graph_index
1482
1843
self._parents = parents
1483
1844
self.has_graph = parents
1484
1845
self._is_locked = is_locked
1846
self._inconsistency_fatal = inconsistency_fatal
1847
if track_external_parent_refs:
1848
self._key_dependencies = knit._KeyRefs(
1849
track_new_keys=track_new_keys)
1851
self._key_dependencies = None
1486
1853
def add_records(self, records, random_id=False):
1487
1854
"""Add multiple records to the index.
1637
2031
delta_end = int(bits[3])
1638
2032
return node[0], start, stop, basis_end, delta_end
2034
def scan_unvalidated_index(self, graph_index):
2035
"""Inform this _GCGraphIndex that there is an unvalidated index.
2037
This allows this _GCGraphIndex to keep track of any missing
2038
compression parents we may want to have filled in to make those
2039
indices valid. It also allows _GCGraphIndex to track any new keys.
2041
:param graph_index: A GraphIndex
2043
key_dependencies = self._key_dependencies
2044
if key_dependencies is None:
2046
for node in graph_index.iter_all_entries():
2047
# Add parent refs from graph_index (and discard parent refs
2048
# that the graph_index has).
2049
key_dependencies.add_references(node[1], node[3][0])
1641
2052
from bzrlib._groupcompress_py import (
1643
2054
apply_delta_to_source,
1644
2055
encode_base128_int,
1645
2056
decode_base128_int,
2057
decode_copy_instruction,
1646
2058
LinesDeltaIndex,
added
removed
bzrlib/groupcompress.py
