161
164
# 'unconsumed_tail'
163
166
# Do we have enough bytes already?
164
if num_bytes is not None and len(self._content) >= num_bytes:
166
if num_bytes is None and self._z_content_decompressor is None:
167
# We must have already decompressed everything
167
if len(self._content) >= num_bytes:
169
169
# If we got this far, and don't have a decompressor, something is wrong
170
170
if self._z_content_decompressor is None:
171
171
raise AssertionError(
172
172
'No decompressor to decompress %d bytes' % num_bytes)
173
173
remaining_decomp = self._z_content_decompressor.unconsumed_tail
174
if num_bytes is None:
176
# We don't know how much is left, but we'll decompress it all
177
self._content += self._z_content_decompressor.decompress(
179
# Note: There's what I consider a bug in zlib.decompressobj
180
# If you pass back in the entire unconsumed_tail, only
181
# this time you don't pass a max-size, it doesn't
182
# change the unconsumed_tail back to None/''.
183
# However, we know we are done with the whole stream
184
self._z_content_decompressor = None
185
# XXX: Why is this the only place in this routine we set this?
186
self._content_length = len(self._content)
188
if not remaining_decomp:
189
raise AssertionError('Nothing left to decompress')
190
needed_bytes = num_bytes - len(self._content)
191
# We always set max_size to 32kB over the minimum needed, so that
192
# zlib will give us as much as we really want.
193
# TODO: If this isn't good enough, we could make a loop here,
194
# that keeps expanding the request until we get enough
195
self._content += self._z_content_decompressor.decompress(
196
remaining_decomp, needed_bytes + _ZLIB_DECOMP_WINDOW)
197
if len(self._content) < num_bytes:
198
raise AssertionError('%d bytes wanted, only %d available'
199
% (num_bytes, len(self._content)))
200
if not self._z_content_decompressor.unconsumed_tail:
201
# The stream is finished
202
self._z_content_decompressor = None
174
if not remaining_decomp:
175
raise AssertionError('Nothing left to decompress')
176
needed_bytes = num_bytes - len(self._content)
177
# We always set max_size to 32kB over the minimum needed, so that
178
# zlib will give us as much as we really want.
179
# TODO: If this isn't good enough, we could make a loop here,
180
# that keeps expanding the request until we get enough
181
self._content += self._z_content_decompressor.decompress(
182
remaining_decomp, needed_bytes + _ZLIB_DECOMP_WINDOW)
183
if len(self._content) < num_bytes:
184
raise AssertionError('%d bytes wanted, only %d available'
185
% (num_bytes, len(self._content)))
186
if not self._z_content_decompressor.unconsumed_tail:
187
# The stream is finished
188
self._z_content_decompressor = None
204
190
def _parse_bytes(self, bytes, pos):
205
191
"""Read the various lengths from the header.
273
259
bytes = apply_delta_to_source(self._content, content_start, end)
262
def set_chunked_content(self, content_chunks, length):
263
"""Set the content of this block to the given chunks."""
264
# If we have lots of short lines, it is may be more efficient to join
265
# the content ahead of time. If the content is <10MiB, we don't really
266
# care about the extra memory consumption, so we can just pack it and
267
# be done. However, timing showed 18s => 17.9s for repacking 1k revs of
268
# mysql, which is below the noise margin
269
self._content_length = length
270
self._content_chunks = content_chunks
272
self._z_content = None
276
274
def set_content(self, content):
277
275
"""Set the content of this block."""
278
276
self._content_length = len(content)
279
277
self._content = content
280
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):
290
compressor = zlib.compressobj(zlib.Z_DEFAULT_COMPRESSION)
291
compressed_chunks = map(compressor.compress, self._content_chunks)
292
compressed_chunks.append(compressor.flush())
293
self._z_content = ''.join(compressed_chunks)
294
self._z_content_length = len(self._z_content)
296
def _create_z_content(self):
297
if self._z_content is not None:
300
self._create_z_content_using_lzma()
302
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)
282
308
def to_bytes(self):
283
309
"""Encode the information into a byte stream."""
