13
13
# You should have received a copy of the GNU General Public License
14
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20
17
from bzrlib.lazy_import import lazy_import
21
18
lazy_import(globals(), """
22
19
from itertools import izip
25
24
from bzrlib import (
38
29
from bzrlib.index import (
39
33
CombinedGraphIndex,
40
34
GraphIndexPrefixAdapter,
42
from bzrlib.knit import (
48
from bzrlib import tsort
36
from bzrlib.knit import KnitGraphIndex, _PackAccess, _KnitData
37
from bzrlib.osutils import rand_chars
38
from bzrlib.pack import ContainerWriter
39
from bzrlib.store import revision
50
41
from bzrlib import (
55
revision as _mod_revision,
58
from bzrlib.decorators import needs_write_lock, only_raises
59
from bzrlib.btree_index import (
63
from bzrlib.index import (
67
from bzrlib.lock import LogicalLockResult
54
from bzrlib.decorators import needs_read_lock, needs_write_lock
68
55
from bzrlib.repofmt.knitrepo import KnitRepository
69
56
from bzrlib.repository import (
71
59
MetaDirRepositoryFormat,
73
RepositoryWriteLockResult,
77
from bzrlib.trace import (
62
import bzrlib.revision as _mod_revision
63
from bzrlib.store.revision.knit import KnitRevisionStore
64
from bzrlib.store.versioned import VersionedFileStore
65
from bzrlib.trace import mutter, note, warning
84
68
class PackCommitBuilder(CommitBuilder):
85
69
"""A subclass of CommitBuilder to add texts with pack semantics.
87
71
Specifically this uses one knit object rather than one knit object per
88
72
added text, reducing memory and object pressure.
91
def __init__(self, repository, parents, config, timestamp=None,
92
timezone=None, committer=None, revprops=None,
94
CommitBuilder.__init__(self, repository, parents, config,
95
timestamp=timestamp, timezone=timezone, committer=committer,
96
revprops=revprops, revision_id=revision_id)
97
self._file_graph = graph.Graph(
98
repository._pack_collection.text_index.combined_index)
100
def _heads(self, file_id, revision_ids):
101
keys = [(file_id, revision_id) for revision_id in revision_ids]
102
return set([key[1] for key in self._file_graph.heads(keys)])
75
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
76
return self.repository._pack_collection._add_text_to_weave(file_id,
77
self._new_revision_id, new_lines, parents, nostore_sha,
105
81
class PackRootCommitBuilder(RootCommitBuilder):
106
82
"""A subclass of RootCommitBuilder to add texts with pack semantics.
108
84
Specifically this uses one knit object rather than one knit object per
109
85
added text, reducing memory and object pressure.
112
def __init__(self, repository, parents, config, timestamp=None,
113
timezone=None, committer=None, revprops=None,
115
CommitBuilder.__init__(self, repository, parents, config,
116
timestamp=timestamp, timezone=timezone, committer=committer,
117
revprops=revprops, revision_id=revision_id)
118
self._file_graph = graph.Graph(
119
repository._pack_collection.text_index.combined_index)
121
def _heads(self, file_id, revision_ids):
122
keys = [(file_id, revision_id) for revision_id in revision_ids]
123
return set([key[1] for key in self._file_graph.heads(keys)])
88
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
89
return self.repository._pack_collection._add_text_to_weave(file_id,
90
self._new_revision_id, new_lines, parents, nostore_sha,
126
94
class Pack(object):
263
170
return not self.__eq__(other)
265
172
def __repr__(self):
266
return "<%s.%s object at 0x%x, %s, %s" % (
267
self.__class__.__module__, self.__class__.__name__, id(self),
268
self.pack_transport, self.name)
271
class ResumedPack(ExistingPack):
273
def __init__(self, name, revision_index, inventory_index, text_index,
274
signature_index, upload_transport, pack_transport, index_transport,
275
pack_collection, chk_index=None):
276
"""Create a ResumedPack object."""
277
ExistingPack.__init__(self, pack_transport, name, revision_index,
278
inventory_index, text_index, signature_index,
280
self.upload_transport = upload_transport
281
self.index_transport = index_transport
282
self.index_sizes = [None, None, None, None]
284
('revision', revision_index),
285
('inventory', inventory_index),
286
('text', text_index),
287
('signature', signature_index),
289
if chk_index is not None:
290
indices.append(('chk', chk_index))
291
self.index_sizes.append(None)
292
for index_type, index in indices:
293
offset = self.index_offset(index_type)
294
self.index_sizes[offset] = index._size
295
self.index_class = pack_collection._index_class
296
self._pack_collection = pack_collection
297
self._state = 'resumed'
298
# XXX: perhaps check that the .pack file exists?
300
def access_tuple(self):
301
if self._state == 'finished':
302
return Pack.access_tuple(self)
303
elif self._state == 'resumed':
304
return self.upload_transport, self.file_name()
306
raise AssertionError(self._state)
309
self.upload_transport.delete(self.file_name())
310
indices = [self.revision_index, self.inventory_index, self.text_index,
311
self.signature_index]
312
if self.chk_index is not None:
313
indices.append(self.chk_index)
314
for index in indices:
315
index._transport.delete(index._name)
318
self._check_references()
319
index_types = ['revision', 'inventory', 'text', 'signature']
320
if self.chk_index is not None:
321
index_types.append('chk')
322
for index_type in index_types:
323
old_name = self.index_name(index_type, self.name)
324
new_name = '../indices/' + old_name
325
self.upload_transport.rename(old_name, new_name)
326
self._replace_index_with_readonly(index_type)
327
new_name = '../packs/' + self.file_name()
328
self.upload_transport.rename(self.file_name(), new_name)
329
self._state = 'finished'
331
def _get_external_refs(self, index):
332
"""Return compression parents for this index that are not present.
334
This returns any compression parents that are referenced by this index,
335
which are not contained *in* this index. They may be present elsewhere.
337
return index.external_references(1)
173
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
174
id(self), self.transport, self.name)
340
177
class NewPack(Pack):
341
178
"""An in memory proxy for a pack which is being created."""
343
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
180
# A map of index 'type' to the file extension and position in the
182
index_definitions = {
183
'revision': ('.rix', 0),
184
'inventory': ('.iix', 1),
186
'signature': ('.six', 3),
189
def __init__(self, upload_transport, index_transport, pack_transport,
344
191
"""Create a NewPack instance.
346
:param pack_collection: A PackCollection into which this is being inserted.
193
:param upload_transport: A writable transport for the pack to be
194
incrementally uploaded to.
195
:param index_transport: A writable transport for the pack's indices to
196
be written to when the pack is finished.
197
:param pack_transport: A writable transport for the pack to be renamed
198
to when the upload is complete. This *must* be the same as
199
upload_transport.clone('../packs').
347
200
:param upload_suffix: An optional suffix to be given to any temporary
348
201
files created during the pack creation. e.g '.autopack'
349
:param file_mode: Unix permissions for newly created file.
