153
139
texts/deltas (via (fileid, revisionid) tuples).
154
140
:param signature_index: A GraphIndex for determining what signatures are
155
141
present in the Pack and accessing the locations of their texts.
156
:param chk_index: A GraphIndex for accessing content by CHK, if the
159
143
self.revision_index = revision_index
160
144
self.inventory_index = inventory_index
161
145
self.text_index = text_index
162
146
self.signature_index = signature_index
163
self.chk_index = chk_index
165
148
def access_tuple(self):
166
149
"""Return a tuple (transport, name) for the pack content."""
167
150
return self.pack_transport, self.file_name()
169
def _check_references(self):
170
"""Make sure our external references are present.
172
Packs are allowed to have deltas whose base is not in the pack, but it
173
must be present somewhere in this collection. It is not allowed to
174
have deltas based on a fallback repository.
175
(See <https://bugs.launchpad.net/bzr/+bug/288751>)
178
for (index_name, external_refs, index) in [
180
self._get_external_refs(self.text_index),
181
self._pack_collection.text_index.combined_index),
183
self._get_external_refs(self.inventory_index),
184
self._pack_collection.inventory_index.combined_index),
186
missing = external_refs.difference(
187
k for (idx, k, v, r) in
188
index.iter_entries(external_refs))
190
missing_items[index_name] = sorted(list(missing))
192
from pprint import pformat
193
raise errors.BzrCheckError(
194
"Newly created pack file %r has delta references to "
195
"items not in its repository:\n%s"
196
% (self, pformat(missing_items)))
198
152
def file_name(self):
199
153
"""Get the file name for the pack on disk."""
200
154
return self.name + '.pack'
226
172
"""The text index is the name + .tix."""
227
173
return self.index_name('text', name)
229
def _replace_index_with_readonly(self, index_type):
230
unlimited_cache = False
231
if index_type == 'chk':
232
unlimited_cache = True
233
setattr(self, index_type + '_index',
234
self.index_class(self.index_transport,
235
self.index_name(index_type, self.name),
236
self.index_sizes[self.index_offset(index_type)],
237
unlimited_cache=unlimited_cache))
175
def _external_compression_parents_of_texts(self):
178
for node in self.text_index.iter_all_entries():
180
refs.update(node[3][1])
240
184
class ExistingPack(Pack):
241
185
"""An in memory proxy for an existing .pack and its disk indices."""
243
187
def __init__(self, pack_transport, name, revision_index, inventory_index,
244
text_index, signature_index, chk_index=None):
188
text_index, signature_index):
245
189
"""Create an ExistingPack object.
247
191
:param pack_transport: The transport where the pack file resides.
248
192
:param name: The name of the pack on disk in the pack_transport.
250
194
Pack.__init__(self, revision_index, inventory_index, text_index,
251
signature_index, chk_index)
253
197
self.pack_transport = pack_transport
254
198
if None in (revision_index, inventory_index, text_index,
262
206
return not self.__eq__(other)
264
208
def __repr__(self):
265
return "<%s.%s object at 0x%x, %s, %s" % (
266
self.__class__.__module__, self.__class__.__name__, id(self),
267
self.pack_transport, self.name)
270
class ResumedPack(ExistingPack):
272
def __init__(self, name, revision_index, inventory_index, text_index,
273
signature_index, upload_transport, pack_transport, index_transport,
274
pack_collection, chk_index=None):
275
"""Create a ResumedPack object."""
276
ExistingPack.__init__(self, pack_transport, name, revision_index,
277
inventory_index, text_index, signature_index,
279
self.upload_transport = upload_transport
280
self.index_transport = index_transport
281
self.index_sizes = [None, None, None, None]
283
('revision', revision_index),
284
('inventory', inventory_index),
285
('text', text_index),
286
('signature', signature_index),
288
if chk_index is not None:
289
indices.append(('chk', chk_index))
290
self.index_sizes.append(None)
291
for index_type, index in indices:
292
offset = self.index_offset(index_type)
293
self.index_sizes[offset] = index._size
294
self.index_class = pack_collection._index_class
295
self._pack_collection = pack_collection
296
self._state = 'resumed'
297
# XXX: perhaps check that the .pack file exists?
299
def access_tuple(self):
300
if self._state == 'finished':
301
return Pack.access_tuple(self)
302
elif self._state == 'resumed':
303
return self.upload_transport, self.file_name()
305
raise AssertionError(self._state)
308
self.upload_transport.delete(self.file_name())
309
indices = [self.revision_index, self.inventory_index, self.text_index,
310
self.signature_index]
311
if self.chk_index is not None:
312
indices.append(self.chk_index)
313
for index in indices:
314
index._transport.delete(index._name)
317
self._check_references()
318
index_types = ['revision', 'inventory', 'text', 'signature']
319
if self.chk_index is not None:
320
index_types.append('chk')
321
for index_type in index_types:
322
old_name = self.index_name(index_type, self.name)
323
new_name = '../indices/' + old_name
324
self.upload_transport.rename(old_name, new_name)
325
self._replace_index_with_readonly(index_type)
326
new_name = '../packs/' + self.file_name()
327
self.upload_transport.rename(self.file_name(), new_name)
328
self._state = 'finished'
330
def _get_external_refs(self, index):
331
"""Return compression parents for this index that are not present.
333
This returns any compression parents that are referenced by this index,
334
which are not contained *in* this index. They may be present elsewhere.
336
return index.external_references(1)
209
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
210
id(self), self.pack_transport, self.name)
339
213
class NewPack(Pack):
340
214
"""An in memory proxy for a pack which is being created."""
342
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
216
# A map of index 'type' to the file extension and position in the
218
index_definitions = {
219
'revision': ('.rix', 0),
220
'inventory': ('.iix', 1),
222
'signature': ('.six', 3),
225
def __init__(self, upload_transport, index_transport, pack_transport,
226
upload_suffix='', file_mode=None, index_builder_class=None,
343
228
"""Create a NewPack instance.
345
:param pack_collection: A PackCollection into which this is being inserted.
230
:param upload_transport: A writable transport for the pack to be
231
incrementally uploaded to.
232
:param index_transport: A writable transport for the pack's indices to
233
be written to when the pack is finished.
234
:param pack_transport: A writable transport for the pack to be renamed
235
to when the upload is complete. This *must* be the same as
236
upload_transport.clone('../packs').
346
237
:param upload_suffix: An optional suffix to be given to any temporary
347
238
files created during the pack creation. e.g '.autopack'
348
:param file_mode: Unix permissions for newly created file.
239
:param file_mode: An optional file mode to create the new files with.
240
:param index_builder_class: Required keyword parameter - the class of
241
index builder to use.