284
compress = zlib.compress
286
compress = pylzma.compress
287
if self._z_content is None:
288
if self._content is None:
289
raise AssertionError('Nothing to compress')
290
self._z_content = compress(self._content)
291
self._z_content_length = len(self._z_content)
310
self._create_z_content()
293
312
header = self.GCB_LZ_HEADER
300
319
return ''.join(chunks)
321
def _dump(self, include_text=False):
322
"""Take this block, and spit out a human-readable structure.
324
:param include_text: Inserts also include text bits, chose whether you
325
want this displayed in the dump or not.
326
:return: A dump of the given block. The layout is something like:
327
[('f', length), ('d', delta_length, text_length, [delta_info])]
328
delta_info := [('i', num_bytes, text), ('c', offset, num_bytes),
331
self._ensure_content()
334
while pos < self._content_length:
335
kind = self._content[pos]
337
if kind not in ('f', 'd'):
338
raise ValueError('invalid kind character: %r' % (kind,))
339
content_len, len_len = decode_base128_int(
340
self._content[pos:pos + 5])
342
if content_len + pos > self._content_length:
343
raise ValueError('invalid content_len %d for record @ pos %d'
344
% (content_len, pos - len_len - 1))
345
if kind == 'f': # Fulltext
347
text = self._content[pos:pos+content_len]
348
result.append(('f', content_len, text))
350
result.append(('f', content_len))
351
elif kind == 'd': # Delta
352
delta_content = self._content[pos:pos+content_len]
354
# The first entry in a delta is the decompressed length
355
decomp_len, delta_pos = decode_base128_int(delta_content)
356
result.append(('d', content_len, decomp_len, delta_info))
358
while delta_pos < content_len:
359
c = ord(delta_content[delta_pos])
363
delta_pos) = decode_copy_instruction(delta_content, c,
366
text = self._content[offset:offset+length]
367
delta_info.append(('c', offset, length, text))
369
delta_info.append(('c', offset, length))
370
measured_len += length
373
txt = delta_content[delta_pos:delta_pos+c]
376
delta_info.append(('i', c, txt))
379
if delta_pos != content_len:
380
raise ValueError('Delta consumed a bad number of bytes:'
381
' %d != %d' % (delta_pos, content_len))
382
if measured_len != decomp_len:
383
raise ValueError('Delta claimed fulltext was %d bytes, but'
384
' extraction resulted in %d bytes'
385
% (decomp_len, measured_len))
303
390
class _LazyGroupCompressFactory(object):
304
391
"""Yield content from a GroupCompressBlock on demand."""
443
537
# time (self._block._content) is a little expensive.
444
538
self._block._ensure_content(self._last_byte)
446
def _check_rebuild_block(self):
540
def _check_rebuild_action(self):
447
541
"""Check to see if our block should be repacked."""
448
542
total_bytes_used = 0
449
543
last_byte_used = 0
450
544
for factory in self._factories:
451
545
total_bytes_used += factory._end - factory._start
452
last_byte_used = max(last_byte_used, factory._end)
453
# If we are using most of the bytes from the block, we have nothing
454
# else to check (currently more that 1/2)
546
if last_byte_used < factory._end:
547
last_byte_used = factory._end
548
# If we are using more than half of the bytes from the block, we have
549
# nothing else to check
455
550
if total_bytes_used * 2 >= self._block._content_length:
457
# Can we just strip off the trailing bytes? If we are going to be
458
# transmitting more than 50% of the front of the content, go ahead
551
return None, last_byte_used, total_bytes_used
552
# We are using less than 50% of the content. Is the content we are
553
# using at the beginning of the block? If so, we can just trim the
554
# tail, rather than rebuilding from scratch.
459
555
if total_bytes_used * 2 > last_byte_used:
460
self._trim_block(last_byte_used)
556
return 'trim', last_byte_used, total_bytes_used
463
558
# We are using a small amount of the data, and it isn't just packed
464
559
# nicely at the front, so rebuild the content.
471
566
# expanding many deltas into fulltexts, as well.
472
567
# If we build a cheap enough 'strip', then we could try a strip,
473
568
# if that expands the content, we then rebuild.
474
self._rebuild_block()
569
return 'rebuild', last_byte_used, total_bytes_used
571
def check_is_well_utilized(self):
572
"""Is the current block considered 'well utilized'?