351
203
# The relative locations of the packs are constrained, but all are
352
204
# passed in because the caller has them, so as to avoid object churn.
353
index_builder_class = pack_collection._index_builder_class
354
if pack_collection.chk_index is not None:
355
chk_index = index_builder_class(reference_lists=0)
358
205
Pack.__init__(self,
359
206
# Revisions: parents list, no text compression.
360
index_builder_class(reference_lists=1),
207
InMemoryGraphIndex(reference_lists=1),
361
208
# Inventory: We want to map compression only, but currently the
362
209
# knit code hasn't been updated enough to understand that, so we
363
210
# have a regular 2-list index giving parents and compression
365
index_builder_class(reference_lists=2),
212
InMemoryGraphIndex(reference_lists=2),
366
213
# Texts: compression and per file graph, for all fileids - so two
367
214
# reference lists and two elements in the key tuple.
368
index_builder_class(reference_lists=2, key_elements=2),
215
InMemoryGraphIndex(reference_lists=2, key_elements=2),
369
216
# Signatures: Just blobs to store, no compression, no parents
371
index_builder_class(reference_lists=0),
372
# CHK based storage - just blobs, no compression or parents.
218
InMemoryGraphIndex(reference_lists=0),
375
self._pack_collection = pack_collection
376
# When we make readonly indices, we need this.
377
self.index_class = pack_collection._index_class
378
220
# where should the new pack be opened
379
self.upload_transport = pack_collection._upload_transport
221
self.upload_transport = upload_transport
380
222
# where are indices written out to
381
self.index_transport = pack_collection._index_transport
223
self.index_transport = index_transport
382
224
# where is the pack renamed to when it is finished?
383
self.pack_transport = pack_collection._pack_transport
384
# What file mode to upload the pack and indices with.
385
self._file_mode = file_mode
225
self.pack_transport = pack_transport
386
226
# tracks the content written to the .pack file.
387
self._hash = osutils.md5()
388
# a tuple with the length in bytes of the indices, once the pack
389
# is finalised. (rev, inv, text, sigs, chk_if_in_use)
227
self._hash = md5.new()
228
# a four-tuple with the length in bytes of the indices, once the pack
229
# is finalised. (rev, inv, text, sigs)
390
230
self.index_sizes = None
391
231
# How much data to cache when writing packs. Note that this is not
392
232
# synchronised with reads, because it's not in the transport layer, so
751
532
def open_pack(self):
752
533
"""Open a pack for the pack we are creating."""
753
new_pack = self._pack_collection.pack_factory(self._pack_collection,
754
upload_suffix=self.suffix,
755
file_mode=self._pack_collection.repo.bzrdir._get_file_mode())
756
# We know that we will process all nodes in order, and don't need to
757
# query, so don't combine any indices spilled to disk until we are done
758
new_pack.revision_index.set_optimize(combine_backing_indices=False)
759
new_pack.inventory_index.set_optimize(combine_backing_indices=False)
760
new_pack.text_index.set_optimize(combine_backing_indices=False)
761
new_pack.signature_index.set_optimize(combine_backing_indices=False)
764
def _update_pack_order(self, entries, index_to_pack_map):
765
"""Determine how we want our packs to be ordered.
767
This changes the sort order of the self.packs list so that packs unused
768
by 'entries' will be at the end of the list, so that future requests
769
can avoid probing them. Used packs will be at the front of the
770
self.packs list, in the order of their first use in 'entries'.
772
:param entries: A list of (index, ...) tuples
773
:param index_to_pack_map: A mapping from index objects to pack objects.
777
for entry in entries:
779
if index not in seen_indexes:
780
packs.append(index_to_pack_map[index])
781
seen_indexes.add(index)
782
if len(packs) == len(self.packs):
783
if 'pack' in debug.debug_flags:
784
mutter('Not changing pack list, all packs used.')
786
seen_packs = set(packs)
787
for pack in self.packs:
788
if pack not in seen_packs:
534
return NewPack(self._pack_collection._upload_transport,
535
self._pack_collection._index_transport,
536
self._pack_collection._pack_transport, upload_suffix=self.suffix)
538
def _create_pack_from_packs(self):
539
self.pb.update("Opening pack", 0, 5)
540
new_pack = self.open_pack()
541
# buffer data - we won't be reading-back during the pack creation and
542
# this makes a significant difference on sftp pushes.
543
new_pack.set_write_cache_size(1024*1024)
791
544
if 'pack' in debug.debug_flags:
792
old_names = [p.access_tuple()[1] for p in self.packs]
793
new_names = [p.access_tuple()[1] for p in packs]
794
mutter('Reordering packs\nfrom: %s\n to: %s',
795
old_names, new_names)
798
def _copy_revision_texts(self):
799
"""Copy revision data to the new pack."""
545
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
546
for a_pack in self.packs]
547
if self.revision_ids is not None:
548
rev_count = len(self.revision_ids)
551
mutter('%s: create_pack: creating pack from source packs: '
552
'%s%s %s revisions wanted %s t=0',
553
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
554
plain_pack_list, rev_count)
800
555
# select revisions
801
556
if self.revision_ids:
802
557
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
804
559
revision_keys = None
805
561
# select revision keys
806
revision_index_map, revision_indices = self._pack_map_and_index_list(
808
revision_nodes = self._index_contents(revision_indices, revision_keys)
809
revision_nodes = list(revision_nodes)
810
self._update_pack_order(revision_nodes, revision_index_map)
562
revision_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
563
self.packs, 'revision_index')[0]
564
revision_nodes = self._pack_collection._index_contents(revision_index_map, revision_keys)
811
565
# copy revision keys and adjust values
812
566
self.pb.update("Copying revision texts", 1)
813
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
814
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
815
self.new_pack.revision_index, readv_group_iter, total_items))
567
list(self._copy_nodes_graph(revision_nodes, revision_index_map,
568
new_pack._writer, new_pack.revision_index))
816
569
if 'pack' in debug.debug_flags:
817
570
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
818
time.ctime(), self._pack_collection._upload_transport.base,
819
self.new_pack.random_name,
820
self.new_pack.revision_index.key_count(),
821
time.time() - self.new_pack.start_time)
822
self._revision_keys = revision_keys
824
def _copy_inventory_texts(self):
825
"""Copy the inventory texts to the new pack.
827
self._revision_keys is used to determine what inventories to copy.
829
Sets self._text_filter appropriately.
