263
170
return not self.__eq__(other)
265
172
def __repr__(self):
266
return "<%s.%s object at 0x%x, %s, %s" % (
267
self.__class__.__module__, self.__class__.__name__, id(self),
268
self.pack_transport, self.name)
271
class ResumedPack(ExistingPack):
273
def __init__(self, name, revision_index, inventory_index, text_index,
274
signature_index, upload_transport, pack_transport, index_transport,
275
pack_collection, chk_index=None):
276
"""Create a ResumedPack object."""
277
ExistingPack.__init__(self, pack_transport, name, revision_index,
278
inventory_index, text_index, signature_index,
280
self.upload_transport = upload_transport
281
self.index_transport = index_transport
282
self.index_sizes = [None, None, None, None]
284
('revision', revision_index),
285
('inventory', inventory_index),
286
('text', text_index),
287
('signature', signature_index),
289
if chk_index is not None:
290
indices.append(('chk', chk_index))
291
self.index_sizes.append(None)
292
for index_type, index in indices:
293
offset = self.index_offset(index_type)
294
self.index_sizes[offset] = index._size
295
self.index_class = pack_collection._index_class
296
self._pack_collection = pack_collection
297
self._state = 'resumed'
298
# XXX: perhaps check that the .pack file exists?
300
def access_tuple(self):
301
if self._state == 'finished':
302
return Pack.access_tuple(self)
303
elif self._state == 'resumed':
304
return self.upload_transport, self.file_name()
306
raise AssertionError(self._state)
309
self.upload_transport.delete(self.file_name())
310
indices = [self.revision_index, self.inventory_index, self.text_index,
311
self.signature_index]
312
if self.chk_index is not None:
313
indices.append(self.chk_index)
314
for index in indices:
315
index._transport.delete(index._name)
318
self._check_references()
319
index_types = ['revision', 'inventory', 'text', 'signature']
320
if self.chk_index is not None:
321
index_types.append('chk')
322
for index_type in index_types:
323
old_name = self.index_name(index_type, self.name)
324
new_name = '../indices/' + old_name
325
self.upload_transport.rename(old_name, new_name)
326
self._replace_index_with_readonly(index_type)
327
new_name = '../packs/' + self.file_name()
328
self.upload_transport.rename(self.file_name(), new_name)
329
self._state = 'finished'
331
def _get_external_refs(self, index):
332
"""Return compression parents for this index that are not present.
334
This returns any compression parents that are referenced by this index,
335
which are not contained *in* this index. They may be present elsewhere.
337
return index.external_references(1)
173
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
174
id(self), self.transport, self.name)
340
177
class NewPack(Pack):
341
178
"""An in memory proxy for a pack which is being created."""
343
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
180
# A map of index 'type' to the file extension and position in the
182
index_definitions = {
183
'revision': ('.rix', 0),
184
'inventory': ('.iix', 1),
186
'signature': ('.six', 3),
189
def __init__(self, upload_transport, index_transport, pack_transport,
344
191
"""Create a NewPack instance.
346
:param pack_collection: A PackCollection into which this is being inserted.
193
:param upload_transport: A writable transport for the pack to be
194
incrementally uploaded to.
195
:param index_transport: A writable transport for the pack's indices to
196
be written to when the pack is finished.
197
:param pack_transport: A writable transport for the pack to be renamed
198
to when the upload is complete. This *must* be the same as
199
upload_transport.clone('../packs').
347
200
:param upload_suffix: An optional suffix to be given to any temporary
348
201
files created during the pack creation. e.g '.autopack'
349
:param file_mode: Unix permissions for newly created file.
351
203
# The relative locations of the packs are constrained, but all are
352
204
# passed in because the caller has them, so as to avoid object churn.
353
index_builder_class = pack_collection._index_builder_class
354
if pack_collection.chk_index is not None:
355
chk_index = index_builder_class(reference_lists=0)
358
205
Pack.__init__(self,
359
206
# Revisions: parents list, no text compression.
360
index_builder_class(reference_lists=1),
207
InMemoryGraphIndex(reference_lists=1),
361
208
# Inventory: We want to map compression only, but currently the
362
209
# knit code hasn't been updated enough to understand that, so we
363
210
# have a regular 2-list index giving parents and compression
365
index_builder_class(reference_lists=2),
212
InMemoryGraphIndex(reference_lists=2),
366
213
# Texts: compression and per file graph, for all fileids - so two
367
214
# reference lists and two elements in the key tuple.
368
index_builder_class(reference_lists=2, key_elements=2),
215
InMemoryGraphIndex(reference_lists=2, key_elements=2),
369
216
# Signatures: Just blobs to store, no compression, no parents
371
index_builder_class(reference_lists=0),
372
# CHK based storage - just blobs, no compression or parents.
218
InMemoryGraphIndex(reference_lists=0),
375
self._pack_collection = pack_collection
376
# When we make readonly indices, we need this.
377
self.index_class = pack_collection._index_class
378
220
# where should the new pack be opened
379
self.upload_transport = pack_collection._upload_transport
221
self.upload_transport = upload_transport
380
222
# where are indices written out to
381
self.index_transport = pack_collection._index_transport
223
self.index_transport = index_transport
382
224
# where is the pack renamed to when it is finished?
383
self.pack_transport = pack_collection._pack_transport
384
# What file mode to upload the pack and indices with.
385
self._file_mode = file_mode
225
self.pack_transport = pack_transport
386
226
# tracks the content written to the .pack file.
387
self._hash = osutils.md5()
388
# a tuple with the length in bytes of the indices, once the pack
389
# is finalised. (rev, inv, text, sigs, chk_if_in_use)
227
self._hash = md5.new()
228
# a four-tuple with the length in bytes of the indices, once the pack
229
# is finalised. (rev, inv, text, sigs)
390
230
self.index_sizes = None
391
231
# How much data to cache when writing packs. Note that this is not
392
232
# synchronised with reads, because it's not in the transport layer, so
595
428
Future searches on the aggregate index will seach this new index
596
429
before all previously inserted indices.
598
431
:param index: An Index for the pack.
599
432
:param pack: A Pack instance.
601
434
# expose it to the index map
602
435
self.index_to_pack[index] = pack.access_tuple()
603
436
# put it at the front of the linear index list
604
self.combined_index.insert_index(0, index, pack.name)
437
self.combined_index.insert_index(0, index)
606
439
def add_writable_index(self, index, pack):
607
440
"""Add an index which is able to have data added to it.
609
442
There can be at most one writable index at any time. Any
610
443
modifications made to the knit are put into this index.
612
445
:param index: An index from the pack parameter.
613
446
:param pack: A Pack instance.
615
if self.add_callback is not None:
616
raise AssertionError(
617
"%s already has a writable index through %s" % \
618
(self, self.add_callback))
448
assert self.add_callback is None, \
449
"%s already has a writable index through %s" % \
450
(self, self.add_callback)
619
451
# allow writing: queue writes to a new index
620
452
self.add_index(index, pack)
621
453
# Updates the index to packs mapping as a side effect,
622
self.data_access.set_writer(pack._writer, index, pack.access_tuple())
454
self.knit_access.set_writer(pack._writer, index, pack.access_tuple())
623
455
self.add_callback = index.add_nodes
626
458
"""Reset all the aggregate data to nothing."""
627
self.data_access.set_writer(None, None, (None, None))
459
self.knit_access.set_writer(None, None, (None, None))
628
460
self.index_to_pack.clear()
629
461
del self.combined_index._indices[:]
630
del self.combined_index._index_names[:]
631
462
self.add_callback = None
633
def remove_index(self, index):
464
def remove_index(self, index, pack):
634
465
"""Remove index from the indices used to answer queries.
636
467
:param index: An index from the pack parameter.
468
:param pack: A Pack instance.
638
470
del self.index_to_pack[index]
639
pos = self.combined_index._indices.index(index)
640
del self.combined_index._indices[pos]
641
del self.combined_index._index_names[pos]
471
self.combined_index._indices.remove(index)
642
472
if (self.add_callback is not None and
643
473
getattr(index, 'add_nodes', None) == self.add_callback):
644
474
self.add_callback = None
645
self.data_access.set_writer(None, None, (None, None))
648
class Packer(object):
649
"""Create a pack from packs."""
651
def __init__(self, pack_collection, packs, suffix, revision_ids=None,
655
:param pack_collection: A RepositoryPackCollection object where the
656
new pack is being written to.
657
:param packs: The packs to combine.
658
:param suffix: The suffix to use on the temporary files for the pack.
659
:param revision_ids: Revision ids to limit the pack to.
660
:param reload_func: A function to call if a pack file/index goes
661
missing. The side effect of calling this function should be to
662
update self.packs. See also AggregateIndex
666
self.revision_ids = revision_ids
667
# The pack object we are creating.
669
self._pack_collection = pack_collection
670
self._reload_func = reload_func
671
# The index layer keys for the revisions being copied. None for 'all
673
self._revision_keys = None
674
# What text keys to copy. None for 'all texts'. This is set by
675
# _copy_inventory_texts
676
self._text_filter = None
679
def _extra_init(self):
680
"""A template hook to allow extending the constructor trivially."""