242
:param index_class: Required keyword parameter - the class of index
350
245
# The relative locations of the packs are constrained, but all are
351
246
# passed in because the caller has them, so as to avoid object churn.
352
index_builder_class = pack_collection._index_builder_class
353
if pack_collection.chk_index is not None:
354
chk_index = index_builder_class(reference_lists=0)
357
247
Pack.__init__(self,
358
248
# Revisions: parents list, no text compression.
359
249
index_builder_class(reference_lists=1),
368
258
# Signatures: Just blobs to store, no compression, no parents
370
260
index_builder_class(reference_lists=0),
371
# CHK based storage - just blobs, no compression or parents.
374
self._pack_collection = pack_collection
375
262
# When we make readonly indices, we need this.
376
self.index_class = pack_collection._index_class
263
self.index_class = index_class
377
264
# where should the new pack be opened
378
self.upload_transport = pack_collection._upload_transport
265
self.upload_transport = upload_transport
379
266
# where are indices written out to
380
self.index_transport = pack_collection._index_transport
267
self.index_transport = index_transport
381
268
# where is the pack renamed to when it is finished?
382
self.pack_transport = pack_collection._pack_transport
269
self.pack_transport = pack_transport
383
270
# What file mode to upload the pack and indices with.
384
271
self._file_mode = file_mode
385
272
# tracks the content written to the .pack file.
386
273
self._hash = osutils.md5()
387
# a tuple with the length in bytes of the indices, once the pack
388
# is finalised. (rev, inv, text, sigs, chk_if_in_use)
274
# a four-tuple with the length in bytes of the indices, once the pack
275
# is finalised. (rev, inv, text, sigs)
389
276
self.index_sizes = None
390
277
# How much data to cache when writing packs. Note that this is not
391
278
# synchronised with reads, because it's not in the transport layer, so
476
352
- stores the index size tuple for the pack in the index_sizes
479
self.finish_content()
481
self._check_references()
357
self._write_data('', flush=True)
358
self.name = self._hash.hexdigest()
483
360
# XXX: It'd be better to write them all to temporary names, then
484
361
# rename them all into place, so that the window when only some are
485
362
# visible is smaller. On the other hand none will be seen until
486
363
# they're in the names list.
487
364
self.index_sizes = [None, None, None, None]
488
self._write_index('revision', self.revision_index, 'revision', suspend)
489
self._write_index('inventory', self.inventory_index, 'inventory',
491
self._write_index('text', self.text_index, 'file texts', suspend)
365
self._write_index('revision', self.revision_index, 'revision')
366
self._write_index('inventory', self.inventory_index, 'inventory')
367
self._write_index('text', self.text_index, 'file texts')
492
368
self._write_index('signature', self.signature_index,
493
'revision signatures', suspend)
494
if self.chk_index is not None:
495
self.index_sizes.append(None)
496
self._write_index('chk', self.chk_index,
497
'content hash bytes', suspend)
369
'revision signatures')
498
370
self.write_stream.close()
499
371
# Note that this will clobber an existing pack with the same name,
500
372
# without checking for hash collisions. While this is undesirable this
526
397
self._hash.update(bytes)
527
398
self._buffer[:] = [[], 0]
529
def _get_external_refs(self, index):
530
return index._external_references()
400
def index_name(self, index_type, name):
401
"""Get the disk name of an index type for pack name 'name'."""
402
return name + NewPack.index_definitions[index_type][0]
404
def index_offset(self, index_type):
405
"""Get the position in a index_size array for a given index type."""
406
return NewPack.index_definitions[index_type][1]
408
def _replace_index_with_readonly(self, index_type):
409
setattr(self, index_type + '_index',
410
self.index_class(self.index_transport,
411
self.index_name(index_type, self.name),
412
self.index_sizes[self.index_offset(index_type)]))
532
414
def set_write_cache_size(self, size):
533
415
self._cache_limit = size
535
def _write_index(self, index_type, index, label, suspend=False):
417
def _write_index(self, index_type, index, label):
536
418
"""Write out an index.
538
420
:param index_type: The type of index to write - e.g. 'revision'.
540
422
:param label: What label to give the index e.g. 'revision'.
542
424
index_name = self.index_name(index_type, self.name)
544
transport = self.upload_transport
546
transport = self.index_transport
547
self.index_sizes[self.index_offset(index_type)] = transport.put_file(
548
index_name, index.finish(), mode=self._file_mode)
425
self.index_sizes[self.index_offset(index_type)] = \
426
self.index_transport.put_file(index_name, index.finish(),
427
mode=self._file_mode)
549
428
if 'pack' in debug.debug_flags:
550
429
# XXX: size might be interesting?
551
430
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
552
431
time.ctime(), label, self.upload_transport.base,
553
432
self.random_name, time.time() - self.start_time)
554
# Replace the writable index on this object with a readonly,
433
# Replace the writable index on this object with a readonly,
555
434
# presently unloaded index. We should alter
556
435
# the index layer to make its finish() error if add_node is
557
436
# subsequently used. RBC
566
445
such as 'revision index'.
568
447
A CombinedIndex provides an index on a single key space built up
569
from several on-disk indices. The AggregateIndex builds on this
448
from several on-disk indices. The AggregateIndex builds on this
570
449
to provide a knit access layer, and allows having up to one writable
571
450
index within the collection.
573
452
# XXX: Probably 'can be written to' could/should be separated from 'acts
574
453
# like a knit index' -- mbp 20071024
576
def __init__(self, reload_func=None, flush_func=None):
577
"""Create an AggregateIndex.
579
:param reload_func: A function to call if we find we are missing an
580
index. Should have the form reload_func() => True if the list of
581
active pack files has changed.
583
self._reload_func = reload_func
456
"""Create an AggregateIndex."""
584
457
self.index_to_pack = {}
585
self.combined_index = CombinedGraphIndex([], reload_func=reload_func)
586
self.data_access = _DirectPackAccess(self.index_to_pack,
587
reload_func=reload_func,
588
flush_func=flush_func)
458
self.combined_index = CombinedGraphIndex([])
459
self.data_access = _DirectPackAccess(self.index_to_pack)
460
self.add_callback = None
462
def replace_indices(self, index_to_pack, indices):
463
"""Replace the current mappings with fresh ones.
465
This should probably not be used eventually, rather incremental add and
466
removal of indices. It has been added during refactoring of existing
469
:param index_to_pack: A mapping from index objects to
470
(transport, name) tuples for the pack file data.
471
:param indices: A list of indices.
473
# refresh the revision pack map dict without replacing the instance.
474
self.index_to_pack.clear()
475
self.index_to_pack.update(index_to_pack)
476
# XXX: API break - clearly a 'replace' method would be good?
477
self.combined_index._indices[:] = indices
478
# the current add nodes callback for the current writable index if
589
480
self.add_callback = None
591
482
def add_index(self, index, pack):
678
562
def _extra_init(self):
679
563
"""A template hook to allow extending the constructor trivially."""
681
def _pack_map_and_index_list(self, index_attribute):
682
"""Convert a list of packs to an index pack map and index list.
684
:param index_attribute: The attribute that the desired index is found
686
:return: A tuple (map, list) where map contains the dict from
687
index:pack_tuple, and list contains the indices in the preferred
692
for pack_obj in self.packs:
693
index = getattr(pack_obj, index_attribute)
694
indices.append(index)
695
pack_map[index] = pack_obj
696
return pack_map, indices
698
def _index_contents(self, indices, key_filter=None):
699
"""Get an iterable of the index contents from a pack_map.
701
:param indices: The list of indices to query
702
:param key_filter: An optional filter to limit the keys returned.
704
all_index = CombinedGraphIndex(indices)
705
if key_filter is None:
706
return all_index.iter_all_entries()
708
return all_index.iter_entries(key_filter)
710
565
def pack(self, pb=None):
711
566
"""Create a new pack by reading data from other packs.
713
568
This does little more than a bulk copy of data. One key difference
714
569
is that data with the same item key across multiple packs is elided
715
570
from the output. The new pack is written into the current pack store
716
along with its indices, and the name added to the pack names. The
571
along with its indices, and the name added to the pack names. The
717
572
source packs are not altered and are not required to be in the current
750
604
def open_pack(self):
751
605
"""Open a pack for the pack we are creating."""
752
new_pack = self._pack_collection.pack_factory(self._pack_collection,
753
upload_suffix=self.suffix,
754
file_mode=self._pack_collection.repo.bzrdir._get_file_mode())
755
# We know that we will process all nodes in order, and don't need to
756
# query, so don't combine any indices spilled to disk until we are done
757
new_pack.revision_index.set_optimize(combine_backing_indices=False)
758
new_pack.inventory_index.set_optimize(combine_backing_indices=False)
759
new_pack.text_index.set_optimize(combine_backing_indices=False)
760
new_pack.signature_index.set_optimize(combine_backing_indices=False)
763
def _update_pack_order(self, entries, index_to_pack_map):
764
"""Determine how we want our packs to be ordered.