571
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
572
new_pack.revision_index.key_count(),
573
time.time() - new_pack.start_time)
831
574
# select inventory keys
832
inv_keys = self._revision_keys # currently the same keyspace, and note that
575
inv_keys = revision_keys # currently the same keyspace, and note that
833
576
# querying for keys here could introduce a bug where an inventory item
834
577
# is missed, so do not change it to query separately without cross
835
578
# checking like the text key check below.
836
inventory_index_map, inventory_indices = self._pack_map_and_index_list(
838
inv_nodes = self._index_contents(inventory_indices, inv_keys)
579
inventory_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
580
self.packs, 'inventory_index')[0]
581
inv_nodes = self._pack_collection._index_contents(inventory_index_map, inv_keys)
839
582
# copy inventory keys and adjust values
840
583
# XXX: Should be a helper function to allow different inv representation
842
585
self.pb.update("Copying inventory texts", 2)
843
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
844
# Only grab the output lines if we will be processing them
845
output_lines = bool(self.revision_ids)
846
inv_lines = self._copy_nodes_graph(inventory_index_map,
847
self.new_pack._writer, self.new_pack.inventory_index,
848
readv_group_iter, total_items, output_lines=output_lines)
586
inv_lines = self._copy_nodes_graph(inv_nodes, inventory_index_map,
587
new_pack._writer, new_pack.inventory_index, output_lines=True)
849
588
if self.revision_ids:
850
self._process_inventory_lines(inv_lines)
589
fileid_revisions = self._pack_collection.repo._find_file_ids_from_xml_inventory_lines(
590
inv_lines, self.revision_ids)
592
for fileid, file_revids in fileid_revisions.iteritems():
594
[(fileid, file_revid) for file_revid in file_revids])
852
596
# eat the iterator to cause it to execute.
854
self._text_filter = None
855
599
if 'pack' in debug.debug_flags:
856
600
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
857
time.ctime(), self._pack_collection._upload_transport.base,
858
self.new_pack.random_name,
859
self.new_pack.inventory_index.key_count(),
860
time.time() - self.new_pack.start_time)
862
def _copy_text_texts(self):
601
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
602
new_pack.inventory_index.key_count(),
603
time.time() - new_pack.start_time)
863
604
# select text keys
864
text_index_map, text_nodes = self._get_text_nodes()
865
if self._text_filter is not None:
605
text_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
606
self.packs, 'text_index')[0]
607
text_nodes = self._pack_collection._index_contents(text_index_map, text_filter)
608
if text_filter is not None:
866
609
# We could return the keys copied as part of the return value from
867
610
# _copy_nodes_graph but this doesn't work all that well with the
868
611
# need to get line output too, so we check separately, and as we're
1005
689
# linear scan up the pack
1006
690
pack_readv_requests.sort()
1008
pack_obj = index_map[index]
1009
transport, path = pack_obj.access_tuple()
1011
reader = pack.make_readv_reader(transport, path,
1012
[offset[0:2] for offset in pack_readv_requests])
1013
except errors.NoSuchFile:
1014
if self._reload_func is not None:
692
transport, path = index_map[index]
693
reader = pack.make_readv_reader(transport, path,
694
[offset[0:2] for offset in pack_readv_requests])
1017
695
for (names, read_func), (_1, _2, (key, eol_flag)) in \
1018
696
izip(reader.iter_records(), pack_readv_requests):
1019
697
raw_data = read_func(None)
1020
698
# check the header only
1021
if output_lines is not None:
1022
output_lines(knit._parse_record(key[-1], raw_data)[0])
1024
df, _ = knit._parse_record_header(key, raw_data)
699
df, _ = knit_data._parse_record_header(key[-1], raw_data)
1026
701
pos, size = writer.add_bytes_record(raw_data, names)
1027
702
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
1028
703
pb.update("Copied record", record_index)
1029
704
record_index += 1
1031
def _copy_nodes_graph(self, index_map, writer, write_index,
1032
readv_group_iter, total_items, output_lines=False):
706
def _copy_nodes_graph(self, nodes, index_map, writer, write_index,
1033
708
"""Copy knit nodes between packs.
1035
710
:param output_lines: Return lines present in the copied data as
1036
an iterator of line,version_id.
1038
713
pb = ui.ui_factory.nested_progress_bar()
1040
for result in self._do_copy_nodes_graph(index_map, writer,
1041
write_index, output_lines, pb, readv_group_iter, total_items):
1044
# Python 2.4 does not permit try:finally: in a generator.
715
return self._do_copy_nodes_graph(nodes, index_map, writer,
716
write_index, output_lines, pb)
1050
def _do_copy_nodes_graph(self, index_map, writer, write_index,
1051
output_lines, pb, readv_group_iter, total_items):
720
def _do_copy_nodes_graph(self, nodes, index_map, writer, write_index,
1052
722
# for record verification
1053
knit = KnitVersionedFiles(None, None)
723
knit_data = _KnitData(None)
1054
724
# for line extraction when requested (inventories only)
1055
725
if output_lines:
1056
factory = KnitPlainFactory()
726
factory = knit.KnitPlainFactory()
727
# plan a readv on each source pack:
729
nodes = sorted(nodes)
730
# how to map this into knit.py - or knit.py into this?
731
# we don't want the typical knit logic, we want grouping by pack
732
# at this point - perhaps a helper library for the following code
733
# duplication points?
1057
735
record_index = 0
1058
pb.update("Copied record", record_index, total_items)
1059
for index, readv_vector, node_vector in readv_group_iter:
736
pb.update("Copied record", record_index, len(nodes))
737
for index, key, value, references in nodes:
738
if index not in request_groups:
739
request_groups[index] = []
740
request_groups[index].append((key, value, references))
741
for index, items in request_groups.iteritems():
742
pack_readv_requests = []
743
for key, value, references in items:
744
# ---- KnitGraphIndex.get_position
745
bits = value[1:].split(' ')
746
offset, length = int(bits[0]), int(bits[1])
747
pack_readv_requests.append((offset, length, (key, value[0], references)))
748
# linear scan up the pack
749
pack_readv_requests.sort()
1061
pack_obj = index_map[index]
1062
transport, path = pack_obj.access_tuple()
1064
reader = pack.make_readv_reader(transport, path, readv_vector)
1065
except errors.NoSuchFile:
1066
if self._reload_func is not None:
1069
for (names, read_func), (key, eol_flag, references) in \
1070
izip(reader.iter_records(), node_vector):
751
transport, path = index_map[index]
752
reader = pack.make_readv_reader(transport, path,
753
[offset[0:2] for offset in pack_readv_requests])
754
for (names, read_func), (_1, _2, (key, eol_flag, references)) in \
755
izip(reader.iter_records(), pack_readv_requests):
1071
756
raw_data = read_func(None)
1072
757
if output_lines:
1073
758
# read the entire thing
1074
content, _ = knit._parse_record(key[-1], raw_data)
759
content, _ = knit_data._parse_record(key[-1], raw_data)
1075
760
if len(references[-1]) == 0:
1076
761
line_iterator = factory.get_fulltext_content(content)
1078
763
line_iterator = factory.get_linedelta_content(content)
1079
764
for line in line_iterator:
1082
767
# check the header only
1083
df, _ = knit._parse_record_header(key, raw_data)
768
df, _ = knit_data._parse_record_header(key[-1], raw_data)
1085
770
pos, size = writer.add_bytes_record(raw_data, names)
1086
771
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1087
772
pb.update("Copied record", record_index)
1088
773
record_index += 1
1090
def _get_text_nodes(self):
1091
text_index_map, text_indices = self._pack_map_and_index_list(
1093
return text_index_map, self._index_contents(text_indices,
1096
def _least_readv_node_readv(self, nodes):
1097
"""Generate request groups for nodes using the least readv's.
1099
:param nodes: An iterable of graph index nodes.
1100
:return: Total node count and an iterator of the data needed to perform
1101
readvs to obtain the data for nodes. Each item yielded by the
1102
iterator is a tuple with:
1103
index, readv_vector, node_vector. readv_vector is a list ready to
1104
hand to the transport readv method, and node_vector is a list of
1105
(key, eol_flag, references) for the node retrieved by the
1106
matching readv_vector.
