261
170
return not self.__eq__(other)
263
172
def __repr__(self):
264
return "<%s.%s object at 0x%x, %s, %s" % (
265
self.__class__.__module__, self.__class__.__name__, id(self),
266
self.pack_transport, self.name)
269
class ResumedPack(ExistingPack):
271
def __init__(self, name, revision_index, inventory_index, text_index,
272
signature_index, upload_transport, pack_transport, index_transport,
273
pack_collection, chk_index=None):
274
"""Create a ResumedPack object."""
275
ExistingPack.__init__(self, pack_transport, name, revision_index,
276
inventory_index, text_index, signature_index,
278
self.upload_transport = upload_transport
279
self.index_transport = index_transport
280
self.index_sizes = [None, None, None, None]
282
('revision', revision_index),
283
('inventory', inventory_index),
284
('text', text_index),
285
('signature', signature_index),
287
if chk_index is not None:
288
indices.append(('chk', chk_index))
289
self.index_sizes.append(None)
290
for index_type, index in indices:
291
offset = self.index_offset(index_type)
292
self.index_sizes[offset] = index._size
293
self.index_class = pack_collection._index_class
294
self._pack_collection = pack_collection
295
self._state = 'resumed'
296
# XXX: perhaps check that the .pack file exists?
298
def access_tuple(self):
299
if self._state == 'finished':
300
return Pack.access_tuple(self)
301
elif self._state == 'resumed':
302
return self.upload_transport, self.file_name()
304
raise AssertionError(self._state)
307
self.upload_transport.delete(self.file_name())
308
indices = [self.revision_index, self.inventory_index, self.text_index,
309
self.signature_index]
310
if self.chk_index is not None:
311
indices.append(self.chk_index)
312
for index in indices:
313
index._transport.delete(index._name)
316
self._check_references()
317
index_types = ['revision', 'inventory', 'text', 'signature']
318
if self.chk_index is not None:
319
index_types.append('chk')
320
for index_type in index_types:
321
old_name = self.index_name(index_type, self.name)
322
new_name = '../indices/' + old_name
323
self.upload_transport.rename(old_name, new_name)
324
self._replace_index_with_readonly(index_type)
325
new_name = '../packs/' + self.file_name()
326
self.upload_transport.rename(self.file_name(), new_name)
327
self._state = 'finished'
329
def _get_external_refs(self, index):
330
"""Return compression parents for this index that are not present.
332
This returns any compression parents that are referenced by this index,
333
which are not contained *in* this index. They may be present elsewhere.
335
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)
338
177
class NewPack(Pack):
339
178
"""An in memory proxy for a pack which is being created."""
341
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,
342
191
"""Create a NewPack instance.
344
: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').
345
200
:param upload_suffix: An optional suffix to be given to any temporary
346
201
files created during the pack creation. e.g '.autopack'
347
:param file_mode: Unix permissions for newly created file.
349
203
# The relative locations of the packs are constrained, but all are
350
204
# passed in because the caller has them, so as to avoid object churn.
351
index_builder_class = pack_collection._index_builder_class
352
if pack_collection.chk_index is not None:
353
chk_index = index_builder_class(reference_lists=0)
356
205
Pack.__init__(self,
357
206
# Revisions: parents list, no text compression.
358
index_builder_class(reference_lists=1),
207
InMemoryGraphIndex(reference_lists=1),
359
208
# Inventory: We want to map compression only, but currently the
360
209
# knit code hasn't been updated enough to understand that, so we
361
210
# have a regular 2-list index giving parents and compression
363
index_builder_class(reference_lists=2),
212
InMemoryGraphIndex(reference_lists=2),
364
213
# Texts: compression and per file graph, for all fileids - so two
365
214
# reference lists and two elements in the key tuple.
366
index_builder_class(reference_lists=2, key_elements=2),
215
InMemoryGraphIndex(reference_lists=2, key_elements=2),
367
216
# Signatures: Just blobs to store, no compression, no parents
369
index_builder_class(reference_lists=0),
370
# CHK based storage - just blobs, no compression or parents.
218
InMemoryGraphIndex(reference_lists=0),
373
self._pack_collection = pack_collection
374
# When we make readonly indices, we need this.
375
self.index_class = pack_collection._index_class
376
220
# where should the new pack be opened
377
self.upload_transport = pack_collection._upload_transport
221
self.upload_transport = upload_transport
378
222
# where are indices written out to
379
self.index_transport = pack_collection._index_transport
223
self.index_transport = index_transport
380
224
# where is the pack renamed to when it is finished?
381
self.pack_transport = pack_collection._pack_transport
382
# What file mode to upload the pack and indices with.
383
self._file_mode = file_mode
225
self.pack_transport = pack_transport
384
226
# tracks the content written to the .pack file.
385
self._hash = osutils.md5()
386
# a tuple with the length in bytes of the indices, once the pack
387
# 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)
388
230
self.index_sizes = None
389
231
# How much data to cache when writing packs. Note that this is not
390
232
# synchronised with reads, because it's not in the transport layer, so
593
418
Future searches on the aggregate index will seach this new index
594
419
before all previously inserted indices.
596
421
:param index: An Index for the pack.
597
422
:param pack: A Pack instance.
599
424
# expose it to the index map
600
425
self.index_to_pack[index] = pack.access_tuple()
601
426
# put it at the front of the linear index list
602
self.combined_index.insert_index(0, index, pack.name)
427
self.combined_index.insert_index(0, index)
604
429
def add_writable_index(self, index, pack):
605
430
"""Add an index which is able to have data added to it.
607
There can be at most one writable index at any time. Any
608
modifications made to the knit are put into this index.
610
432
:param index: An index from the pack parameter.
611
433
:param pack: A Pack instance.
613
if self.add_callback is not None:
614
raise AssertionError(
615
"%s already has a writable index through %s" % \
616
(self, self.add_callback))
435
assert self.add_callback is None, \
436
"%s already has a writable index through %s" % \
437
(self, self.add_callback)
617
438
# allow writing: queue writes to a new index
618
439
self.add_index(index, pack)
619
440
# Updates the index to packs mapping as a side effect,
620
self.data_access.set_writer(pack._writer, index, pack.access_tuple())
441
self.knit_access.set_writer(pack._writer, index, pack.access_tuple())
621
442
self.add_callback = index.add_nodes
624
445
"""Reset all the aggregate data to nothing."""
625
self.data_access.set_writer(None, None, (None, None))
446
self.knit_access.set_writer(None, None, (None, None))
626
447
self.index_to_pack.clear()
627
448
del self.combined_index._indices[:]
628
del self.combined_index._index_names[:]
629
449
self.add_callback = None
631
def remove_index(self, index):
451
def remove_index(self, index, pack):
632
452
"""Remove index from the indices used to answer queries.
634
454
:param index: An index from the pack parameter.
455
:param pack: A Pack instance.
636
457
del self.index_to_pack[index]
637
pos = self.combined_index._indices.index(index)
638
del self.combined_index._indices[pos]
639
del self.combined_index._index_names[pos]
458
self.combined_index._indices.remove(index)
640
459
if (self.add_callback is not None and
641
460
getattr(index, 'add_nodes', None) == self.add_callback):
642
461
self.add_callback = None
643
self.data_access.set_writer(None, None, (None, None))
646
class Packer(object):
647
"""Create a pack from packs."""
649
def __init__(self, pack_collection, packs, suffix, revision_ids=None,
653
:param pack_collection: A RepositoryPackCollection object where the
654
new pack is being written to.
655
:param packs: The packs to combine.
656
:param suffix: The suffix to use on the temporary files for the pack.
657
:param revision_ids: Revision ids to limit the pack to.
658
:param reload_func: A function to call if a pack file/index goes
659
missing. The side effect of calling this function should be to
660
update self.packs. See also AggregateIndex
664
self.revision_ids = revision_ids
665
# The pack object we are creating.
667
self._pack_collection = pack_collection
668
self._reload_func = reload_func
669
# The index layer keys for the revisions being copied. None for 'all
671
self._revision_keys = None
672
# What text keys to copy. None for 'all texts'. This is set by
673
# _copy_inventory_texts
674
self._text_filter = None
677
def _extra_init(self):
678
"""A template hook to allow extending the constructor trivially."""
680
def _pack_map_and_index_list(self, index_attribute):
681
"""Convert a list of packs to an index pack map and index list.
683
:param index_attribute: The attribute that the desired index is found
685
:return: A tuple (map, list) where map contains the dict from
686
index:pack_tuple, and list contains the indices in the preferred
691
for pack_obj in self.packs:
692
index = getattr(pack_obj, index_attribute)
693
indices.append(index)
694
pack_map[index] = pack_obj
695
return pack_map, indices
697
def _index_contents(self, indices, key_filter=None):
698
"""Get an iterable of the index contents from a pack_map.
700
:param indices: The list of indices to query
701
:param key_filter: An optional filter to limit the keys returned.
703
all_index = CombinedGraphIndex(indices)
704
if key_filter is None:
705
return all_index.iter_all_entries()
707
return all_index.iter_entries(key_filter)
709
def pack(self, pb=None):
710
"""Create a new pack by reading data from other packs.
712
This does little more than a bulk copy of data. One key difference
713
is that data with the same item key across multiple packs is elided
714
from the output. The new pack is written into the current pack store
715
along with its indices, and the name added to the pack names. The
716
source packs are not altered and are not required to be in the current
719
:param pb: An optional progress bar to use. A nested bar is created if
721
:return: A Pack object, or None if nothing was copied.
723
# open a pack - using the same name as the last temporary file
724
# - which has already been flushed, so its safe.
725
# XXX: - duplicate code warning with start_write_group; fix before
726
# considering 'done'.
727
if self._pack_collection._new_pack is not None:
728
raise errors.BzrError('call to %s.pack() while another pack is'
730
% (self.__class__.__name__,))
731
if self.revision_ids is not None:
732
if len(self.revision_ids) == 0:
733
# silly fetch request.
736
self.revision_ids = frozenset(self.revision_ids)
737
self.revision_keys = frozenset((revid,) for revid in
740
self.pb = ui.ui_factory.nested_progress_bar()
744
return self._create_pack_from_packs()
750
"""Open a pack for the pack we are creating."""