682
def _pack_map_and_index_list(self, index_attribute):
683
"""Convert a list of packs to an index pack map and index list.
685
:param index_attribute: The attribute that the desired index is found
687
:return: A tuple (map, list) where map contains the dict from
688
index:pack_tuple, and list contains the indices in the preferred
693
for pack_obj in self.packs:
694
index = getattr(pack_obj, index_attribute)
695
indices.append(index)
696
pack_map[index] = pack_obj
697
return pack_map, indices
699
def _index_contents(self, indices, key_filter=None):
700
"""Get an iterable of the index contents from a pack_map.
702
:param indices: The list of indices to query
703
:param key_filter: An optional filter to limit the keys returned.
705
all_index = CombinedGraphIndex(indices)
706
if key_filter is None:
707
return all_index.iter_all_entries()
709
return all_index.iter_entries(key_filter)
711
def pack(self, pb=None):
712
"""Create a new pack by reading data from other packs.
714
This does little more than a bulk copy of data. One key difference
715
is that data with the same item key across multiple packs is elided
716
from the output. The new pack is written into the current pack store
717
along with its indices, and the name added to the pack names. The
718
source packs are not altered and are not required to be in the current
721
:param pb: An optional progress bar to use. A nested bar is created if
723
:return: A Pack object, or None if nothing was copied.
725
# open a pack - using the same name as the last temporary file
726
# - which has already been flushed, so its safe.
727
# XXX: - duplicate code warning with start_write_group; fix before
728
# considering 'done'.
729
if self._pack_collection._new_pack is not None:
730
raise errors.BzrError('call to %s.pack() while another pack is'
732
% (self.__class__.__name__,))
733
if self.revision_ids is not None:
734
if len(self.revision_ids) == 0:
735
# silly fetch request.
738
self.revision_ids = frozenset(self.revision_ids)
739
self.revision_keys = frozenset((revid,) for revid in
742
self.pb = ui.ui_factory.nested_progress_bar()
746
return self._create_pack_from_packs()
752
"""Open a pack for the pack we are creating."""
753
new_pack = self._pack_collection.pack_factory(self._pack_collection,
754
upload_suffix=self.suffix,
755
file_mode=self._pack_collection.repo.bzrdir._get_file_mode())
756
# We know that we will process all nodes in order, and don't need to
757
# query, so don't combine any indices spilled to disk until we are done
758
new_pack.revision_index.set_optimize(combine_backing_indices=False)
759
new_pack.inventory_index.set_optimize(combine_backing_indices=False)
760
new_pack.text_index.set_optimize(combine_backing_indices=False)
761
new_pack.signature_index.set_optimize(combine_backing_indices=False)
764
def _update_pack_order(self, entries, index_to_pack_map):
765
"""Determine how we want our packs to be ordered.
767
This changes the sort order of the self.packs list so that packs unused
768
by 'entries' will be at the end of the list, so that future requests
769
can avoid probing them. Used packs will be at the front of the
770
self.packs list, in the order of their first use in 'entries'.
772
:param entries: A list of (index, ...) tuples
773
:param index_to_pack_map: A mapping from index objects to pack objects.
777
for entry in entries:
779
if index not in seen_indexes:
780
packs.append(index_to_pack_map[index])
781
seen_indexes.add(index)
782
if len(packs) == len(self.packs):
783
if 'pack' in debug.debug_flags:
784
mutter('Not changing pack list, all packs used.')
786
seen_packs = set(packs)
787
for pack in self.packs:
788
if pack not in seen_packs:
791
if 'pack' in debug.debug_flags:
792
old_names = [p.access_tuple()[1] for p in self.packs]
793
new_names = [p.access_tuple()[1] for p in packs]
794
mutter('Reordering packs\nfrom: %s\n to: %s',
795
old_names, new_names)
798
def _copy_revision_texts(self):
799
"""Copy revision data to the new pack."""
801
if self.revision_ids:
802
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
805
# select revision keys
806
revision_index_map, revision_indices = self._pack_map_and_index_list(
808
revision_nodes = self._index_contents(revision_indices, revision_keys)
809
revision_nodes = list(revision_nodes)
810
self._update_pack_order(revision_nodes, revision_index_map)
811
# copy revision keys and adjust values
812
self.pb.update("Copying revision texts", 1)
813
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
814
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
815
self.new_pack.revision_index, readv_group_iter, total_items))
816
if 'pack' in debug.debug_flags:
817
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
818
time.ctime(), self._pack_collection._upload_transport.base,
819
self.new_pack.random_name,
820
self.new_pack.revision_index.key_count(),
821
time.time() - self.new_pack.start_time)
822
self._revision_keys = revision_keys
824
def _copy_inventory_texts(self):
825
"""Copy the inventory texts to the new pack.
827
self._revision_keys is used to determine what inventories to copy.
829
Sets self._text_filter appropriately.
831
# select inventory keys
832
inv_keys = self._revision_keys # currently the same keyspace, and note that
833
# querying for keys here could introduce a bug where an inventory item
834
# is missed, so do not change it to query separately without cross
835
# checking like the text key check below.
836
inventory_index_map, inventory_indices = self._pack_map_and_index_list(
838
inv_nodes = self._index_contents(inventory_indices, inv_keys)
839
# copy inventory keys and adjust values
840
# XXX: Should be a helper function to allow different inv representation
842
self.pb.update("Copying inventory texts", 2)
843
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
844
# Only grab the output lines if we will be processing them
845
output_lines = bool(self.revision_ids)
846
inv_lines = self._copy_nodes_graph(inventory_index_map,
847
self.new_pack._writer, self.new_pack.inventory_index,
848
readv_group_iter, total_items, output_lines=output_lines)
849
if self.revision_ids:
850
self._process_inventory_lines(inv_lines)
852
# eat the iterator to cause it to execute.
854
self._text_filter = None
855
if 'pack' in debug.debug_flags:
856
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
857
time.ctime(), self._pack_collection._upload_transport.base,
858
self.new_pack.random_name,
859
self.new_pack.inventory_index.key_count(),
860
time.time() - self.new_pack.start_time)
862
def _copy_text_texts(self):
864
text_index_map, text_nodes = self._get_text_nodes()
865
if self._text_filter is not None:
866
# We could return the keys copied as part of the return value from
867
# _copy_nodes_graph but this doesn't work all that well with the
868
# need to get line output too, so we check separately, and as we're
869
# going to buffer everything anyway, we check beforehand, which
870
# saves reading knit data over the wire when we know there are
872
text_nodes = set(text_nodes)
873
present_text_keys = set(_node[1] for _node in text_nodes)
874
missing_text_keys = set(self._text_filter) - present_text_keys
875
if missing_text_keys:
876
# TODO: raise a specific error that can handle many missing
878
mutter("missing keys during fetch: %r", missing_text_keys)
879
a_missing_key = missing_text_keys.pop()
880
raise errors.RevisionNotPresent(a_missing_key[1],
882
# copy text keys and adjust values
883
self.pb.update("Copying content texts", 3)
884
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
885
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
886
self.new_pack.text_index, readv_group_iter, total_items))
887
self._log_copied_texts()
889
def _create_pack_from_packs(self):
890
self.pb.update("Opening pack", 0, 5)
891
self.new_pack = self.open_pack()
892
new_pack = self.new_pack
893
# buffer data - we won't be reading-back during the pack creation and
894
# this makes a significant difference on sftp pushes.
895
new_pack.set_write_cache_size(1024*1024)
896
if 'pack' in debug.debug_flags:
897
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
898
for a_pack in self.packs]
899
if self.revision_ids is not None:
900
rev_count = len(self.revision_ids)
903
mutter('%s: create_pack: creating pack from source packs: '
904
'%s%s %s revisions wanted %s t=0',
905
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
906
plain_pack_list, rev_count)
907
self._copy_revision_texts()
908
self._copy_inventory_texts()
909
self._copy_text_texts()
910
# select signature keys
911
signature_filter = self._revision_keys # same keyspace
912
signature_index_map, signature_indices = self._pack_map_and_index_list(
914
signature_nodes = self._index_contents(signature_indices,
916
# copy signature keys and adjust values
917
self.pb.update("Copying signature texts", 4)
918
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
919
new_pack.signature_index)
920
if 'pack' in debug.debug_flags:
921
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
922
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
923
new_pack.signature_index.key_count(),
924
time.time() - new_pack.start_time)
926
# NB XXX: how to check CHK references are present? perhaps by yielding
927
# the items? How should that interact with stacked repos?
928
if new_pack.chk_index is not None:
930
if 'pack' in debug.debug_flags:
931
mutter('%s: create_pack: chk content copied: %s%s %d items t+%6.3fs',
932
time.ctime(), self._pack_collection._upload_transport.base,
933
new_pack.random_name,
934
new_pack.chk_index.key_count(),
935
time.time() - new_pack.start_time)
936
new_pack._check_references()
937
if not self._use_pack(new_pack):
940
self.pb.update("Finishing pack", 5)
942
self._pack_collection.allocate(new_pack)
945
def _copy_chks(self, refs=None):
946
# XXX: Todo, recursive follow-pointers facility when fetching some
948
chk_index_map, chk_indices = self._pack_map_and_index_list(
950
chk_nodes = self._index_contents(chk_indices, refs)
952
# TODO: This isn't strictly tasteful as we are accessing some private
953
# variables (_serializer). Perhaps a better way would be to have
954
# Repository._deserialise_chk_node()
955
search_key_func = chk_map.search_key_registry.get(
956
self._pack_collection.repo._serializer.search_key_name)
957
def accumlate_refs(lines):
958
# XXX: move to a generic location
960
bytes = ''.join(lines)
961
node = chk_map._deserialise(bytes, ("unknown",), search_key_func)
962
new_refs.update(node.refs())
963
self._copy_nodes(chk_nodes, chk_index_map, self.new_pack._writer,
964
self.new_pack.chk_index, output_lines=accumlate_refs)
967
def _copy_nodes(self, nodes, index_map, writer, write_index,
969
"""Copy knit nodes between packs with no graph references.