1108
# group by pack so we do one readv per pack
1109
nodes = sorted(nodes)
1112
for index, key, value, references in nodes:
1113
if index not in request_groups:
1114
request_groups[index] = []
1115
request_groups[index].append((key, value, references))
1117
for index, items in request_groups.iteritems():
1118
pack_readv_requests = []
1119
for key, value, references in items:
1120
# ---- KnitGraphIndex.get_position
1121
bits = value[1:].split(' ')
1122
offset, length = int(bits[0]), int(bits[1])
1123
pack_readv_requests.append(
1124
((offset, length), (key, value[0], references)))
1125
# linear scan up the pack to maximum range combining.
1126
pack_readv_requests.sort()
1127
# split out the readv and the node data.
1128
pack_readv = [readv for readv, node in pack_readv_requests]
1129
node_vector = [node for readv, node in pack_readv_requests]
1130
result.append((index, pack_readv, node_vector))
1131
return total, result
1133
def _log_copied_texts(self):
1134
if 'pack' in debug.debug_flags:
1135
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
1136
time.ctime(), self._pack_collection._upload_transport.base,
1137
self.new_pack.random_name,
1138
self.new_pack.text_index.key_count(),
1139
time.time() - self.new_pack.start_time)
1141
def _process_inventory_lines(self, inv_lines):
1142
"""Use up the inv_lines generator and setup a text key filter."""
1143
repo = self._pack_collection.repo
1144
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
1145
inv_lines, self.revision_keys)
1147
for fileid, file_revids in fileid_revisions.iteritems():
1148
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
1149
self._text_filter = text_filter
1151
def _revision_node_readv(self, revision_nodes):
1152
"""Return the total revisions and the readv's to issue.
1154
:param revision_nodes: The revision index contents for the packs being
1155
incorporated into the new pack.
1156
:return: As per _least_readv_node_readv.
1158
return self._least_readv_node_readv(revision_nodes)
1160
def _use_pack(self, new_pack):
1161
"""Return True if new_pack should be used.
1163
:param new_pack: The pack that has just been created.
1164
:return: True if the pack should be used.
1166
return new_pack.data_inserted()
1169
class OptimisingPacker(Packer):
1170
"""A packer which spends more time to create better disk layouts."""
1172
def _revision_node_readv(self, revision_nodes):
1173
"""Return the total revisions and the readv's to issue.
1175
This sort places revisions in topological order with the ancestors
1178
:param revision_nodes: The revision index contents for the packs being
1179
incorporated into the new pack.
1180
:return: As per _least_readv_node_readv.
1182
# build an ancestors dict
1185
for index, key, value, references in revision_nodes:
1186
ancestors[key] = references[0]
1187
by_key[key] = (index, value, references)
1188
order = tsort.topo_sort(ancestors)
1190
# Single IO is pathological, but it will work as a starting point.
1192
for key in reversed(order):
1193
index, value, references = by_key[key]
1194
# ---- KnitGraphIndex.get_position
1195
bits = value[1:].split(' ')
1196
offset, length = int(bits[0]), int(bits[1])
1198
(index, [(offset, length)], [(key, value[0], references)]))
1199
# TODO: combine requests in the same index that are in ascending order.
1200
return total, requests
1202
def open_pack(self):
1203
"""Open a pack for the pack we are creating."""
1204
new_pack = super(OptimisingPacker, self).open_pack()
1205
# Turn on the optimization flags for all the index builders.
1206
new_pack.revision_index.set_optimize(for_size=True)
1207
new_pack.inventory_index.set_optimize(for_size=True)
1208
new_pack.text_index.set_optimize(for_size=True)
1209
new_pack.signature_index.set_optimize(for_size=True)
1213
776
class ReconcilePacker(Packer):
1214
777
"""A packer which regenerates indices etc as it copies.
1216
779
This is used by ``bzr reconcile`` to cause parent text pointers to be
1220
def _extra_init(self):
1221
self._data_changed = False
1223
def _process_inventory_lines(self, inv_lines):
1224
"""Generate a text key reference map rather for reconciling with."""
1225
repo = self._pack_collection.repo
1226
refs = repo._find_text_key_references_from_xml_inventory_lines(
1228
self._text_refs = refs
1229
# during reconcile we:
1230
# - convert unreferenced texts to full texts
1231
# - correct texts which reference a text not copied to be full texts
1232
# - copy all others as-is but with corrected parents.
1233
# - so at this point we don't know enough to decide what becomes a full
1235
self._text_filter = None
1237
def _copy_text_texts(self):
1238
"""generate what texts we should have and then copy."""
1239
self.pb.update("Copying content texts", 3)
1240
# we have three major tasks here:
1241
# 1) generate the ideal index
1242
repo = self._pack_collection.repo
1243
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1244
_1, key, _2, refs in
1245
self.new_pack.revision_index.iter_all_entries()])
1246
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1247
# 2) generate a text_nodes list that contains all the deltas that can
1248
# be used as-is, with corrected parents.
1251
discarded_nodes = []
1252
NULL_REVISION = _mod_revision.NULL_REVISION
1253
text_index_map, text_nodes = self._get_text_nodes()
1254
for node in text_nodes:
1260
ideal_parents = tuple(ideal_index[node[1]])
1262
discarded_nodes.append(node)
1263
self._data_changed = True
1265
if ideal_parents == (NULL_REVISION,):
1267
if ideal_parents == node[3][0]:
1269
ok_nodes.append(node)
1270
elif ideal_parents[0:1] == node[3][0][0:1]:
1271
# the left most parent is the same, or there are no parents
1272
# today. Either way, we can preserve the representation as
1273
# long as we change the refs to be inserted.
1274
self._data_changed = True
1275
ok_nodes.append((node[0], node[1], node[2],
1276
(ideal_parents, node[3][1])))
1277
self._data_changed = True
1279
# Reinsert this text completely
1280
bad_texts.append((node[1], ideal_parents))
1281
self._data_changed = True
1282
# we're finished with some data.
1285
# 3) bulk copy the ok data
1286
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1287
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1288
self.new_pack.text_index, readv_group_iter, total_items))