751
new_pack = self._pack_collection.pack_factory(self._pack_collection,
752
upload_suffix=self.suffix,
753
file_mode=self._pack_collection.repo.bzrdir._get_file_mode())
754
# We know that we will process all nodes in order, and don't need to
755
# query, so don't combine any indices spilled to disk until we are done
756
new_pack.revision_index.set_optimize(combine_backing_indices=False)
757
new_pack.inventory_index.set_optimize(combine_backing_indices=False)
758
new_pack.text_index.set_optimize(combine_backing_indices=False)
759
new_pack.signature_index.set_optimize(combine_backing_indices=False)
762
def _update_pack_order(self, entries, index_to_pack_map):
763
"""Determine how we want our packs to be ordered.
765
This changes the sort order of the self.packs list so that packs unused
766
by 'entries' will be at the end of the list, so that future requests
767
can avoid probing them. Used packs will be at the front of the
768
self.packs list, in the order of their first use in 'entries'.
770
:param entries: A list of (index, ...) tuples
771
:param index_to_pack_map: A mapping from index objects to pack objects.
775
for entry in entries:
777
if index not in seen_indexes:
778
packs.append(index_to_pack_map[index])
779
seen_indexes.add(index)
780
if len(packs) == len(self.packs):
781
if 'pack' in debug.debug_flags:
782
mutter('Not changing pack list, all packs used.')
784
seen_packs = set(packs)
785
for pack in self.packs:
786
if pack not in seen_packs:
789
if 'pack' in debug.debug_flags:
790
old_names = [p.access_tuple()[1] for p in self.packs]
791
new_names = [p.access_tuple()[1] for p in packs]
792
mutter('Reordering packs\nfrom: %s\n to: %s',
793
old_names, new_names)
796
def _copy_revision_texts(self):
797
"""Copy revision data to the new pack."""
799
if self.revision_ids:
800
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
803
# select revision keys
804
revision_index_map, revision_indices = self._pack_map_and_index_list(
806
revision_nodes = self._index_contents(revision_indices, revision_keys)
807
revision_nodes = list(revision_nodes)
808
self._update_pack_order(revision_nodes, revision_index_map)
809
# copy revision keys and adjust values
810
self.pb.update("Copying revision texts", 1)
811
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
812
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
813
self.new_pack.revision_index, readv_group_iter, total_items))
814
if 'pack' in debug.debug_flags:
815
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
816
time.ctime(), self._pack_collection._upload_transport.base,
817
self.new_pack.random_name,
818
self.new_pack.revision_index.key_count(),
819
time.time() - self.new_pack.start_time)
820
self._revision_keys = revision_keys
822
def _copy_inventory_texts(self):
823
"""Copy the inventory texts to the new pack.
825
self._revision_keys is used to determine what inventories to copy.
827
Sets self._text_filter appropriately.
829
# select inventory keys
830
inv_keys = self._revision_keys # currently the same keyspace, and note that
831
# querying for keys here could introduce a bug where an inventory item
832
# is missed, so do not change it to query separately without cross
833
# checking like the text key check below.
834
inventory_index_map, inventory_indices = self._pack_map_and_index_list(
836
inv_nodes = self._index_contents(inventory_indices, inv_keys)
837
# copy inventory keys and adjust values
838
# XXX: Should be a helper function to allow different inv representation
840
self.pb.update("Copying inventory texts", 2)
841
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
842
# Only grab the output lines if we will be processing them
843
output_lines = bool(self.revision_ids)
844
inv_lines = self._copy_nodes_graph(inventory_index_map,
845
self.new_pack._writer, self.new_pack.inventory_index,
846
readv_group_iter, total_items, output_lines=output_lines)
847
if self.revision_ids:
848
self._process_inventory_lines(inv_lines)
850
# eat the iterator to cause it to execute.
852
self._text_filter = None
853
if 'pack' in debug.debug_flags:
854
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
855
time.ctime(), self._pack_collection._upload_transport.base,
856
self.new_pack.random_name,
857
self.new_pack.inventory_index.key_count(),
858
time.time() - self.new_pack.start_time)
860
def _copy_text_texts(self):
862
text_index_map, text_nodes = self._get_text_nodes()
863
if self._text_filter is not None:
864
# We could return the keys copied as part of the return value from
865
# _copy_nodes_graph but this doesn't work all that well with the
866
# need to get line output too, so we check separately, and as we're
867
# going to buffer everything anyway, we check beforehand, which
868
# saves reading knit data over the wire when we know there are
870
text_nodes = set(text_nodes)
871
present_text_keys = set(_node[1] for _node in text_nodes)
872
missing_text_keys = set(self._text_filter) - present_text_keys
873
if missing_text_keys:
874
# TODO: raise a specific error that can handle many missing
876
mutter("missing keys during fetch: %r", missing_text_keys)
877
a_missing_key = missing_text_keys.pop()
878
raise errors.RevisionNotPresent(a_missing_key[1],
880
# copy text keys and adjust values
881
self.pb.update("Copying content texts", 3)
882
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
883
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
884
self.new_pack.text_index, readv_group_iter, total_items))
885
self._log_copied_texts()
887
def _create_pack_from_packs(self):
888
self.pb.update("Opening pack", 0, 5)
889
self.new_pack = self.open_pack()
890
new_pack = self.new_pack
891
# buffer data - we won't be reading-back during the pack creation and
892
# this makes a significant difference on sftp pushes.
893
new_pack.set_write_cache_size(1024*1024)
894
if 'pack' in debug.debug_flags:
895
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
896
for a_pack in self.packs]
897
if self.revision_ids is not None:
898
rev_count = len(self.revision_ids)
901
mutter('%s: create_pack: creating pack from source packs: '
902
'%s%s %s revisions wanted %s t=0',
903
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
904
plain_pack_list, rev_count)
905
self._copy_revision_texts()
906
self._copy_inventory_texts()
907
self._copy_text_texts()
908
# select signature keys
909
signature_filter = self._revision_keys # same keyspace
910
signature_index_map, signature_indices = self._pack_map_and_index_list(
912
signature_nodes = self._index_contents(signature_indices,
914
# copy signature keys and adjust values
915
self.pb.update("Copying signature texts", 4)
916
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
917
new_pack.signature_index)
918
if 'pack' in debug.debug_flags:
919
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
920
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
921
new_pack.signature_index.key_count(),
922
time.time() - new_pack.start_time)
924
# NB XXX: how to check CHK references are present? perhaps by yielding
925
# the items? How should that interact with stacked repos?
926
if new_pack.chk_index is not None:
928
if 'pack' in debug.debug_flags:
929
mutter('%s: create_pack: chk content copied: %s%s %d items t+%6.3fs',
930
time.ctime(), self._pack_collection._upload_transport.base,
931
new_pack.random_name,
932
new_pack.chk_index.key_count(),
933
time.time() - new_pack.start_time)
934
new_pack._check_references()
935
if not self._use_pack(new_pack):
938
self.pb.update("Finishing pack", 5)
940
self._pack_collection.allocate(new_pack)
943
def _copy_chks(self, refs=None):
944
# XXX: Todo, recursive follow-pointers facility when fetching some
946
chk_index_map, chk_indices = self._pack_map_and_index_list(
948
chk_nodes = self._index_contents(chk_indices, refs)
950
# TODO: This isn't strictly tasteful as we are accessing some private
951
# variables (_serializer). Perhaps a better way would be to have
952
# Repository._deserialise_chk_node()
953
search_key_func = chk_map.search_key_registry.get(
954
self._pack_collection.repo._serializer.search_key_name)
955
def accumlate_refs(lines):
956
# XXX: move to a generic location
958
bytes = ''.join(lines)
959
node = chk_map._deserialise(bytes, ("unknown",), search_key_func)
960
new_refs.update(node.refs())
961
self._copy_nodes(chk_nodes, chk_index_map, self.new_pack._writer,
962
self.new_pack.chk_index, output_lines=accumlate_refs)
965
def _copy_nodes(self, nodes, index_map, writer, write_index,
967
"""Copy knit nodes between packs with no graph references.
969
:param output_lines: Output full texts of copied items.
971
pb = ui.ui_factory.nested_progress_bar()
973
return self._do_copy_nodes(nodes, index_map, writer,
974
write_index, pb, output_lines=output_lines)
978
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb,
980
# for record verification
981
knit = KnitVersionedFiles(None, None)
982
# plan a readv on each source pack:
984
nodes = sorted(nodes)
985
# how to map this into knit.py - or knit.py into this?
986
# we don't want the typical knit logic, we want grouping by pack
987
# at this point - perhaps a helper library for the following code
988
# duplication points?
990
for index, key, value in nodes:
991
if index not in request_groups:
992
request_groups[index] = []
993
request_groups[index].append((key, value))
995
pb.update("Copied record", record_index, len(nodes))
996
for index, items in request_groups.iteritems():
997
pack_readv_requests = []
998
for key, value in items:
999
# ---- KnitGraphIndex.get_position
1000
bits = value[1:].split(' ')
1001
offset, length = int(bits[0]), int(bits[1])
1002
pack_readv_requests.append((offset, length, (key, value[0])))
1003
# linear scan up the pack
1004
pack_readv_requests.sort()
1006
pack_obj = index_map[index]
1007
transport, path = pack_obj.access_tuple()
1009
reader = pack.make_readv_reader(transport, path,
1010
[offset[0:2] for offset in pack_readv_requests])
1011
except errors.NoSuchFile:
1012
if self._reload_func is not None:
1015
for (names, read_func), (_1, _2, (key, eol_flag)) in \
1016
izip(reader.iter_records(), pack_readv_requests):
1017
raw_data = read_func(None)
1018
# check the header only
1019
if output_lines is not None:
1020
output_lines(knit._parse_record(key[-1], raw_data)[0])
1022
df, _ = knit._parse_record_header(key, raw_data)
1024
pos, size = writer.add_bytes_record(raw_data, names)
1025
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
1026
pb.update("Copied record", record_index)
1029
def _copy_nodes_graph(self, index_map, writer, write_index,
1030
readv_group_iter, total_items, output_lines=False):
1031
"""Copy knit nodes between packs.
1033
:param output_lines: Return lines present in the copied data as
1034
an iterator of line,version_id.