766
This changes the sort order of the self.packs list so that packs unused
767
by 'entries' will be at the end of the list, so that future requests
768
can avoid probing them. Used packs will be at the front of the
769
self.packs list, in the order of their first use in 'entries'.
771
:param entries: A list of (index, ...) tuples
772
:param index_to_pack_map: A mapping from index objects to pack objects.
776
for entry in entries:
778
if index not in seen_indexes:
779
packs.append(index_to_pack_map[index])
780
seen_indexes.add(index)
781
if len(packs) == len(self.packs):
782
if 'pack' in debug.debug_flags:
783
mutter('Not changing pack list, all packs used.')
785
seen_packs = set(packs)
786
for pack in self.packs:
787
if pack not in seen_packs:
790
if 'pack' in debug.debug_flags:
791
old_names = [p.access_tuple()[1] for p in self.packs]
792
new_names = [p.access_tuple()[1] for p in packs]
793
mutter('Reordering packs\nfrom: %s\n to: %s',
794
old_names, new_names)
606
return NewPack(self._pack_collection._upload_transport,
607
self._pack_collection._index_transport,
608
self._pack_collection._pack_transport, upload_suffix=self.suffix,
609
file_mode=self._pack_collection.repo.bzrdir._get_file_mode(),
610
index_builder_class=self._pack_collection._index_builder_class,
611
index_class=self._pack_collection._index_class)
797
613
def _copy_revision_texts(self):
798
614
"""Copy revision data to the new pack."""
941
756
self._pack_collection.allocate(new_pack)
944
def _copy_chks(self, refs=None):
945
# XXX: Todo, recursive follow-pointers facility when fetching some
947
chk_index_map, chk_indices = self._pack_map_and_index_list(
949
chk_nodes = self._index_contents(chk_indices, refs)
951
# TODO: This isn't strictly tasteful as we are accessing some private
952
# variables (_serializer). Perhaps a better way would be to have
953
# Repository._deserialise_chk_node()
954
search_key_func = chk_map.search_key_registry.get(
955
self._pack_collection.repo._serializer.search_key_name)
956
def accumlate_refs(lines):
957
# XXX: move to a generic location
959
bytes = ''.join(lines)
960
node = chk_map._deserialise(bytes, ("unknown",), search_key_func)
961
new_refs.update(node.refs())
962
self._copy_nodes(chk_nodes, chk_index_map, self.new_pack._writer,
963
self.new_pack.chk_index, output_lines=accumlate_refs)
966
def _copy_nodes(self, nodes, index_map, writer, write_index,
968
"""Copy knit nodes between packs with no graph references.
970
:param output_lines: Output full texts of copied items.
759
def _copy_nodes(self, nodes, index_map, writer, write_index):
760
"""Copy knit nodes between packs with no graph references."""
972
761
pb = ui.ui_factory.nested_progress_bar()
974
763
return self._do_copy_nodes(nodes, index_map, writer,
975
write_index, pb, output_lines=output_lines)
979
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb,
768
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb):
981
769
# for record verification
982
770
knit = KnitVersionedFiles(None, None)
983
771
# plan a readv on each source pack:
1004
792
# linear scan up the pack
1005
793
pack_readv_requests.sort()
1007
pack_obj = index_map[index]
1008
transport, path = pack_obj.access_tuple()
1010
reader = pack.make_readv_reader(transport, path,
1011
[offset[0:2] for offset in pack_readv_requests])
1012
except errors.NoSuchFile:
1013
if self._reload_func is not None:
795
transport, path = index_map[index]
796
reader = pack.make_readv_reader(transport, path,
797
[offset[0:2] for offset in pack_readv_requests])
1016
798
for (names, read_func), (_1, _2, (key, eol_flag)) in \
1017
799
izip(reader.iter_records(), pack_readv_requests):
1018
800
raw_data = read_func(None)
1019
801
# check the header only
1020
if output_lines is not None:
1021
output_lines(knit._parse_record(key[-1], raw_data)[0])
1023
df, _ = knit._parse_record_header(key, raw_data)
802
df, _ = knit._parse_record_header(key, raw_data)
1025
804
pos, size = writer.add_bytes_record(raw_data, names)
1026
805
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
1027
806
pb.update("Copied record", record_index)
1087
860
record_index += 1
1089
862
def _get_text_nodes(self):
1090
text_index_map, text_indices = self._pack_map_and_index_list(
1092
return text_index_map, self._index_contents(text_indices,
863
text_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
864
self.packs, 'text_index')[0]
865
return text_index_map, self._pack_collection._index_contents(text_index_map,
1093
866
self._text_filter)
1095
868
def _least_readv_node_readv(self, nodes):
1096
869
"""Generate request groups for nodes using the least readv's.
1098
871
:param nodes: An iterable of graph index nodes.
1099
872
:return: Total node count and an iterator of the data needed to perform
1100
873
readvs to obtain the data for nodes. Each item yielded by the
1101
874
iterator is a tuple with:
1102
875
index, readv_vector, node_vector. readv_vector is a list ready to
1103
876
hand to the transport readv method, and node_vector is a list of
1104
(key, eol_flag, references) for the node retrieved by the
877
(key, eol_flag, references) for the the node retrieved by the
1105
878
matching readv_vector.
1107
880
# group by pack so we do one readv per pack
1383
1150
self._pack_transport = pack_transport
1384
1151
self._index_builder_class = index_builder_class
1385
1152
self._index_class = index_class
1386
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3,
1153
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
1388
1154
self.packs = []
1389
1155
# name:Pack mapping
1391
1156
self._packs_by_name = {}
1392
1157
# the previous pack-names content
1393
1158
self._packs_at_load = None
1394
1159
# when a pack is being created by this object, the state of that pack.
1395
1160
self._new_pack = None
1396
1161
# aggregated revision index data
1397
flush = self._flush_new_pack
1398
self.revision_index = AggregateIndex(self.reload_pack_names, flush)
1399
self.inventory_index = AggregateIndex(self.reload_pack_names, flush)
1400
self.text_index = AggregateIndex(self.reload_pack_names, flush)
1401
self.signature_index = AggregateIndex(self.reload_pack_names, flush)
1402
all_indices = [self.revision_index, self.inventory_index,
1403
self.text_index, self.signature_index]
1405
self.chk_index = AggregateIndex(self.reload_pack_names, flush)
1406
all_indices.append(self.chk_index)
1408
# used to determine if we're using a chk_index elsewhere.
1409
self.chk_index = None
1410
# Tell all the CombinedGraphIndex objects about each other, so they can
1411
# share hints about which pack names to search first.
1412
all_combined = [agg_idx.combined_index for agg_idx in all_indices]
1413
for combined_idx in all_combined:
1414
combined_idx.set_sibling_indices(
1415
set(all_combined).difference([combined_idx]))
1417
self._resumed_packs = []
1420
return '%s(%r)' % (self.__class__.__name__, self.repo)
1162
self.revision_index = AggregateIndex()
1163
self.inventory_index = AggregateIndex()
1164
self.text_index = AggregateIndex()
1165
self.signature_index = AggregateIndex()
1422
1167
def add_pack_to_memory(self, pack):
1423
1168
"""Make a Pack object available to the repository to satisfy queries.
1425
1170
:param pack: A Pack object.