1289
# 4) adhoc copy all the other texts.
1290
# We have to topologically insert all texts otherwise we can fail to
1291
# reconcile when parts of a single delta chain are preserved intact,
1292
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1293
# reinserted, and if d3 has incorrect parents it will also be
1294
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1295
# copied), so we will try to delta, but d2 is not currently able to be
1296
# extracted because it's basis d1 is not present. Topologically sorting
1297
# addresses this. The following generates a sort for all the texts that
1298
# are being inserted without having to reference the entire text key
1299
# space (we only topo sort the revisions, which is smaller).
1300
topo_order = tsort.topo_sort(ancestors)
1301
rev_order = dict(zip(topo_order, range(len(topo_order))))
1302
bad_texts.sort(key=lambda key:rev_order.get(key[0][1], 0))
1303
transaction = repo.get_transaction()
1304
file_id_index = GraphIndexPrefixAdapter(
1305
self.new_pack.text_index,
1307
add_nodes_callback=self.new_pack.text_index.add_nodes)
1308
data_access = _DirectPackAccess(
1309
{self.new_pack.text_index:self.new_pack.access_tuple()})
1310
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1311
self.new_pack.access_tuple())
1312
output_texts = KnitVersionedFiles(
1313
_KnitGraphIndex(self.new_pack.text_index,
1314
add_callback=self.new_pack.text_index.add_nodes,
1315
deltas=True, parents=True, is_locked=repo.is_locked),
1316
data_access=data_access, max_delta_chain=200)
1317
for key, parent_keys in bad_texts:
1318
# We refer to the new pack to delta data being output.
1319
# A possible improvement would be to catch errors on short reads
1320
# and only flush then.
1321
self.new_pack.flush()
1323
for parent_key in parent_keys:
1324
if parent_key[0] != key[0]:
1325
# Graph parents must match the fileid
1326
raise errors.BzrError('Mismatched key parent %r:%r' %
1328
parents.append(parent_key[1])
1329
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1330
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1331
output_texts.add_lines(key, parent_keys, text_lines,
1332
random_id=True, check_content=False)
1333
# 5) check that nothing inserted has a reference outside the keyspace.
1334
missing_text_keys = self.new_pack.text_index._external_references()
1335
if missing_text_keys:
1336
raise errors.BzrCheckError('Reference to missing compression parents %r'
1337
% (missing_text_keys,))
1338
self._log_copied_texts()
1340
def _use_pack(self, new_pack):
1341
"""Override _use_pack to check for reconcile having changed content."""
1342
# XXX: we might be better checking this at the copy time.
1343
original_inventory_keys = set()
1344
inv_index = self._pack_collection.inventory_index.combined_index
1345
for entry in inv_index.iter_all_entries():
1346
original_inventory_keys.add(entry[1])
1347
new_inventory_keys = set()
1348
for entry in new_pack.inventory_index.iter_all_entries():
1349
new_inventory_keys.add(entry[1])
1350
if new_inventory_keys != original_inventory_keys:
1351
self._data_changed = True
1352
return new_pack.data_inserted() and self._data_changed
1355
784
class RepositoryPackCollection(object):
1356
"""Management of packs within a repository.
1358
:ivar _names: map of {pack_name: (index_size,)}
1361
pack_factory = NewPack
1362
resumed_pack_factory = ResumedPack
785
"""Management of packs within a repository."""
1364
787
def __init__(self, repo, transport, index_transport, upload_transport,
1365
pack_transport, index_builder_class, index_class,
1367
789
"""Create a new RepositoryPackCollection.
1369
:param transport: Addresses the repository base directory
791
:param transport: Addresses the repository base directory
1370
792
(typically .bzr/repository/).
1371
793
:param index_transport: Addresses the directory containing indices.
1372
794
:param upload_transport: Addresses the directory into which packs are written
1373
795
while they're being created.
1374
796
:param pack_transport: Addresses the directory of existing complete packs.
1375
:param index_builder_class: The index builder class to use.
1376
:param index_class: The index class to use.
1377
:param use_chk_index: Whether to setup and manage a CHK index.
1379
# XXX: This should call self.reset()
1380
798
self.repo = repo
1381
799
self.transport = transport
1382
800
self._index_transport = index_transport
1383
801
self._upload_transport = upload_transport
1384
802
self._pack_transport = pack_transport
1385
self._index_builder_class = index_builder_class
1386
self._index_class = index_class
1387
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3,
803
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
1390
805
# name:Pack mapping
1392
806
self._packs_by_name = {}
1393
807
# the previous pack-names content
1394
808
self._packs_at_load = None
1395
809
# when a pack is being created by this object, the state of that pack.
1396
810
self._new_pack = None
1397
811
# aggregated revision index data
1398
flush = self._flush_new_pack
1399
self.revision_index = AggregateIndex(self.reload_pack_names, flush)
1400
self.inventory_index = AggregateIndex(self.reload_pack_names, flush)
1401
self.text_index = AggregateIndex(self.reload_pack_names, flush)
1402
self.signature_index = AggregateIndex(self.reload_pack_names, flush)
1403
all_indices = [self.revision_index, self.inventory_index,
1404
self.text_index, self.signature_index]
1406
self.chk_index = AggregateIndex(self.reload_pack_names, flush)
1407
all_indices.append(self.chk_index)
1409
# used to determine if we're using a chk_index elsewhere.
1410
self.chk_index = None
1411
# Tell all the CombinedGraphIndex objects about each other, so they can
1412
# share hints about which pack names to search first.
1413
all_combined = [agg_idx.combined_index for agg_idx in all_indices]
1414
for combined_idx in all_combined:
1415
combined_idx.set_sibling_indices(
1416
set(all_combined).difference([combined_idx]))
1418
self._resumed_packs = []
1421
return '%s(%r)' % (self.__class__.__name__, self.repo)
812
self.revision_index = AggregateIndex()
813
self.inventory_index = AggregateIndex()
814
self.text_index = AggregateIndex()
815
self.signature_index = AggregateIndex()
1423
817
def add_pack_to_memory(self, pack):
1424
818
"""Make a Pack object available to the repository to satisfy queries.
1426
820
:param pack: A Pack object.
1428
if pack.name in self._packs_by_name:
1429
raise AssertionError(
1430
'pack %s already in _packs_by_name' % (pack.name,))
822
assert pack.name not in self._packs_by_name
1431
823
self.packs.append(pack)
1432
824
self._packs_by_name[pack.name] = pack
1433
825
self.revision_index.add_index(pack.revision_index, pack)
1434
826
self.inventory_index.add_index(pack.inventory_index, pack)
1435
827
self.text_index.add_index(pack.text_index, pack)
1436
828
self.signature_index.add_index(pack.signature_index, pack)
1437
if self.chk_index is not None:
1438
self.chk_index.add_index(pack.chk_index, pack)
830
def _add_text_to_weave(self, file_id, revision_id, new_lines, parents,
831
nostore_sha, random_revid):
832
file_id_index = GraphIndexPrefixAdapter(
833
self.text_index.combined_index,
835
add_nodes_callback=self.text_index.add_callback)
836
self.repo._text_knit._index._graph_index = file_id_index
837
self.repo._text_knit._index._add_callback = file_id_index.add_nodes
838
return self.repo._text_knit.add_lines_with_ghosts(
839
revision_id, parents, new_lines, nostore_sha=nostore_sha,
840
random_id=random_revid, check_content=False)[0:2]
1440
842
def all_packs(self):
1441
843
"""Return a list of all the Pack objects this repository has.