971
:param output_lines: Output full texts of copied items.
973
pb = ui.ui_factory.nested_progress_bar()
975
return self._do_copy_nodes(nodes, index_map, writer,
976
write_index, pb, output_lines=output_lines)
980
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb,
982
# for record verification
983
knit = KnitVersionedFiles(None, None)
984
# plan a readv on each source pack:
986
nodes = sorted(nodes)
987
# how to map this into knit.py - or knit.py into this?
988
# we don't want the typical knit logic, we want grouping by pack
989
# at this point - perhaps a helper library for the following code
990
# duplication points?
992
for index, key, value in nodes:
993
if index not in request_groups:
994
request_groups[index] = []
995
request_groups[index].append((key, value))
997
pb.update("Copied record", record_index, len(nodes))
998
for index, items in request_groups.iteritems():
999
pack_readv_requests = []
1000
for key, value in items:
1001
# ---- KnitGraphIndex.get_position
1002
bits = value[1:].split(' ')
1003
offset, length = int(bits[0]), int(bits[1])
1004
pack_readv_requests.append((offset, length, (key, value[0])))
1005
# linear scan up the pack
1006
pack_readv_requests.sort()
1008
pack_obj = index_map[index]
1009
transport, path = pack_obj.access_tuple()
1011
reader = pack.make_readv_reader(transport, path,
1012
[offset[0:2] for offset in pack_readv_requests])
1013
except errors.NoSuchFile:
1014
if self._reload_func is not None:
1017
for (names, read_func), (_1, _2, (key, eol_flag)) in \
1018
izip(reader.iter_records(), pack_readv_requests):
1019
raw_data = read_func(None)
1020
# check the header only
1021
if output_lines is not None:
1022
output_lines(knit._parse_record(key[-1], raw_data)[0])
1024
df, _ = knit._parse_record_header(key, raw_data)
1026
pos, size = writer.add_bytes_record(raw_data, names)
1027
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
1028
pb.update("Copied record", record_index)
1031
def _copy_nodes_graph(self, index_map, writer, write_index,
1032
readv_group_iter, total_items, output_lines=False):
1033
"""Copy knit nodes between packs.
1035
:param output_lines: Return lines present in the copied data as
1036
an iterator of line,version_id.
1038
pb = ui.ui_factory.nested_progress_bar()
1040
for result in self._do_copy_nodes_graph(index_map, writer,
1041
write_index, output_lines, pb, readv_group_iter, total_items):
1044
# Python 2.4 does not permit try:finally: in a generator.
1050
def _do_copy_nodes_graph(self, index_map, writer, write_index,
1051
output_lines, pb, readv_group_iter, total_items):
1052
# for record verification
1053
knit = KnitVersionedFiles(None, None)
1054
# for line extraction when requested (inventories only)
1056
factory = KnitPlainFactory()
1058
pb.update("Copied record", record_index, total_items)
1059
for index, readv_vector, node_vector in readv_group_iter:
1061
pack_obj = index_map[index]
1062
transport, path = pack_obj.access_tuple()
1064
reader = pack.make_readv_reader(transport, path, readv_vector)
1065
except errors.NoSuchFile:
1066
if self._reload_func is not None:
1069
for (names, read_func), (key, eol_flag, references) in \
1070
izip(reader.iter_records(), node_vector):
1071
raw_data = read_func(None)
1073
# read the entire thing
1074
content, _ = knit._parse_record(key[-1], raw_data)
1075
if len(references[-1]) == 0:
1076
line_iterator = factory.get_fulltext_content(content)
1078
line_iterator = factory.get_linedelta_content(content)
1079
for line in line_iterator:
1082
# check the header only
1083
df, _ = knit._parse_record_header(key, raw_data)
1085
pos, size = writer.add_bytes_record(raw_data, names)
1086
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1087
pb.update("Copied record", record_index)
1090
def _get_text_nodes(self):
1091
text_index_map, text_indices = self._pack_map_and_index_list(
1093
return text_index_map, self._index_contents(text_indices,
1096
def _least_readv_node_readv(self, nodes):
1097
"""Generate request groups for nodes using the least readv's.
1099
:param nodes: An iterable of graph index nodes.
1100
:return: Total node count and an iterator of the data needed to perform
1101
readvs to obtain the data for nodes. Each item yielded by the
1102
iterator is a tuple with:
1103
index, readv_vector, node_vector. readv_vector is a list ready to
1104
hand to the transport readv method, and node_vector is a list of
1105
(key, eol_flag, references) for the node retrieved by the
1106
matching readv_vector.
1108
# group by pack so we do one readv per pack
1109
nodes = sorted(nodes)
1112
for index, key, value, references in nodes:
1113
if index not in request_groups:
1114
request_groups[index] = []
1115
request_groups[index].append((key, value, references))
1117
for index, items in request_groups.iteritems():
1118
pack_readv_requests = []
1119
for key, value, references in items:
1120
# ---- KnitGraphIndex.get_position
1121
bits = value[1:].split(' ')
1122
offset, length = int(bits[0]), int(bits[1])
1123
pack_readv_requests.append(
1124
((offset, length), (key, value[0], references)))
1125
# linear scan up the pack to maximum range combining.
1126
pack_readv_requests.sort()
1127
# split out the readv and the node data.
1128
pack_readv = [readv for readv, node in pack_readv_requests]
1129
node_vector = [node for readv, node in pack_readv_requests]
1130
result.append((index, pack_readv, node_vector))
1131
return total, result
1133
def _log_copied_texts(self):
1134
if 'pack' in debug.debug_flags:
1135
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
1136
time.ctime(), self._pack_collection._upload_transport.base,
1137
self.new_pack.random_name,
1138
self.new_pack.text_index.key_count(),
1139
time.time() - self.new_pack.start_time)
1141
def _process_inventory_lines(self, inv_lines):
1142
"""Use up the inv_lines generator and setup a text key filter."""
1143
repo = self._pack_collection.repo
1144
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
1145
inv_lines, self.revision_keys)
1147
for fileid, file_revids in fileid_revisions.iteritems():
1148
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
1149
self._text_filter = text_filter
1151
def _revision_node_readv(self, revision_nodes):
1152
"""Return the total revisions and the readv's to issue.
1154
:param revision_nodes: The revision index contents for the packs being
1155
incorporated into the new pack.
1156
:return: As per _least_readv_node_readv.
1158
return self._least_readv_node_readv(revision_nodes)
1160
def _use_pack(self, new_pack):
1161
"""Return True if new_pack should be used.
1163
:param new_pack: The pack that has just been created.
1164
:return: True if the pack should be used.
1166
return new_pack.data_inserted()
1169
class OptimisingPacker(Packer):
1170
"""A packer which spends more time to create better disk layouts."""
1172
def _revision_node_readv(self, revision_nodes):
1173
"""Return the total revisions and the readv's to issue.
1175
This sort places revisions in topological order with the ancestors
1178
:param revision_nodes: The revision index contents for the packs being
1179
incorporated into the new pack.
1180
:return: As per _least_readv_node_readv.
1182
# build an ancestors dict
1185
for index, key, value, references in revision_nodes:
1186
ancestors[key] = references[0]
1187
by_key[key] = (index, value, references)
1188
order = tsort.topo_sort(ancestors)
1190
# Single IO is pathological, but it will work as a starting point.
1192
for key in reversed(order):
1193
index, value, references = by_key[key]
1194
# ---- KnitGraphIndex.get_position
1195
bits = value[1:].split(' ')
1196
offset, length = int(bits[0]), int(bits[1])
1198
(index, [(offset, length)], [(key, value[0], references)]))
1199
# TODO: combine requests in the same index that are in ascending order.
1200
return total, requests
1202
def open_pack(self):
1203
"""Open a pack for the pack we are creating."""
1204
new_pack = super(OptimisingPacker, self).open_pack()
1205
# Turn on the optimization flags for all the index builders.
1206
new_pack.revision_index.set_optimize(for_size=True)
1207
new_pack.inventory_index.set_optimize(for_size=True)
1208
new_pack.text_index.set_optimize(for_size=True)
1209
new_pack.signature_index.set_optimize(for_size=True)
1213
class ReconcilePacker(Packer):
1214
"""A packer which regenerates indices etc as it copies.
