206
258
return not self.__eq__(other)
208
260
def __repr__(self):
209
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
210
id(self), self.pack_transport, self.name)
261
return "<%s.%s object at 0x%x, %s, %s" % (
262
self.__class__.__module__, self.__class__.__name__, id(self),
263
self.pack_transport, self.name)
266
class ResumedPack(ExistingPack):
268
def __init__(self, name, revision_index, inventory_index, text_index,
269
signature_index, upload_transport, pack_transport, index_transport,
270
pack_collection, chk_index=None):
271
"""Create a ResumedPack object."""
272
ExistingPack.__init__(self, pack_transport, name, revision_index,
273
inventory_index, text_index, signature_index,
275
self.upload_transport = upload_transport
276
self.index_transport = index_transport
277
self.index_sizes = [None, None, None, None]
279
('revision', revision_index),
280
('inventory', inventory_index),
281
('text', text_index),
282
('signature', signature_index),
284
if chk_index is not None:
285
indices.append(('chk', chk_index))
286
self.index_sizes.append(None)
287
for index_type, index in indices:
288
offset = self.index_offset(index_type)
289
self.index_sizes[offset] = index._size
290
self.index_class = pack_collection._index_class
291
self._pack_collection = pack_collection
292
self._state = 'resumed'
293
# XXX: perhaps check that the .pack file exists?
295
def access_tuple(self):
296
if self._state == 'finished':
297
return Pack.access_tuple(self)
298
elif self._state == 'resumed':
299
return self.upload_transport, self.file_name()
301
raise AssertionError(self._state)
304
self.upload_transport.delete(self.file_name())
305
indices = [self.revision_index, self.inventory_index, self.text_index,
306
self.signature_index]
307
if self.chk_index is not None:
308
indices.append(self.chk_index)
309
for index in indices:
310
index._transport.delete(index._name)
313
self._check_references()
314
index_types = ['revision', 'inventory', 'text', 'signature']
315
if self.chk_index is not None:
316
index_types.append('chk')
317
for index_type in index_types:
318
old_name = self.index_name(index_type, self.name)
319
new_name = '../indices/' + old_name
320
self.upload_transport.move(old_name, new_name)
321
self._replace_index_with_readonly(index_type)
322
new_name = '../packs/' + self.file_name()
323
self.upload_transport.move(self.file_name(), new_name)
324
self._state = 'finished'
326
def _get_external_refs(self, index):
327
"""Return compression parents for this index that are not present.
329
This returns any compression parents that are referenced by this index,
330
which are not contained *in* this index. They may be present elsewhere.
332
return index.external_references(1)
213
335
class NewPack(Pack):
214
336
"""An in memory proxy for a pack which is being created."""
216
# A map of index 'type' to the file extension and position in the
218
index_definitions = {
219
'revision': ('.rix', 0),
220
'inventory': ('.iix', 1),
222
'signature': ('.six', 3),
225
def __init__(self, upload_transport, index_transport, pack_transport,
226
upload_suffix='', file_mode=None, index_builder_class=None,
338
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
228
339
"""Create a NewPack instance.
230
:param upload_transport: A writable transport for the pack to be
231
incrementally uploaded to.
232
:param index_transport: A writable transport for the pack's indices to
233
be written to when the pack is finished.
234
:param pack_transport: A writable transport for the pack to be renamed
235
to when the upload is complete. This *must* be the same as
236
upload_transport.clone('../packs').
341
:param pack_collection: A PackCollection into which this is being inserted.
237
342
:param upload_suffix: An optional suffix to be given to any temporary
238
343
files created during the pack creation. e.g '.autopack'
239
:param file_mode: An optional file mode to create the new files with.
240
:param index_builder_class: Required keyword parameter - the class of
241
index builder to use.
242
:param index_class: Required keyword parameter - the class of index
344
:param file_mode: Unix permissions for newly created file.
245
346
# The relative locations of the packs are constrained, but all are
246
347
# passed in because the caller has them, so as to avoid object churn.
348
index_builder_class = pack_collection._index_builder_class
349
if pack_collection.chk_index is not None:
350
chk_index = index_builder_class(reference_lists=0)
247
353
Pack.__init__(self,
248
354
# Revisions: parents list, no text compression.
249
355
index_builder_class(reference_lists=1),
642
743
Sets self._text_filter appropriately.
644
# select inventory keys
645
inv_keys = self._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._pack_collection._packs_list_to_pack_map_and_index_list(
650
self.packs, 'inventory_index')[0]
651
inv_nodes = self._pack_collection._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
self.pb.update("Copying inventory texts", 2)
656
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
657
# Only grab the output lines if we will be processing them
658
output_lines = bool(self.revision_ids)
659
inv_lines = self._copy_nodes_graph(inventory_index_map,
660
self.new_pack._writer, self.new_pack.inventory_index,
661
readv_group_iter, total_items, output_lines=output_lines)
662
if self.revision_ids:
663
self._process_inventory_lines(inv_lines)
665
# eat the iterator to cause it to execute.
667
self._text_filter = None
668
if 'pack' in debug.debug_flags:
669
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
670
time.ctime(), self._pack_collection._upload_transport.base,
671
self.new_pack.random_name,
672
self.new_pack.inventory_index.key_count(),
673
time.time() - self.new_pack.start_time)
745
raise NotImplementedError(self._copy_inventory_texts)
675
747
def _copy_text_texts(self):
677
text_index_map, text_nodes = self._get_text_nodes()
678
if self._text_filter is not None:
679
# We could return the keys copied as part of the return value from
680
# _copy_nodes_graph but this doesn't work all that well with the
681
# need to get line output too, so we check separately, and as we're
682
# going to buffer everything anyway, we check beforehand, which
683
# saves reading knit data over the wire when we know there are
685
text_nodes = set(text_nodes)
686
present_text_keys = set(_node[1] for _node in text_nodes)
687
missing_text_keys = set(self._text_filter) - present_text_keys
688
if missing_text_keys:
689
# TODO: raise a specific error that can handle many missing
691
a_missing_key = missing_text_keys.pop()
692
raise errors.RevisionNotPresent(a_missing_key[1],
694
# copy text keys and adjust values
695
self.pb.update("Copying content texts", 3)
696
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
697
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
698
self.new_pack.text_index, readv_group_iter, total_items))
699
self._log_copied_texts()
701
def _check_references(self):
702
"""Make sure our external refereneces are present."""
703
external_refs = self.new_pack._external_compression_parents_of_texts()
705
index = self._pack_collection.text_index.combined_index
706
found_items = list(index.iter_entries(external_refs))
707
if len(found_items) != len(external_refs):
708
found_keys = set(k for idx, k, refs, value in found_items)
709
missing_items = external_refs - found_keys
710
missing_file_id, missing_revision_id = missing_items.pop()
711
raise errors.RevisionNotPresent(missing_revision_id,
748
raise NotImplementedError(self._copy_text_texts)
714
750
def _create_pack_from_packs(self):
715
self.pb.update("Opening pack", 0, 5)
716
self.new_pack = self.open_pack()
717
new_pack = self.new_pack
718
# buffer data - we won't be reading-back during the pack creation and
719
# this makes a significant difference on sftp pushes.
720
new_pack.set_write_cache_size(1024*1024)
721
if 'pack' in debug.debug_flags:
722
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
723
for a_pack in self.packs]
724
if self.revision_ids is not None:
725
rev_count = len(self.revision_ids)
728
mutter('%s: create_pack: creating pack from source packs: '
729
'%s%s %s revisions wanted %s t=0',
730
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
731
plain_pack_list, rev_count)
732
self._copy_revision_texts()
733
self._copy_inventory_texts()
734
self._copy_text_texts()
735
# select signature keys
736
signature_filter = self._revision_keys # same keyspace
737
signature_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
738
self.packs, 'signature_index')[0]
739
signature_nodes = self._pack_collection._index_contents(signature_index_map,
741
# copy signature keys and adjust values
742
self.pb.update("Copying signature texts", 4)
743
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
744
new_pack.signature_index)
745
if 'pack' in debug.debug_flags:
746
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
747
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
748
new_pack.signature_index.key_count(),
749
time.time() - new_pack.start_time)
750
self._check_references()
751
if not self._use_pack(new_pack):
754
self.pb.update("Finishing pack", 5)
756
self._pack_collection.allocate(new_pack)
759
def _copy_nodes(self, nodes, index_map, writer, write_index):
760
"""Copy knit nodes between packs with no graph references."""