1427
1172
if pack.name in self._packs_by_name:
1428
raise AssertionError(
1429
'pack %s already in _packs_by_name' % (pack.name,))
1173
raise AssertionError()
1430
1174
self.packs.append(pack)
1431
1175
self._packs_by_name[pack.name] = pack
1432
1176
self.revision_index.add_index(pack.revision_index, pack)
1433
1177
self.inventory_index.add_index(pack.inventory_index, pack)
1434
1178
self.text_index.add_index(pack.text_index, pack)
1435
1179
self.signature_index.add_index(pack.signature_index, pack)
1436
if self.chk_index is not None:
1437
self.chk_index.add_index(pack.chk_index, pack)
1439
1181
def all_packs(self):
1440
1182
"""Return a list of all the Pack objects this repository has.
1462
1204
in synchronisation with certain steps. Otherwise the names collection
1463
1205
is not flushed.
1465
:return: Something evaluating true if packing took place.
1207
:return: True if packing took place.
1469
return self._do_autopack()
1470
except errors.RetryAutopack:
1471
# If we get a RetryAutopack exception, we should abort the
1472
# current action, and retry.
1475
def _do_autopack(self):
1476
1209
# XXX: Should not be needed when the management of indices is sane.
1477
1210
total_revisions = self.revision_index.combined_index.key_count()
1478
1211
total_packs = len(self._names)
1479
1212
if self._max_pack_count(total_revisions) >= total_packs:
1214
# XXX: the following may want to be a class, to pack with a given
1216
mutter('Auto-packing repository %s, which has %d pack files, '
1217
'containing %d revisions into %d packs.', self, total_packs,
1218
total_revisions, self._max_pack_count(total_revisions))
1481
1219
# determine which packs need changing
1482
1220
pack_distribution = self.pack_distribution(total_revisions)
1483
1221
existing_packs = []
1491
1229
# group their data with the relevant commit, and that may
1492
1230
# involve rewriting ancient history - which autopack tries to
1493
1231
# avoid. Alternatively we could not group the data but treat
1494
# each of these as having a single revision, and thus add
1232
# each of these as having a single revision, and thus add
1495
1233
# one revision for each to the total revision count, to get
1496
1234
# a matching distribution.
1498
1236
existing_packs.append((revision_count, pack))
1499
1237
pack_operations = self.plan_autopack_combinations(
1500
1238
existing_packs, pack_distribution)
1501
num_new_packs = len(pack_operations)
1502
num_old_packs = sum([len(po[1]) for po in pack_operations])
1503
num_revs_affected = sum([po[0] for po in pack_operations])
1504
mutter('Auto-packing repository %s, which has %d pack files, '
1505
'containing %d revisions. Packing %d files into %d affecting %d'
1506
' revisions', self, total_packs, total_revisions, num_old_packs,
1507
num_new_packs, num_revs_affected)
1508
result = self._execute_pack_operations(pack_operations,
1509
reload_func=self._restart_autopack)
1510
mutter('Auto-packing repository %s completed', self)
1239
self._execute_pack_operations(pack_operations)
1513
def _execute_pack_operations(self, pack_operations, _packer_class=Packer,
1242
def _execute_pack_operations(self, pack_operations, _packer_class=Packer):
1515
1243
"""Execute a series of pack operations.
1517
1245
:param pack_operations: A list of [revision_count, packs_to_combine].
1518
1246
:param _packer_class: The class of packer to use (default: Packer).
1519
:return: The new pack names.
1521
1249
for revision_count, packs in pack_operations:
1522
1250
# we may have no-ops from the setup logic
1523
1251
if len(packs) == 0:
1525
packer = _packer_class(self, packs, '.autopack',
1526
reload_func=reload_func)
1529
except errors.RetryWithNewPacks:
1530
# An exception is propagating out of this context, make sure
1531
# this packer has cleaned up. Packer() doesn't set its new_pack
1532
# state into the RepositoryPackCollection object, so we only
1533
# have access to it directly here.
1534
if packer.new_pack is not None:
1535
packer.new_pack.abort()
1253
_packer_class(self, packs, '.autopack').pack()
1537
1254
for pack in packs:
1538
1255
self._remove_pack_from_memory(pack)
1539
1256
# record the newly available packs and stop advertising the old
1541
to_be_obsoleted = []
1542
for _, packs in pack_operations:
1543
to_be_obsoleted.extend(packs)
1544
result = self._save_pack_names(clear_obsolete_packs=True,
1545
obsolete_packs=to_be_obsoleted)
1548
def _flush_new_pack(self):
1549
if self._new_pack is not None:
1550
self._new_pack.flush()
1258
self._save_pack_names(clear_obsolete_packs=True)
1259
# Move the old packs out of the way now they are no longer referenced.
1260
for revision_count, packs in pack_operations:
1261
self._obsolete_packs(packs)
1552
1263
def lock_names(self):
1553
1264
"""Acquire the mutex around the pack-names index.
1555
1266
This cannot be used in the middle of a read-only transaction on the
1558
1269
self.repo.control_files.lock_write()
1560
def _already_packed(self):
1561
"""Is the collection already packed?"""
1562
return not (self.repo._format.pack_compresses or (len(self._names) > 1))
1564
def pack(self, hint=None, clean_obsolete_packs=False):
1565
1272
"""Pack the pack collection totally."""
1566
1273
self.ensure_loaded()
1567
1274
total_packs = len(self._names)
1568
if self._already_packed():
1276
# This is arguably wrong because we might not be optimal, but for
1277
# now lets leave it in. (e.g. reconcile -> one pack. But not
1570
1280
total_revisions = self.revision_index.combined_index.key_count()
1571
1281
# XXX: the following may want to be a class, to pack with a given
1573
1283
mutter('Packing repository %s, which has %d pack files, '
1574
'containing %d revisions with hint %r.', self, total_packs,
1575
total_revisions, hint)
1284
'containing %d revisions into 1 packs.', self, total_packs,
1576
1286
# determine which packs need changing
1287
pack_distribution = [1]
1577
1288
pack_operations = [[0, []]]
1578
1289
for pack in self.all_packs():
1579
if hint is None or pack.name in hint:
1580
# Either no hint was provided (so we are packing everything),
1581
# or this pack was included in the hint.
1582
pack_operations[-1][0] += pack.get_revision_count()
1583
pack_operations[-1][1].append(pack)
1290
pack_operations[-1][0] += pack.get_revision_count()
1291
pack_operations[-1][1].append(pack)
1584
1292
self._execute_pack_operations(pack_operations, OptimisingPacker)
1586
if clean_obsolete_packs:
1587
self._clear_obsolete_packs()
1589
1294
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1590
1295
"""Plan a pack operation.
1678
1375
inv_index = self._make_index(name, '.iix')
1679
1376
txt_index = self._make_index(name, '.tix')
1680
1377
sig_index = self._make_index(name, '.six')
1681
if self.chk_index is not None:
1682
chk_index = self._make_index(name, '.cix', unlimited_cache=True)
1685
1378
result = ExistingPack(self._pack_transport, name, rev_index,
1686
inv_index, txt_index, sig_index, chk_index)
1379
inv_index, txt_index, sig_index)
1687
1380
self.add_pack_to_memory(result)
1690
def _resume_pack(self, name):
1691
"""Get a suspended Pack object by name.
1693
:param name: The name of the pack - e.g. '123456'
1694
:return: A Pack object.
1696
if not re.match('[a-f0-9]{32}', name):
1697
# Tokens should be md5sums of the suspended pack file, i.e. 32 hex
1699
raise errors.UnresumableWriteGroup(
1700
self.repo, [name], 'Malformed write group token')
1702
rev_index = self._make_index(name, '.rix', resume=True)
1703
inv_index = self._make_index(name, '.iix', resume=True)
1704
txt_index = self._make_index(name, '.tix', resume=True)
1705
sig_index = self._make_index(name, '.six', resume=True)
1706
if self.chk_index is not None:
1707
chk_index = self._make_index(name, '.cix', resume=True,
1708
unlimited_cache=True)
1711
result = self.resumed_pack_factory(name, rev_index, inv_index,
1712
txt_index, sig_index, self._upload_transport,
1713
self._pack_transport, self._index_transport, self,
1714
chk_index=chk_index)
1715
except errors.NoSuchFile, e:
1716
raise errors.UnresumableWriteGroup(self.repo, [name], str(e))
1717
self.add_pack_to_memory(result)
1718
self._resumed_packs.append(result)
1721
1383
def allocate(self, a_new_pack):
1722
1384
"""Allocate name in the list of packs.