1492
890
# group their data with the relevant commit, and that may
1493
891
# involve rewriting ancient history - which autopack tries to
1494
892
# avoid. Alternatively we could not group the data but treat
1495
# each of these as having a single revision, and thus add
893
# each of these as having a single revision, and thus add
1496
894
# one revision for each to the total revision count, to get
1497
895
# a matching distribution.
1499
897
existing_packs.append((revision_count, pack))
1500
898
pack_operations = self.plan_autopack_combinations(
1501
899
existing_packs, pack_distribution)
1502
num_new_packs = len(pack_operations)
1503
num_old_packs = sum([len(po[1]) for po in pack_operations])
1504
num_revs_affected = sum([po[0] for po in pack_operations])
1505
mutter('Auto-packing repository %s, which has %d pack files, '
1506
'containing %d revisions. Packing %d files into %d affecting %d'
1507
' revisions', self, total_packs, total_revisions, num_old_packs,
1508
num_new_packs, num_revs_affected)
1509
result = self._execute_pack_operations(pack_operations,
1510
reload_func=self._restart_autopack)
1511
mutter('Auto-packing repository %s completed', self)
900
self._execute_pack_operations(pack_operations)
1514
def _execute_pack_operations(self, pack_operations, _packer_class=Packer,
903
def _execute_pack_operations(self, pack_operations):
1516
904
"""Execute a series of pack operations.
1518
906
:param pack_operations: A list of [revision_count, packs_to_combine].
1519
:param _packer_class: The class of packer to use (default: Packer).
1520
:return: The new pack names.
1522
909
for revision_count, packs in pack_operations:
1523
910
# we may have no-ops from the setup logic
1524
911
if len(packs) == 0:
1526
packer = _packer_class(self, packs, '.autopack',
1527
reload_func=reload_func)
1530
except errors.RetryWithNewPacks:
1531
# An exception is propagating out of this context, make sure
1532
# this packer has cleaned up. Packer() doesn't set its new_pack
1533
# state into the RepositoryPackCollection object, so we only
1534
# have access to it directly here.
1535
if packer.new_pack is not None:
1536
packer.new_pack.abort()
913
Packer(self, packs, '.autopack').pack()
1538
914
for pack in packs:
1539
915
self._remove_pack_from_memory(pack)
1540
916
# record the newly available packs and stop advertising the old
1542
to_be_obsoleted = []
1543
for _, packs in pack_operations:
1544
to_be_obsoleted.extend(packs)
1545
result = self._save_pack_names(clear_obsolete_packs=True,
1546
obsolete_packs=to_be_obsoleted)
1549
def _flush_new_pack(self):
1550
if self._new_pack is not None:
1551
self._new_pack.flush()
918
self._save_pack_names(clear_obsolete_packs=True)
919
# Move the old packs out of the way now they are no longer referenced.
920
for revision_count, packs in pack_operations:
921
self._obsolete_packs(packs)
1553
923
def lock_names(self):
1554
924
"""Acquire the mutex around the pack-names index.
1556
926
This cannot be used in the middle of a read-only transaction on the
1559
929
self.repo.control_files.lock_write()
1561
def _already_packed(self):
1562
"""Is the collection already packed?"""
1563
return not (self.repo._format.pack_compresses or (len(self._names) > 1))
1565
def pack(self, hint=None, clean_obsolete_packs=False):
1566
932
"""Pack the pack collection totally."""
1567
933
self.ensure_loaded()
1568
934
total_packs = len(self._names)
1569
if self._already_packed():
1571
937
total_revisions = self.revision_index.combined_index.key_count()
1572
938
# XXX: the following may want to be a class, to pack with a given
1574
940
mutter('Packing repository %s, which has %d pack files, '
1575
'containing %d revisions with hint %r.', self, total_packs,
1576
total_revisions, hint)
941
'containing %d revisions into 1 packs.', self, total_packs,
1577
943
# determine which packs need changing
944
pack_distribution = [1]
1578
945
pack_operations = [[0, []]]
1579
946
for pack in self.all_packs():
1580
if hint is None or pack.name in hint:
1581
# Either no hint was provided (so we are packing everything),
1582
# or this pack was included in the hint.
1583
pack_operations[-1][0] += pack.get_revision_count()
1584
pack_operations[-1][1].append(pack)
1585
self._execute_pack_operations(pack_operations, OptimisingPacker)
1587
if clean_obsolete_packs:
1588
self._clear_obsolete_packs()
947
revision_count = pack.get_revision_count()
948
pack_operations[-1][0] += revision_count
949
pack_operations[-1][1].append(pack)
950
self._execute_pack_operations(pack_operations)
1590
952
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1591
953
"""Plan a pack operation.
1875
1157
self._packs_by_name = {}
1876
1158
self._packs_at_load = None
1160
def _make_index_map(self, index_suffix):
1161
"""Return information on existing indices.
1163
:param suffix: Index suffix added to pack name.
1165
:returns: (pack_map, indices) where indices is a list of GraphIndex
1166
objects, and pack_map is a mapping from those objects to the
1167
pack tuple they describe.
1169
# TODO: stop using this; it creates new indices unnecessarily.
1170
self.ensure_loaded()
1171
suffix_map = {'.rix': 'revision_index',
1172
'.six': 'signature_index',
1173
'.iix': 'inventory_index',
1174
'.tix': 'text_index',
1176
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1177
suffix_map[index_suffix])
1179
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1180
"""Convert a list of packs to an index pack map and index list.
1182
:param packs: The packs list to process.
1183
:param index_attribute: The attribute that the desired index is found
1185
:return: A tuple (map, list) where map contains the dict from
1186
index:pack_tuple, and lsit contains the indices in the same order
1192
index = getattr(pack, index_attribute)
1193
indices.append(index)
1194
pack_map[index] = (pack.pack_transport, pack.file_name())
1195
return pack_map, indices
1197
def _index_contents(self, pack_map, key_filter=None):
1198
"""Get an iterable of the index contents from a pack_map.
1200
:param pack_map: A map from indices to pack details.
1201
:param key_filter: An optional filter to limit the
1204
indices = [index for index in pack_map.iterkeys()]
1205
all_index = CombinedGraphIndex(indices)
1206
if key_filter is None:
1207
return all_index.iter_all_entries()
1209
return all_index.iter_entries(key_filter)
1878
1211
def _unlock_names(self):
1879
1212
"""Release the mutex around the pack-names index."""
1880
1213
self.repo.control_files.unlock()
1882
def _diff_pack_names(self):
1883
"""Read the pack names from disk, and compare it to the one in memory.
1885
:return: (disk_nodes, deleted_nodes, new_nodes)
1886
disk_nodes The final set of nodes that should be referenced
1887
deleted_nodes Nodes which have been removed from when we started
1888
new_nodes Nodes that are newly introduced
1890
# load the disk nodes across
1892
for index, key, value in self._iter_disk_pack_index():
1893
disk_nodes.add((key, value))
1894
orig_disk_nodes = set(disk_nodes)
1896
# do a two-way diff against our original content
1897
current_nodes = set()
1898
for name, sizes in self._names.iteritems():
1900
((name, ), ' '.join(str(size) for size in sizes)))
1902
# Packs no longer present in the repository, which were present when we
1903
# locked the repository
1904
deleted_nodes = self._packs_at_load - current_nodes
1905
# Packs which this process is adding
1906
new_nodes = current_nodes - self._packs_at_load
1908
# Update the disk_nodes set to include the ones we are adding, and
1909
# remove the ones which were removed by someone else
1910
disk_nodes.difference_update(deleted_nodes)
1911
disk_nodes.update(new_nodes)
1913
return disk_nodes, deleted_nodes, new_nodes, orig_disk_nodes
1915
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1916
"""Given the correct set of pack files, update our saved info.