574
This heuristic asks if the current block considers itself to be a fully
575
developed group, rather than just a loose collection of data.
577
if len(self._factories) == 1:
578
# A block of length 1 could be improved by combining with other
579
# groups - don't look deeper. Even larger than max size groups
580
# could compress well with adjacent versions of the same thing.
582
action, last_byte_used, total_bytes_used = self._check_rebuild_action()
583
block_size = self._block._content_length
584
if total_bytes_used < block_size * self._max_cut_fraction:
585
# This block wants to trim itself small enough that we want to
586
# consider it under-utilized.
588
# TODO: This code is meant to be the twin of _insert_record_stream's
589
# 'start_new_block' logic. It would probably be better to factor
590
# out that logic into a shared location, so that it stays
592
# We currently assume a block is properly utilized whenever it is >75%
593
# of the size of a 'full' block. In normal operation, a block is
594
# considered full when it hits 4MB of same-file content. So any block
595
# >3MB is 'full enough'.
596
# The only time this isn't true is when a given block has large-object
597
# content. (a single file >4MB, etc.)
598
# Under these circumstances, we allow a block to grow to
599
# 2 x largest_content. Which means that if a given block had a large
600
# object, it may actually be under-utilized. However, given that this
601
# is 'pack-on-the-fly' it is probably reasonable to not repack large
602
# content blobs on-the-fly. Note that because we return False for all
603
# 1-item blobs, we will repack them; we may wish to reevaluate our
604
# treatment of large object blobs in the future.
605
if block_size >= self._full_enough_block_size:
607
# If a block is <3MB, it still may be considered 'full' if it contains
608
# mixed content. The current rule is 2MB of mixed content is considered
609
# full. So check to see if this block contains mixed content, and
610
# set the threshold appropriately.
612
for factory in self._factories:
613
prefix = factory.key[:-1]
614
if common_prefix is None:
615
common_prefix = prefix
616
elif prefix != common_prefix:
617
# Mixed content, check the size appropriately
618
if block_size >= self._full_enough_mixed_block_size:
621
# The content failed both the mixed check and the single-content check
622
# so obviously it is not fully utilized
623
# TODO: there is one other constraint that isn't being checked
624
# namely, that the entries in the block are in the appropriate
625
# order. For example, you could insert the entries in exactly
626
# reverse groupcompress order, and we would think that is ok.
627
# (all the right objects are in one group, and it is fully
628
# utilized, etc.) For now, we assume that case is rare,
629
# especially since we should always fetch in 'groupcompress'
633
def _check_rebuild_block(self):
634
action, last_byte_used, total_bytes_used = self._check_rebuild_action()
638
self._trim_block(last_byte_used)
639
elif action == 'rebuild':
640
self._rebuild_block()
642
raise ValueError('unknown rebuild action: %r' % (action,))
476
644
def _wire_bytes(self):
477
645
"""Return a byte stream suitable for transmitting over the wire."""
864
1041
versioned_files.stream.close()
1044
class _BatchingBlockFetcher(object):
1045
"""Fetch group compress blocks in batches.
1047
:ivar total_bytes: int of expected number of bytes needed to fetch the
1048
currently pending batch.
1051
def __init__(self, gcvf, locations):
1053
self.locations = locations
1055
self.batch_memos = {}
1056
self.memos_to_get = []
1057
self.total_bytes = 0
1058
self.last_read_memo = None
1061
def add_key(self, key):
1062
"""Add another to key to fetch.