761
pb = ui.ui_factory.nested_progress_bar()
763
return self._do_copy_nodes(nodes, index_map, writer,
768
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb):
769
# for record verification
770
knit = KnitVersionedFiles(None, None)
771
# plan a readv on each source pack:
773
nodes = sorted(nodes)
774
# how to map this into knit.py - or knit.py into this?
775
# we don't want the typical knit logic, we want grouping by pack
776
# at this point - perhaps a helper library for the following code
777
# duplication points?
779
for index, key, value in nodes:
780
if index not in request_groups:
781
request_groups[index] = []
782
request_groups[index].append((key, value))
784
pb.update("Copied record", record_index, len(nodes))
785
for index, items in request_groups.iteritems():
786
pack_readv_requests = []
787
for key, value in items:
788
# ---- KnitGraphIndex.get_position
789
bits = value[1:].split(' ')
790
offset, length = int(bits[0]), int(bits[1])
791
pack_readv_requests.append((offset, length, (key, value[0])))
792
# linear scan up the pack
793
pack_readv_requests.sort()
795
transport, path = index_map[index]
796
reader = pack.make_readv_reader(transport, path,
797
[offset[0:2] for offset in pack_readv_requests])
798
for (names, read_func), (_1, _2, (key, eol_flag)) in \
799
izip(reader.iter_records(), pack_readv_requests):
800
raw_data = read_func(None)
801
# check the header only
802
df, _ = knit._parse_record_header(key, raw_data)
804
pos, size = writer.add_bytes_record(raw_data, names)
805
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
806
pb.update("Copied record", record_index)
809
def _copy_nodes_graph(self, index_map, writer, write_index,
810
readv_group_iter, total_items, output_lines=False):
811
"""Copy knit nodes between packs.
813
:param output_lines: Return lines present in the copied data as
814
an iterator of line,version_id.
816
pb = ui.ui_factory.nested_progress_bar()
818
for result in self._do_copy_nodes_graph(index_map, writer,
819
write_index, output_lines, pb, readv_group_iter, total_items):
822
# Python 2.4 does not permit try:finally: in a generator.
828
def _do_copy_nodes_graph(self, index_map, writer, write_index,
829
output_lines, pb, readv_group_iter, total_items):
830
# for record verification
831
knit = KnitVersionedFiles(None, None)
832
# for line extraction when requested (inventories only)
834
factory = KnitPlainFactory()
836
pb.update("Copied record", record_index, total_items)
837
for index, readv_vector, node_vector in readv_group_iter:
839
transport, path = index_map[index]
840
reader = pack.make_readv_reader(transport, path, readv_vector)
841
for (names, read_func), (key, eol_flag, references) in \
842
izip(reader.iter_records(), node_vector):
843
raw_data = read_func(None)
845
# read the entire thing
846
content, _ = knit._parse_record(key[-1], raw_data)
847
if len(references[-1]) == 0:
848
line_iterator = factory.get_fulltext_content(content)
850
line_iterator = factory.get_linedelta_content(content)
851
for line in line_iterator:
854
# check the header only
855
df, _ = knit._parse_record_header(key, raw_data)
857
pos, size = writer.add_bytes_record(raw_data, names)
858
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
859
pb.update("Copied record", record_index)
862
def _get_text_nodes(self):
863
text_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
864
self.packs, 'text_index')[0]
865
return text_index_map, self._pack_collection._index_contents(text_index_map,
868
def _least_readv_node_readv(self, nodes):
869
"""Generate request groups for nodes using the least readv's.
871
:param nodes: An iterable of graph index nodes.
872
:return: Total node count and an iterator of the data needed to perform
873
readvs to obtain the data for nodes. Each item yielded by the
874
iterator is a tuple with:
875
index, readv_vector, node_vector. readv_vector is a list ready to
876
hand to the transport readv method, and node_vector is a list of
877
(key, eol_flag, references) for the the node retrieved by the
878
matching readv_vector.
880
# group by pack so we do one readv per pack
881
nodes = sorted(nodes)
884
for index, key, value, references in nodes:
885
if index not in request_groups:
886
request_groups[index] = []
887
request_groups[index].append((key, value, references))
889
for index, items in request_groups.iteritems():
890
pack_readv_requests = []
891
for key, value, references in items:
892
# ---- KnitGraphIndex.get_position
893
bits = value[1:].split(' ')
894
offset, length = int(bits[0]), int(bits[1])
895
pack_readv_requests.append(
896
((offset, length), (key, value[0], references)))
897
# linear scan up the pack to maximum range combining.
898
pack_readv_requests.sort()
899
# split out the readv and the node data.
900
pack_readv = [readv for readv, node in pack_readv_requests]
901
node_vector = [node for readv, node in pack_readv_requests]
902
result.append((index, pack_readv, node_vector))
751
raise NotImplementedError(self._create_pack_from_packs)
905
753
def _log_copied_texts(self):
906
754
if 'pack' in debug.debug_flags:
938
767
return new_pack.data_inserted()
941
class OptimisingPacker(Packer):
942
"""A packer which spends more time to create better disk layouts."""
944
def _revision_node_readv(self, revision_nodes):
945
"""Return the total revisions and the readv's to issue.
947
This sort places revisions in topological order with the ancestors
950
:param revision_nodes: The revision index contents for the packs being
951
incorporated into the new pack.
952
:return: As per _least_readv_node_readv.
954
# build an ancestors dict
957
for index, key, value, references in revision_nodes:
958
ancestors[key] = references[0]
959
by_key[key] = (index, value, references)
960
order = tsort.topo_sort(ancestors)
962
# Single IO is pathological, but it will work as a starting point.
964
for key in reversed(order):
965
index, value, references = by_key[key]
966
# ---- KnitGraphIndex.get_position
967
bits = value[1:].split(' ')
968
offset, length = int(bits[0]), int(bits[1])
970
(index, [(offset, length)], [(key, value[0], references)]))
971
# TODO: combine requests in the same index that are in ascending order.
972
return total, requests
975
class ReconcilePacker(Packer):
976
"""A packer which regenerates indices etc as it copies.
978
This is used by ``bzr reconcile`` to cause parent text pointers to be
982
def _extra_init(self):
983
self._data_changed = False
985
def _process_inventory_lines(self, inv_lines):
986
"""Generate a text key reference map rather for reconciling with."""
987
repo = self._pack_collection.repo
988
refs = repo._find_text_key_references_from_xml_inventory_lines(
990
self._text_refs = refs
991
# during reconcile we:
992
# - convert unreferenced texts to full texts
993
# - correct texts which reference a text not copied to be full texts
994
# - copy all others as-is but with corrected parents.
995
# - so at this point we don't know enough to decide what becomes a full
997
self._text_filter = None
999
def _copy_text_texts(self):
1000
"""generate what texts we should have and then copy."""