1036
pb = ui.ui_factory.nested_progress_bar()
1038
for result in self._do_copy_nodes_graph(index_map, writer,
1039
write_index, output_lines, pb, readv_group_iter, total_items):
1042
# Python 2.4 does not permit try:finally: in a generator.
1048
def _do_copy_nodes_graph(self, index_map, writer, write_index,
1049
output_lines, pb, readv_group_iter, total_items):
1050
# for record verification
1051
knit = KnitVersionedFiles(None, None)
1052
# for line extraction when requested (inventories only)
1054
factory = KnitPlainFactory()
1056
pb.update("Copied record", record_index, total_items)
1057
for index, readv_vector, node_vector in readv_group_iter:
1059
pack_obj = index_map[index]
1060
transport, path = pack_obj.access_tuple()
1062
reader = pack.make_readv_reader(transport, path, readv_vector)
1063
except errors.NoSuchFile:
1064
if self._reload_func is not None:
1067
for (names, read_func), (key, eol_flag, references) in \
1068
izip(reader.iter_records(), node_vector):
1069
raw_data = read_func(None)
1071
# read the entire thing
1072
content, _ = knit._parse_record(key[-1], raw_data)
1073
if len(references[-1]) == 0:
1074
line_iterator = factory.get_fulltext_content(content)
1076
line_iterator = factory.get_linedelta_content(content)
1077
for line in line_iterator:
1080
# check the header only
1081
df, _ = knit._parse_record_header(key, raw_data)
1083
pos, size = writer.add_bytes_record(raw_data, names)
1084
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1085
pb.update("Copied record", record_index)
1088
def _get_text_nodes(self):
1089
text_index_map, text_indices = self._pack_map_and_index_list(
1091
return text_index_map, self._index_contents(text_indices,
1094
def _least_readv_node_readv(self, nodes):
1095
"""Generate request groups for nodes using the least readv's.
1097
:param nodes: An iterable of graph index nodes.
1098
:return: Total node count and an iterator of the data needed to perform
1099
readvs to obtain the data for nodes. Each item yielded by the
1100
iterator is a tuple with:
1101
index, readv_vector, node_vector. readv_vector is a list ready to
1102
hand to the transport readv method, and node_vector is a list of
1103
(key, eol_flag, references) for the node retrieved by the
1104
matching readv_vector.
1106
# group by pack so we do one readv per pack
1107
nodes = sorted(nodes)
1110
for index, key, value, references in nodes:
1111
if index not in request_groups:
1112
request_groups[index] = []
1113
request_groups[index].append((key, value, references))
1115
for index, items in request_groups.iteritems():
1116
pack_readv_requests = []
1117
for key, value, references in items:
1118
# ---- KnitGraphIndex.get_position
1119
bits = value[1:].split(' ')
1120
offset, length = int(bits[0]), int(bits[1])
1121
pack_readv_requests.append(
1122
((offset, length), (key, value[0], references)))
1123
# linear scan up the pack to maximum range combining.
1124
pack_readv_requests.sort()
1125
# split out the readv and the node data.
1126
pack_readv = [readv for readv, node in pack_readv_requests]
1127
node_vector = [node for readv, node in pack_readv_requests]
1128
result.append((index, pack_readv, node_vector))
1129
return total, result
1131
def _log_copied_texts(self):
1132
if 'pack' in debug.debug_flags:
1133
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
1134
time.ctime(), self._pack_collection._upload_transport.base,
1135
self.new_pack.random_name,
1136
self.new_pack.text_index.key_count(),
1137
time.time() - self.new_pack.start_time)
1139
def _process_inventory_lines(self, inv_lines):
1140
"""Use up the inv_lines generator and setup a text key filter."""
1141
repo = self._pack_collection.repo
1142
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
1143
inv_lines, self.revision_keys)
1145
for fileid, file_revids in fileid_revisions.iteritems():
1146
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
1147
self._text_filter = text_filter
1149
def _revision_node_readv(self, revision_nodes):
1150
"""Return the total revisions and the readv's to issue.
1152
:param revision_nodes: The revision index contents for the packs being
1153
incorporated into the new pack.
1154
:return: As per _least_readv_node_readv.
1156
return self._least_readv_node_readv(revision_nodes)
1158
def _use_pack(self, new_pack):
1159
"""Return True if new_pack should be used.
1161
:param new_pack: The pack that has just been created.
1162
:return: True if the pack should be used.
1164
return new_pack.data_inserted()
1167
class OptimisingPacker(Packer):
1168
"""A packer which spends more time to create better disk layouts."""
1170
def _revision_node_readv(self, revision_nodes):
1171
"""Return the total revisions and the readv's to issue.
1173
This sort places revisions in topological order with the ancestors
1176
:param revision_nodes: The revision index contents for the packs being
1177
incorporated into the new pack.
1178
:return: As per _least_readv_node_readv.
1180
# build an ancestors dict
1183
for index, key, value, references in revision_nodes:
1184
ancestors[key] = references[0]
1185
by_key[key] = (index, value, references)
1186
order = tsort.topo_sort(ancestors)
1188
# Single IO is pathological, but it will work as a starting point.
1190
for key in reversed(order):
1191
index, value, references = by_key[key]
1192
# ---- KnitGraphIndex.get_position
1193
bits = value[1:].split(' ')
1194
offset, length = int(bits[0]), int(bits[1])
1196
(index, [(offset, length)], [(key, value[0], references)]))
1197
# TODO: combine requests in the same index that are in ascending order.
1198
return total, requests
1200
def open_pack(self):
1201
"""Open a pack for the pack we are creating."""
1202
new_pack = super(OptimisingPacker, self).open_pack()
1203
# Turn on the optimization flags for all the index builders.
1204
new_pack.revision_index.set_optimize(for_size=True)
1205
new_pack.inventory_index.set_optimize(for_size=True)
1206
new_pack.text_index.set_optimize(for_size=True)
1207
new_pack.signature_index.set_optimize(for_size=True)
1211
class ReconcilePacker(Packer):
1212
"""A packer which regenerates indices etc as it copies.
1214
This is used by ``bzr reconcile`` to cause parent text pointers to be
1218
def _extra_init(self):
1219
self._data_changed = False
1221
def _process_inventory_lines(self, inv_lines):
1222
"""Generate a text key reference map rather for reconciling with."""
1223
repo = self._pack_collection.repo
1224
refs = repo._find_text_key_references_from_xml_inventory_lines(
1226
self._text_refs = refs
1227
# during reconcile we:
1228
# - convert unreferenced texts to full texts
1229
# - correct texts which reference a text not copied to be full texts
1230
# - copy all others as-is but with corrected parents.
1231
# - so at this point we don't know enough to decide what becomes a full
1233
self._text_filter = None
1235
def _copy_text_texts(self):
1236
"""generate what texts we should have and then copy."""
1237
self.pb.update("Copying content texts", 3)
1238
# we have three major tasks here:
1239
# 1) generate the ideal index
1240
repo = self._pack_collection.repo
1241
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1242
_1, key, _2, refs in
1243
self.new_pack.revision_index.iter_all_entries()])
1244
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1245
# 2) generate a text_nodes list that contains all the deltas that can
1246
# be used as-is, with corrected parents.
1249
discarded_nodes = []
1250
NULL_REVISION = _mod_revision.NULL_REVISION
1251
text_index_map, text_nodes = self._get_text_nodes()
1252
for node in text_nodes:
1258
ideal_parents = tuple(ideal_index[node[1]])
1260
discarded_nodes.append(node)
1261
self._data_changed = True
1263
if ideal_parents == (NULL_REVISION,):
1265
if ideal_parents == node[3][0]:
1267
ok_nodes.append(node)
1268
elif ideal_parents[0:1] == node[3][0][0:1]:
1269
# the left most parent is the same, or there are no parents
1270
# today. Either way, we can preserve the representation as
1271
# long as we change the refs to be inserted.
1272
self._data_changed = True
1273
ok_nodes.append((node[0], node[1], node[2],
1274
(ideal_parents, node[3][1])))
1275
self._data_changed = True
1277
# Reinsert this text completely
1278
bad_texts.append((node[1], ideal_parents))
1279
self._data_changed = True
1280
# we're finished with some data.
1283
# 3) bulk copy the ok data
1284
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1285
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1286
self.new_pack.text_index, readv_group_iter, total_items))
1287
# 4) adhoc copy all the other texts.
1288
# We have to topologically insert all texts otherwise we can fail to
1289
# reconcile when parts of a single delta chain are preserved intact,
1290
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1291
# reinserted, and if d3 has incorrect parents it will also be
1292
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1293
# copied), so we will try to delta, but d2 is not currently able to be
1294
# extracted because it's basis d1 is not present. Topologically sorting
1295
# addresses this. The following generates a sort for all the texts that
1296
# are being inserted without having to reference the entire text key
1297
# space (we only topo sort the revisions, which is smaller).
1298
topo_order = tsort.topo_sort(ancestors)
1299
rev_order = dict(zip(topo_order, range(len(topo_order))))
1300
bad_texts.sort(key=lambda key:rev_order.get(key[0][1], 0))
1301
transaction = repo.get_transaction()
1302
file_id_index = GraphIndexPrefixAdapter(
1303
self.new_pack.text_index,
1305
add_nodes_callback=self.new_pack.text_index.add_nodes)
1306
data_access = _DirectPackAccess(
1307
{self.new_pack.text_index:self.new_pack.access_tuple()})
1308
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1309
self.new_pack.access_tuple())
1310
output_texts = KnitVersionedFiles(
1311
_KnitGraphIndex(self.new_pack.text_index,
1312
add_callback=self.new_pack.text_index.add_nodes,
1313
deltas=True, parents=True, is_locked=repo.is_locked),
1314
data_access=data_access, max_delta_chain=200)
1315
for key, parent_keys in bad_texts:
1316
# We refer to the new pack to delta data being output.
1317
# A possible improvement would be to catch errors on short reads
1318
# and only flush then.
1319
self.new_pack.flush()
1321
for parent_key in parent_keys:
1322
if parent_key[0] != key[0]:
1323
# Graph parents must match the fileid
1324
raise errors.BzrError('Mismatched key parent %r:%r' %
1326
parents.append(parent_key[1])
1327
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1328
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1329
output_texts.add_lines(key, parent_keys, text_lines,
1330
random_id=True, check_content=False)
1331
# 5) check that nothing inserted has a reference outside the keyspace.
1332
missing_text_keys = self.new_pack.text_index._external_references()
1333
if missing_text_keys:
1334
raise errors.BzrCheckError('Reference to missing compression parents %r'
1335
% (missing_text_keys,))
1336
self._log_copied_texts()
1338
def _use_pack(self, new_pack):
1339
"""Override _use_pack to check for reconcile having changed content."""