1741
1403
return self._index_class(self.transport, 'pack-names', None
1742
1404
).iter_all_entries()
1744
def _make_index(self, name, suffix, resume=False, unlimited_cache=False):
1406
def _make_index(self, name, suffix):
1745
1407
size_offset = self._suffix_offsets[suffix]
1746
1408
index_name = name + suffix
1748
transport = self._upload_transport
1749
index_size = transport.stat(index_name).st_size
1751
transport = self._index_transport
1752
index_size = self._names[name][size_offset]
1753
return self._index_class(transport, index_name, index_size,
1754
unlimited_cache=unlimited_cache)
1409
index_size = self._names[name][size_offset]
1410
return self._index_class(
1411
self._index_transport, index_name, index_size)
1756
1413
def _max_pack_count(self, total_revisions):
1757
1414
"""Return the maximum number of packs to use for total revisions.
1759
1416
:param total_revisions: The total number of revisions in the
1785
1442
:param return: None.
1787
1444
for pack in packs:
1789
pack.pack_transport.rename(pack.file_name(),
1790
'../obsolete_packs/' + pack.file_name())
1791
except (errors.PathError, errors.TransportError), e:
1792
# TODO: Should these be warnings or mutters?
1793
mutter("couldn't rename obsolete pack, skipping it:\n%s"
1445
pack.pack_transport.rename(pack.file_name(),
1446
'../obsolete_packs/' + pack.file_name())
1795
1447
# TODO: Probably needs to know all possible indices for this pack
1796
1448
# - or maybe list the directory and move all indices matching this
1797
1449
# name whether we recognize it or not?
1798
suffixes = ['.iix', '.six', '.tix', '.rix']
1799
if self.chk_index is not None:
1800
suffixes.append('.cix')
1801
for suffix in suffixes:
1803
self._index_transport.rename(pack.name + suffix,
1804
'../obsolete_packs/' + pack.name + suffix)
1805
except (errors.PathError, errors.TransportError), e:
1806
mutter("couldn't rename obsolete index, skipping it:\n%s"
1450
for suffix in ('.iix', '.six', '.tix', '.rix'):
1451
self._index_transport.rename(pack.name + suffix,
1452
'../obsolete_packs/' + pack.name + suffix)
1809
1454
def pack_distribution(self, total_revisions):
1810
1455
"""Generate a list of the number of revisions to put in each pack.
1836
1481
self._remove_pack_indices(pack)
1837
1482
self.packs.remove(pack)
1839
def _remove_pack_indices(self, pack, ignore_missing=False):
1840
"""Remove the indices for pack from the aggregated indices.
1842
:param ignore_missing: Suppress KeyErrors from calling remove_index.
1844
for index_type in Pack.index_definitions.keys():
1845
attr_name = index_type + '_index'
1846
aggregate_index = getattr(self, attr_name)
1847
if aggregate_index is not None:
1848
pack_index = getattr(pack, attr_name)
1850
aggregate_index.remove_index(pack_index)
1484
def _remove_pack_indices(self, pack):
1485
"""Remove the indices for pack from the aggregated indices."""
1486
self.revision_index.remove_index(pack.revision_index, pack)
1487
self.inventory_index.remove_index(pack.inventory_index, pack)
1488
self.text_index.remove_index(pack.text_index, pack)
1489
self.signature_index.remove_index(pack.signature_index, pack)
1856
1491
def reset(self):
1857
1492
"""Clear all cached data."""
1858
1493
# cached revision data
1494
self.repo._revision_knit = None
1859
1495
self.revision_index.clear()
1860
1496
# cached signature data
1497
self.repo._signature_knit = None
1861
1498
self.signature_index.clear()
1862
1499
# cached file text data
1863
1500
self.text_index.clear()
1501
self.repo._text_knit = None
1864
1502
# cached inventory data
1865
1503
self.inventory_index.clear()
1867
if self.chk_index is not None:
1868
self.chk_index.clear()
1869
1504
# remove the open pack
1870
1505
self._new_pack = None
1871
1506
# information about packs.
1874
1509
self._packs_by_name = {}
1875
1510
self._packs_at_load = None
1512
def _make_index_map(self, index_suffix):
1513
"""Return information on existing indices.
1515
:param suffix: Index suffix added to pack name.
1517
:returns: (pack_map, indices) where indices is a list of GraphIndex
1518
objects, and pack_map is a mapping from those objects to the
1519
pack tuple they describe.
1521
# TODO: stop using this; it creates new indices unnecessarily.
1522
self.ensure_loaded()
1523
suffix_map = {'.rix': 'revision_index',
1524
'.six': 'signature_index',
1525
'.iix': 'inventory_index',
1526
'.tix': 'text_index',
1528
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1529
suffix_map[index_suffix])
1531
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1532
"""Convert a list of packs to an index pack map and index list.
1534
:param packs: The packs list to process.
1535
:param index_attribute: The attribute that the desired index is found
1537
:return: A tuple (map, list) where map contains the dict from
1538
index:pack_tuple, and lsit contains the indices in the same order
1544
index = getattr(pack, index_attribute)
1545
indices.append(index)
1546
pack_map[index] = (pack.pack_transport, pack.file_name())
1547
return pack_map, indices
1549
def _index_contents(self, pack_map, key_filter=None):
1550
"""Get an iterable of the index contents from a pack_map.
1552
:param pack_map: A map from indices to pack details.
1553
:param key_filter: An optional filter to limit the
1556
indices = [index for index in pack_map.iterkeys()]
1557
all_index = CombinedGraphIndex(indices)
1558
if key_filter is None:
1559
return all_index.iter_all_entries()
1561
return all_index.iter_entries(key_filter)
1877
1563
def _unlock_names(self):
1878
1564
"""Release the mutex around the pack-names index."""
1879
1565
self.repo.control_files.unlock()
1881
def _diff_pack_names(self):
1882
"""Read the pack names from disk, and compare it to the one in memory.
1884
:return: (disk_nodes, deleted_nodes, new_nodes)
1885
disk_nodes The final set of nodes that should be referenced
1886
deleted_nodes Nodes which have been removed from when we started
1887
new_nodes Nodes that are newly introduced
1889
# load the disk nodes across
1891
for index, key, value in self._iter_disk_pack_index():
1892
disk_nodes.add((key, value))
1893
orig_disk_nodes = set(disk_nodes)
1895
# do a two-way diff against our original content
1896
current_nodes = set()
1897
for name, sizes in self._names.iteritems():
1899
((name, ), ' '.join(str(size) for size in sizes)))
1901
# Packs no longer present in the repository, which were present when we
1902
# locked the repository
1903
deleted_nodes = self._packs_at_load - current_nodes
1904
# Packs which this process is adding
1905
new_nodes = current_nodes - self._packs_at_load
1907
# Update the disk_nodes set to include the ones we are adding, and
1908
# remove the ones which were removed by someone else
1909
disk_nodes.difference_update(deleted_nodes)
1910
disk_nodes.update(new_nodes)
1912
return disk_nodes, deleted_nodes, new_nodes, orig_disk_nodes
1914
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1915
"""Given the correct set of pack files, update our saved info.