1064
:return: The estimated number of bytes needed to fetch the batch so
1067
self.keys.append(key)
1068
index_memo, _, _, _ = self.locations[key]
1069
read_memo = index_memo[0:3]
1070
# Three possibilities for this read_memo:
1071
# - it's already part of this batch; or
1072
# - it's not yet part of this batch, but is already cached; or
1073
# - it's not yet part of this batch and will need to be fetched.
1074
if read_memo in self.batch_memos:
1075
# This read memo is already in this batch.
1076
return self.total_bytes
1078
cached_block = self.gcvf._group_cache[read_memo]
1080
# This read memo is new to this batch, and the data isn't cached
1082
self.batch_memos[read_memo] = None
1083
self.memos_to_get.append(read_memo)
1084
byte_length = read_memo[2]
1085
self.total_bytes += byte_length
1087
# This read memo is new to this batch, but cached.
1088
# Keep a reference to the cached block in batch_memos because it's
1089
# certain that we'll use it when this batch is processed, but
1090
# there's a risk that it would fall out of _group_cache between now
1092
self.batch_memos[read_memo] = cached_block
1093
return self.total_bytes
1095
def _flush_manager(self):
1096
if self.manager is not None:
1097
for factory in self.manager.get_record_stream():
1100
self.last_read_memo = None
1102
def yield_factories(self, full_flush=False):
1103
"""Yield factories for keys added since the last yield. They will be
1104
returned in the order they were added via add_key.
1106
:param full_flush: by default, some results may not be returned in case
1107
they can be part of the next batch. If full_flush is True, then
1108
all results are returned.
1110
if self.manager is None and not self.keys:
1112
# Fetch all memos in this batch.
1113
blocks = self.gcvf._get_blocks(self.memos_to_get)
1114
# Turn blocks into factories and yield them.
1115
memos_to_get_stack = list(self.memos_to_get)
1116
memos_to_get_stack.reverse()
1117
for key in self.keys:
1118
index_memo, _, parents, _ = self.locations[key]
1119
read_memo = index_memo[:3]
1120
if self.last_read_memo != read_memo:
1121
# We are starting a new block. If we have a
1122
# manager, we have found everything that fits for
1123
# now, so yield records
1124
for factory in self._flush_manager():
1126
# Now start a new manager.
1127
if memos_to_get_stack and memos_to_get_stack[-1] == read_memo:
1128
# The next block from _get_blocks will be the block we
1130
block_read_memo, block = blocks.next()
1131
if block_read_memo != read_memo:
1132
raise AssertionError(
1133
"block_read_memo out of sync with read_memo"
1134
"(%r != %r)" % (block_read_memo, read_memo))
1135
self.batch_memos[read_memo] = block
1136
memos_to_get_stack.pop()
1138
block = self.batch_memos[read_memo]
1139
self.manager = _LazyGroupContentManager(block)
1140
self.last_read_memo = read_memo
1141
start, end = index_memo[3:5]
1142
self.manager.add_factory(key, parents, start, end)
1144
for factory in self._flush_manager():
1147
self.batch_memos.clear()
1148
del self.memos_to_get[:]
1149
self.total_bytes = 0
867
1152
class GroupCompressVersionedFiles(VersionedFiles):
868
1153
"""A group-compress based VersionedFiles implementation."""
870
def __init__(self, index, access, delta=True):
1155
def __init__(self, index, access, delta=True, _unadded_refs=None):
871
1156
"""Create a GroupCompressVersionedFiles object.
873
1158
:param index: The index object storing access and graph data.
874
1159
:param access: The access object storing raw data.
875
1160
:param delta: Whether to delta compress or just entropy compress.
1161
:param _unadded_refs: private parameter, don't use.
877
1163
self._index = index
878
1164
self._access = access
879
1165
self._delta = delta
880
self._unadded_refs = {}
1166
if _unadded_refs is None:
1168
self._unadded_refs = _unadded_refs
881
1169
self._group_cache = LRUSizeCache(max_size=50*1024*1024)
882
1170
self._fallback_vfs = []
1172
def without_fallbacks(self):
1173
"""Return a clone of this object without any fallbacks configured."""
1174
return GroupCompressVersionedFiles(self._index, self._access,
1175
self._delta, _unadded_refs=dict(self._unadded_refs))
884
1177
def add_lines(self, key, parents, lines, parent_texts=None,
885
1178
left_matching_blocks=None, nostore_sha=None, random_id=False,
886
1179
check_content=True):
931
1224
nostore_sha=nostore_sha))[0]
932
1225
return sha1, length, None
1227
def _add_text(self, key, parents, text, nostore_sha=None, random_id=False):
1228
"""See VersionedFiles._add_text()."""