1340
# XXX: we might be better checking this at the copy time.
1341
original_inventory_keys = set()
1342
inv_index = self._pack_collection.inventory_index.combined_index
1343
for entry in inv_index.iter_all_entries():
1344
original_inventory_keys.add(entry[1])
1345
new_inventory_keys = set()
1346
for entry in new_pack.inventory_index.iter_all_entries():
1347
new_inventory_keys.add(entry[1])
1348
if new_inventory_keys != original_inventory_keys:
1349
self._data_changed = True
1350
return new_pack.data_inserted() and self._data_changed
462
self.knit_access.set_writer(None, None, (None, None))
1353
465
class RepositoryPackCollection(object):
1354
"""Management of packs within a repository.
1356
:ivar _names: map of {pack_name: (index_size,)}
1359
pack_factory = NewPack
1360
resumed_pack_factory = ResumedPack
466
"""Management of packs within a repository."""
1362
468
def __init__(self, repo, transport, index_transport, upload_transport,
1363
pack_transport, index_builder_class, index_class,
1365
470
"""Create a new RepositoryPackCollection.
1367
:param transport: Addresses the repository base directory
472
:param transport: Addresses the repository base directory
1368
473
(typically .bzr/repository/).
1369
474
:param index_transport: Addresses the directory containing indices.
1370
475
:param upload_transport: Addresses the directory into which packs are written
1371
476
while they're being created.
1372
477
:param pack_transport: Addresses the directory of existing complete packs.
1373
:param index_builder_class: The index builder class to use.
1374
:param index_class: The index class to use.
1375
:param use_chk_index: Whether to setup and manage a CHK index.
1377
# XXX: This should call self.reset()
1378
479
self.repo = repo
1379
480
self.transport = transport
1380
481
self._index_transport = index_transport
1381
482
self._upload_transport = upload_transport
1382
483
self._pack_transport = pack_transport
1383
self._index_builder_class = index_builder_class
1384
self._index_class = index_class
1385
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3,
484
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
1388
486
# name:Pack mapping
1390
self._packs_by_name = {}
1391
488
# the previous pack-names content
1392
489
self._packs_at_load = None
1393
490
# when a pack is being created by this object, the state of that pack.
1394
491
self._new_pack = None
1395
492
# aggregated revision index data
1396
flush = self._flush_new_pack
1397
self.revision_index = AggregateIndex(self.reload_pack_names, flush)
1398
self.inventory_index = AggregateIndex(self.reload_pack_names, flush)
1399
self.text_index = AggregateIndex(self.reload_pack_names, flush)
1400
self.signature_index = AggregateIndex(self.reload_pack_names, flush)
1401
all_indices = [self.revision_index, self.inventory_index,
1402
self.text_index, self.signature_index]
1404
self.chk_index = AggregateIndex(self.reload_pack_names, flush)
1405
all_indices.append(self.chk_index)
1407
# used to determine if we're using a chk_index elsewhere.
1408
self.chk_index = None
1409
# Tell all the CombinedGraphIndex objects about each other, so they can
1410
# share hints about which pack names to search first.
1411
all_combined = [agg_idx.combined_index for agg_idx in all_indices]
1412
for combined_idx in all_combined:
1413
combined_idx.set_sibling_indices(
1414
set(all_combined).difference([combined_idx]))
1416
self._resumed_packs = []
1419
return '%s(%r)' % (self.__class__.__name__, self.repo)
493
self.revision_index = AggregateIndex()
494
self.inventory_index = AggregateIndex()
495
self.text_index = AggregateIndex()
496
self.signature_index = AggregateIndex()
1421
498
def add_pack_to_memory(self, pack):
1422
499
"""Make a Pack object available to the repository to satisfy queries.
1424
501
:param pack: A Pack object.
1426
if pack.name in self._packs_by_name:
1427
raise AssertionError(
1428
'pack %s already in _packs_by_name' % (pack.name,))
1429
503
self.packs.append(pack)
1430
self._packs_by_name[pack.name] = pack
504
assert pack.name not in self._packs
505
self._packs[pack.name] = pack
1431
506
self.revision_index.add_index(pack.revision_index, pack)
1432
507
self.inventory_index.add_index(pack.inventory_index, pack)
1433
508
self.text_index.add_index(pack.text_index, pack)
1434
509
self.signature_index.add_index(pack.signature_index, pack)
1435
if self.chk_index is not None:
1436
self.chk_index.add_index(pack.chk_index, pack)
511
def _add_text_to_weave(self, file_id, revision_id, new_lines, parents,
512
nostore_sha, random_revid):
513
file_id_index = GraphIndexPrefixAdapter(
514
self.text_index.combined_index,
516
add_nodes_callback=self.text_index.add_callback)
517
self.repo._text_knit._index._graph_index = file_id_index
518
self.repo._text_knit._index._add_callback = file_id_index.add_nodes
519
return self.repo._text_knit.add_lines_with_ghosts(
520
revision_id, parents, new_lines, nostore_sha=nostore_sha,
521
random_id=random_revid, check_content=False)[0:2]
1438
523
def all_packs(self):
1439
524
"""Return a list of all the Pack objects this repository has.
1490
571
# group their data with the relevant commit, and that may
1491
572
# involve rewriting ancient history - which autopack tries to
1492
573
# avoid. Alternatively we could not group the data but treat
1493
# each of these as having a single revision, and thus add
574
# each of these as having a single revision, and thus add
1494
575
# one revision for each to the total revision count, to get
1495
576
# a matching distribution.
1497
578
existing_packs.append((revision_count, pack))
1498
579
pack_operations = self.plan_autopack_combinations(
1499
580
existing_packs, pack_distribution)
1500
num_new_packs = len(pack_operations)
1501
num_old_packs = sum([len(po[1]) for po in pack_operations])
1502
num_revs_affected = sum([po[0] for po in pack_operations])
1503
mutter('Auto-packing repository %s, which has %d pack files, '
1504
'containing %d revisions. Packing %d files into %d affecting %d'
1505
' revisions', self, total_packs, total_revisions, num_old_packs,
1506
num_new_packs, num_revs_affected)
1507
result = self._execute_pack_operations(pack_operations,
1508
reload_func=self._restart_autopack)
1509
mutter('Auto-packing repository %s completed', self)
1512
def _execute_pack_operations(self, pack_operations, _packer_class=Packer,
581
self._execute_pack_operations(pack_operations)
584
def create_pack_from_packs(self, packs, suffix, revision_ids=None):
585
"""Create a new pack by reading data from other packs.
587
This does little more than a bulk copy of data. One key difference
588
is that data with the same item key across multiple packs is elided
589
from the output. The new pack is written into the current pack store
590
along with its indices, and the name added to the pack names. The
591
source packs are not altered and are not required to be in the current
594
:param packs: An iterable of Packs to combine.
595
:param revision_ids: Either None, to copy all data, or a list
596
of revision_ids to limit the copied data to the data they
598
:return: A Pack object, or None if nothing was copied.
600
# open a pack - using the same name as the last temporary file
601
# - which has already been flushed, so its safe.
602
# XXX: - duplicate code warning with start_write_group; fix before
603
# considering 'done'.
604
if self._new_pack is not None:
605
raise errors.BzrError('call to create_pack_from_packs while '
606
'another pack is being written.')
607
if revision_ids is not None and len(revision_ids) == 0:
608
# silly fetch request.
610
new_pack = NewPack(self._upload_transport, self._index_transport,
611
self._pack_transport, upload_suffix=suffix)
612
# buffer data - we won't be reading-back during the pack creation and
613
# this makes a significant difference on sftp pushes.
614
new_pack.set_write_cache_size(1024*1024)
615
if 'fetch' in debug.debug_flags:
616
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
618
if revision_ids is not None:
619
rev_count = len(revision_ids)
622
mutter('%s: create_pack: creating pack from source packs: '
623
'%s%s %s revisions wanted %s t=0',
624
time.ctime(), self._upload_transport.base, new_pack.random_name,
625
plain_pack_list, rev_count)
628
revision_keys = [(revision_id,) for revision_id in revision_ids]
632
# select revision keys
633
revision_index_map = self._packs_list_to_pack_map_and_index_list(
634
packs, 'revision_index')[0]
635
revision_nodes = self._index_contents(revision_index_map, revision_keys)
636
# copy revision keys and adjust values
637
list(self._copy_nodes_graph(revision_nodes, revision_index_map,
638
new_pack._writer, new_pack.revision_index))
639
if 'fetch' in debug.debug_flags:
640
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
641
time.ctime(), self._upload_transport.base, new_pack.random_name,
642
new_pack.revision_index.key_count(),
643
time.time() - new_pack.start_time)
644
# select inventory keys
645
inv_keys = revision_keys # currently the same keyspace, and note that
646
# querying for keys here could introduce a bug where an inventory item
647
# is missed, so do not change it to query separately without cross
648
# checking like the text key check below.
649
inventory_index_map = self._packs_list_to_pack_map_and_index_list(
650
packs, 'inventory_index')[0]
651
inv_nodes = self._index_contents(inventory_index_map, inv_keys)
652
# copy inventory keys and adjust values
653
# XXX: Should be a helper function to allow different inv representation
655
inv_lines = self._copy_nodes_graph(inv_nodes, inventory_index_map,
656
new_pack._writer, new_pack.inventory_index, output_lines=True)
658
fileid_revisions = self.repo._find_file_ids_from_xml_inventory_lines(
659
inv_lines, revision_ids)
661
for fileid, file_revids in fileid_revisions.iteritems():
663
[(fileid, file_revid) for file_revid in file_revids])