1216
This is used by ``bzr reconcile`` to cause parent text pointers to be
1220
def _extra_init(self):
1221
self._data_changed = False
1223
def _process_inventory_lines(self, inv_lines):
1224
"""Generate a text key reference map rather for reconciling with."""
1225
repo = self._pack_collection.repo
1226
refs = repo._find_text_key_references_from_xml_inventory_lines(
1228
self._text_refs = refs
1229
# during reconcile we:
1230
# - convert unreferenced texts to full texts
1231
# - correct texts which reference a text not copied to be full texts
1232
# - copy all others as-is but with corrected parents.
1233
# - so at this point we don't know enough to decide what becomes a full
1235
self._text_filter = None
1237
def _copy_text_texts(self):
1238
"""generate what texts we should have and then copy."""
1239
self.pb.update("Copying content texts", 3)
1240
# we have three major tasks here:
1241
# 1) generate the ideal index
1242
repo = self._pack_collection.repo
1243
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1244
_1, key, _2, refs in
1245
self.new_pack.revision_index.iter_all_entries()])
1246
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1247
# 2) generate a text_nodes list that contains all the deltas that can
1248
# be used as-is, with corrected parents.
1251
discarded_nodes = []
1252
NULL_REVISION = _mod_revision.NULL_REVISION
1253
text_index_map, text_nodes = self._get_text_nodes()
1254
for node in text_nodes:
1260
ideal_parents = tuple(ideal_index[node[1]])
1262
discarded_nodes.append(node)
1263
self._data_changed = True
1265
if ideal_parents == (NULL_REVISION,):
1267
if ideal_parents == node[3][0]:
1269
ok_nodes.append(node)
1270
elif ideal_parents[0:1] == node[3][0][0:1]:
1271
# the left most parent is the same, or there are no parents
1272
# today. Either way, we can preserve the representation as
1273
# long as we change the refs to be inserted.
1274
self._data_changed = True
1275
ok_nodes.append((node[0], node[1], node[2],
1276
(ideal_parents, node[3][1])))
1277
self._data_changed = True
1279
# Reinsert this text completely
1280
bad_texts.append((node[1], ideal_parents))
1281
self._data_changed = True
1282
# we're finished with some data.
1285
# 3) bulk copy the ok data
1286
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1287
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1288
self.new_pack.text_index, readv_group_iter, total_items))
1289
# 4) adhoc copy all the other texts.
1290
# We have to topologically insert all texts otherwise we can fail to
1291
# reconcile when parts of a single delta chain are preserved intact,
1292
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1293
# reinserted, and if d3 has incorrect parents it will also be
1294
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1295
# copied), so we will try to delta, but d2 is not currently able to be
1296
# extracted because it's basis d1 is not present. Topologically sorting
1297
# addresses this. The following generates a sort for all the texts that
1298
# are being inserted without having to reference the entire text key
1299
# space (we only topo sort the revisions, which is smaller).
1300
topo_order = tsort.topo_sort(ancestors)
1301
rev_order = dict(zip(topo_order, range(len(topo_order))))
1302
bad_texts.sort(key=lambda key:rev_order.get(key[0][1], 0))
1303
transaction = repo.get_transaction()
1304
file_id_index = GraphIndexPrefixAdapter(
1305
self.new_pack.text_index,
1307
add_nodes_callback=self.new_pack.text_index.add_nodes)
1308
data_access = _DirectPackAccess(
1309
{self.new_pack.text_index:self.new_pack.access_tuple()})
1310
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1311
self.new_pack.access_tuple())
1312
output_texts = KnitVersionedFiles(
1313
_KnitGraphIndex(self.new_pack.text_index,
1314
add_callback=self.new_pack.text_index.add_nodes,
1315
deltas=True, parents=True, is_locked=repo.is_locked),
1316
data_access=data_access, max_delta_chain=200)
1317
for key, parent_keys in bad_texts:
1318
# We refer to the new pack to delta data being output.
1319
# A possible improvement would be to catch errors on short reads
1320
# and only flush then.
1321
self.new_pack.flush()
1323
for parent_key in parent_keys:
1324
if parent_key[0] != key[0]:
1325
# Graph parents must match the fileid
1326
raise errors.BzrError('Mismatched key parent %r:%r' %
1328
parents.append(parent_key[1])
1329
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1330
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1331
output_texts.add_lines(key, parent_keys, text_lines,
1332
random_id=True, check_content=False)
1333
# 5) check that nothing inserted has a reference outside the keyspace.
1334
missing_text_keys = self.new_pack.text_index._external_references()
1335
if missing_text_keys:
1336
raise errors.BzrCheckError('Reference to missing compression parents %r'
1337
% (missing_text_keys,))
1338
self._log_copied_texts()
1340
def _use_pack(self, new_pack):
1341
"""Override _use_pack to check for reconcile having changed content."""
1342
# XXX: we might be better checking this at the copy time.
1343
original_inventory_keys = set()
1344
inv_index = self._pack_collection.inventory_index.combined_index
1345
for entry in inv_index.iter_all_entries():
1346
original_inventory_keys.add(entry[1])
1347
new_inventory_keys = set()
1348
for entry in new_pack.inventory_index.iter_all_entries():
1349
new_inventory_keys.add(entry[1])
1350
if new_inventory_keys != original_inventory_keys:
1351
self._data_changed = True
1352
return new_pack.data_inserted() and self._data_changed
475
self.knit_access.set_writer(None, None, (None, None))
1355
478
class RepositoryPackCollection(object):
1356
"""Management of packs within a repository.
1358
:ivar _names: map of {pack_name: (index_size,)}
1361
pack_factory = NewPack
1362
resumed_pack_factory = ResumedPack
479
"""Management of packs within a repository."""
1364
481
def __init__(self, repo, transport, index_transport, upload_transport,
1365
pack_transport, index_builder_class, index_class,
1367
483
"""Create a new RepositoryPackCollection.
1369
:param transport: Addresses the repository base directory
485
:param transport: Addresses the repository base directory
1370
486
(typically .bzr/repository/).
1371
487
:param index_transport: Addresses the directory containing indices.
1372
488
:param upload_transport: Addresses the directory into which packs are written
1373
489
while they're being created.
1374
490
:param pack_transport: Addresses the directory of existing complete packs.
1375
:param index_builder_class: The index builder class to use.
1376
:param index_class: The index class to use.
1377
:param use_chk_index: Whether to setup and manage a CHK index.
1379
# XXX: This should call self.reset()
1380
492
self.repo = repo
1381
493
self.transport = transport
1382
494
self._index_transport = index_transport
1383
495
self._upload_transport = upload_transport
1384
496
self._pack_transport = pack_transport
1385
self._index_builder_class = index_builder_class
1386
self._index_class = index_class
1387
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3,
497
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
1390
499
# name:Pack mapping
1392
500
self._packs_by_name = {}
1393
501
# the previous pack-names content
1394
502
self._packs_at_load = None
1395
503
# when a pack is being created by this object, the state of that pack.
1396
504
self._new_pack = None
1397
505
# aggregated revision index data
1398
flush = self._flush_new_pack
1399
self.revision_index = AggregateIndex(self.reload_pack_names, flush)
1400
self.inventory_index = AggregateIndex(self.reload_pack_names, flush)
1401
self.text_index = AggregateIndex(self.reload_pack_names, flush)
1402
self.signature_index = AggregateIndex(self.reload_pack_names, flush)
1403
all_indices = [self.revision_index, self.inventory_index,
1404
self.text_index, self.signature_index]
1406
self.chk_index = AggregateIndex(self.reload_pack_names, flush)
1407
all_indices.append(self.chk_index)
1409
# used to determine if we're using a chk_index elsewhere.
1410
self.chk_index = None
1411
# Tell all the CombinedGraphIndex objects about each other, so they can
1412
# share hints about which pack names to search first.
1413
all_combined = [agg_idx.combined_index for agg_idx in all_indices]
1414
for combined_idx in all_combined:
1415
combined_idx.set_sibling_indices(
1416
set(all_combined).difference([combined_idx]))
1418
self._resumed_packs = []
1421
return '%s(%r)' % (self.__class__.__name__, self.repo)
506
self.revision_index = AggregateIndex()
507
self.inventory_index = AggregateIndex()
508
self.text_index = AggregateIndex()
509
self.signature_index = AggregateIndex()
1423
511
def add_pack_to_memory(self, pack):
1424
512
"""Make a Pack object available to the repository to satisfy queries.
1426
514
:param pack: A Pack object.