1229
self._index._check_write_ok()
1230
self._check_add(key, None, random_id, check_content=False)
1231
if text.__class__ is not str:
1232
raise errors.BzrBadParameterUnicode("text")
1234
# The caller might pass None if there is no graph data, but kndx
1235
# indexes can't directly store that, so we give them
1236
# an empty tuple instead.
1238
# double handling for now. Make it work until then.
1240
record = FulltextContentFactory(key, parents, None, text)
1241
sha1 = list(self._insert_record_stream([record], random_id=random_id,
1242
nostore_sha=nostore_sha))[0]
1243
return sha1, length, None
934
1245
def add_fallback_versioned_files(self, a_versioned_files):
935
1246
"""Add a source of texts for texts not present in this knit.
941
1252
def annotate(self, key):
942
1253
"""See VersionedFiles.annotate."""
944
parent_map = self.get_parent_map([key])
946
raise errors.RevisionNotPresent(key, self)
947
if parent_map[key] is not None:
948
search = graph._make_breadth_first_searcher([key])
952
present, ghosts = search.next_with_ghosts()
953
except StopIteration:
956
parent_map = self.get_parent_map(keys)
959
parent_map = {key:()}
960
head_cache = _mod_graph.FrozenHeadsCache(graph)
962
reannotate = annotate.reannotate
963
for record in self.get_record_stream(keys, 'topological', True):
965
chunks = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
966
parent_lines = [parent_cache[parent] for parent in parent_map[key]]
967
parent_cache[key] = list(
968
reannotate(parent_lines, chunks, key, None, head_cache))
969
return parent_cache[key]
971
def check(self, progress_bar=None):
1254
ann = annotate.Annotator(self)
1255
return ann.annotate_flat(key)
1257
def get_annotator(self):
1258
return annotate.Annotator(self)
1260
def check(self, progress_bar=None, keys=None):
972
1261
"""See VersionedFiles.check()."""
974
for record in self.get_record_stream(keys, 'unordered', True):
975
record.get_bytes_as('fulltext')
1264
for record in self.get_record_stream(keys, 'unordered', True):
1265
record.get_bytes_as('fulltext')
1267
return self.get_record_stream(keys, 'unordered', True)
1269
def clear_cache(self):
1270
"""See VersionedFiles.clear_cache()"""
1271
self._group_cache.clear()
1272
self._index._graph_index.clear_cache()
1273
self._index._int_cache.clear()
977
1275
def _check_add(self, key, lines, random_id, check_content):
978
1276
"""check that version_id and lines are safe to add."""
1021
1335
missing.difference_update(set(new_result))
1022
1336
return result, source_results
1024
def _get_block(self, index_memo):
1025
read_memo = index_memo[0:3]
1028
block = self._group_cache[read_memo]
1031
zdata = self._access.get_raw_records([read_memo]).next()
1032
# decompress - whole thing - this is not a bug, as it
1033
# permits caching. We might want to store the partially
1034
# decompresed group and decompress object, so that recent
1035
# texts are not penalised by big groups.
1036
block = GroupCompressBlock.from_bytes(zdata)
1037
self._group_cache[read_memo] = block
1039
# print len(zdata), len(plain)
1040
# parse - requires split_lines, better to have byte offsets
1041
# here (but not by much - we only split the region for the
1042
# recipe, and we often want to end up with lines anyway.
1338
def _get_blocks(self, read_memos):
1339
"""Get GroupCompressBlocks for the given read_memos.