665
# eat the iterator to cause it to execute.
668
if 'fetch' in debug.debug_flags:
669
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
670
time.ctime(), self._upload_transport.base, new_pack.random_name,
671
new_pack.inventory_index.key_count(),
672
time.time() - new_pack.start_time)
674
text_index_map = self._packs_list_to_pack_map_and_index_list(
675
packs, 'text_index')[0]
676
text_nodes = self._index_contents(text_index_map, text_filter)
677
if text_filter is not None:
678
# We could return the keys copied as part of the return value from
679
# _copy_nodes_graph but this doesn't work all that well with the
680
# need to get line output too, so we check separately, and as we're
681
# going to buffer everything anyway, we check beforehand, which
682
# saves reading knit data over the wire when we know there are
684
text_nodes = set(text_nodes)
685
present_text_keys = set(_node[1] for _node in text_nodes)
686
missing_text_keys = set(text_filter) - present_text_keys
687
if missing_text_keys:
688
# TODO: raise a specific error that can handle many missing
690
a_missing_key = missing_text_keys.pop()
691
raise errors.RevisionNotPresent(a_missing_key[1],
693
# copy text keys and adjust values
694
list(self._copy_nodes_graph(text_nodes, text_index_map,
695
new_pack._writer, new_pack.text_index))
696
if 'fetch' in debug.debug_flags:
697
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
698
time.ctime(), self._upload_transport.base, new_pack.random_name,
699
new_pack.text_index.key_count(),
700
time.time() - new_pack.start_time)
701
# select signature keys
702
signature_filter = revision_keys # same keyspace
703
signature_index_map = self._packs_list_to_pack_map_and_index_list(
704
packs, 'signature_index')[0]
705
signature_nodes = self._index_contents(signature_index_map,
707
# copy signature keys and adjust values
708
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
709
new_pack.signature_index)
710
if 'fetch' in debug.debug_flags:
711
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
712
time.ctime(), self._upload_transport.base, new_pack.random_name,
713
new_pack.signature_index.key_count(),
714
time.time() - new_pack.start_time)
715
if not new_pack.data_inserted():
719
self.allocate(new_pack)
722
def _execute_pack_operations(self, pack_operations):
1514
723
"""Execute a series of pack operations.
1516
725
:param pack_operations: A list of [revision_count, packs_to_combine].
1517
:param _packer_class: The class of packer to use (default: Packer).
1518
:return: The new pack names.
1520
728
for revision_count, packs in pack_operations:
1521
729
# we may have no-ops from the setup logic
1522
730
if len(packs) == 0:
1524
packer = _packer_class(self, packs, '.autopack',
1525
reload_func=reload_func)
1528
except errors.RetryWithNewPacks:
1529
# An exception is propagating out of this context, make sure
1530
# this packer has cleaned up. Packer() doesn't set its new_pack
1531
# state into the RepositoryPackCollection object, so we only
1532
# have access to it directly here.
1533
if packer.new_pack is not None:
1534
packer.new_pack.abort()
732
# have a progress bar?
733
self.create_pack_from_packs(packs, '.autopack')
1536
734
for pack in packs:
1537
self._remove_pack_from_memory(pack)
735
self.remove_pack_from_memory(pack)
1538
736
# record the newly available packs and stop advertising the old
1540
to_be_obsoleted = []
1541
for _, packs in pack_operations:
1542
to_be_obsoleted.extend(packs)
1543
result = self._save_pack_names(clear_obsolete_packs=True,
1544
obsolete_packs=to_be_obsoleted)
1547
def _flush_new_pack(self):
1548
if self._new_pack is not None:
1549
self._new_pack.flush()
738
self._save_pack_names()
739
# Move the old packs out of the way now they are no longer referenced.
740
for revision_count, packs in pack_operations:
741
self._obsolete_packs(packs)
1551
743
def lock_names(self):
1552
744
"""Acquire the mutex around the pack-names index.
1554
746
This cannot be used in the middle of a read-only transaction on the
1557
749
self.repo.control_files.lock_write()
1559
def _already_packed(self):
1560
"""Is the collection already packed?"""
1561
return not (self.repo._format.pack_compresses or (len(self._names) > 1))
1563
def pack(self, hint=None, clean_obsolete_packs=False):
1564
752
"""Pack the pack collection totally."""
1565
753
self.ensure_loaded()
1566
754
total_packs = len(self._names)
1567
if self._already_packed():
1569
757
total_revisions = self.revision_index.combined_index.key_count()
1570
758
# XXX: the following may want to be a class, to pack with a given
1572
760
mutter('Packing repository %s, which has %d pack files, '
1573
'containing %d revisions with hint %r.', self, total_packs,
1574
total_revisions, hint)
761
'containing %d revisions into 1 packs.', self, total_packs,
1575
763
# determine which packs need changing
764
pack_distribution = [1]
1576
765
pack_operations = [[0, []]]
1577
766
for pack in self.all_packs():
1578
if hint is None or pack.name in hint:
1579
# Either no hint was provided (so we are packing everything),
1580
# or this pack was included in the hint.
1581
pack_operations[-1][0] += pack.get_revision_count()
1582
pack_operations[-1][1].append(pack)
1583
self._execute_pack_operations(pack_operations, OptimisingPacker)
1585
if clean_obsolete_packs:
1586
self._clear_obsolete_packs()
767
revision_count = pack.get_revision_count()
768
pack_operations[-1][0] += revision_count
769
pack_operations[-1][1].append(pack)
770
self._execute_pack_operations(pack_operations)
1588
772
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1589
773
"""Plan a pack operation.
1591
775
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1593
:param pack_distribution: A list with the number of revisions desired
777
:parma pack_distribution: A list with the number of revisions desired
1596
780
if len(existing_packs) <= len(pack_distribution):
1825
1028
"""Return a tuple with the transport and file name for a pack name."""
1826
1029
return self._pack_transport, name + '.pack'
1828
def _remove_pack_from_memory(self, pack):
1031
def remove_pack_from_memory(self, pack):
1829
1032
"""Remove pack from the packs accessed by this repository.
1831
1034
Only affects memory state, until self._save_pack_names() is invoked.
1833
1036
self._names.pop(pack.name)
1834
self._packs_by_name.pop(pack.name)
1037
self._packs.pop(pack.name)
1835
1038
self._remove_pack_indices(pack)
1836
self.packs.remove(pack)
1838
def _remove_pack_indices(self, pack, ignore_missing=False):
1839
"""Remove the indices for pack from the aggregated indices.
1841
:param ignore_missing: Suppress KeyErrors from calling remove_index.
1843
for index_type in Pack.index_definitions.keys():
1844
attr_name = index_type + '_index'
1845
aggregate_index = getattr(self, attr_name)
1846
if aggregate_index is not None:
1847
pack_index = getattr(pack, attr_name)
1849
aggregate_index.remove_index(pack_index)
1040
def _remove_pack_indices(self, pack):
1041
"""Remove the indices for pack from the aggregated indices."""
1042
self.revision_index.remove_index(pack.revision_index, pack)
1043
self.inventory_index.remove_index(pack.inventory_index, pack)
1044
self.text_index.remove_index(pack.text_index, pack)
1045
self.signature_index.remove_index(pack.signature_index, pack)
1855
1047
def reset(self):
1856
1048
"""Clear all cached data."""
1857
1049
# cached revision data
1050
self.repo._revision_knit = None
1858
1051
self.revision_index.clear()
1859
1052
# cached signature data
1053
self.repo._signature_knit = None
1860
1054
self.signature_index.clear()
1861
1055
# cached file text data
1862
1056
self.text_index.clear()
1057
self.repo._text_knit = None
1863
1058
# cached inventory data
1864
1059
self.inventory_index.clear()
1866
if self.chk_index is not None:
1867
self.chk_index.clear()
1868
1060
# remove the open pack
1869
1061
self._new_pack = None
1870
1062
# information about packs.
1871
1063
self._names = None
1872
1064
self.packs = []
1873
self._packs_by_name = {}
1874
1066
self._packs_at_load = None
1876
def _unlock_names(self):
1068
def _make_index_map(self, index_suffix):
1069
"""Return information on existing indices.
1071
:param suffix: Index suffix added to pack name.
1073
:returns: (pack_map, indices) where indices is a list of GraphIndex
1074
objects, and pack_map is a mapping from those objects to the
1075
pack tuple they describe.
1077
# TODO: stop using this; it creates new indices unnecessarily.
1078
self.ensure_loaded()
1079
suffix_map = {'.rix': 'revision_index',
1080
'.six': 'signature_index',
1081
'.iix': 'inventory_index',
1082
'.tix': 'text_index',
1084
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1085
suffix_map[index_suffix])
1087
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1088
"""Convert a list of packs to an index pack map and index list.
1090
:param packs: The packs list to process.
1091
:param index_attribute: The attribute that the desired index is found
1093
:return: A tuple (map, list) where map contains the dict from
1094
index:pack_tuple, and lsit contains the indices in the same order
1100
index = getattr(pack, index_attribute)
1101
indices.append(index)
1102
pack_map[index] = (pack.pack_transport, pack.file_name())
1103
return pack_map, indices
1105
def _index_contents(self, pack_map, key_filter=None):
1106
"""Get an iterable of the index contents from a pack_map.
1108
:param pack_map: A map from indices to pack details.
1109
:param key_filter: An optional filter to limit the
1112
indices = [index for index in pack_map.iterkeys()]
1113
all_index = CombinedGraphIndex(indices)
1114
if key_filter is None:
1115
return all_index.iter_all_entries()
1117
return all_index.iter_entries(key_filter)
1119
def release_names(self):
1877
1120
"""Release the mutex around the pack-names index."""
1878
1121
self.repo.control_files.unlock()
1880
def _diff_pack_names(self):
1881
"""Read the pack names from disk, and compare it to the one in memory.