1001
self.pb.update("Copying content texts", 3)
1002
# we have three major tasks here:
1003
# 1) generate the ideal index
1004
repo = self._pack_collection.repo
1005
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1006
_1, key, _2, refs in
1007
self.new_pack.revision_index.iter_all_entries()])
1008
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1009
# 2) generate a text_nodes list that contains all the deltas that can
1010
# be used as-is, with corrected parents.
1013
discarded_nodes = []
1014
NULL_REVISION = _mod_revision.NULL_REVISION
1015
text_index_map, text_nodes = self._get_text_nodes()
1016
for node in text_nodes:
1022
ideal_parents = tuple(ideal_index[node[1]])
1024
discarded_nodes.append(node)
1025
self._data_changed = True
1027
if ideal_parents == (NULL_REVISION,):
1029
if ideal_parents == node[3][0]:
1031
ok_nodes.append(node)
1032
elif ideal_parents[0:1] == node[3][0][0:1]:
1033
# the left most parent is the same, or there are no parents
1034
# today. Either way, we can preserve the representation as
1035
# long as we change the refs to be inserted.
1036
self._data_changed = True
1037
ok_nodes.append((node[0], node[1], node[2],
1038
(ideal_parents, node[3][1])))
1039
self._data_changed = True
1041
# Reinsert this text completely
1042
bad_texts.append((node[1], ideal_parents))
1043
self._data_changed = True
1044
# we're finished with some data.
1047
# 3) bulk copy the ok data
1048
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1049
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1050
self.new_pack.text_index, readv_group_iter, total_items))
1051
# 4) adhoc copy all the other texts.
1052
# We have to topologically insert all texts otherwise we can fail to
1053
# reconcile when parts of a single delta chain are preserved intact,
1054
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1055
# reinserted, and if d3 has incorrect parents it will also be
1056
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1057
# copied), so we will try to delta, but d2 is not currently able to be
1058
# extracted because it's basis d1 is not present. Topologically sorting
1059
# addresses this. The following generates a sort for all the texts that
1060
# are being inserted without having to reference the entire text key
1061
# space (we only topo sort the revisions, which is smaller).
1062
topo_order = tsort.topo_sort(ancestors)
1063
rev_order = dict(zip(topo_order, range(len(topo_order))))
1064
bad_texts.sort(key=lambda key:rev_order[key[0][1]])
1065
transaction = repo.get_transaction()
1066
file_id_index = GraphIndexPrefixAdapter(
1067
self.new_pack.text_index,
1069
add_nodes_callback=self.new_pack.text_index.add_nodes)
1070
data_access = _DirectPackAccess(
1071
{self.new_pack.text_index:self.new_pack.access_tuple()})
1072
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1073
self.new_pack.access_tuple())
1074
output_texts = KnitVersionedFiles(
1075
_KnitGraphIndex(self.new_pack.text_index,
1076
add_callback=self.new_pack.text_index.add_nodes,
1077
deltas=True, parents=True, is_locked=repo.is_locked),
1078
data_access=data_access, max_delta_chain=200)
1079
for key, parent_keys in bad_texts:
1080
# We refer to the new pack to delta data being output.
1081
# A possible improvement would be to catch errors on short reads
1082
# and only flush then.
1083
self.new_pack.flush()
1085
for parent_key in parent_keys:
1086
if parent_key[0] != key[0]:
1087
# Graph parents must match the fileid
1088
raise errors.BzrError('Mismatched key parent %r:%r' %
1090
parents.append(parent_key[1])
1091
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1092
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1093
output_texts.add_lines(key, parent_keys, text_lines,
1094
random_id=True, check_content=False)
1095
# 5) check that nothing inserted has a reference outside the keyspace.
1096
missing_text_keys = self.new_pack._external_compression_parents_of_texts()
1097
if missing_text_keys:
1098
raise errors.BzrError('Reference to missing compression parents %r'
1099
% (missing_text_keys,))
1100
self._log_copied_texts()
1102
def _use_pack(self, new_pack):
1103
"""Override _use_pack to check for reconcile having changed content."""
1104
# XXX: we might be better checking this at the copy time.
1105
original_inventory_keys = set()
1106
inv_index = self._pack_collection.inventory_index.combined_index
1107
for entry in inv_index.iter_all_entries():
1108
original_inventory_keys.add(entry[1])
1109
new_inventory_keys = set()
1110
for entry in new_pack.inventory_index.iter_all_entries():
1111
new_inventory_keys.add(entry[1])
1112
if new_inventory_keys != original_inventory_keys:
1113
self._data_changed = True
1114
return new_pack.data_inserted() and self._data_changed
1117
770
class RepositoryPackCollection(object):
1118
771
"""Management of packs within a repository.
1120
773
:ivar _names: map of {pack_name: (index_size,)}
777
resumed_pack_factory = None
778
normal_packer_class = None
779
optimising_packer_class = None
1123
781
def __init__(self, repo, transport, index_transport, upload_transport,
1124
pack_transport, index_builder_class, index_class):
782
pack_transport, index_builder_class, index_class,
1125
784
"""Create a new RepositoryPackCollection.
1127
:param transport: Addresses the repository base directory
786
:param transport: Addresses the repository base directory
1128
787
(typically .bzr/repository/).
1129
788
:param index_transport: Addresses the directory containing indices.
1130
789
:param upload_transport: Addresses the directory into which packs are written
1219
910
# group their data with the relevant commit, and that may
1220
911
# involve rewriting ancient history - which autopack tries to
1221
912
# avoid. Alternatively we could not group the data but treat
1222
# each of these as having a single revision, and thus add
913
# each of these as having a single revision, and thus add
1223
914
# one revision for each to the total revision count, to get
1224
915
# a matching distribution.
1226
917
existing_packs.append((revision_count, pack))
1227
918
pack_operations = self.plan_autopack_combinations(
1228
919
existing_packs, pack_distribution)
1229
self._execute_pack_operations(pack_operations)
920
num_new_packs = len(pack_operations)
921
num_old_packs = sum([len(po[1]) for po in pack_operations])
922
num_revs_affected = sum([po[0] for po in pack_operations])
923
mutter('Auto-packing repository %s, which has %d pack files, '
924
'containing %d revisions. Packing %d files into %d affecting %d'
925
' revisions', self, total_packs, total_revisions, num_old_packs,
926
num_new_packs, num_revs_affected)
927
result = self._execute_pack_operations(pack_operations, packer_class=self.normal_packer_class,
928
reload_func=self._restart_autopack)
929
mutter('Auto-packing repository %s completed', self)
1232
def _execute_pack_operations(self, pack_operations, _packer_class=Packer):
932
def _execute_pack_operations(self, pack_operations, packer_class,
1233
934
"""Execute a series of pack operations.
1235
936
:param pack_operations: A list of [revision_count, packs_to_combine].
1236
:param _packer_class: The class of packer to use (default: Packer).
937
:param packer_class: The class of packer to use
938
:return: The new pack names.
1239
940
for revision_count, packs in pack_operations:
1240
941
# we may have no-ops from the setup logic
1241
942
if len(packs) == 0:
1243
_packer_class(self, packs, '.autopack').pack()
944
packer = packer_class(self, packs, '.autopack',
945
reload_func=reload_func)
947
result = packer.pack()
948
except errors.RetryWithNewPacks:
949
# An exception is propagating out of this context, make sure
950
# this packer has cleaned up. Packer() doesn't set its new_pack
951
# state into the RepositoryPackCollection object, so we only
952
# have access to it directly here.
953
if packer.new_pack is not None:
954
packer.new_pack.abort()
1244
958
for pack in packs:
1245
959
self._remove_pack_from_memory(pack)
1246
960
# record the newly available packs and stop advertising the old
1248
self._save_pack_names(clear_obsolete_packs=True)
1249
# Move the old packs out of the way now they are no longer referenced.
1250
for revision_count, packs in pack_operations:
1251
self._obsolete_packs(packs)
963
for _, packs in pack_operations:
964
to_be_obsoleted.extend(packs)
965
result = self._save_pack_names(clear_obsolete_packs=True,
966
obsolete_packs=to_be_obsoleted)
969
def _flush_new_pack(self):
970
if self._new_pack is not None:
971
self._new_pack.flush()
1253
973
def lock_names(self):
1254
974
"""Acquire the mutex around the pack-names index.