1341
:returns: a series of (read_memo, block) pairs, in the order they were
1345
for read_memo in read_memos:
1347
block = self._group_cache[read_memo]
1351
cached[read_memo] = block
1353
not_cached_seen = set()
1354
for read_memo in read_memos:
1355
if read_memo in cached:
1356
# Don't fetch what we already have
1358
if read_memo in not_cached_seen:
1359
# Don't try to fetch the same data twice
1361
not_cached.append(read_memo)
1362
not_cached_seen.add(read_memo)
1363
raw_records = self._access.get_raw_records(not_cached)
1364
for read_memo in read_memos:
1366
yield read_memo, cached[read_memo]
1368
# Read the block, and cache it.
1369
zdata = raw_records.next()
1370
block = GroupCompressBlock.from_bytes(zdata)
1371
self._group_cache[read_memo] = block
1372
cached[read_memo] = block
1373
yield read_memo, block
1045
1375
def get_missing_compression_parent_keys(self):
1046
1376
"""Return the keys of missing compression parents.
1212
1542
unadded_keys, source_result)
1213
1543
for key in missing:
1214
1544
yield AbsentContentFactory(key)
1216
last_read_memo = None
1217
# TODO: This works fairly well at batching up existing groups into a
1218
# streamable format, and possibly allowing for taking one big
1219
# group and splitting it when it isn't fully utilized.
1220
# However, it doesn't allow us to find under-utilized groups and
1221
# combine them into a bigger group on the fly.
1222
# (Consider the issue with how chk_map inserts texts
1223
# one-at-a-time.) This could be done at insert_record_stream()
1224
# time, but it probably would decrease the number of
1225
# bytes-on-the-wire for fetch.
1545
# Batch up as many keys as we can until either:
1546
# - we encounter an unadded ref, or
1547
# - we run out of keys, or
1548
# - the total bytes to retrieve for this batch > BATCH_SIZE
1549
batcher = _BatchingBlockFetcher(self, locations)
1226
1550
for source, keys in source_keys:
1227
1551
if source is self:
1228
1552
for key in keys:
1229
1553
if key in self._unadded_refs:
1230
if manager is not None:
1231
for factory in manager.get_record_stream():
1233
last_read_memo = manager = None
1554
# Flush batch, then yield unadded ref from
1556
for factory in batcher.yield_factories(full_flush=True):
1234
1558
bytes, sha1 = self._compressor.extract(key)
1235
1559
parents = self._unadded_refs[key]
1236
1560
yield FulltextContentFactory(key, parents, sha1, bytes)
1238
index_memo, _, parents, (method, _) = locations[key]
1239
read_memo = index_memo[0:3]
1240
if last_read_memo != read_memo:
1241
# We are starting a new block. If we have a
1242
# manager, we have found everything that fits for
1243
# now, so yield records
1244
if manager is not None:
1245
for factory in manager.get_record_stream():
1247
# Now start a new manager
1248
block = self._get_block(index_memo)
1249
manager = _LazyGroupContentManager(block)
1250
last_read_memo = read_memo
1251
start, end = index_memo[3:5]
1252
manager.add_factory(key, parents, start, end)
1562
if batcher.add_key(key) > BATCH_SIZE:
1563
# Ok, this batch is big enough. Yield some results.
1564
for factory in batcher.yield_factories():
1254
if manager is not None:
1255
for factory in manager.get_record_stream():
1257
last_read_memo = manager = None
1567
for factory in batcher.yield_factories(full_flush=True):
1258
1569
for record in source.get_record_stream(keys, ordering,
1259
1570
include_delta_closure):
1261
if manager is not None:
1262
for factory in manager.get_record_stream():
1572
for factory in batcher.yield_factories(full_flush=True):
1265
1575
def get_sha1s(self, keys):
1266
1576
"""See VersionedFiles.get_sha1s()."""
1352
1663
if reuse_blocks:
1353
1664
# If the reuse_blocks flag is set, check to see if we can just
1354
1665
# copy a groupcompress block as-is.
1666
# We only check on the first record (groupcompress-block) not
1667
# on all of the (groupcompress-block-ref) entries.
1668
# The reuse_this_block flag is then kept for as long as
1669
if record.storage_kind == 'groupcompress-block':
1670
# Check to see if we really want to re-use this block
1671
insert_manager = record._manager
1672
reuse_this_block = insert_manager.check_is_well_utilized()
1674
reuse_this_block = False
1675
if reuse_this_block:
1676
# We still want to reuse this block
1355
1677
if record.storage_kind == 'groupcompress-block':
1356
1678
# Insert the raw block into the target repo
1357
1679
insert_manager = record._manager
1358
insert_manager._check_rebuild_block()
1359
1680
bytes = record._manager._block.to_bytes()
1360
1681
_, start, length = self._access.add_raw_records(
1361
1682
[(None, len(bytes))], bytes)[0]
1493
1825
:param add_callback: If not None, allow additions to the index and call
1494
1826
this callback with a list of added GraphIndex nodes:
1495
1827
[(node, value, node_refs), ...]