1883
:return: (disk_nodes, deleted_nodes, new_nodes)
1884
disk_nodes The final set of nodes that should be referenced
1885
deleted_nodes Nodes which have been removed from when we started
1886
new_nodes Nodes that are newly introduced
1888
# load the disk nodes across
1890
for index, key, value in self._iter_disk_pack_index():
1891
disk_nodes.add((key, value))
1892
orig_disk_nodes = set(disk_nodes)
1894
# do a two-way diff against our original content
1895
current_nodes = set()
1896
for name, sizes in self._names.iteritems():
1898
((name, ), ' '.join(str(size) for size in sizes)))
1900
# Packs no longer present in the repository, which were present when we
1901
# locked the repository
1902
deleted_nodes = self._packs_at_load - current_nodes
1903
# Packs which this process is adding
1904
new_nodes = current_nodes - self._packs_at_load
1906
# Update the disk_nodes set to include the ones we are adding, and
1907
# remove the ones which were removed by someone else
1908
disk_nodes.difference_update(deleted_nodes)
1909
disk_nodes.update(new_nodes)
1911
return disk_nodes, deleted_nodes, new_nodes, orig_disk_nodes
1913
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1914
"""Given the correct set of pack files, update our saved info.
1916
:return: (removed, added, modified)
1917
removed pack names removed from self._names
1918
added pack names added to self._names
1919
modified pack names that had changed value
1924
## self._packs_at_load = disk_nodes
1123
def _save_pack_names(self):
1124
"""Save the list of packs.
1126
This will take out the mutex around the pack names list for the
1127
duration of the method call. If concurrent updates have been made, a
1128
three-way merge between the current list and the current in memory list
1133
builder = GraphIndexBuilder()
1134
# load the disk nodes across
1136
for index, key, value in self._iter_disk_pack_index():
1137
disk_nodes.add((key, value))
1138
# do a two-way diff against our original content
1139
current_nodes = set()
1140
for name, sizes in self._names.iteritems():
1142
((name, ), ' '.join(str(size) for size in sizes)))
1143
deleted_nodes = self._packs_at_load - current_nodes
1144
new_nodes = current_nodes - self._packs_at_load
1145
disk_nodes.difference_update(deleted_nodes)
1146
disk_nodes.update(new_nodes)
1147
# TODO: handle same-name, index-size-changes here -
1148
# e.g. use the value from disk, not ours, *unless* we're the one
1150
for key, value in disk_nodes:
1151
builder.add_node(key, value)
1152
self.transport.put_file('pack-names', builder.finish())
1153
# move the baseline forward
1154
self._packs_at_load = disk_nodes
1156
self.release_names()
1157
# synchronise the memory packs list with what we just wrote:
1925
1158
new_names = dict(disk_nodes)
1926
1159
# drop no longer present nodes
1927
1160
for pack in self.all_packs():
1928
1161
if (pack.name,) not in new_names:
1929
removed.append(pack.name)
1930
self._remove_pack_from_memory(pack)
1162
self.remove_pack_from_memory(pack)
1931
1163
# add new nodes/refresh existing ones
1932
1164
for key, value in disk_nodes:
2076
1196
self._new_pack)
2077
1197
self.text_index.add_writable_index(self._new_pack.text_index,
2078
1198
self._new_pack)
2079
self._new_pack.text_index.set_optimize(combine_backing_indices=False)
2080
1199
self.signature_index.add_writable_index(self._new_pack.signature_index,
2081
1200
self._new_pack)
2082
if self.chk_index is not None:
2083
self.chk_index.add_writable_index(self._new_pack.chk_index,
2085
self.repo.chk_bytes._index._add_callback = self.chk_index.add_callback
2086
self._new_pack.chk_index.set_optimize(combine_backing_indices=False)
2088
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
2089
self.repo.revisions._index._add_callback = self.revision_index.add_callback
2090
self.repo.signatures._index._add_callback = self.signature_index.add_callback
2091
self.repo.texts._index._add_callback = self.text_index.add_callback
1202
# reused revision and signature knits may need updating
1203
if self.repo._revision_knit is not None:
1204
self.repo._revision_knit._index._add_callback = \
1205
self.revision_index.add_callback
1206
if self.repo._signature_knit is not None:
1207
self.repo._signature_knit._index._add_callback = \
1208
self.signature_index.add_callback
1209
# create a reused knit object for text addition in commit.
1210
self.repo._text_knit = self.repo.weave_store.get_weave_or_empty(
2093
1213
def _abort_write_group(self):
2094
1214
# FIXME: just drop the transient index.
2095
1215
# forget what names there are
2096
if self._new_pack is not None:
2097
operation = cleanup.OperationWithCleanups(self._new_pack.abort)
2098
operation.add_cleanup(setattr, self, '_new_pack', None)
2099
# If we aborted while in the middle of finishing the write
2100
# group, _remove_pack_indices could fail because the indexes are
2101
# already gone. But they're not there we shouldn't fail in this
2102
# case, so we pass ignore_missing=True.
2103
operation.add_cleanup(self._remove_pack_indices, self._new_pack,
2104
ignore_missing=True)
2105
operation.run_simple()
2106
for resumed_pack in self._resumed_packs:
2107
operation = cleanup.OperationWithCleanups(resumed_pack.abort)
2108
# See comment in previous finally block.
2109
operation.add_cleanup(self._remove_pack_indices, resumed_pack,
2110
ignore_missing=True)
2111
operation.run_simple()
2112
del self._resumed_packs[:]
2114
def _remove_resumed_pack_indices(self):
2115
for resumed_pack in self._resumed_packs:
2116
self._remove_pack_indices(resumed_pack)
2117
del self._resumed_packs[:]
2119
def _check_new_inventories(self):
2120
"""Detect missing inventories in this write group.
2122
:returns: list of strs, summarising any problems found. If the list is
2123
empty no problems were found.
2125
# The base implementation does no checks. GCRepositoryPackCollection
1216
self._new_pack.abort()
1217
self._remove_pack_indices(self._new_pack)
1218
self._new_pack = None
1219
self.repo._text_knit = None
2129
1221
def _commit_write_group(self):
2131
for prefix, versioned_file in (
2132
('revisions', self.repo.revisions),
2133
('inventories', self.repo.inventories),
2134
('texts', self.repo.texts),
2135
('signatures', self.repo.signatures),
2137
missing = versioned_file.get_missing_compression_parent_keys()
2138
all_missing.update([(prefix,) + key for key in missing])
2140
raise errors.BzrCheckError(
2141
"Repository %s has missing compression parent(s) %r "
2142
% (self.repo, sorted(all_missing)))
2143
problems = self._check_new_inventories()
2145
problems_summary = '\n'.join(problems)
2146
raise errors.BzrCheckError(
2147
"Cannot add revision(s) to repository: " + problems_summary)
2148
1222
self._remove_pack_indices(self._new_pack)
2149
any_new_content = False
2150
1223
if self._new_pack.data_inserted():
2151
1224
# get all the data to disk and read to use
2152
1225
self._new_pack.finish()
2153
1226
self.allocate(self._new_pack)
2154
1227
self._new_pack = None
2155
any_new_content = True
2157
self._new_pack.abort()
2158
self._new_pack = None
2159
for resumed_pack in self._resumed_packs:
2160
# XXX: this is a pretty ugly way to turn the resumed pack into a
2161
# properly committed pack.
2162
self._names[resumed_pack.name] = None
2163
self._remove_pack_from_memory(resumed_pack)
2164
resumed_pack.finish()
2165
self.allocate(resumed_pack)
2166
any_new_content = True
2167
del self._resumed_packs[:]
2169
result = self.autopack()
1228
if not self.autopack():
2171
1229
# when autopack takes no steps, the names list is still
2173
return self._save_pack_names()
2177
def _suspend_write_group(self):
2178
tokens = [pack.name for pack in self._resumed_packs]
2179
self._remove_pack_indices(self._new_pack)
2180
if self._new_pack.data_inserted():
2181
# get all the data to disk and read to use
2182
self._new_pack.finish(suspend=True)
2183
tokens.append(self._new_pack.name)
2184
self._new_pack = None
1231
self._save_pack_names()
2186
1233
self._new_pack.abort()
2187
self._new_pack = None
2188
self._remove_resumed_pack_indices()
2191
def _resume_write_group(self, tokens):
2192
for token in tokens:
2193
self._resume_pack(token)
1234
self.repo._text_knit = None
1237
class KnitPackRevisionStore(KnitRevisionStore):
1238
"""An object to adapt access from RevisionStore's to use KnitPacks.
1240
This class works by replacing the original RevisionStore.
1241
We need to do this because the KnitPackRevisionStore is less
1242
isolated in its layering - it uses services from the repo.
1245
def __init__(self, repo, transport, revisionstore):
1246
"""Create a KnitPackRevisionStore on repo with revisionstore.
1248
This will store its state in the Repository, use the
1249
indices FileNames to provide a KnitGraphIndex,
1250
and at the end of transactions write new indices.
1252
KnitRevisionStore.__init__(self, revisionstore.versioned_file_store)
1254
self._serializer = revisionstore._serializer
1255
self.transport = transport
1257
def get_revision_file(self, transaction):
1258
"""Get the revision versioned file object."""
1259
if getattr(self.repo, '_revision_knit', None) is not None:
1260
return self.repo._revision_knit
1261
self.repo._packs.ensure_loaded()
1262
add_callback = self.repo._packs.revision_index.add_callback
1263
# setup knit specific objects
1264
knit_index = KnitGraphIndex(
1265
self.repo._packs.revision_index.combined_index,
1266
add_callback=add_callback)
1267
self.repo._revision_knit = knit.KnitVersionedFile(
1268
'revisions', self.transport.clone('..'),
1269
self.repo.control_files._file_mode,
1270
create=False, access_mode=self.repo._access_mode(),
1271
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1272
access_method=self.repo._packs.revision_index.knit_access)
1273
return self.repo._revision_knit
1275
def get_signature_file(self, transaction):
1276
"""Get the signature versioned file object."""
1277
if getattr(self.repo, '_signature_knit', None) is not None:
1278
return self.repo._signature_knit
1279
self.repo._packs.ensure_loaded()
1280
add_callback = self.repo._packs.signature_index.add_callback
1281
# setup knit specific objects
1282
knit_index = KnitGraphIndex(
1283
self.repo._packs.signature_index.combined_index,
1284
add_callback=add_callback, parents=False)
1285
self.repo._signature_knit = knit.KnitVersionedFile(
1286
'signatures', self.transport.clone('..'),
1287
self.repo.control_files._file_mode,
1288
create=False, access_mode=self.repo._access_mode(),
1289
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1290
access_method=self.repo._packs.signature_index.knit_access)
1291
return self.repo._signature_knit
1294
class KnitPackTextStore(VersionedFileStore):
1295
"""An object to adapt access from VersionedFileStore's to use KnitPacks.