1428
if pack.name in self._packs_by_name:
1429
raise AssertionError(
1430
'pack %s already in _packs_by_name' % (pack.name,))
516
assert pack.name not in self._packs_by_name
1431
517
self.packs.append(pack)
1432
518
self._packs_by_name[pack.name] = pack
1433
519
self.revision_index.add_index(pack.revision_index, pack)
1434
520
self.inventory_index.add_index(pack.inventory_index, pack)
1435
521
self.text_index.add_index(pack.text_index, pack)
1436
522
self.signature_index.add_index(pack.signature_index, pack)
1437
if self.chk_index is not None:
1438
self.chk_index.add_index(pack.chk_index, pack)
524
def _add_text_to_weave(self, file_id, revision_id, new_lines, parents,
525
nostore_sha, random_revid):
526
file_id_index = GraphIndexPrefixAdapter(
527
self.text_index.combined_index,
529
add_nodes_callback=self.text_index.add_callback)
530
self.repo._text_knit._index._graph_index = file_id_index
531
self.repo._text_knit._index._add_callback = file_id_index.add_nodes
532
return self.repo._text_knit.add_lines_with_ghosts(
533
revision_id, parents, new_lines, nostore_sha=nostore_sha,
534
random_id=random_revid, check_content=False)[0:2]
1440
536
def all_packs(self):
1441
537
"""Return a list of all the Pack objects this repository has.
1492
584
# group their data with the relevant commit, and that may
1493
585
# involve rewriting ancient history - which autopack tries to
1494
586
# avoid. Alternatively we could not group the data but treat
1495
# each of these as having a single revision, and thus add
587
# each of these as having a single revision, and thus add
1496
588
# one revision for each to the total revision count, to get
1497
589
# a matching distribution.
1499
591
existing_packs.append((revision_count, pack))
1500
592
pack_operations = self.plan_autopack_combinations(
1501
593
existing_packs, pack_distribution)
1502
num_new_packs = len(pack_operations)
1503
num_old_packs = sum([len(po[1]) for po in pack_operations])
1504
num_revs_affected = sum([po[0] for po in pack_operations])
1505
mutter('Auto-packing repository %s, which has %d pack files, '
1506
'containing %d revisions. Packing %d files into %d affecting %d'
1507
' revisions', self, total_packs, total_revisions, num_old_packs,
1508
num_new_packs, num_revs_affected)
1509
result = self._execute_pack_operations(pack_operations,
1510
reload_func=self._restart_autopack)
1511
mutter('Auto-packing repository %s completed', self)
1514
def _execute_pack_operations(self, pack_operations, _packer_class=Packer,
594
self._execute_pack_operations(pack_operations)
597
def create_pack_from_packs(self, packs, suffix, revision_ids=None):
598
"""Create a new pack by reading data from other packs.
600
This does little more than a bulk copy of data. One key difference
601
is that data with the same item key across multiple packs is elided
602
from the output. The new pack is written into the current pack store
603
along with its indices, and the name added to the pack names. The
604
source packs are not altered and are not required to be in the current
607
:param packs: An iterable of Packs to combine.
608
:param revision_ids: Either None, to copy all data, or a list
609
of revision_ids to limit the copied data to the data they
611
:return: A Pack object, or None if nothing was copied.
613
# open a pack - using the same name as the last temporary file
614
# - which has already been flushed, so its safe.
615
# XXX: - duplicate code warning with start_write_group; fix before
616
# considering 'done'.
617
if self._new_pack is not None:
618
raise errors.BzrError('call to create_pack_from_packs while '
619
'another pack is being written.')
620
if revision_ids is not None and len(revision_ids) == 0:
621
# silly fetch request.
623
new_pack = NewPack(self._upload_transport, self._index_transport,
624
self._pack_transport, upload_suffix=suffix)
625
# buffer data - we won't be reading-back during the pack creation and
626
# this makes a significant difference on sftp pushes.
627
new_pack.set_write_cache_size(1024*1024)
628
if 'pack' in debug.debug_flags:
629
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
631
if revision_ids is not None:
632
rev_count = len(revision_ids)
635
mutter('%s: create_pack: creating pack from source packs: '
636
'%s%s %s revisions wanted %s t=0',
637
time.ctime(), self._upload_transport.base, new_pack.random_name,
638
plain_pack_list, rev_count)
641
revision_keys = [(revision_id,) for revision_id in revision_ids]
645
# select revision keys
646
revision_index_map = self._packs_list_to_pack_map_and_index_list(
647
packs, 'revision_index')[0]
648
revision_nodes = self._index_contents(revision_index_map, revision_keys)
649
# copy revision keys and adjust values
650
list(self._copy_nodes_graph(revision_nodes, revision_index_map,
651
new_pack._writer, new_pack.revision_index))
652
if 'pack' in debug.debug_flags:
653
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
654
time.ctime(), self._upload_transport.base, new_pack.random_name,
655
new_pack.revision_index.key_count(),
656
time.time() - new_pack.start_time)
657
# select inventory keys
658
inv_keys = revision_keys # currently the same keyspace, and note that
659
# querying for keys here could introduce a bug where an inventory item
660
# is missed, so do not change it to query separately without cross
661
# checking like the text key check below.
662
inventory_index_map = self._packs_list_to_pack_map_and_index_list(
663
packs, 'inventory_index')[0]
664
inv_nodes = self._index_contents(inventory_index_map, inv_keys)
665
# copy inventory keys and adjust values
666
# XXX: Should be a helper function to allow different inv representation
668
inv_lines = self._copy_nodes_graph(inv_nodes, inventory_index_map,
669
new_pack._writer, new_pack.inventory_index, output_lines=True)
671
fileid_revisions = self.repo._find_file_ids_from_xml_inventory_lines(
672
inv_lines, revision_ids)
674
for fileid, file_revids in fileid_revisions.iteritems():
676
[(fileid, file_revid) for file_revid in file_revids])
678
# eat the iterator to cause it to execute.
681
if 'pack' in debug.debug_flags:
682
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
683
time.ctime(), self._upload_transport.base, new_pack.random_name,
684
new_pack.inventory_index.key_count(),
685
time.time() - new_pack.start_time)
687
text_index_map = self._packs_list_to_pack_map_and_index_list(
688
packs, 'text_index')[0]
689
text_nodes = self._index_contents(text_index_map, text_filter)
690
if text_filter is not None:
691
# We could return the keys copied as part of the return value from
692
# _copy_nodes_graph but this doesn't work all that well with the
693
# need to get line output too, so we check separately, and as we're
694
# going to buffer everything anyway, we check beforehand, which
695
# saves reading knit data over the wire when we know there are
697
text_nodes = set(text_nodes)
698
present_text_keys = set(_node[1] for _node in text_nodes)
699
missing_text_keys = set(text_filter) - present_text_keys
700
if missing_text_keys:
701
# TODO: raise a specific error that can handle many missing
703
a_missing_key = missing_text_keys.pop()
704
raise errors.RevisionNotPresent(a_missing_key[1],
706
# copy text keys and adjust values
707
list(self._copy_nodes_graph(text_nodes, text_index_map,
708
new_pack._writer, new_pack.text_index))
709
if 'pack' in debug.debug_flags:
710
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
711
time.ctime(), self._upload_transport.base, new_pack.random_name,
712
new_pack.text_index.key_count(),
713
time.time() - new_pack.start_time)
714
# select signature keys
715
signature_filter = revision_keys # same keyspace
716
signature_index_map = self._packs_list_to_pack_map_and_index_list(
717
packs, 'signature_index')[0]
718
signature_nodes = self._index_contents(signature_index_map,
720
# copy signature keys and adjust values
721
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
722
new_pack.signature_index)
723
if 'pack' in debug.debug_flags:
724
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
725
time.ctime(), self._upload_transport.base, new_pack.random_name,
726
new_pack.signature_index.key_count(),
727
time.time() - new_pack.start_time)
728
if not new_pack.data_inserted():
732
self.allocate(new_pack)
735
def _execute_pack_operations(self, pack_operations):
1516
736
"""Execute a series of pack operations.
1518
738
:param pack_operations: A list of [revision_count, packs_to_combine].
1519
:param _packer_class: The class of packer to use (default: Packer).
1520
:return: The new pack names.
1522
741
for revision_count, packs in pack_operations:
1523
742
# we may have no-ops from the setup logic
1524
743
if len(packs) == 0:
1526
packer = _packer_class(self, packs, '.autopack',
1527
reload_func=reload_func)
1530
except errors.RetryWithNewPacks:
1531
# An exception is propagating out of this context, make sure
1532
# this packer has cleaned up. Packer() doesn't set its new_pack
1533
# state into the RepositoryPackCollection object, so we only
1534
# have access to it directly here.
1535
if packer.new_pack is not None:
1536
packer.new_pack.abort()
745
# have a progress bar?
746
self.create_pack_from_packs(packs, '.autopack')
1538
747
for pack in packs:
1539
748
self._remove_pack_from_memory(pack)
1540
749
# record the newly available packs and stop advertising the old
1542
to_be_obsoleted = []
1543
for _, packs in pack_operations:
1544
to_be_obsoleted.extend(packs)
1545
result = self._save_pack_names(clear_obsolete_packs=True,
1546
obsolete_packs=to_be_obsoleted)
1549
def _flush_new_pack(self):
1550
if self._new_pack is not None:
1551
self._new_pack.flush()
751
self._save_pack_names()
752
# Move the old packs out of the way now they are no longer referenced.
753
for revision_count, packs in pack_operations:
754
self._obsolete_packs(packs)
1553
756
def lock_names(self):
1554
757
"""Acquire the mutex around the pack-names index.