1918
:return: (removed, added, modified)
1919
removed pack names removed from self._names
1920
added pack names added to self._names
1921
modified pack names that had changed value
1926
## self._packs_at_load = disk_nodes
1215
def _save_pack_names(self, clear_obsolete_packs=False):
1216
"""Save the list of packs.
1218
This will take out the mutex around the pack names list for the
1219
duration of the method call. If concurrent updates have been made, a
1220
three-way merge between the current list and the current in memory list
1223
:param clear_obsolete_packs: If True, clear out the contents of the
1224
obsolete_packs directory.
1228
builder = GraphIndexBuilder()
1229
# load the disk nodes across
1231
for index, key, value in self._iter_disk_pack_index():
1232
disk_nodes.add((key, value))
1233
# do a two-way diff against our original content
1234
current_nodes = set()
1235
for name, sizes in self._names.iteritems():
1237
((name, ), ' '.join(str(size) for size in sizes)))
1238
deleted_nodes = self._packs_at_load - current_nodes
1239
new_nodes = current_nodes - self._packs_at_load
1240
disk_nodes.difference_update(deleted_nodes)
1241
disk_nodes.update(new_nodes)
1242
# TODO: handle same-name, index-size-changes here -
1243
# e.g. use the value from disk, not ours, *unless* we're the one
1245
for key, value in disk_nodes:
1246
builder.add_node(key, value)
1247
self.transport.put_file('pack-names', builder.finish())
1248
# move the baseline forward
1249
self._packs_at_load = disk_nodes
1250
# now clear out the obsolete packs directory
1251
if clear_obsolete_packs:
1252
self.transport.clone('obsolete_packs').delete_multi(
1253
self.transport.list_dir('obsolete_packs'))
1255
self._unlock_names()
1256
# synchronise the memory packs list with what we just wrote:
1927
1257
new_names = dict(disk_nodes)
1928
1258
# drop no longer present nodes
1929
1259
for pack in self.all_packs():
1930
1260
if (pack.name,) not in new_names:
1931
removed.append(pack.name)
1932
1261
self._remove_pack_from_memory(pack)
1933
1262
# add new nodes/refresh existing ones
1934
1263
for key, value in disk_nodes:
1948
1277
self._remove_pack_from_memory(self.get_pack_by_name(name))
1949
1278
self._names[name] = sizes
1950
1279
self.get_pack_by_name(name)
1951
modified.append(name)
1954
1282
self._names[name] = sizes
1955
1283
self.get_pack_by_name(name)
1957
return removed, added, modified
1959
def _save_pack_names(self, clear_obsolete_packs=False, obsolete_packs=None):
1960
"""Save the list of packs.
1962
This will take out the mutex around the pack names list for the
1963
duration of the method call. If concurrent updates have been made, a
1964
three-way merge between the current list and the current in memory list
1967
:param clear_obsolete_packs: If True, clear out the contents of the
1968
obsolete_packs directory.
1969
:param obsolete_packs: Packs that are obsolete once the new pack-names
1970
file has been written.
1971
:return: A list of the names saved that were not previously on disk.
1973
already_obsolete = []
1976
builder = self._index_builder_class()
1977
(disk_nodes, deleted_nodes, new_nodes,
1978
orig_disk_nodes) = self._diff_pack_names()
1979
# TODO: handle same-name, index-size-changes here -
1980
# e.g. use the value from disk, not ours, *unless* we're the one
1982
for key, value in disk_nodes:
1983
builder.add_node(key, value)
1984
self.transport.put_file('pack-names', builder.finish(),
1985
mode=self.repo.bzrdir._get_file_mode())
1986
self._packs_at_load = disk_nodes
1987
if clear_obsolete_packs:
1990
to_preserve = set([o.name for o in obsolete_packs])
1991
already_obsolete = self._clear_obsolete_packs(to_preserve)
1993
self._unlock_names()
1994
# synchronise the memory packs list with what we just wrote:
1995
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1997
# TODO: We could add one more condition here. "if o.name not in
1998
# orig_disk_nodes and o != the new_pack we haven't written to
1999
# disk yet. However, the new pack object is not easily
2000
# accessible here (it would have to be passed through the
2001
# autopacking code, etc.)
2002
obsolete_packs = [o for o in obsolete_packs
2003
if o.name not in already_obsolete]
2004
self._obsolete_packs(obsolete_packs)
2005
return [new_node[0][0] for new_node in new_nodes]
2007
def reload_pack_names(self):
2008
"""Sync our pack listing with what is present in the repository.
2010
This should be called when we find out that something we thought was
2011
present is now missing. This happens when another process re-packs the
2014
:return: True if the in-memory list of packs has been altered at all.
2016
# The ensure_loaded call is to handle the case where the first call
2017
# made involving the collection was to reload_pack_names, where we
2018
# don't have a view of disk contents. Its a bit of a bandaid, and
2019
# causes two reads of pack-names, but its a rare corner case not struck
2020
# with regular push/pull etc.
2021
first_read = self.ensure_loaded()
2024
# out the new value.
2025
(disk_nodes, deleted_nodes, new_nodes,
2026
orig_disk_nodes) = self._diff_pack_names()
2027
# _packs_at_load is meant to be the explicit list of names in
2028
# 'pack-names' at then start. As such, it should not contain any
2029
# pending names that haven't been written out yet.
2030
self._packs_at_load = orig_disk_nodes
2032
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
2033
if removed or added or modified:
2037
def _restart_autopack(self):
2038
"""Reload the pack names list, and restart the autopack code."""
2039
if not self.reload_pack_names():
2040
# Re-raise the original exception, because something went missing
2041
# and a restart didn't find it
2043
raise errors.RetryAutopack(self.repo, False, sys.exc_info())
2045
def _clear_obsolete_packs(self, preserve=None):
2046
"""Delete everything from the obsolete-packs directory.
2048
:return: A list of pack identifiers (the filename without '.pack') that
2049
were found in obsolete_packs.
2052
obsolete_pack_transport = self.transport.clone('obsolete_packs')
2053
if preserve is None:
2055
for filename in obsolete_pack_transport.list_dir('.'):
2056
name, ext = osutils.splitext(filename)
2059
if name in preserve:
2062
obsolete_pack_transport.delete(filename)
2063
except (errors.PathError, errors.TransportError), e:
2064
warning("couldn't delete obsolete pack, skipping it:\n%s"
2068
1285
def _start_write_group(self):