1917
:return: (removed, added, modified)
1918
removed pack names removed from self._names
1919
added pack names added to self._names
1920
modified pack names that had changed value
1925
## self._packs_at_load = disk_nodes
1567
def _save_pack_names(self, clear_obsolete_packs=False):
1568
"""Save the list of packs.
1570
This will take out the mutex around the pack names list for the
1571
duration of the method call. If concurrent updates have been made, a
1572
three-way merge between the current list and the current in memory list
1575
:param clear_obsolete_packs: If True, clear out the contents of the
1576
obsolete_packs directory.
1580
builder = self._index_builder_class()
1581
# load the disk nodes across
1583
for index, key, value in self._iter_disk_pack_index():
1584
disk_nodes.add((key, value))
1585
# do a two-way diff against our original content
1586
current_nodes = set()
1587
for name, sizes in self._names.iteritems():
1589
((name, ), ' '.join(str(size) for size in sizes)))
1590
deleted_nodes = self._packs_at_load - current_nodes
1591
new_nodes = current_nodes - self._packs_at_load
1592
disk_nodes.difference_update(deleted_nodes)
1593
disk_nodes.update(new_nodes)
1594
# TODO: handle same-name, index-size-changes here -
1595
# e.g. use the value from disk, not ours, *unless* we're the one
1597
for key, value in disk_nodes:
1598
builder.add_node(key, value)
1599
self.transport.put_file('pack-names', builder.finish(),
1600
mode=self.repo.bzrdir._get_file_mode())
1601
# move the baseline forward
1602
self._packs_at_load = disk_nodes
1603
if clear_obsolete_packs:
1604
self._clear_obsolete_packs()
1606
self._unlock_names()
1607
# synchronise the memory packs list with what we just wrote:
1926
1608
new_names = dict(disk_nodes)
1927
1609
# drop no longer present nodes
1928
1610
for pack in self.all_packs():
1929
1611
if (pack.name,) not in new_names:
1930
removed.append(pack.name)
1931
1612
self._remove_pack_from_memory(pack)
1932
1613
# add new nodes/refresh existing ones
1933
1614
for key, value in disk_nodes:
1947
1628
self._remove_pack_from_memory(self.get_pack_by_name(name))
1948
1629
self._names[name] = sizes
1949
1630
self.get_pack_by_name(name)
1950
modified.append(name)
1953
1633
self._names[name] = sizes
1954
1634
self.get_pack_by_name(name)
1956
return removed, added, modified
1958
def _save_pack_names(self, clear_obsolete_packs=False, obsolete_packs=None):
1959
"""Save the list of packs.
1961
This will take out the mutex around the pack names list for the
1962
duration of the method call. If concurrent updates have been made, a
1963
three-way merge between the current list and the current in memory list
1966
:param clear_obsolete_packs: If True, clear out the contents of the
1967
obsolete_packs directory.
1968
:param obsolete_packs: Packs that are obsolete once the new pack-names
1969
file has been written.
1970
:return: A list of the names saved that were not previously on disk.
1972
already_obsolete = []
1975
builder = self._index_builder_class()
1976
(disk_nodes, deleted_nodes, new_nodes,
1977
orig_disk_nodes) = self._diff_pack_names()
1978
# TODO: handle same-name, index-size-changes here -
1979
# e.g. use the value from disk, not ours, *unless* we're the one
1981
for key, value in disk_nodes:
1982
builder.add_node(key, value)
1983
self.transport.put_file('pack-names', builder.finish(),
1984
mode=self.repo.bzrdir._get_file_mode())
1985
self._packs_at_load = disk_nodes
1986
if clear_obsolete_packs:
1989
to_preserve = set([o.name for o in obsolete_packs])
1990
already_obsolete = self._clear_obsolete_packs(to_preserve)
1992
self._unlock_names()
1993
# synchronise the memory packs list with what we just wrote:
1994
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1996
# TODO: We could add one more condition here. "if o.name not in
1997
# orig_disk_nodes and o != the new_pack we haven't written to
1998
# disk yet. However, the new pack object is not easily
1999
# accessible here (it would have to be passed through the
2000
# autopacking code, etc.)
2001
obsolete_packs = [o for o in obsolete_packs
2002
if o.name not in already_obsolete]
2003
self._obsolete_packs(obsolete_packs)
2004
return [new_node[0][0] for new_node in new_nodes]
2006
def reload_pack_names(self):
2007
"""Sync our pack listing with what is present in the repository.
2009
This should be called when we find out that something we thought was
2010
present is now missing. This happens when another process re-packs the
2013
:return: True if the in-memory list of packs has been altered at all.
2015
# The ensure_loaded call is to handle the case where the first call
2016
# made involving the collection was to reload_pack_names, where we
2017
# don't have a view of disk contents. Its a bit of a bandaid, and
2018
# causes two reads of pack-names, but its a rare corner case not struck
2019
# with regular push/pull etc.
2020
first_read = self.ensure_loaded()
2023
# out the new value.
2024
(disk_nodes, deleted_nodes, new_nodes,
2025
orig_disk_nodes) = self._diff_pack_names()
2026
# _packs_at_load is meant to be the explicit list of names in
2027
# 'pack-names' at then start. As such, it should not contain any
2028
# pending names that haven't been written out yet.
2029
self._packs_at_load = orig_disk_nodes
2031
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
2032
if removed or added or modified:
2036
def _restart_autopack(self):
2037
"""Reload the pack names list, and restart the autopack code."""
2038
if not self.reload_pack_names():
2039
# Re-raise the original exception, because something went missing
2040
# and a restart didn't find it
2042
raise errors.RetryAutopack(self.repo, False, sys.exc_info())
2044
def _clear_obsolete_packs(self, preserve=None):
1636
def _clear_obsolete_packs(self):
2045
1637
"""Delete everything from the obsolete-packs directory.
2047
:return: A list of pack identifiers (the filename without '.pack') that
2048
were found in obsolete_packs.
2051
1639
obsolete_pack_transport = self.transport.clone('obsolete_packs')
2052
if preserve is None:
2054
1640
for filename in obsolete_pack_transport.list_dir('.'):
2055
name, ext = osutils.splitext(filename)
2058
if name in preserve:
2061
1642
obsolete_pack_transport.delete(filename)
2062
1643
except (errors.PathError, errors.TransportError), e:
2063
warning("couldn't delete obsolete pack, skipping it:\n%s"
1644
warning("couldn't delete obsolete pack, skipping it:\n%s" % (e,))
2067
1646
def _start_write_group(self):
2068
1647
# Do not permit preparation for writing if we're not in a 'write lock'.
2069
1648
if not self.repo.is_write_locked():
2070
1649
raise errors.NotWriteLocked(self)
2071
self._new_pack = self.pack_factory(self, upload_suffix='.pack',
2072
file_mode=self.repo.bzrdir._get_file_mode())
1650
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1651
self._pack_transport, upload_suffix='.pack',
1652
file_mode=self.repo.bzrdir._get_file_mode(),
1653
index_builder_class=self._index_builder_class,
1654
index_class=self._index_class)
2073
1655
# allow writing: queue writes to a new index
2074
1656
self.revision_index.add_writable_index(self._new_pack.revision_index,
2075
1657
self._new_pack)
2095
1671
# FIXME: just drop the transient index.
2096
1672
# forget what names there are
2097
1673
if self._new_pack is not None:
2098
operation = cleanup.OperationWithCleanups(self._new_pack.abort)
2099
operation.add_cleanup(setattr, self, '_new_pack', None)
2100
# If we aborted while in the middle of finishing the write
2101
# group, _remove_pack_indices could fail because the indexes are
2102
# already gone. But they're not there we shouldn't fail in this
2103
# case, so we pass ignore_missing=True.
2104
operation.add_cleanup(self._remove_pack_indices, self._new_pack,
2105
ignore_missing=True)
2106
operation.run_simple()
2107
for resumed_pack in self._resumed_packs:
2108
operation = cleanup.OperationWithCleanups(resumed_pack.abort)
2109
# See comment in previous finally block.
2110
operation.add_cleanup(self._remove_pack_indices, resumed_pack,
2111
ignore_missing=True)
2112
operation.run_simple()
2113
del self._resumed_packs[:]
2115
def _remove_resumed_pack_indices(self):
2116
for resumed_pack in self._resumed_packs:
2117
self._remove_pack_indices(resumed_pack)
2118
del self._resumed_packs[:]
2120
def _check_new_inventories(self):
2121
"""Detect missing inventories in this write group.