1556
759
This cannot be used in the middle of a read-only transaction on the
1559
762
self.repo.control_files.lock_write()
1561
def _already_packed(self):
1562
"""Is the collection already packed?"""
1563
return not (self.repo._format.pack_compresses or (len(self._names) > 1))
1565
def pack(self, hint=None, clean_obsolete_packs=False):
1566
765
"""Pack the pack collection totally."""
1567
766
self.ensure_loaded()
1568
767
total_packs = len(self._names)
1569
if self._already_packed():
1571
770
total_revisions = self.revision_index.combined_index.key_count()
1572
771
# XXX: the following may want to be a class, to pack with a given
1574
773
mutter('Packing repository %s, which has %d pack files, '
1575
'containing %d revisions with hint %r.', self, total_packs,
1576
total_revisions, hint)
774
'containing %d revisions into 1 packs.', self, total_packs,
1577
776
# determine which packs need changing
777
pack_distribution = [1]
1578
778
pack_operations = [[0, []]]
1579
779
for pack in self.all_packs():
1580
if hint is None or pack.name in hint:
1581
# Either no hint was provided (so we are packing everything),
1582
# or this pack was included in the hint.
1583
pack_operations[-1][0] += pack.get_revision_count()
1584
pack_operations[-1][1].append(pack)
1585
self._execute_pack_operations(pack_operations, OptimisingPacker)
1587
if clean_obsolete_packs:
1588
self._clear_obsolete_packs()
780
revision_count = pack.get_revision_count()
781
pack_operations[-1][0] += revision_count
782
pack_operations[-1][1].append(pack)
783
self._execute_pack_operations(pack_operations)
1590
785
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1591
786
"""Plan a pack operation.
2078
1209
self._new_pack)
2079
1210
self.text_index.add_writable_index(self._new_pack.text_index,
2080
1211
self._new_pack)
2081
self._new_pack.text_index.set_optimize(combine_backing_indices=False)
2082
1212
self.signature_index.add_writable_index(self._new_pack.signature_index,
2083
1213
self._new_pack)
2084
if self.chk_index is not None:
2085
self.chk_index.add_writable_index(self._new_pack.chk_index,
2087
self.repo.chk_bytes._index._add_callback = self.chk_index.add_callback
2088
self._new_pack.chk_index.set_optimize(combine_backing_indices=False)
2090
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
2091
self.repo.revisions._index._add_callback = self.revision_index.add_callback
2092
self.repo.signatures._index._add_callback = self.signature_index.add_callback
2093
self.repo.texts._index._add_callback = self.text_index.add_callback
1215
# reused revision and signature knits may need updating
1217
# "Hysterical raisins. client code in bzrlib grabs those knits outside
1218
# of write groups and then mutates it inside the write group."
1219
if self.repo._revision_knit is not None:
1220
self.repo._revision_knit._index._add_callback = \
1221
self.revision_index.add_callback
1222
if self.repo._signature_knit is not None:
1223
self.repo._signature_knit._index._add_callback = \
1224
self.signature_index.add_callback
1225
# create a reused knit object for text addition in commit.
1226
self.repo._text_knit = self.repo.weave_store.get_weave_or_empty(
2095
1229
def _abort_write_group(self):
2096
1230
# FIXME: just drop the transient index.
2097
1231
# forget what names there are
2098
if self._new_pack is not None:
2099
operation = cleanup.OperationWithCleanups(self._new_pack.abort)
2100
operation.add_cleanup(setattr, self, '_new_pack', None)
2101
# If we aborted while in the middle of finishing the write
2102
# group, _remove_pack_indices could fail because the indexes are
2103
# already gone. But they're not there we shouldn't fail in this
2104
# case, so we pass ignore_missing=True.
2105
operation.add_cleanup(self._remove_pack_indices, self._new_pack,
2106
ignore_missing=True)
2107
operation.run_simple()
2108
for resumed_pack in self._resumed_packs:
2109
operation = cleanup.OperationWithCleanups(resumed_pack.abort)
2110
# See comment in previous finally block.
2111
operation.add_cleanup(self._remove_pack_indices, resumed_pack,
2112
ignore_missing=True)
2113
operation.run_simple()
2114
del self._resumed_packs[:]
2116
def _remove_resumed_pack_indices(self):
2117
for resumed_pack in self._resumed_packs:
2118
self._remove_pack_indices(resumed_pack)
2119
del self._resumed_packs[:]
2121
def _check_new_inventories(self):
2122
"""Detect missing inventories in this write group.
2124
:returns: list of strs, summarising any problems found. If the list is
2125
empty no problems were found.
2127
# The base implementation does no checks. GCRepositoryPackCollection
1232
self._new_pack.abort()
1233
self._remove_pack_indices(self._new_pack)
1234
self._new_pack = None
1235
self.repo._text_knit = None
2131
1237
def _commit_write_group(self):
2133
for prefix, versioned_file in (
2134
('revisions', self.repo.revisions),
2135
('inventories', self.repo.inventories),
2136
('texts', self.repo.texts),
2137
('signatures', self.repo.signatures),
2139
missing = versioned_file.get_missing_compression_parent_keys()
2140
all_missing.update([(prefix,) + key for key in missing])
2142
raise errors.BzrCheckError(
2143
"Repository %s has missing compression parent(s) %r "
2144
% (self.repo, sorted(all_missing)))
2145
problems = self._check_new_inventories()
2147
problems_summary = '\n'.join(problems)
2148
raise errors.BzrCheckError(
2149
"Cannot add revision(s) to repository: " + problems_summary)
2150
1238
self._remove_pack_indices(self._new_pack)
2151
any_new_content = False
2152
1239
if self._new_pack.data_inserted():
2153
1240
# get all the data to disk and read to use
2154
1241
self._new_pack.finish()
2155
1242
self.allocate(self._new_pack)
2156
1243
self._new_pack = None
2157
any_new_content = True
2159
self._new_pack.abort()
2160
self._new_pack = None
2161
for resumed_pack in self._resumed_packs:
2162
# XXX: this is a pretty ugly way to turn the resumed pack into a
2163
# properly committed pack.
2164
self._names[resumed_pack.name] = None
2165
self._remove_pack_from_memory(resumed_pack)
2166
resumed_pack.finish()
2167
self.allocate(resumed_pack)
2168
any_new_content = True
2169
del self._resumed_packs[:]
2171
result = self.autopack()
1244
if not self.autopack():
2173
1245
# when autopack takes no steps, the names list is still
2175
return self._save_pack_names()
2179
def _suspend_write_group(self):
2180
tokens = [pack.name for pack in self._resumed_packs]
2181
self._remove_pack_indices(self._new_pack)
2182
if self._new_pack.data_inserted():
2183
# get all the data to disk and read to use
2184
self._new_pack.finish(suspend=True)
2185
tokens.append(self._new_pack.name)
2186
self._new_pack = None
1247
self._save_pack_names()
2188
1249
self._new_pack.abort()
2189
self._new_pack = None
2190
self._remove_resumed_pack_indices()
2193
def _resume_write_group(self, tokens):
2194
for token in tokens:
2195
self._resume_pack(token)
1250
self.repo._text_knit = None
1253
class KnitPackRevisionStore(KnitRevisionStore):
1254
"""An object to adapt access from RevisionStore's to use KnitPacks.
1256
This class works by replacing the original RevisionStore.
1257
We need to do this because the KnitPackRevisionStore is less
1258
isolated in its layering - it uses services from the repo.
1261
def __init__(self, repo, transport, revisionstore):
1262
"""Create a KnitPackRevisionStore on repo with revisionstore.
1264
This will store its state in the Repository, use the
1265
indices to provide a KnitGraphIndex,
1266
and at the end of transactions write new indices.
1268
KnitRevisionStore.__init__(self, revisionstore.versioned_file_store)
1270
self._serializer = revisionstore._serializer
1271
self.transport = transport
1273
def get_revision_file(self, transaction):
1274
"""Get the revision versioned file object."""
1275
if getattr(self.repo, '_revision_knit', None) is not None:
1276
return self.repo._revision_knit
1277
self.repo._pack_collection.ensure_loaded()
1278
add_callback = self.repo._pack_collection.revision_index.add_callback
1279
# setup knit specific objects
1280
knit_index = KnitGraphIndex(
1281
self.repo._pack_collection.revision_index.combined_index,
1282
add_callback=add_callback)
1283
self.repo._revision_knit = knit.KnitVersionedFile(
1284
'revisions', self.transport.clone('..'),
1285
self.repo.control_files._file_mode,
1286
create=False, access_mode=self.repo._access_mode(),
1287
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1288
access_method=self.repo._pack_collection.revision_index.knit_access)
1289
return self.repo._revision_knit
1291
def get_signature_file(self, transaction):
1292
"""Get the signature versioned file object."""