1256
976
This cannot be used in the middle of a read-only transaction on the
1259
979
self.repo.control_files.lock_write()
981
def _already_packed(self):
982
"""Is the collection already packed?"""
983
return not (self.repo._format.pack_compresses or (len(self._names) > 1))
985
def pack(self, hint=None, clean_obsolete_packs=False):
1262
986
"""Pack the pack collection totally."""
1263
987
self.ensure_loaded()
1264
988
total_packs = len(self._names)
1266
# This is arguably wrong because we might not be optimal, but for
1267
# now lets leave it in. (e.g. reconcile -> one pack. But not
989
if self._already_packed():
1270
991
total_revisions = self.revision_index.combined_index.key_count()
1271
992
# XXX: the following may want to be a class, to pack with a given
1273
994
mutter('Packing repository %s, which has %d pack files, '
1274
'containing %d revisions into 1 packs.', self, total_packs,
995
'containing %d revisions with hint %r.', self, total_packs,
996
total_revisions, hint)
999
self._try_pack_operations(hint)
1000
except RetryPackOperations:
1004
if clean_obsolete_packs:
1005
self._clear_obsolete_packs()
1007
def _try_pack_operations(self, hint):
1008
"""Calculate the pack operations based on the hint (if any), and
1276
1011
# determine which packs need changing
1277
pack_distribution = [1]
1278
1012
pack_operations = [[0, []]]
1279
1013
for pack in self.all_packs():
1280
pack_operations[-1][0] += pack.get_revision_count()
1281
pack_operations[-1][1].append(pack)
1282
self._execute_pack_operations(pack_operations, OptimisingPacker)
1014
if hint is None or pack.name in hint:
1015
# Either no hint was provided (so we are packing everything),
1016
# or this pack was included in the hint.
1017
pack_operations[-1][0] += pack.get_revision_count()
1018
pack_operations[-1][1].append(pack)
1019
self._execute_pack_operations(pack_operations,
1020
packer_class=self.optimising_packer_class,
1021
reload_func=self._restart_pack_operations)
1284
1023
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1285
1024
"""Plan a pack operation.
1498
1321
self._packs_by_name = {}
1499
1322
self._packs_at_load = None
1501
def _make_index_map(self, index_suffix):
1502
"""Return information on existing indices.
1504
:param suffix: Index suffix added to pack name.
1506
:returns: (pack_map, indices) where indices is a list of GraphIndex
1507
objects, and pack_map is a mapping from those objects to the
1508
pack tuple they describe.
1510
# TODO: stop using this; it creates new indices unnecessarily.
1511
self.ensure_loaded()
1512
suffix_map = {'.rix': 'revision_index',
1513
'.six': 'signature_index',
1514
'.iix': 'inventory_index',
1515
'.tix': 'text_index',
1517
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1518
suffix_map[index_suffix])
1520
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1521
"""Convert a list of packs to an index pack map and index list.
1523
:param packs: The packs list to process.
1524
:param index_attribute: The attribute that the desired index is found
1526
:return: A tuple (map, list) where map contains the dict from
1527
index:pack_tuple, and lsit contains the indices in the same order
1533
index = getattr(pack, index_attribute)
1534
indices.append(index)
1535
pack_map[index] = (pack.pack_transport, pack.file_name())
1536
return pack_map, indices
1538
def _index_contents(self, pack_map, key_filter=None):
1539
"""Get an iterable of the index contents from a pack_map.
1541
:param pack_map: A map from indices to pack details.
1542
:param key_filter: An optional filter to limit the
1545
indices = [index for index in pack_map.iterkeys()]
1546
all_index = CombinedGraphIndex(indices)
1547
if key_filter is None:
1548
return all_index.iter_all_entries()
1550
return all_index.iter_entries(key_filter)
1552
1324
def _unlock_names(self):
1553
1325
"""Release the mutex around the pack-names index."""
1554
1326
self.repo.control_files.unlock()
1556
def _save_pack_names(self, clear_obsolete_packs=False):
1557
"""Save the list of packs.
1559
This will take out the mutex around the pack names list for the
1560
duration of the method call. If concurrent updates have been made, a
1561
three-way merge between the current list and the current in memory list
1564
:param clear_obsolete_packs: If True, clear out the contents of the
1565
obsolete_packs directory.
1569
builder = self._index_builder_class()
1570
# load the disk nodes across
1572
for index, key, value in self._iter_disk_pack_index():
1573
disk_nodes.add((key, value))
1574
# do a two-way diff against our original content
1575
current_nodes = set()
1576
for name, sizes in self._names.iteritems():
1578
((name, ), ' '.join(str(size) for size in sizes)))
1579
deleted_nodes = self._packs_at_load - current_nodes
1580
new_nodes = current_nodes - self._packs_at_load
1581
disk_nodes.difference_update(deleted_nodes)
1582
disk_nodes.update(new_nodes)
1583
# TODO: handle same-name, index-size-changes here -
1584
# e.g. use the value from disk, not ours, *unless* we're the one
1586
for key, value in disk_nodes:
1587
builder.add_node(key, value)
1588
self.transport.put_file('pack-names', builder.finish(),
1589
mode=self.repo.bzrdir._get_file_mode())
1590
# move the baseline forward
1591
self._packs_at_load = disk_nodes
1592
if clear_obsolete_packs:
1593
self._clear_obsolete_packs()
1595
self._unlock_names()
1596
# synchronise the memory packs list with what we just wrote:
1328
def _diff_pack_names(self):
1329
"""Read the pack names from disk, and compare it to the one in memory.
1331
:return: (disk_nodes, deleted_nodes, new_nodes)
1332
disk_nodes The final set of nodes that should be referenced
1333
deleted_nodes Nodes which have been removed from when we started
1334
new_nodes Nodes that are newly introduced
1336
# load the disk nodes across
1338
for index, key, value in self._iter_disk_pack_index():
1339
disk_nodes.add((key, value))
1340
orig_disk_nodes = set(disk_nodes)
1342
# do a two-way diff against our original content
1343
current_nodes = set()
1344
for name, sizes in self._names.iteritems():
1346
((name, ), ' '.join(str(size) for size in sizes)))
1348
# Packs no longer present in the repository, which were present when we
1349
# locked the repository
1350
deleted_nodes = self._packs_at_load - current_nodes
1351
# Packs which this process is adding
1352
new_nodes = current_nodes - self._packs_at_load
1354
# Update the disk_nodes set to include the ones we are adding, and
1355
# remove the ones which were removed by someone else
1356
disk_nodes.difference_update(deleted_nodes)
1357
disk_nodes.update(new_nodes)
1359
return disk_nodes, deleted_nodes, new_nodes, orig_disk_nodes
1361
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1362
"""Given the correct set of pack files, update our saved info.
1364
:return: (removed, added, modified)
1365
removed pack names removed from self._names
1366
added pack names added to self._names
1367
modified pack names that had changed value
1372
## self._packs_at_load = disk_nodes
1597
1373
new_names = dict(disk_nodes)
1598
1374
# drop no longer present nodes
1599
1375
for pack in self.all_packs():
1600
1376
if (pack.name,) not in new_names:
1377
removed.append(pack.name)
1601
1378
self._remove_pack_from_memory(pack)
1602
1379
# add new nodes/refresh existing ones
1603
1380
for key, value in disk_nodes:
1613
1390
# disk index because the set values are the same, unless
1614
1391
# the only index shows up as deleted by the set difference
1615
1392
# - which it may. Until there is a specific test for this,
1616
# assume its broken. RBC 20071017.
1393
# assume it's broken. RBC 20071017.
1617
1394
self._remove_pack_from_memory(self.get_pack_by_name(name))
1618
1395
self._names[name] = sizes
1619
1396
self.get_pack_by_name(name)
1397
modified.append(name)
1622
1400
self._names[name] = sizes
1623
1401
self.get_pack_by_name(name)
1625
def _clear_obsolete_packs(self):
1403
return removed, added, modified
1405
def _save_pack_names(self, clear_obsolete_packs=False, obsolete_packs=None):
1406
"""Save the list of packs.