2123
:returns: list of strs, summarising any problems found. If the list is
2124
empty no problems were found.
2126
# The base implementation does no checks. GCRepositoryPackCollection
1674
self._new_pack.abort()
1675
self._remove_pack_indices(self._new_pack)
1676
self._new_pack = None
1677
self.repo._text_knit = None
2130
1679
def _commit_write_group(self):
2132
for prefix, versioned_file in (
2133
('revisions', self.repo.revisions),
2134
('inventories', self.repo.inventories),
2135
('texts', self.repo.texts),
2136
('signatures', self.repo.signatures),
2138
missing = versioned_file.get_missing_compression_parent_keys()
2139
all_missing.update([(prefix,) + key for key in missing])
2141
raise errors.BzrCheckError(
2142
"Repository %s has missing compression parent(s) %r "
2143
% (self.repo, sorted(all_missing)))
2144
problems = self._check_new_inventories()
2146
problems_summary = '\n'.join(problems)
2147
raise errors.BzrCheckError(
2148
"Cannot add revision(s) to repository: " + problems_summary)
2149
1680
self._remove_pack_indices(self._new_pack)
2150
any_new_content = False
2151
1681
if self._new_pack.data_inserted():
2152
1682
# get all the data to disk and read to use
2153
1683
self._new_pack.finish()
2154
1684
self.allocate(self._new_pack)
2155
1685
self._new_pack = None
2156
any_new_content = True
2158
self._new_pack.abort()
2159
self._new_pack = None
2160
for resumed_pack in self._resumed_packs:
2161
# XXX: this is a pretty ugly way to turn the resumed pack into a
2162
# properly committed pack.
2163
self._names[resumed_pack.name] = None
2164
self._remove_pack_from_memory(resumed_pack)
2165
resumed_pack.finish()
2166
self.allocate(resumed_pack)
2167
any_new_content = True
2168
del self._resumed_packs[:]
2170
result = self.autopack()
1686
if not self.autopack():
2172
1687
# when autopack takes no steps, the names list is still
2174
return self._save_pack_names()
2178
def _suspend_write_group(self):
2179
tokens = [pack.name for pack in self._resumed_packs]
2180
self._remove_pack_indices(self._new_pack)
2181
if self._new_pack.data_inserted():
2182
# get all the data to disk and read to use
2183
self._new_pack.finish(suspend=True)
2184
tokens.append(self._new_pack.name)
2185
self._new_pack = None
1689
self._save_pack_names()
2187
1691
self._new_pack.abort()
2188
1692
self._new_pack = None
2189
self._remove_resumed_pack_indices()
2192
def _resume_write_group(self, tokens):
2193
for token in tokens:
2194
self._resume_pack(token)
1693
self.repo._text_knit = None
2197
1696
class KnitPackRepository(KnitRepository):
2198
1697
"""Repository with knit objects stored inside pack containers.
2200
1699
The layering for a KnitPackRepository is:
2202
1701
Graph | HPSS | Repository public layer |
2257
1753
deltas=True, parents=True, is_locked=self.is_locked),
2258
1754
data_access=self._pack_collection.text_index.data_access,
2259
1755
max_delta_chain=200)
2260
if _format.supports_chks:
2261
# No graph, no compression:- references from chks are between
2262
# different objects not temporal versions of the same; and without
2263
# some sort of temporal structure knit compression will just fail.
2264
self.chk_bytes = KnitVersionedFiles(
2265
_KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2266
add_callback=self._pack_collection.chk_index.add_callback,
2267
deltas=False, parents=False, is_locked=self.is_locked),
2268
data_access=self._pack_collection.chk_index.data_access,
2271
self.chk_bytes = None
2272
1756
# True when the repository object is 'write locked' (as opposed to the
2273
# physical lock only taken out around changes to the pack-names list.)
1757
# physical lock only taken out around changes to the pack-names list.)
2274
1758
# Another way to represent this would be a decorator around the control
2275
1759
# files object that presents logical locks as physical ones - if this
2276
1760
# gets ugly consider that alternative design. RBC 20071011
2280
1764
self._reconcile_does_inventory_gc = True
2281
1765
self._reconcile_fixes_text_parents = True
2282
1766
self._reconcile_backsup_inventory = False
1767
self._fetch_order = 'unordered'
2284
def _warn_if_deprecated(self, branch=None):
1769
def _warn_if_deprecated(self):
2285
1770
# This class isn't deprecated, but one sub-format is
2286
1771
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2287
super(KnitPackRepository, self)._warn_if_deprecated(branch)
1772
from bzrlib import repository
1773
if repository._deprecation_warning_done:
1775
repository._deprecation_warning_done = True
1776
warning("Format %s for %s is deprecated - please use"
1777
" 'bzr upgrade --1.6.1-rich-root'"
1778
% (self._format, self.bzrdir.transport.base))
2289
1780
def _abort_write_group(self):
2290
self.revisions._index._key_dependencies.clear()
2291
1781
self._pack_collection._abort_write_group()
2293
def _get_source(self, to_format):
2294
if to_format.network_name() == self._format.network_name():
2295
return KnitPackStreamSource(self, to_format)
2296
return super(KnitPackRepository, self)._get_source(to_format)
1783
def _find_inconsistent_revision_parents(self):
1784
"""Find revisions with incorrectly cached parents.
1786
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1787
parents-in-revision).
1789
if not self.is_locked():
1790
raise errors.ObjectNotLocked(self)
1791
pb = ui.ui_factory.nested_progress_bar()
1794
revision_nodes = self._pack_collection.revision_index \
1795
.combined_index.iter_all_entries()
1796
index_positions = []
1797
# Get the cached index values for all revisions, and also the location
1798
# in each index of the revision text so we can perform linear IO.
1799
for index, key, value, refs in revision_nodes:
1800
pos, length = value[1:].split(' ')
1801
index_positions.append((index, int(pos), key[0],
1802
tuple(parent[0] for parent in refs[0])))
1803
pb.update("Reading revision index.", 0, 0)
1804
index_positions.sort()
1805
batch_count = len(index_positions) / 1000 + 1
1806
pb.update("Checking cached revision graph.", 0, batch_count)
1807
for offset in xrange(batch_count):
1808
pb.update("Checking cached revision graph.", offset)
1809
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1812
rev_ids = [item[2] for item in to_query]
1813
revs = self.get_revisions(rev_ids)
1814
for revision, item in zip(revs, to_query):
1815
index_parents = item[3]
1816
rev_parents = tuple(revision.parent_ids)
1817
if index_parents != rev_parents:
1818
result.append((revision.revision_id, index_parents, rev_parents))
1823
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
1824
def get_parents(self, revision_ids):
1825
"""See graph._StackedParentsProvider.get_parents."""