1293
if getattr(self.repo, '_signature_knit', None) is not None:
1294
return self.repo._signature_knit
1295
self.repo._pack_collection.ensure_loaded()
1296
add_callback = self.repo._pack_collection.signature_index.add_callback
1297
# setup knit specific objects
1298
knit_index = KnitGraphIndex(
1299
self.repo._pack_collection.signature_index.combined_index,
1300
add_callback=add_callback, parents=False)
1301
self.repo._signature_knit = knit.KnitVersionedFile(
1302
'signatures', self.transport.clone('..'),
1303
self.repo.control_files._file_mode,
1304
create=False, access_mode=self.repo._access_mode(),
1305
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1306
access_method=self.repo._pack_collection.signature_index.knit_access)
1307
return self.repo._signature_knit
1310
class KnitPackTextStore(VersionedFileStore):
1311
"""Presents a TextStore abstraction on top of packs.
1313
This class works by replacing the original VersionedFileStore.
1314
We need to do this because the KnitPackRevisionStore is less
1315
isolated in its layering - it uses services from the repo and shares them
1316
with all the data written in a single write group.
1319
def __init__(self, repo, transport, weavestore):
1320
"""Create a KnitPackTextStore on repo with weavestore.
1322
This will store its state in the Repository, use the
1323
indices FileNames to provide a KnitGraphIndex,
1324
and at the end of transactions write new indices.
1326
# don't call base class constructor - it's not suitable.
1327
# no transient data stored in the transaction
1329
self._precious = False
1331
self.transport = transport
1332
self.weavestore = weavestore
1333
# XXX for check() which isn't updated yet
1334
self._transport = weavestore._transport
1336
def get_weave_or_empty(self, file_id, transaction):
1337
"""Get a 'Knit' backed by the .tix indices.
1339
The transaction parameter is ignored.
1341
self.repo._pack_collection.ensure_loaded()
1342
add_callback = self.repo._pack_collection.text_index.add_callback
1343
# setup knit specific objects
1344
file_id_index = GraphIndexPrefixAdapter(
1345
self.repo._pack_collection.text_index.combined_index,
1346
(file_id, ), 1, add_nodes_callback=add_callback)
1347
knit_index = KnitGraphIndex(file_id_index,
1348
add_callback=file_id_index.add_nodes,
1349
deltas=True, parents=True)
1350
return knit.KnitVersionedFile('text:' + file_id,
1351
self.transport.clone('..'),
1354
access_method=self.repo._pack_collection.text_index.knit_access,
1355
factory=knit.KnitPlainFactory())
1357
get_weave = get_weave_or_empty
1360
"""Generate a list of the fileids inserted, for use by check."""
1361
self.repo._pack_collection.ensure_loaded()
1363
for index, key, value, refs in \
1364
self.repo._pack_collection.text_index.combined_index.iter_all_entries():
1369
class InventoryKnitThunk(object):
1370
"""An object to manage thunking get_inventory_weave to pack based knits."""
1372
def __init__(self, repo, transport):
1373
"""Create an InventoryKnitThunk for repo at transport.
1375
This will store its state in the Repository, use the
1376
indices FileNames to provide a KnitGraphIndex,
1377
and at the end of transactions write a new index..
1380
self.transport = transport
1382
def get_weave(self):
1383
"""Get a 'Knit' that contains inventory data."""
1384
self.repo._pack_collection.ensure_loaded()
1385
add_callback = self.repo._pack_collection.inventory_index.add_callback
1386
# setup knit specific objects
1387
knit_index = KnitGraphIndex(
1388
self.repo._pack_collection.inventory_index.combined_index,
1389
add_callback=add_callback, deltas=True, parents=True)
1390
return knit.KnitVersionedFile(
1391
'inventory', self.transport.clone('..'),
1392
self.repo.control_files._file_mode,
1393
create=False, access_mode=self.repo._access_mode(),
1394
index=knit_index, delta=True, factory=knit.KnitPlainFactory(),
1395
access_method=self.repo._pack_collection.inventory_index.knit_access)
2198
1398
class KnitPackRepository(KnitRepository):
2199
"""Repository with knit objects stored inside pack containers.
2201
The layering for a KnitPackRepository is:
2203
Graph | HPSS | Repository public layer |
2204
===================================================
2205
Tuple based apis below, string based, and key based apis above
2206
---------------------------------------------------
2208
Provides .texts, .revisions etc
2209
This adapts the N-tuple keys to physical knit records which only have a
2210
single string identifier (for historical reasons), which in older formats
2211
was always the revision_id, and in the mapped code for packs is always
2212
the last element of key tuples.
2213
---------------------------------------------------
2215
A separate GraphIndex is used for each of the
2216
texts/inventories/revisions/signatures contained within each individual
2217
pack file. The GraphIndex layer works in N-tuples and is unaware of any
2219
===================================================
2223
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
1399
"""Experimental graph-knit using repository."""
1401
def __init__(self, _format, a_bzrdir, control_files, _revision_store,
1402
control_store, text_store, _commit_builder_class, _serializer):
2225
1403
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2226
_commit_builder_class, _serializer)
2227
index_transport = self._transport.clone('indices')
2228
self._pack_collection = RepositoryPackCollection(self, self._transport,
1404
_revision_store, control_store, text_store, _commit_builder_class,
1406
index_transport = control_files._transport.clone('indices')
1407
self._pack_collection = RepositoryPackCollection(self, control_files._transport,
2229
1408
index_transport,
2230
self._transport.clone('upload'),
2231
self._transport.clone('packs'),
2232
_format.index_builder_class,
2233
_format.index_class,
2234
use_chk_index=self._format.supports_chks,
2236
self.inventories = KnitVersionedFiles(
2237
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2238
add_callback=self._pack_collection.inventory_index.add_callback,
2239
deltas=True, parents=True, is_locked=self.is_locked),
2240
data_access=self._pack_collection.inventory_index.data_access,
2241
max_delta_chain=200)
2242
self.revisions = KnitVersionedFiles(
2243
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2244
add_callback=self._pack_collection.revision_index.add_callback,
2245
deltas=False, parents=True, is_locked=self.is_locked,
2246
track_external_parent_refs=True),
2247
data_access=self._pack_collection.revision_index.data_access,
2249
self.signatures = KnitVersionedFiles(
2250
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2251
add_callback=self._pack_collection.signature_index.add_callback,
2252
deltas=False, parents=False, is_locked=self.is_locked),
2253
data_access=self._pack_collection.signature_index.data_access,
2255
self.texts = KnitVersionedFiles(
2256
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2257
add_callback=self._pack_collection.text_index.add_callback,
2258
deltas=True, parents=True, is_locked=self.is_locked),
2259
data_access=self._pack_collection.text_index.data_access,
2260
max_delta_chain=200)
2261
if _format.supports_chks:
2262
# No graph, no compression:- references from chks are between
2263
# different objects not temporal versions of the same; and without
2264
# some sort of temporal structure knit compression will just fail.
2265
self.chk_bytes = KnitVersionedFiles(
2266
_KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2267
add_callback=self._pack_collection.chk_index.add_callback,
2268
deltas=False, parents=False, is_locked=self.is_locked),
2269
data_access=self._pack_collection.chk_index.data_access,
2272
self.chk_bytes = None
1409
control_files._transport.clone('upload'),
1410
control_files._transport.clone('packs'))
1411
self._revision_store = KnitPackRevisionStore(self, index_transport, self._revision_store)
1412
self.weave_store = KnitPackTextStore(self, index_transport, self.weave_store)
1413
self._inv_thunk = InventoryKnitThunk(self, index_transport)
2273
1414
# True when the repository object is 'write locked' (as opposed to the
2274
# physical lock only taken out around changes to the pack-names list.)
1415
# physical lock only taken out around changes to the pack-names list.)
2275
1416
# Another way to represent this would be a decorator around the control
2276
1417
# files object that presents logical locks as physical ones - if this
2277
1418
# gets ugly consider that alternative design. RBC 20071011
2278
1419
self._write_lock_count = 0
2279
1420
self._transaction = None
2281
self._reconcile_does_inventory_gc = True
2282
self._reconcile_fixes_text_parents = True
2283
self._reconcile_backsup_inventory = False
2285
def _warn_if_deprecated(self, branch=None):
2286
# This class isn't deprecated, but one sub-format is
2287
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2288
super(KnitPackRepository, self)._warn_if_deprecated(branch)
1422
self._reconcile_does_inventory_gc = False
1423
self._reconcile_fixes_text_parents = False
2290
1425
def _abort_write_group(self):
2291
self.revisions._index._key_dependencies.clear()
2292
1426
self._pack_collection._abort_write_group()
2294
def _get_source(self, to_format):
2295
if to_format.network_name() == self._format.network_name():
2296
return KnitPackStreamSource(self, to_format)
2297
return super(KnitPackRepository, self)._get_source(to_format)
1428
def _access_mode(self):
1429
"""Return 'w' or 'r' for depending on whether a write lock is active.
1431
This method is a helper for the Knit-thunking support objects.
1433
if self.is_write_locked():
1437
def get_parents(self, revision_ids):
1438
"""See StackedParentsProvider.get_parents.