1828
:param track_external_parent_refs: As keys are added, keep track of the
1829
keys they reference, so that we can query get_missing_parents(),
1831
:param inconsistency_fatal: When asked to add records that are already
1832
present, and the details are inconsistent with the existing
1833
record, raise an exception instead of warning (and skipping the
1497
1836
self._add_callback = add_callback
1498
1837
self._graph_index = graph_index
1499
1838
self._parents = parents
1500
1839
self.has_graph = parents
1501
1840
self._is_locked = is_locked
1841
self._inconsistency_fatal = inconsistency_fatal
1842
# GroupCompress records tend to have the same 'group' start + offset
1843
# repeated over and over, this creates a surplus of ints
1844
self._int_cache = {}
1845
if track_external_parent_refs:
1846
self._key_dependencies = knit._KeyRefs(
1847
track_new_keys=track_new_keys)
1849
self._key_dependencies = None
1503
1851
def add_records(self, records, random_id=False):
1504
1852
"""Add multiple records to the index.
1535
1883
if not random_id:
1536
1884
present_nodes = self._get_entries(keys)
1537
1885
for (index, key, value, node_refs) in present_nodes:
1538
if node_refs != keys[key][1]:
1539
raise errors.KnitCorrupt(self, "inconsistent details in add_records"
1540
": %s %s" % ((value, node_refs), keys[key]))
1886
# Sometimes these are passed as a list rather than a tuple
1887
node_refs = static_tuple.as_tuples(node_refs)
1888
passed = static_tuple.as_tuples(keys[key])
1889
if node_refs != passed[1]:
1890
details = '%s %s %s' % (key, (value, node_refs), passed)
1891
if self._inconsistency_fatal:
1892
raise errors.KnitCorrupt(self, "inconsistent details"
1893
" in add_records: %s" %
1896
trace.warning("inconsistent details in skipped"
1897
" record: %s", details)
1648
2026
"""Convert an index value to position details."""
1649
2027
bits = node[2].split(' ')
1650
2028
# It would be nice not to read the entire gzip.
2029
# start and stop are put into _int_cache because they are very common.
2030
# They define the 'group' that an entry is in, and many groups can have
2031
# thousands of objects.
2032
# Branching Launchpad, for example, saves ~600k integers, at 12 bytes
2033
# each, or about 7MB. Note that it might be even more when you consider
2034
# how PyInt is allocated in separate slabs. And you can't return a slab
2035
# 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
1651
2038
start = int(bits[0])
2039
start = self._int_cache.setdefault(start, start)
1652
2040
stop = int(bits[1])
2041
stop = self._int_cache.setdefault(stop, stop)
1653
2042
basis_end = int(bits[2])
1654
2043
delta_end = int(bits[3])
1655
return node[0], start, stop, basis_end, delta_end
2044
# We can't use StaticTuple here, because node[0] is a BTreeGraphIndex
2046
return (node[0], start, stop, basis_end, delta_end)
2048
def scan_unvalidated_index(self, graph_index):
2049
"""Inform this _GCGraphIndex that there is an unvalidated index.
2051
This allows this _GCGraphIndex to keep track of any missing
2052
compression parents we may want to have filled in to make those
2053
indices valid. It also allows _GCGraphIndex to track any new keys.
2055
:param graph_index: A GraphIndex
2057
key_dependencies = self._key_dependencies
2058
if key_dependencies is None:
2060
for node in graph_index.iter_all_entries():
2061
# Add parent refs from graph_index (and discard parent refs
2062
# that the graph_index has).
2063
key_dependencies.add_references(node[1], node[3][0])
1658
2066
from bzrlib._groupcompress_py import (