1826
parent_map = self.get_parent_map(revision_ids)
1827
return [parent_map.get(r, None) for r in revision_ids]
2298
1829
def _make_parents_provider(self):
2299
1830
return graph.CachingParentsProvider(self)
2301
1832
def _refresh_data(self):
2302
if not self.is_locked():
2304
self._pack_collection.reload_pack_names()
1833
if self._write_lock_count == 1 or (
1834
self.control_files._lock_count == 1 and
1835
self.control_files._lock_mode == 'r'):
1836
# forget what names there are
1837
self._pack_collection.reset()
1838
# XXX: Better to do an in-memory merge when acquiring a new lock -
1839
# factor out code from _save_pack_names.
1840
self._pack_collection.ensure_loaded()
2306
1842
def _start_write_group(self):
2307
1843
self._pack_collection._start_write_group()
2309
1845
def _commit_write_group(self):
2310
hint = self._pack_collection._commit_write_group()
2311
self.revisions._index._key_dependencies.clear()
2314
def suspend_write_group(self):
2315
# XXX check self._write_group is self.get_transaction()?
2316
tokens = self._pack_collection._suspend_write_group()
2317
self.revisions._index._key_dependencies.clear()
2318
self._write_group = None
2321
def _resume_write_group(self, tokens):
2322
self._start_write_group()
2324
self._pack_collection._resume_write_group(tokens)
2325
except errors.UnresumableWriteGroup:
2326
self._abort_write_group()
2328
for pack in self._pack_collection._resumed_packs:
2329
self.revisions._index.scan_unvalidated_index(pack.revision_index)
1846
return self._pack_collection._commit_write_group()
2331
1848
def get_transaction(self):
2332
1849
if self._write_lock_count:
2341
1858
return self._write_lock_count
2343
1860
def lock_write(self, token=None):
2344
locked = self.is_locked()
2345
if not self._write_lock_count and locked:
1861
if not self._write_lock_count and self.is_locked():
2346
1862
raise errors.ReadOnlyError(self)
2347
1863
self._write_lock_count += 1
2348
1864
if self._write_lock_count == 1:
2349
1865
self._transaction = transactions.WriteTransaction()
2351
if 'relock' in debug.debug_flags and self._prev_lock == 'w':
2352
note('%r was write locked again', self)
2353
self._prev_lock = 'w'
2354
1866
for repo in self._fallback_repositories:
2355
1867
# Writes don't affect fallback repos
2356
1868
repo.lock_read()
2357
self._refresh_data()
2358
return RepositoryWriteLockResult(self.unlock, None)
1869
self._refresh_data()
2360
1871
def lock_read(self):
2361
locked = self.is_locked()
2362
1872
if self._write_lock_count:
2363
1873
self._write_lock_count += 1
2365
1875
self.control_files.lock_read()
2367
if 'relock' in debug.debug_flags and self._prev_lock == 'r':
2368
note('%r was read locked again', self)
2369
self._prev_lock = 'r'
2370
1876
for repo in self._fallback_repositories:
1877
# Writes don't affect fallback repos
2371
1878
repo.lock_read()
2372
self._refresh_data()
1879
self._refresh_data()
2375
1881
def leave_lock_in_place(self):
2376
1882
# not supported - raise an error
2416
1917
transaction = self._transaction
2417
1918
self._transaction = None
2418
1919
transaction.finish()
1920
for repo in self._fallback_repositories:
2420
1923
self.control_files.unlock()
2422
if not self.is_locked():
2423
1924
for repo in self._fallback_repositories:
2427
class KnitPackStreamSource(StreamSource):
2428
"""A StreamSource used to transfer data between same-format KnitPack repos.
2430
This source assumes:
2431
1) Same serialization format for all objects
2432
2) Same root information
2433
3) XML format inventories
2434
4) Atomic inserts (so we can stream inventory texts before text
2439
def __init__(self, from_repository, to_format):
2440
super(KnitPackStreamSource, self).__init__(from_repository, to_format)
2441
self._text_keys = None
2442
self._text_fetch_order = 'unordered'
2444
def _get_filtered_inv_stream(self, revision_ids):
2445
from_repo = self.from_repository
2446
parent_ids = from_repo._find_parent_ids_of_revisions(revision_ids)
2447
parent_keys = [(p,) for p in parent_ids]
2448
find_text_keys = from_repo._find_text_key_references_from_xml_inventory_lines
2449
parent_text_keys = set(find_text_keys(
2450
from_repo._inventory_xml_lines_for_keys(parent_keys)))
2451
content_text_keys = set()
2452
knit = KnitVersionedFiles(None, None)
2453
factory = KnitPlainFactory()
2454
def find_text_keys_from_content(record):
2455
if record.storage_kind not in ('knit-delta-gz', 'knit-ft-gz'):
2456
raise ValueError("Unknown content storage kind for"
2457
" inventory text: %s" % (record.storage_kind,))
2458
# It's a knit record, it has a _raw_record field (even if it was
2459
# reconstituted from a network stream).
2460
raw_data = record._raw_record
2461
# read the entire thing
2462
revision_id = record.key[-1]
2463
content, _ = knit._parse_record(revision_id, raw_data)
2464
if record.storage_kind == 'knit-delta-gz':
2465
line_iterator = factory.get_linedelta_content(content)
2466
elif record.storage_kind == 'knit-ft-gz':
2467
line_iterator = factory.get_fulltext_content(content)
2468
content_text_keys.update(find_text_keys(
2469
[(line, revision_id) for line in line_iterator]))
2470
revision_keys = [(r,) for r in revision_ids]
2471
def _filtered_inv_stream():
2472
source_vf = from_repo.inventories
2473
stream = source_vf.get_record_stream(revision_keys,
2475
for record in stream:
2476
if record.storage_kind == 'absent':
2477
raise errors.NoSuchRevision(from_repo, record.key)
2478
find_text_keys_from_content(record)
2480
self._text_keys = content_text_keys - parent_text_keys
2481
return ('inventories', _filtered_inv_stream())
2483
def _get_text_stream(self):
2484
# Note: We know we don't have to handle adding root keys, because both
2485
# the source and target are the identical network name.
2486
text_stream = self.from_repository.texts.get_record_stream(
2487
self._text_keys, self._text_fetch_order, False)
2488
return ('texts', text_stream)
2490
def get_stream(self, search):
2491
revision_ids = search.get_keys()
2492
for stream_info in self._fetch_revision_texts(revision_ids):
2494
self._revision_keys = [(rev_id,) for rev_id in revision_ids]
2495
yield self._get_filtered_inv_stream(revision_ids)
2496
yield self._get_text_stream()
2500
1928
class RepositoryFormatPack(MetaDirRepositoryFormat):
2501
1929
"""Format logic for pack structured repositories.
2838
2285
def get_format_string(self):
2839
2286
"""See RepositoryFormat.get_format_string()."""
2840
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2287
return "Bazaar development format 2 (needs bzr.dev from before 1.8)\n"
2842
2289
def get_format_description(self):
2843
2290
"""See RepositoryFormat.get_format_description()."""
2844
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2847
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2848
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2850
1.6-rich-root with B+Tree indices.
2853
repository_class = KnitPackRepository
2854
_commit_builder_class = PackRootCommitBuilder
2855
rich_root_data = True
2856
supports_tree_reference = False # no subtrees
2857
supports_external_lookups = True
2858
# What index classes to use
2859
index_builder_class = BTreeBuilder
2860
index_class = BTreeGraphIndex
2863
def _serializer(self):
2864
return xml6.serializer_v6
2866
def _get_matching_bzrdir(self):
2867
return bzrdir.format_registry.make_bzrdir(
2870
def _ignore_setting_bzrdir(self, format):
2291
return ("Development repository format, currently the same as "
2292
"1.6.1 with B+Trees.\n")
2294
def check_conversion_target(self, target_format):
2873
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2875
def get_format_string(self):
2876
"""See RepositoryFormat.get_format_string()."""
2877
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2879
def get_format_description(self):
2880
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2883
2298
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2884
2299
"""A subtrees development repository.
added
removed
bzrlib/repofmt/pack_repo.py