1408
This will take out the mutex around the pack names list for the
1409
duration of the method call. If concurrent updates have been made, a
1410
three-way merge between the current list and the current in memory list
1413
:param clear_obsolete_packs: If True, clear out the contents of the
1414
obsolete_packs directory.
1415
:param obsolete_packs: Packs that are obsolete once the new pack-names
1416
file has been written.
1417
:return: A list of the names saved that were not previously on disk.
1419
already_obsolete = []
1422
builder = self._index_builder_class()
1423
(disk_nodes, deleted_nodes, new_nodes,
1424
orig_disk_nodes) = self._diff_pack_names()
1425
# TODO: handle same-name, index-size-changes here -
1426
# e.g. use the value from disk, not ours, *unless* we're the one
1428
for key, value in disk_nodes:
1429
builder.add_node(key, value)
1430
self.transport.put_file('pack-names', builder.finish(),
1431
mode=self.repo.bzrdir._get_file_mode())
1432
self._packs_at_load = disk_nodes
1433
if clear_obsolete_packs:
1436
to_preserve = set([o.name for o in obsolete_packs])
1437
already_obsolete = self._clear_obsolete_packs(to_preserve)
1439
self._unlock_names()
1440
# synchronise the memory packs list with what we just wrote:
1441
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1443
# TODO: We could add one more condition here. "if o.name not in
1444
# orig_disk_nodes and o != the new_pack we haven't written to
1445
# disk yet. However, the new pack object is not easily
1446
# accessible here (it would have to be passed through the
1447
# autopacking code, etc.)
1448
obsolete_packs = [o for o in obsolete_packs
1449
if o.name not in already_obsolete]
1450
self._obsolete_packs(obsolete_packs)
1451
return [new_node[0][0] for new_node in new_nodes]
1453
def reload_pack_names(self):
1454
"""Sync our pack listing with what is present in the repository.
1456
This should be called when we find out that something we thought was
1457
present is now missing. This happens when another process re-packs the
1460
:return: True if the in-memory list of packs has been altered at all.
1462
# The ensure_loaded call is to handle the case where the first call
1463
# made involving the collection was to reload_pack_names, where we
1464
# don't have a view of disk contents. It's a bit of a bandaid, and
1465
# causes two reads of pack-names, but it's a rare corner case not
1466
# struck with regular push/pull etc.
1467
first_read = self.ensure_loaded()
1470
# out the new value.
1471
(disk_nodes, deleted_nodes, new_nodes,
1472
orig_disk_nodes) = self._diff_pack_names()
1473
# _packs_at_load is meant to be the explicit list of names in
1474
# 'pack-names' at then start. As such, it should not contain any
1475
# pending names that haven't been written out yet.
1476
self._packs_at_load = orig_disk_nodes
1478
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1479
if removed or added or modified:
1483
def _restart_autopack(self):
1484
"""Reload the pack names list, and restart the autopack code."""
1485
if not self.reload_pack_names():
1486
# Re-raise the original exception, because something went missing
1487
# and a restart didn't find it
1489
raise errors.RetryAutopack(self.repo, False, sys.exc_info())
1491
def _restart_pack_operations(self):
1492
"""Reload the pack names list, and restart the autopack code."""
1493
if not self.reload_pack_names():
1494
# Re-raise the original exception, because something went missing
1495
# and a restart didn't find it
1497
raise RetryPackOperations(self.repo, False, sys.exc_info())
1499
def _clear_obsolete_packs(self, preserve=None):
1626
1500
"""Delete everything from the obsolete-packs directory.
1502
:return: A list of pack identifiers (the filename without '.pack') that
1503
were found in obsolete_packs.
1628
1506
obsolete_pack_transport = self.transport.clone('obsolete_packs')
1629
for filename in obsolete_pack_transport.list_dir('.'):
1507
if preserve is None:
1510
obsolete_pack_files = obsolete_pack_transport.list_dir('.')
1511
except errors.NoSuchFile:
1513
for filename in obsolete_pack_files:
1514
name, ext = osutils.splitext(filename)
1517
if name in preserve:
1631
1520
obsolete_pack_transport.delete(filename)
1632
1521
except (errors.PathError, errors.TransportError), e:
1633
warning("couldn't delete obsolete pack, skipping it:\n%s" % (e,))
1522
warning("couldn't delete obsolete pack, skipping it:\n%s"
1635
1526
def _start_write_group(self):
1636
1527
# Do not permit preparation for writing if we're not in a 'write lock'.
1637
1528
if not self.repo.is_write_locked():
1638
1529
raise errors.NotWriteLocked(self)
1639
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1640
self._pack_transport, upload_suffix='.pack',
1641
file_mode=self.repo.bzrdir._get_file_mode(),
1642
index_builder_class=self._index_builder_class,
1643
index_class=self._index_class)
1530
self._new_pack = self.pack_factory(self, upload_suffix='.pack',
1531
file_mode=self.repo.bzrdir._get_file_mode())
1644
1532
# allow writing: queue writes to a new index
1645
1533
self.revision_index.add_writable_index(self._new_pack.revision_index,
1646
1534
self._new_pack)
1660
1554
# FIXME: just drop the transient index.
1661
1555
# forget what names there are
1662
1556
if self._new_pack is not None:
1663
self._new_pack.abort()
1664
self._remove_pack_indices(self._new_pack)
1665
self._new_pack = None
1666
self.repo._text_knit = None
1557
operation = cleanup.OperationWithCleanups(self._new_pack.abort)
1558
operation.add_cleanup(setattr, self, '_new_pack', None)
1559
# If we aborted while in the middle of finishing the write
1560
# group, _remove_pack_indices could fail because the indexes are
1561
# already gone. But they're not there we shouldn't fail in this
1562
# case, so we pass ignore_missing=True.
1563
operation.add_cleanup(self._remove_pack_indices, self._new_pack,
1564
ignore_missing=True)
1565
operation.run_simple()
1566
for resumed_pack in self._resumed_packs:
1567
operation = cleanup.OperationWithCleanups(resumed_pack.abort)
1568
# See comment in previous finally block.
1569
operation.add_cleanup(self._remove_pack_indices, resumed_pack,
1570
ignore_missing=True)
1571
operation.run_simple()
1572
del self._resumed_packs[:]
1574
def _remove_resumed_pack_indices(self):
1575
for resumed_pack in self._resumed_packs:
1576
self._remove_pack_indices(resumed_pack)
1577
del self._resumed_packs[:]
1579
def _check_new_inventories(self):
1580
"""Detect missing inventories in this write group.
1582
:returns: list of strs, summarising any problems found. If the list is
1583
empty no problems were found.