1440
This implementation accesses the combined revision index to provide
1443
index = self._pack_collection.revision_index.combined_index
1445
for revision_id in revision_ids:
1446
if revision_id != _mod_revision.NULL_REVISION:
1447
search_keys.add((revision_id,))
1448
found_parents = {_mod_revision.NULL_REVISION:[]}
1449
for index, key, value, refs in index.iter_entries(search_keys):
1452
parents = (_mod_revision.NULL_REVISION,)
1454
parents = tuple(parent[0] for parent in parents)
1455
found_parents[key[0]] = parents
1457
for revision_id in revision_ids:
1459
result.append(found_parents[revision_id])
2299
1464
def _make_parents_provider(self):
2300
return graph.CachingParentsProvider(self)
2302
1467
def _refresh_data(self):
2303
if not self.is_locked():
2305
self._pack_collection.reload_pack_names()
1468
if self._write_lock_count == 1 or self.control_files._lock_count == 1:
1469
# forget what names there are
1470
self._pack_collection.reset()
1471
# XXX: Better to do an in-memory merge when acquiring a new lock -
1472
# factor out code from _save_pack_names.
2307
1474
def _start_write_group(self):
2308
1475
self._pack_collection._start_write_group()
2310
1477
def _commit_write_group(self):
2311
hint = self._pack_collection._commit_write_group()
2312
self.revisions._index._key_dependencies.clear()
2315
def suspend_write_group(self):
2316
# XXX check self._write_group is self.get_transaction()?
2317
tokens = self._pack_collection._suspend_write_group()
2318
self.revisions._index._key_dependencies.clear()
2319
self._write_group = None
2322
def _resume_write_group(self, tokens):
2323
self._start_write_group()
2325
self._pack_collection._resume_write_group(tokens)
2326
except errors.UnresumableWriteGroup:
2327
self._abort_write_group()
2329
for pack in self._pack_collection._resumed_packs:
2330
self.revisions._index.scan_unvalidated_index(pack.revision_index)
1478
return self._pack_collection._commit_write_group()
1480
def get_inventory_weave(self):
1481
return self._inv_thunk.get_weave()
2332
1483
def get_transaction(self):
2333
1484
if self._write_lock_count:
2654
1731
def get_format_description(self):
2655
1732
"""See RepositoryFormat.get_format_description()."""
2656
1733
return "Packs containing knits with subtree support\n"
2659
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2660
"""A rich-root, no subtrees parameterized Pack repository.
2662
This repository format uses the xml6 serializer to get:
2663
- support for recording full info about the tree root
2665
This format was introduced in 1.0.
2668
repository_class = KnitPackRepository
2669
_commit_builder_class = PackRootCommitBuilder
2670
rich_root_data = True
2671
supports_tree_reference = False
2673
def _serializer(self):
2674
return xml6.serializer_v6
2675
# What index classes to use
2676
index_builder_class = InMemoryGraphIndex
2677
index_class = GraphIndex
2679
def _get_matching_bzrdir(self):
2680
return bzrdir.format_registry.make_bzrdir(
2683
def _ignore_setting_bzrdir(self, format):
2686
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2688
def get_format_string(self):
2689
"""See RepositoryFormat.get_format_string()."""
2690
return ("Bazaar pack repository format 1 with rich root"
2691
" (needs bzr 1.0)\n")
2693
def get_format_description(self):
2694
"""See RepositoryFormat.get_format_description()."""
2695
return "Packs containing knits with rich root support\n"
2698
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2699
"""Repository that supports external references to allow stacking.
2703
Supports external lookups, which results in non-truncated ghosts after
2704
reconcile compared to pack-0.92 formats.
2707
repository_class = KnitPackRepository
2708
_commit_builder_class = PackCommitBuilder
2709
supports_external_lookups = True
2710
# What index classes to use
2711
index_builder_class = InMemoryGraphIndex
2712
index_class = GraphIndex
2715
def _serializer(self):
2716
return xml5.serializer_v5
2718
def _get_matching_bzrdir(self):
2719
return bzrdir.format_registry.make_bzrdir('1.6')
2721
def _ignore_setting_bzrdir(self, format):
2724
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2726
def get_format_string(self):
2727
"""See RepositoryFormat.get_format_string()."""
2728
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2730
def get_format_description(self):
2731
"""See RepositoryFormat.get_format_description()."""
2732
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2735
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2736
"""A repository with rich roots and stacking.
2738
New in release 1.6.1.
2740
Supports stacking on other repositories, allowing data to be accessed
2741
without being stored locally.
2744
repository_class = KnitPackRepository
2745
_commit_builder_class = PackRootCommitBuilder
2746
rich_root_data = True
2747
supports_tree_reference = False # no subtrees
2748
supports_external_lookups = True
2749
# What index classes to use
2750
index_builder_class = InMemoryGraphIndex
2751
index_class = GraphIndex
2754
def _serializer(self):
2755
return xml6.serializer_v6
2757
def _get_matching_bzrdir(self):
2758
return bzrdir.format_registry.make_bzrdir(
2761
def _ignore_setting_bzrdir(self, format):
2764
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2766
def get_format_string(self):
2767
"""See RepositoryFormat.get_format_string()."""
2768
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2770
def get_format_description(self):
2771
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2774
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2775
"""A repository with rich roots and external references.
2779
Supports external lookups, which results in non-truncated ghosts after
2780
reconcile compared to pack-0.92 formats.
2782
This format was deprecated because the serializer it uses accidentally
2783
supported subtrees, when the format was not intended to. This meant that
2784
someone could accidentally fetch from an incorrect repository.
2787
repository_class = KnitPackRepository
2788
_commit_builder_class = PackRootCommitBuilder
2789
rich_root_data = True
2790
supports_tree_reference = False # no subtrees
2792
supports_external_lookups = True
2793
# What index classes to use
2794
index_builder_class = InMemoryGraphIndex
2795
index_class = GraphIndex
2798
def _serializer(self):
2799
return xml7.serializer_v7
2801
def _get_matching_bzrdir(self):
2802
matching = bzrdir.format_registry.make_bzrdir(
2804
matching.repository_format = self
2807
def _ignore_setting_bzrdir(self, format):
2810
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2812
def get_format_string(self):
2813
"""See RepositoryFormat.get_format_string()."""
2814
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2816
def get_format_description(self):
2817
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2821
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2822
"""A repository with stacking and btree indexes,
2823
without rich roots or subtrees.
2825
This is equivalent to pack-1.6 with B+Tree indices.
2828
repository_class = KnitPackRepository
2829
_commit_builder_class = PackCommitBuilder
2830
supports_external_lookups = True
2831
# What index classes to use
2832
index_builder_class = BTreeBuilder
2833
index_class = BTreeGraphIndex
2836
def _serializer(self):
2837
return xml5.serializer_v5
2839
def _get_matching_bzrdir(self):
2840
return bzrdir.format_registry.make_bzrdir('1.9')
2842
def _ignore_setting_bzrdir(self, format):
2845
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2847
def get_format_string(self):
2848
"""See RepositoryFormat.get_format_string()."""
2849
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2851
def get_format_description(self):
2852
"""See RepositoryFormat.get_format_description()."""
2853
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2856
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2857
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2859
1.6-rich-root with B+Tree indices.
2862
repository_class = KnitPackRepository
2863
_commit_builder_class = PackRootCommitBuilder
2864
rich_root_data = True
2865
supports_tree_reference = False # no subtrees
2866
supports_external_lookups = True
2867
# What index classes to use
2868
index_builder_class = BTreeBuilder
2869
index_class = BTreeGraphIndex
2872
def _serializer(self):
2873
return xml6.serializer_v6
2875
def _get_matching_bzrdir(self):
2876
return bzrdir.format_registry.make_bzrdir(
2879
def _ignore_setting_bzrdir(self, format):
2882
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2884
def get_format_string(self):
2885
"""See RepositoryFormat.get_format_string()."""
2886
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2888
def get_format_description(self):
2889
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2892
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2893
"""A subtrees development repository.
2895
This format should be retained until the second release after bzr 1.7.
2897
1.6.1-subtree[as it might have been] with B+Tree indices.
2899
This is [now] retained until we have a CHK based subtree format in
2903
repository_class = KnitPackRepository
2904
_commit_builder_class = PackRootCommitBuilder
2905
rich_root_data = True
2907
supports_tree_reference = True
2908
supports_external_lookups = True
2909
# What index classes to use
2910
index_builder_class = BTreeBuilder
2911
index_class = BTreeGraphIndex
2914
def _serializer(self):
2915
return xml7.serializer_v7
2917
def _get_matching_bzrdir(self):
2918
return bzrdir.format_registry.make_bzrdir(
2919
'development-subtree')
2921
def _ignore_setting_bzrdir(self, format):
2924
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2926
def get_format_string(self):
2927
"""See RepositoryFormat.get_format_string()."""
2928
return ("Bazaar development format 2 with subtree support "
2929
"(needs bzr.dev from before 1.8)\n")
2931
def get_format_description(self):
2932
"""See RepositoryFormat.get_format_description()."""
2933
return ("Development repository format, currently the same as "
2934
"1.6.1-subtree with B+Tree indices.\n")
added
removed
bzrlib/repofmt/pack_repo.py