1297
This should not live through to production: by production time we should
1298
have fully integrated the new indexing and have new data for the
1299
repository classes; also we may choose not to do a Knit1 compatible
1300
new repository, just a Knit3 one. If neither of these happen, this
1301
should definately be cleaned up before merging.
1303
This class works by replacing the original VersionedFileStore.
1304
We need to do this because the KnitPackRevisionStore is less
1305
isolated in its layering - it uses services from the repo and shares them
1306
with all the data written in a single write group.
1309
def __init__(self, repo, transport, weavestore):
1310
"""Create a KnitPackTextStore on repo with weavestore.
1312
This will store its state in the Repository, use the
1313
indices FileNames to provide a KnitGraphIndex,
1314
and at the end of transactions write new indices.
1316
# don't call base class constructor - its not suitable.
1317
# no transient data stored in the transaction
1319
self._precious = False
1321
self.transport = transport
1322
self.weavestore = weavestore
1323
# XXX for check() which isn't updated yet
1324
self._transport = weavestore._transport
1326
def get_weave_or_empty(self, file_id, transaction):
1327
"""Get a 'Knit' backed by the .tix indices.
1329
The transaction parameter is ignored.
1331
self.repo._packs.ensure_loaded()
1332
add_callback = self.repo._packs.text_index.add_callback
1333
# setup knit specific objects
1334
file_id_index = GraphIndexPrefixAdapter(
1335
self.repo._packs.text_index.combined_index,
1336
(file_id, ), 1, add_nodes_callback=add_callback)
1337
knit_index = KnitGraphIndex(file_id_index,
1338
add_callback=file_id_index.add_nodes,
1339
deltas=True, parents=True)
1340
return knit.KnitVersionedFile('text:' + file_id,
1341
self.transport.clone('..'),
1344
access_method=self.repo._packs.text_index.knit_access,
1345
factory=knit.KnitPlainFactory())
1347
get_weave = get_weave_or_empty
1350
"""Generate a list of the fileids inserted, for use by check."""
1351
self.repo._packs.ensure_loaded()
1353
for index, key, value, refs in \
1354
self.repo._packs.text_index.combined_index.iter_all_entries():
1359
class InventoryKnitThunk(object):
1360
"""An object to manage thunking get_inventory_weave to pack based knits."""
1362
def __init__(self, repo, transport):
1363
"""Create an InventoryKnitThunk for repo at transport.
1365
This will store its state in the Repository, use the
1366
indices FileNames to provide a KnitGraphIndex,
1367
and at the end of transactions write a new index..
1370
self.transport = transport
1372
def get_weave(self):
1373
"""Get a 'Knit' that contains inventory data."""
1374
self.repo._packs.ensure_loaded()
1375
add_callback = self.repo._packs.inventory_index.add_callback
1376
# setup knit specific objects
1377
knit_index = KnitGraphIndex(
1378
self.repo._packs.inventory_index.combined_index,
1379
add_callback=add_callback, deltas=True, parents=True)
1380
return knit.KnitVersionedFile(
1381
'inventory', self.transport.clone('..'),
1382
self.repo.control_files._file_mode,
1383
create=False, access_mode=self.repo._access_mode(),
1384
index=knit_index, delta=True, factory=knit.KnitPlainFactory(),
1385
access_method=self.repo._packs.inventory_index.knit_access)
2196
1388
class KnitPackRepository(KnitRepository):
2197
"""Repository with knit objects stored inside pack containers.
2199
The layering for a KnitPackRepository is:
2201
Graph | HPSS | Repository public layer |
2202
===================================================
2203
Tuple based apis below, string based, and key based apis above
2204
---------------------------------------------------
2206
Provides .texts, .revisions etc
2207
This adapts the N-tuple keys to physical knit records which only have a
2208
single string identifier (for historical reasons), which in older formats
2209
was always the revision_id, and in the mapped code for packs is always
2210
the last element of key tuples.
2211
---------------------------------------------------
2213
A separate GraphIndex is used for each of the
2214
texts/inventories/revisions/signatures contained within each individual
2215
pack file. The GraphIndex layer works in N-tuples and is unaware of any
2217
===================================================
2221
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
1389
"""Experimental graph-knit using repository."""
1391
def __init__(self, _format, a_bzrdir, control_files, _revision_store,
1392
control_store, text_store, _commit_builder_class, _serializer):
2223
1393
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2224
_commit_builder_class, _serializer)
2225
index_transport = self._transport.clone('indices')
2226
self._pack_collection = RepositoryPackCollection(self, self._transport,
1394
_revision_store, control_store, text_store, _commit_builder_class,
1396
index_transport = control_files._transport.clone('indices')
1397
self._packs = RepositoryPackCollection(self, control_files._transport,
2227
1398
index_transport,
2228
self._transport.clone('upload'),
2229
self._transport.clone('packs'),
2230
_format.index_builder_class,
2231
_format.index_class,
2232
use_chk_index=self._format.supports_chks,
2234
self.inventories = KnitVersionedFiles(
2235
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2236
add_callback=self._pack_collection.inventory_index.add_callback,
2237
deltas=True, parents=True, is_locked=self.is_locked),
2238
data_access=self._pack_collection.inventory_index.data_access,
2239
max_delta_chain=200)
2240
self.revisions = KnitVersionedFiles(
2241
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2242
add_callback=self._pack_collection.revision_index.add_callback,
2243
deltas=False, parents=True, is_locked=self.is_locked,
2244
track_external_parent_refs=True),
2245
data_access=self._pack_collection.revision_index.data_access,
2247
self.signatures = KnitVersionedFiles(
2248
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2249
add_callback=self._pack_collection.signature_index.add_callback,
2250
deltas=False, parents=False, is_locked=self.is_locked),
2251
data_access=self._pack_collection.signature_index.data_access,
2253
self.texts = KnitVersionedFiles(
2254
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2255
add_callback=self._pack_collection.text_index.add_callback,
2256
deltas=True, parents=True, is_locked=self.is_locked),
2257
data_access=self._pack_collection.text_index.data_access,
2258
max_delta_chain=200)
2259
if _format.supports_chks:
2260
# No graph, no compression:- references from chks are between
2261
# different objects not temporal versions of the same; and without
2262
# some sort of temporal structure knit compression will just fail.
2263
self.chk_bytes = KnitVersionedFiles(
2264
_KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2265
add_callback=self._pack_collection.chk_index.add_callback,
2266
deltas=False, parents=False, is_locked=self.is_locked),
2267
data_access=self._pack_collection.chk_index.data_access,
2270
self.chk_bytes = None
1399
control_files._transport.clone('upload'),
1400
control_files._transport.clone('packs'))
1401
self._revision_store = KnitPackRevisionStore(self, index_transport, self._revision_store)
1402
self.weave_store = KnitPackTextStore(self, index_transport, self.weave_store)
1403
self._inv_thunk = InventoryKnitThunk(self, index_transport)
2271
1404
# True when the repository object is 'write locked' (as opposed to the
2272
# physical lock only taken out around changes to the pack-names list.)
1405
# physical lock only taken out around changes to the pack-names list.)
2273
1406
# Another way to represent this would be a decorator around the control
2274
1407
# files object that presents logical locks as physical ones - if this
2275
1408
# gets ugly consider that alternative design. RBC 20071011
2276
1409
self._write_lock_count = 0
2277
1410
self._transaction = None
2279
self._reconcile_does_inventory_gc = True
2280
self._reconcile_fixes_text_parents = True
2281
self._reconcile_backsup_inventory = False
2283
def _warn_if_deprecated(self, branch=None):
2284
# This class isn't deprecated, but one sub-format is
2285
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2286
super(KnitPackRepository, self)._warn_if_deprecated(branch)
1412
self._reconcile_does_inventory_gc = False
1413
self._reconcile_fixes_text_parents = False
2288
1415
def _abort_write_group(self):
2289
self.revisions._index._key_dependencies.clear()
2290
self._pack_collection._abort_write_group()
2292
def _get_source(self, to_format):
2293
if to_format.network_name() == self._format.network_name():
2294
return KnitPackStreamSource(self, to_format)
2295
return super(KnitPackRepository, self)._get_source(to_format)
1416
self._packs._abort_write_group()
1418
def _access_mode(self):
1419
"""Return 'w' or 'r' for depending on whether a write lock is active.
1421
This method is a helper for the Knit-thunking support objects.
1423
if self.is_write_locked():
1427
def get_parents(self, revision_ids):
1428
"""See StackedParentsProvider.get_parents.
1430
This implementation accesses the combined revision index to provide
1433
index = self._packs.revision_index.combined_index
1435
for revision_id in revision_ids:
1436
if revision_id != _mod_revision.NULL_REVISION:
1437
search_keys.add((revision_id,))
1438
found_parents = {_mod_revision.NULL_REVISION:[]}
1439
for index, key, value, refs in index.iter_entries(search_keys):
1442
parents = (_mod_revision.NULL_REVISION,)
1444
parents = tuple(parent[0] for parent in parents)
1445
found_parents[key[0]] = parents
1447
for revision_id in revision_ids:
1449
result.append(found_parents[revision_id])
2297
1454
def _make_parents_provider(self):
2298
return graph.CachingParentsProvider(self)
2300
1457
def _refresh_data(self):
2301
if not self.is_locked():
2303
self._pack_collection.reload_pack_names()
1458
if self._write_lock_count == 1 or self.control_files._lock_count == 1:
1459
# forget what names there are
1461
# XXX: Better to do an in-memory merge when acquiring a new lock -
1462
# factor out code from _save_pack_names.
2305
1464
def _start_write_group(self):
2306
self._pack_collection._start_write_group()
1465
self._packs._start_write_group()
2308
1467
def _commit_write_group(self):
2309
hint = self._pack_collection._commit_write_group()
2310
self.revisions._index._key_dependencies.clear()
2313
def suspend_write_group(self):
2314
# XXX check self._write_group is self.get_transaction()?
2315
tokens = self._pack_collection._suspend_write_group()
2316
self.revisions._index._key_dependencies.clear()
2317
self._write_group = None
2320
def _resume_write_group(self, tokens):
2321
self._start_write_group()
2323
self._pack_collection._resume_write_group(tokens)
2324
except errors.UnresumableWriteGroup:
2325
self._abort_write_group()
2327
for pack in self._pack_collection._resumed_packs:
2328
self.revisions._index.scan_unvalidated_index(pack.revision_index)
1468
return self._packs._commit_write_group()
1470
def get_inventory_weave(self):
1471
return self._inv_thunk.get_weave()
2330
1473
def get_transaction(self):
2331
1474
if self._write_lock_count:
2598
1668
def get_format_string(self):
2599
1669
"""See RepositoryFormat.get_format_string()."""