1585
# The base implementation does no checks. GCRepositoryPackCollection
1668
1589
def _commit_write_group(self):
1591
for prefix, versioned_file in (
1592
('revisions', self.repo.revisions),
1593
('inventories', self.repo.inventories),
1594
('texts', self.repo.texts),
1595
('signatures', self.repo.signatures),
1597
missing = versioned_file.get_missing_compression_parent_keys()
1598
all_missing.update([(prefix,) + key for key in missing])
1600
raise errors.BzrCheckError(
1601
"Repository %s has missing compression parent(s) %r "
1602
% (self.repo, sorted(all_missing)))
1603
problems = self._check_new_inventories()
1605
problems_summary = '\n'.join(problems)
1606
raise errors.BzrCheckError(
1607
"Cannot add revision(s) to repository: " + problems_summary)
1669
1608
self._remove_pack_indices(self._new_pack)
1609
any_new_content = False
1670
1610
if self._new_pack.data_inserted():
1671
1611
# get all the data to disk and read to use
1672
1612
self._new_pack.finish()
1673
1613
self.allocate(self._new_pack)
1674
1614
self._new_pack = None
1675
if not self.autopack():
1615
any_new_content = True
1617
self._new_pack.abort()
1618
self._new_pack = None
1619
for resumed_pack in self._resumed_packs:
1620
# XXX: this is a pretty ugly way to turn the resumed pack into a
1621
# properly committed pack.
1622
self._names[resumed_pack.name] = None
1623
self._remove_pack_from_memory(resumed_pack)
1624
resumed_pack.finish()
1625
self.allocate(resumed_pack)
1626
any_new_content = True
1627
del self._resumed_packs[:]
1629
result = self.autopack()
1676
1631
# when autopack takes no steps, the names list is still
1678
self._save_pack_names()
1633
return self._save_pack_names()
1637
def _suspend_write_group(self):
1638
tokens = [pack.name for pack in self._resumed_packs]
1639
self._remove_pack_indices(self._new_pack)
1640
if self._new_pack.data_inserted():
1641
# get all the data to disk and read to use
1642
self._new_pack.finish(suspend=True)
1643
tokens.append(self._new_pack.name)
1644
self._new_pack = None
1680
1646
self._new_pack.abort()
1681
1647
self._new_pack = None
1682
self.repo._text_knit = None
1685
class KnitPackRepository(KnitRepository):
1648
self._remove_resumed_pack_indices()
1651
def _resume_write_group(self, tokens):
1652
for token in tokens:
1653
self._resume_pack(token)
1656
class PackRepository(MetaDirVersionedFileRepository):
1686
1657
"""Repository with knit objects stored inside pack containers.
1688
1659
The layering for a KnitPackRepository is:
1690
1661
Graph | HPSS | Repository public layer |
1691
1662
===================================================
1692
1663
Tuple based apis below, string based, and key based apis above
1693
1664
---------------------------------------------------
1695
1666
Provides .texts, .revisions etc
1696
1667
This adapts the N-tuple keys to physical knit records which only have a
1697
1668
single string identifier (for historical reasons), which in older formats
1704
1675
pack file. The GraphIndex layer works in N-tuples and is unaware of any
1705
1676
semantic value.
1706
1677
===================================================
1681
# These attributes are inherited from the Repository base class. Setting
1682
# them to None ensures that if the constructor is changed to not initialize
1683
# them, or a subclass fails to call the constructor, that an error will
1684
# occur rather than the system working but generating incorrect data.
1685
_commit_builder_class = None
1710
1688
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
1712
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
1713
_commit_builder_class, _serializer)
1714
index_transport = self._transport.clone('indices')
1715
self._pack_collection = RepositoryPackCollection(self, self._transport,
1717
self._transport.clone('upload'),
1718
self._transport.clone('packs'),
1719
_format.index_builder_class,
1720
_format.index_class)
1721
self.inventories = KnitVersionedFiles(
1722
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
1723
add_callback=self._pack_collection.inventory_index.add_callback,
1724
deltas=True, parents=True, is_locked=self.is_locked),
1725
data_access=self._pack_collection.inventory_index.data_access,
1726
max_delta_chain=200)
1727
self.revisions = KnitVersionedFiles(
1728
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
1729
add_callback=self._pack_collection.revision_index.add_callback,
1730
deltas=False, parents=True, is_locked=self.is_locked),
1731
data_access=self._pack_collection.revision_index.data_access,
1733
self.signatures = KnitVersionedFiles(
1734
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
1735
add_callback=self._pack_collection.signature_index.add_callback,
1736
deltas=False, parents=False, is_locked=self.is_locked),
1737
data_access=self._pack_collection.signature_index.data_access,
1739
self.texts = KnitVersionedFiles(
1740
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
1741
add_callback=self._pack_collection.text_index.add_callback,
1742
deltas=True, parents=True, is_locked=self.is_locked),
1743
data_access=self._pack_collection.text_index.data_access,
1744
max_delta_chain=200)
1745
# True when the repository object is 'write locked' (as opposed to the
1746
# physical lock only taken out around changes to the pack-names list.)
1747
# Another way to represent this would be a decorator around the control
1748
# files object that presents logical locks as physical ones - if this
1749
# gets ugly consider that alternative design. RBC 20071011
1750
self._write_lock_count = 0
1751
self._transaction = None
1753
self._reconcile_does_inventory_gc = True
1690
MetaDirRepository.__init__(self, _format, a_bzrdir, control_files)
1691
self._commit_builder_class = _commit_builder_class
1692
self._serializer = _serializer
1754
1693
self._reconcile_fixes_text_parents = True
1755
self._reconcile_backsup_inventory = False
1756
self._fetch_order = 'unordered'
1694
if self._format.supports_external_lookups:
1695
self._unstacked_provider = graph.CachingParentsProvider(
1696
self._make_parents_provider_unstacked())
1698
self._unstacked_provider = graph.CachingParentsProvider(self)
1699
self._unstacked_provider.disable_cache()
1758
def _warn_if_deprecated(self):
1759
# This class isn't deprecated, but one sub-format is
1760
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
1761
from bzrlib import repository
1762
if repository._deprecation_warning_done:
1764
repository._deprecation_warning_done = True
1765
warning("Format %s for %s is deprecated - please use"
1766
" 'bzr upgrade --1.6.1-rich-root'"
1767
% (self._format, self.bzrdir.transport.base))
1702
def _all_revision_ids(self):
1703
"""See Repository.all_revision_ids()."""
1704
return [key[0] for key in self.revisions.keys()]
1769
1706
def _abort_write_group(self):
1707
self.revisions._index._key_dependencies.clear()
1770
1708
self._pack_collection._abort_write_group()
1772
def _find_inconsistent_revision_parents(self):
1773
"""Find revisions with incorrectly cached parents.
1775
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1776
parents-in-revision).
1778
if not self.is_locked():
1779
raise errors.ObjectNotLocked(self)
1780
pb = ui.ui_factory.nested_progress_bar()
1783
revision_nodes = self._pack_collection.revision_index \
1784
.combined_index.iter_all_entries()
1785
index_positions = []
1786
# Get the cached index values for all revisions, and also the location
1787
# in each index of the revision text so we can perform linear IO.
1788
for index, key, value, refs in revision_nodes:
1789
pos, length = value[1:].split(' ')
1790
index_positions.append((index, int(pos), key[0],
1791
tuple(parent[0] for parent in refs[0])))
1792
pb.update("Reading revision index.", 0, 0)
1793
index_positions.sort()
1794
batch_count = len(index_positions) / 1000 + 1
1795
pb.update("Checking cached revision graph.", 0, batch_count)
1796
for offset in xrange(batch_count):
1797
pb.update("Checking cached revision graph.", offset)
1798
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1801
rev_ids = [item[2] for item in to_query]
1802
revs = self.get_revisions(rev_ids)
1803
for revision, item in zip(revs, to_query):
1804
index_parents = item[3]
1805
rev_parents = tuple(revision.parent_ids)
1806
if index_parents != rev_parents:
1807
result.append((revision.revision_id, index_parents, rev_parents))
1812
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
1813
def get_parents(self, revision_ids):
1814
"""See graph._StackedParentsProvider.get_parents."""
1815
parent_map = self.get_parent_map(revision_ids)
1816
return [parent_map.get(r, None) for r in revision_ids]
1818
1710
def _make_parents_provider(self):
1819
return graph.CachingParentsProvider(self)
1711
if not self._format.supports_external_lookups:
1712
return self._unstacked_provider
1713
return graph.StackedParentsProvider(_LazyListJoin(
1714
[self._unstacked_provider], self._fallback_repositories))
1821
1716
def _refresh_data(self):
1822
if self._write_lock_count == 1 or (
1823
self.control_files._lock_count == 1 and
1824
self.control_files._lock_mode == 'r'):
1825
# forget what names there are
1826
self._pack_collection.reset()
1827
# XXX: Better to do an in-memory merge when acquiring a new lock -
1828
# factor out code from _save_pack_names.
1829
self._pack_collection.ensure_loaded()
1717
if not self.is_locked():
1719
self._pack_collection.reload_pack_names()
1720
self._unstacked_provider.disable_cache()
1721
self._unstacked_provider.enable_cache()
1831
1723
def _start_write_group(self):
1832
1724
self._pack_collection._start_write_group()
1834
1726
def _commit_write_group(self):
1835
return self._pack_collection._commit_write_group()
1727
hint = self._pack_collection._commit_write_group()
1728
self.revisions._index._key_dependencies.clear()
1729
# The commit may have added keys that were previously cached as
1730
# missing, so reset the cache.
1731
self._unstacked_provider.disable_cache()
1732
self._unstacked_provider.enable_cache()
1735
def suspend_write_group(self):
1736
# XXX check self._write_group is self.get_transaction()?
1737
tokens = self._pack_collection._suspend_write_group()
1738
self.revisions._index._key_dependencies.clear()
1739
self._write_group = None
1742
def _resume_write_group(self, tokens):
1743
self._start_write_group()
1745
self._pack_collection._resume_write_group(tokens)
1746
except errors.UnresumableWriteGroup:
1747
self._abort_write_group()
1749
for pack in self._pack_collection._resumed_packs:
1750
self.revisions._index.scan_unvalidated_index(pack.revision_index)
1837
1752
def get_transaction(self):
1838
1753
if self._write_lock_count:
1984
1936
_serializer=self._serializer)
1987
class RepositoryFormatKnitPack1(RepositoryFormatPack):
1988
"""A no-subtrees parameterized Pack repository.
1990
This format was introduced in 0.92.
1993
repository_class = KnitPackRepository
1994
_commit_builder_class = PackCommitBuilder
1996
def _serializer(self):
1997
return xml5.serializer_v5
1998
# What index classes to use
1999
index_builder_class = InMemoryGraphIndex
2000
index_class = GraphIndex
2002
def _get_matching_bzrdir(self):
2003
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2005
def _ignore_setting_bzrdir(self, format):
2008
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2010
def get_format_string(self):
2011
"""See RepositoryFormat.get_format_string()."""
2012
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2014
def get_format_description(self):
2015
"""See RepositoryFormat.get_format_description()."""
2016
return "Packs containing knits without subtree support"
2018
def check_conversion_target(self, target_format):
2022
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2023
"""A subtrees parameterized Pack repository.
2025
This repository format uses the xml7 serializer to get:
2026
- support for recording full info about the tree root
2027
- support for recording tree-references
2029
This format was introduced in 0.92.
2032
repository_class = KnitPackRepository
2033
_commit_builder_class = PackRootCommitBuilder
2034
rich_root_data = True
2035
supports_tree_reference = True
2037
def _serializer(self):
2038
return xml7.serializer_v7
2039
# What index classes to use
2040
index_builder_class = InMemoryGraphIndex
2041
index_class = GraphIndex
2043
def _get_matching_bzrdir(self):
2044
return bzrdir.format_registry.make_bzrdir(
2045
'pack-0.92-subtree')
2047
def _ignore_setting_bzrdir(self, format):
2050
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2052
def check_conversion_target(self, target_format):
2053
if not target_format.rich_root_data:
2054
raise errors.BadConversionTarget(
2055
'Does not support rich root data.', target_format)
2056
if not getattr(target_format, 'supports_tree_reference', False):
2057
raise errors.BadConversionTarget(
2058
'Does not support nested trees', target_format)
2060
def get_format_string(self):
2061
"""See RepositoryFormat.get_format_string()."""
2062
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2064
def get_format_description(self):
2065
"""See RepositoryFormat.get_format_description()."""
2066
return "Packs containing knits with subtree support\n"
2069
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2070
"""A rich-root, no subtrees parameterized Pack repository.
2072
This repository format uses the xml6 serializer to get:
2073
- support for recording full info about the tree root
2075
This format was introduced in 1.0.
2078
repository_class = KnitPackRepository
2079
_commit_builder_class = PackRootCommitBuilder
2080
rich_root_data = True
2081
supports_tree_reference = False
2083
def _serializer(self):
2084
return xml6.serializer_v6
2085
# What index classes to use
2086
index_builder_class = InMemoryGraphIndex
2087
index_class = GraphIndex
2089
def _get_matching_bzrdir(self):
2090
return bzrdir.format_registry.make_bzrdir(
2093
def _ignore_setting_bzrdir(self, format):
2096
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2098
def check_conversion_target(self, target_format):
2099
if not target_format.rich_root_data:
2100
raise errors.BadConversionTarget(
2101
'Does not support rich root data.', target_format)
2103
def get_format_string(self):
2104
"""See RepositoryFormat.get_format_string()."""
2105
return ("Bazaar pack repository format 1 with rich root"
2106
" (needs bzr 1.0)\n")
2108
def get_format_description(self):
2109
"""See RepositoryFormat.get_format_description()."""
2110
return "Packs containing knits with rich root support\n"
2113
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2114
"""Repository that supports external references to allow stacking.
2118
Supports external lookups, which results in non-truncated ghosts after
2119
reconcile compared to pack-0.92 formats.
2122
repository_class = KnitPackRepository
2123
_commit_builder_class = PackCommitBuilder
2124
supports_external_lookups = True
2125
# What index classes to use
2126
index_builder_class = InMemoryGraphIndex
2127
index_class = GraphIndex
2130
def _serializer(self):
2131
return xml5.serializer_v5
2133
def _get_matching_bzrdir(self):
2134
return bzrdir.format_registry.make_bzrdir('1.6')
2136
def _ignore_setting_bzrdir(self, format):
2139
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2141
def get_format_string(self):
2142
"""See RepositoryFormat.get_format_string()."""
2143
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2145
def get_format_description(self):
2146
"""See RepositoryFormat.get_format_description()."""
2147
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2149
def check_conversion_target(self, target_format):
2153
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2154
"""A repository with rich roots and stacking.
2156
New in release 1.6.1.
2158
Supports stacking on other repositories, allowing data to be accessed
2159
without being stored locally.
2162
repository_class = KnitPackRepository
2163
_commit_builder_class = PackRootCommitBuilder
2164
rich_root_data = True
2165
supports_tree_reference = False # no subtrees
2166
supports_external_lookups = True
2167
# What index classes to use
2168
index_builder_class = InMemoryGraphIndex
2169
index_class = GraphIndex
2172
def _serializer(self):
2173
return xml6.serializer_v6
2175
def _get_matching_bzrdir(self):
2176
return bzrdir.format_registry.make_bzrdir(
2179
def _ignore_setting_bzrdir(self, format):
2182
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2184
def check_conversion_target(self, target_format):
2185
if not target_format.rich_root_data:
2186
raise errors.BadConversionTarget(
2187
'Does not support rich root data.', target_format)
2189
def get_format_string(self):
2190
"""See RepositoryFormat.get_format_string()."""
2191
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2193
def get_format_description(self):
2194
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2197
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2198
"""A repository with rich roots and external references.
2202
Supports external lookups, which results in non-truncated ghosts after
2203
reconcile compared to pack-0.92 formats.
2205
This format was deprecated because the serializer it uses accidentally
2206
supported subtrees, when the format was not intended to. This meant that
2207
someone could accidentally fetch from an incorrect repository.
2210
repository_class = KnitPackRepository
2211
_commit_builder_class = PackRootCommitBuilder
2212
rich_root_data = True
2213
supports_tree_reference = False # no subtrees
2215
supports_external_lookups = True
2216
# What index classes to use
2217
index_builder_class = InMemoryGraphIndex
2218
index_class = GraphIndex
2221
def _serializer(self):
2222
return xml7.serializer_v7
2224
def _get_matching_bzrdir(self):
2225
return bzrdir.format_registry.make_bzrdir(
2228
def _ignore_setting_bzrdir(self, format):
2231
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2233
def check_conversion_target(self, target_format):
2234
if not target_format.rich_root_data:
2235
raise errors.BadConversionTarget(
2236
'Does not support rich root data.', target_format)
2238
def get_format_string(self):
2239
"""See RepositoryFormat.get_format_string()."""