2600
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
1670
return "Bazaar Experimental no-subtrees\n"
2602
1672
def get_format_description(self):
2603
1673
"""See RepositoryFormat.get_format_description()."""
2604
return "Packs containing knits without subtree support"
1674
return "Experimental no-subtrees"
1676
def check_conversion_target(self, target_format):
2607
1680
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2608
"""A subtrees parameterized Pack repository.
1681
"""A subtrees parameterised Pack repository.
2610
1683
This repository format uses the xml7 serializer to get:
2611
1684
- support for recording full info about the tree root
2612
1685
- support for recording tree-references
2614
This format was introduced in 0.92.
2617
repository_class = KnitPackRepository
2618
_commit_builder_class = PackRootCommitBuilder
2619
rich_root_data = True
2621
supports_tree_reference = True
2623
def _serializer(self):
2624
return xml7.serializer_v7
2625
# What index classes to use
2626
index_builder_class = InMemoryGraphIndex
2627
index_class = GraphIndex
2629
def _get_matching_bzrdir(self):
2630
return bzrdir.format_registry.make_bzrdir(
2631
'pack-0.92-subtree')
2633
def _ignore_setting_bzrdir(self, format):
2636
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2638
def get_format_string(self):
2639
"""See RepositoryFormat.get_format_string()."""
2640
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2642
def get_format_description(self):
2643
"""See RepositoryFormat.get_format_description()."""
2644
return "Packs containing knits with subtree support\n"
2647
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2648
"""A rich-root, no subtrees parameterized Pack repository.
2650
This repository format uses the xml6 serializer to get:
2651
- support for recording full info about the tree root
2653
This format was introduced in 1.0.
2656
repository_class = KnitPackRepository
2657
_commit_builder_class = PackRootCommitBuilder
2658
rich_root_data = True
2659
supports_tree_reference = False
2661
def _serializer(self):
2662
return xml6.serializer_v6
2663
# What index classes to use
2664
index_builder_class = InMemoryGraphIndex
2665
index_class = GraphIndex
2667
def _get_matching_bzrdir(self):
2668
return bzrdir.format_registry.make_bzrdir(
2671
def _ignore_setting_bzrdir(self, format):
2674
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2676
def get_format_string(self):
2677
"""See RepositoryFormat.get_format_string()."""
2678
return ("Bazaar pack repository format 1 with rich root"
2679
" (needs bzr 1.0)\n")
2681
def get_format_description(self):
2682
"""See RepositoryFormat.get_format_description()."""
2683
return "Packs containing knits with rich root support\n"
2686
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2687
"""Repository that supports external references to allow stacking.
2691
Supports external lookups, which results in non-truncated ghosts after
2692
reconcile compared to pack-0.92 formats.
2695
repository_class = KnitPackRepository
2696
_commit_builder_class = PackCommitBuilder
2697
supports_external_lookups = True
2698
# What index classes to use
2699
index_builder_class = InMemoryGraphIndex
2700
index_class = GraphIndex
2703
def _serializer(self):
2704
return xml5.serializer_v5
2706
def _get_matching_bzrdir(self):
2707
return bzrdir.format_registry.make_bzrdir('1.6')
2709
def _ignore_setting_bzrdir(self, format):
2712
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2714
def get_format_string(self):
2715
"""See RepositoryFormat.get_format_string()."""
2716
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2718
def get_format_description(self):
2719
"""See RepositoryFormat.get_format_description()."""
2720
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2723
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2724
"""A repository with rich roots and stacking.
2726
New in release 1.6.1.
2728
Supports stacking on other repositories, allowing data to be accessed
2729
without being stored locally.
2732
repository_class = KnitPackRepository
2733
_commit_builder_class = PackRootCommitBuilder
2734
rich_root_data = True
2735
supports_tree_reference = False # no subtrees
2736
supports_external_lookups = True
2737
# What index classes to use
2738
index_builder_class = InMemoryGraphIndex
2739
index_class = GraphIndex
2742
def _serializer(self):
2743
return xml6.serializer_v6
2745
def _get_matching_bzrdir(self):
2746
return bzrdir.format_registry.make_bzrdir(
2749
def _ignore_setting_bzrdir(self, format):
2752
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2754
def get_format_string(self):
2755
"""See RepositoryFormat.get_format_string()."""
2756
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2758
def get_format_description(self):
2759
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2762
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2763
"""A repository with rich roots and external references.
2767
Supports external lookups, which results in non-truncated ghosts after
2768
reconcile compared to pack-0.92 formats.
2770
This format was deprecated because the serializer it uses accidentally
2771
supported subtrees, when the format was not intended to. This meant that
2772
someone could accidentally fetch from an incorrect repository.
2775
repository_class = KnitPackRepository
2776
_commit_builder_class = PackRootCommitBuilder
2777
rich_root_data = True
2778
supports_tree_reference = False # no subtrees
2780
supports_external_lookups = True
2781
# What index classes to use
2782
index_builder_class = InMemoryGraphIndex
2783
index_class = GraphIndex
2786
def _serializer(self):
2787
return xml7.serializer_v7
2789
def _get_matching_bzrdir(self):
2790
matching = bzrdir.format_registry.make_bzrdir(
2792
matching.repository_format = self
2795
def _ignore_setting_bzrdir(self, format):
2798
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2800
def get_format_string(self):
2801
"""See RepositoryFormat.get_format_string()."""
2802
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2804
def get_format_description(self):
2805
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2809
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2810
"""A repository with stacking and btree indexes,
2811
without rich roots or subtrees.
2813
This is equivalent to pack-1.6 with B+Tree indices.
2816
repository_class = KnitPackRepository
2817
_commit_builder_class = PackCommitBuilder
2818
supports_external_lookups = True
2819
# What index classes to use
2820
index_builder_class = BTreeBuilder
2821
index_class = BTreeGraphIndex
2824
def _serializer(self):
2825
return xml5.serializer_v5
2827
def _get_matching_bzrdir(self):
2828
return bzrdir.format_registry.make_bzrdir('1.9')
2830
def _ignore_setting_bzrdir(self, format):
2833
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2835
def get_format_string(self):
2836
"""See RepositoryFormat.get_format_string()."""
2837
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2839
def get_format_description(self):
2840
"""See RepositoryFormat.get_format_description()."""
2841
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2844
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2845
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2847
1.6-rich-root with B+Tree indices.
2850
repository_class = KnitPackRepository
2851
_commit_builder_class = PackRootCommitBuilder
2852
rich_root_data = True
2853
supports_tree_reference = False # no subtrees
2854
supports_external_lookups = True
2855
# What index classes to use
2856
index_builder_class = BTreeBuilder
2857
index_class = BTreeGraphIndex
2860
def _serializer(self):
2861
return xml6.serializer_v6
2863
def _get_matching_bzrdir(self):
2864
return bzrdir.format_registry.make_bzrdir(
2867
def _ignore_setting_bzrdir(self, format):
2870
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2872
def get_format_string(self):
2873
"""See RepositoryFormat.get_format_string()."""
2874
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2876
def get_format_description(self):
2877
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2880
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2881
"""A subtrees development repository.
2883
This format should be retained until the second release after bzr 1.7.
2885
1.6.1-subtree[as it might have been] with B+Tree indices.
2887
This is [now] retained until we have a CHK based subtree format in
2891
repository_class = KnitPackRepository
2892
_commit_builder_class = PackRootCommitBuilder
2893
rich_root_data = True
2895
supports_tree_reference = True
2896
supports_external_lookups = True
2897
# What index classes to use
2898
index_builder_class = BTreeBuilder
2899
index_class = BTreeGraphIndex
2902
def _serializer(self):
2903
return xml7.serializer_v7
2905
def _get_matching_bzrdir(self):
2906
return bzrdir.format_registry.make_bzrdir(
2907
'development-subtree')
2909
def _ignore_setting_bzrdir(self, format):
2912
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2914
def get_format_string(self):
2915
"""See RepositoryFormat.get_format_string()."""
2916
return ("Bazaar development format 2 with subtree support "
2917
"(needs bzr.dev from before 1.8)\n")
2919
def get_format_description(self):
2920
"""See RepositoryFormat.get_format_description()."""
2921
return ("Development repository format, currently the same as "
2922
"1.6.1-subtree with B+Tree indices.\n")
1687
This format was introduced in bzr.dev.
1690
repository_class = KnitPackRepository
1691
_commit_builder_class = PackRootCommitBuilder
1692
rich_root_data = True
1693
supports_tree_reference = True
1694
_serializer = xml7.serializer_v7
1696
def _get_matching_bzrdir(self):
1697
return bzrdir.format_registry.make_bzrdir('experimental-subtree')
1699
def _ignore_setting_bzrdir(self, format):
1702
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1704
def check_conversion_target(self, target_format):
1705
if not target_format.rich_root_data:
1706
raise errors.BadConversionTarget(
1707
'Does not support rich root data.', target_format)
1708
if not getattr(target_format, 'supports_tree_reference', False):
1709
raise errors.BadConversionTarget(
1710
'Does not support nested trees', target_format)
1712
def get_format_string(self):
1713
"""See RepositoryFormat.get_format_string()."""
1714
return "Bazaar Experimental subtrees\n"
1716
def get_format_description(self):
1717
"""See RepositoryFormat.get_format_description()."""
1718
return "Experimental subtrees\n"
added
removed
bzrlib/repofmt/pack_repo.py