2240
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2242
def get_format_description(self):
2243
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2247
class RepositoryFormatPackDevelopment2(RepositoryFormatPack):
2248
"""A no-subtrees development repository.
2250
This format should be retained until the second release after bzr 1.7.
2252
This is pack-1.6.1 with B+Tree indices.
2255
repository_class = KnitPackRepository
2256
_commit_builder_class = PackCommitBuilder
2257
supports_external_lookups = True
2258
# What index classes to use
2259
index_builder_class = BTreeBuilder
2260
index_class = BTreeGraphIndex
2263
def _serializer(self):
2264
return xml5.serializer_v5
2266
def _get_matching_bzrdir(self):
2267
return bzrdir.format_registry.make_bzrdir('development2')
2269
def _ignore_setting_bzrdir(self, format):
2272
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2274
def get_format_string(self):
2275
"""See RepositoryFormat.get_format_string()."""
2276
return "Bazaar development format 2 (needs bzr.dev from before 1.8)\n"
2278
def get_format_description(self):
2279
"""See RepositoryFormat.get_format_description()."""
2280
return ("Development repository format, currently the same as "
2281
"1.6.1 with B+Trees.\n")
2283
def check_conversion_target(self, target_format):
2287
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2288
"""A subtrees development repository.
2290
This format should be retained until the second release after bzr 1.7.
2292
1.6.1-subtree[as it might have been] with B+Tree indices.
2295
repository_class = KnitPackRepository
2296
_commit_builder_class = PackRootCommitBuilder
2297
rich_root_data = True
2298
supports_tree_reference = True
2299
supports_external_lookups = True
2300
# What index classes to use
2301
index_builder_class = BTreeBuilder
2302
index_class = BTreeGraphIndex
2305
def _serializer(self):
2306
return xml7.serializer_v7
2308
def _get_matching_bzrdir(self):
2309
return bzrdir.format_registry.make_bzrdir(
2310
'development2-subtree')
2312
def _ignore_setting_bzrdir(self, format):
2315
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2317
def check_conversion_target(self, target_format):
2318
if not target_format.rich_root_data:
2319
raise errors.BadConversionTarget(
2320
'Does not support rich root data.', target_format)
2321
if not getattr(target_format, 'supports_tree_reference', False):
2322
raise errors.BadConversionTarget(
2323
'Does not support nested trees', target_format)
2325
def get_format_string(self):
2326
"""See RepositoryFormat.get_format_string()."""
2327
return ("Bazaar development format 2 with subtree support "
2328
"(needs bzr.dev from before 1.8)\n")
2330
def get_format_description(self):
2331
"""See RepositoryFormat.get_format_description()."""
2332
return ("Development repository format, currently the same as "
2333
"1.6.1-subtree with B+Tree indices.\n")
1939
class RetryPackOperations(errors.RetryWithNewPacks):
1940
"""Raised when we are packing and we find a missing file.
1942
Meant as a signaling exception, to tell the RepositoryPackCollection.pack
1943
code it should try again.
1946
internal_error = True
1948
_fmt = ("Pack files have changed, reload and try pack again."
1949
" context: %(context)s %(orig_error)s")
1952
class _DirectPackAccess(object):
1953
"""Access to data in one or more packs with less translation."""
1955
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
1956
"""Create a _DirectPackAccess object.
1958
:param index_to_packs: A dict mapping index objects to the transport
1959
and file names for obtaining data.
1960
:param reload_func: A function to call if we determine that the pack
1961
files have moved and we need to reload our caches. See
1962
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
1964
self._container_writer = None
1965
self._write_index = None
1966
self._indices = index_to_packs
1967
self._reload_func = reload_func
1968
self._flush_func = flush_func
1970
def add_raw_records(self, key_sizes, raw_data):
1971
"""Add raw knit bytes to a storage area.
1973
The data is spooled to the container writer in one bytes-record per
1976
:param sizes: An iterable of tuples containing the key and size of each
1978
:param raw_data: A bytestring containing the data.
1979
:return: A list of memos to retrieve the record later. Each memo is an
1980
opaque index memo. For _DirectPackAccess the memo is (index, pos,
1981
length), where the index field is the write_index object supplied
1982
to the PackAccess object.
1984
if type(raw_data) is not str:
1985
raise AssertionError(
1986
'data must be plain bytes was %s' % type(raw_data))
1989
for key, size in key_sizes:
1990
p_offset, p_length = self._container_writer.add_bytes_record(
1991
raw_data[offset:offset+size], [])
1993
result.append((self._write_index, p_offset, p_length))
1997
"""Flush pending writes on this access object.
1999
This will flush any buffered writes to a NewPack.
2001
if self._flush_func is not None:
2004
def get_raw_records(self, memos_for_retrieval):
2005
"""Get the raw bytes for a records.
2007
:param memos_for_retrieval: An iterable containing the (index, pos,
2008
length) memo for retrieving the bytes. The Pack access method
2009
looks up the pack to use for a given record in its index_to_pack
2011
:return: An iterator over the bytes of the records.
2013
# first pass, group into same-index requests
2015
current_index = None
2016
for (index, offset, length) in memos_for_retrieval:
2017
if current_index == index:
2018
current_list.append((offset, length))
2020
if current_index is not None:
2021
request_lists.append((current_index, current_list))
2022
current_index = index
2023
current_list = [(offset, length)]
2024
# handle the last entry
2025
if current_index is not None:
2026
request_lists.append((current_index, current_list))
2027
for index, offsets in request_lists:
2029
transport, path = self._indices[index]
2031
# A KeyError here indicates that someone has triggered an index
2032
# reload, and this index has gone missing, we need to start
2034
if self._reload_func is None:
2035
# If we don't have a _reload_func there is nothing that can
2038
raise errors.RetryWithNewPacks(index,
2039
reload_occurred=True,
2040
exc_info=sys.exc_info())
2042
reader = pack.make_readv_reader(transport, path, offsets)
2043
for names, read_func in reader.iter_records():
2044
yield read_func(None)
2045
except errors.NoSuchFile:
2046
# A NoSuchFile error indicates that a pack file has gone
2047
# missing on disk, we need to trigger a reload, and start over.
2048
if self._reload_func is None:
2050
raise errors.RetryWithNewPacks(transport.abspath(path),
2051
reload_occurred=False,
2052
exc_info=sys.exc_info())
2054
def set_writer(self, writer, index, transport_packname):
2055
"""Set a writer to use for adding data."""
2056
if index is not None:
2057
self._indices[index] = transport_packname
2058
self._container_writer = writer
2059
self._write_index = index
2061
def reload_or_raise(self, retry_exc):
2062
"""Try calling the reload function, or re-raise the original exception.
2064
This should be called after _DirectPackAccess raises a
2065
RetryWithNewPacks exception. This function will handle the common logic
2066
of determining when the error is fatal versus being temporary.
2067
It will also make sure that the original exception is raised, rather
2068
than the RetryWithNewPacks exception.
2070
If this function returns, then the calling function should retry
2071
whatever operation was being performed. Otherwise an exception will
2074
:param retry_exc: A RetryWithNewPacks exception.
2077
if self._reload_func is None:
2079
elif not self._reload_func():
2080
# The reload claimed that nothing changed
2081
if not retry_exc.reload_occurred:
2082
# If there wasn't an earlier reload, then we really were
2083
# expecting to find changes. We didn't find them, so this is a
2087
exc_class, exc_value, exc_traceback = retry_exc.exc_info
2088
raise exc_class, exc_value, exc_traceback
added
removed
bzrlib/repofmt/pack_repo.py