2069
1286
# Do not permit preparation for writing if we're not in a 'write lock'.
2070
1287
if not self.repo.is_write_locked():
2071
1288
raise errors.NotWriteLocked(self)
2072
self._new_pack = self.pack_factory(self, upload_suffix='.pack',
2073
file_mode=self.repo.bzrdir._get_file_mode())
1289
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1290
self._pack_transport, upload_suffix='.pack')
2074
1291
# allow writing: queue writes to a new index
2075
1292
self.revision_index.add_writable_index(self._new_pack.revision_index,
2076
1293
self._new_pack)
2078
1295
self._new_pack)
2079
1296
self.text_index.add_writable_index(self._new_pack.text_index,
2080
1297
self._new_pack)
2081
self._new_pack.text_index.set_optimize(combine_backing_indices=False)
2082
1298
self.signature_index.add_writable_index(self._new_pack.signature_index,
2083
1299
self._new_pack)
2084
if self.chk_index is not None:
2085
self.chk_index.add_writable_index(self._new_pack.chk_index,
2087
self.repo.chk_bytes._index._add_callback = self.chk_index.add_callback
2088
self._new_pack.chk_index.set_optimize(combine_backing_indices=False)
2090
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
2091
self.repo.revisions._index._add_callback = self.revision_index.add_callback
2092
self.repo.signatures._index._add_callback = self.signature_index.add_callback
2093
self.repo.texts._index._add_callback = self.text_index.add_callback
1301
# reused revision and signature knits may need updating
1303
# "Hysterical raisins. client code in bzrlib grabs those knits outside
1304
# of write groups and then mutates it inside the write group."
1305
if self.repo._revision_knit is not None:
1306
self.repo._revision_knit._index._add_callback = \
1307
self.revision_index.add_callback
1308
if self.repo._signature_knit is not None:
1309
self.repo._signature_knit._index._add_callback = \
1310
self.signature_index.add_callback
1311
# create a reused knit object for text addition in commit.
1312
self.repo._text_knit = self.repo.weave_store.get_weave_or_empty(
2095
1315
def _abort_write_group(self):
2096
1316
# FIXME: just drop the transient index.
2097
1317
# forget what names there are
2098
if self._new_pack is not None:
2099
operation = cleanup.OperationWithCleanups(self._new_pack.abort)
2100
operation.add_cleanup(setattr, self, '_new_pack', None)
2101
# If we aborted while in the middle of finishing the write
2102
# group, _remove_pack_indices could fail because the indexes are
2103
# already gone. But they're not there we shouldn't fail in this
2104
# case, so we pass ignore_missing=True.
2105
operation.add_cleanup(self._remove_pack_indices, self._new_pack,
2106
ignore_missing=True)
2107
operation.run_simple()
2108
for resumed_pack in self._resumed_packs:
2109
operation = cleanup.OperationWithCleanups(resumed_pack.abort)
2110
# See comment in previous finally block.
2111
operation.add_cleanup(self._remove_pack_indices, resumed_pack,
2112
ignore_missing=True)
2113
operation.run_simple()
2114
del self._resumed_packs[:]
2116
def _remove_resumed_pack_indices(self):
2117
for resumed_pack in self._resumed_packs:
2118
self._remove_pack_indices(resumed_pack)
2119
del self._resumed_packs[:]
2121
def _check_new_inventories(self):
2122
"""Detect missing inventories in this write group.
2124
:returns: list of strs, summarising any problems found. If the list is
2125
empty no problems were found.
2127
# The base implementation does no checks. GCRepositoryPackCollection
1318
self._new_pack.abort()
1319
self._remove_pack_indices(self._new_pack)
1320
self._new_pack = None
1321
self.repo._text_knit = None
2131
1323
def _commit_write_group(self):
2133
for prefix, versioned_file in (
2134
('revisions', self.repo.revisions),
2135
('inventories', self.repo.inventories),
2136
('texts', self.repo.texts),
2137
('signatures', self.repo.signatures),
2139
missing = versioned_file.get_missing_compression_parent_keys()
2140
all_missing.update([(prefix,) + key for key in missing])
2142
raise errors.BzrCheckError(
2143
"Repository %s has missing compression parent(s) %r "
2144
% (self.repo, sorted(all_missing)))
2145
problems = self._check_new_inventories()
2147
problems_summary = '\n'.join(problems)
2148
raise errors.BzrCheckError(
2149
"Cannot add revision(s) to repository: " + problems_summary)
2150
1324
self._remove_pack_indices(self._new_pack)
2151
any_new_content = False
2152
1325
if self._new_pack.data_inserted():
2153
1326
# get all the data to disk and read to use
2154
1327
self._new_pack.finish()
2155
1328
self.allocate(self._new_pack)
2156
1329
self._new_pack = None
2157
any_new_content = True
2159
self._new_pack.abort()
2160
self._new_pack = None
2161
for resumed_pack in self._resumed_packs:
2162
# XXX: this is a pretty ugly way to turn the resumed pack into a
2163
# properly committed pack.
2164
self._names[resumed_pack.name] = None
2165
self._remove_pack_from_memory(resumed_pack)
2166
resumed_pack.finish()
2167
self.allocate(resumed_pack)
2168
any_new_content = True
2169
del self._resumed_packs[:]
2171
result = self.autopack()
1330
if not self.autopack():
2173
1331
# when autopack takes no steps, the names list is still
2175
return self._save_pack_names()
2179
def _suspend_write_group(self):
2180
tokens = [pack.name for pack in self._resumed_packs]
2181
self._remove_pack_indices(self._new_pack)
2182
if self._new_pack.data_inserted():
2183
# get all the data to disk and read to use
2184
self._new_pack.finish(suspend=True)
2185
tokens.append(self._new_pack.name)
2186
self._new_pack = None
1333
self._save_pack_names()
2188
1335
self._new_pack.abort()
2189
1336
self._new_pack = None
2190
self._remove_resumed_pack_indices()
2193
def _resume_write_group(self, tokens):
2194
for token in tokens:
2195
self._resume_pack(token)
1337
self.repo._text_knit = None
1340
class KnitPackRevisionStore(KnitRevisionStore):
1341
"""An object to adapt access from RevisionStore's to use KnitPacks.
1343
This class works by replacing the original RevisionStore.
1344
We need to do this because the KnitPackRevisionStore is less
1345
isolated in its layering - it uses services from the repo.
1348
def __init__(self, repo, transport, revisionstore):
1349
"""Create a KnitPackRevisionStore on repo with revisionstore.
1351
This will store its state in the Repository, use the
1352
indices to provide a KnitGraphIndex,
1353
and at the end of transactions write new indices.
1355
KnitRevisionStore.__init__(self, revisionstore.versioned_file_store)
1357
self._serializer = revisionstore._serializer
1358
self.transport = transport
1360
def get_revision_file(self, transaction):
1361
"""Get the revision versioned file object."""
1362
if getattr(self.repo, '_revision_knit', None) is not None:
1363
return self.repo._revision_knit
1364
self.repo._pack_collection.ensure_loaded()
1365
add_callback = self.repo._pack_collection.revision_index.add_callback
1366
# setup knit specific objects
1367
knit_index = KnitGraphIndex(
1368
self.repo._pack_collection.revision_index.combined_index,
1369
add_callback=add_callback)
1370
self.repo._revision_knit = knit.KnitVersionedFile(
1371
'revisions', self.transport.clone('..'),
1372
self.repo.control_files._file_mode,
1373
create=False, access_mode=self.repo._access_mode(),
1374
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1375
access_method=self.repo._pack_collection.revision_index.knit_access)
1376
return self.repo._revision_knit
1378
def get_signature_file(self, transaction):
1379
"""Get the signature versioned file object."""
1380
if getattr(self.repo, '_signature_knit', None) is not None:
1381
return self.repo._signature_knit
1382
self.repo._pack_collection.ensure_loaded()
1383
add_callback = self.repo._pack_collection.signature_index.add_callback
1384
# setup knit specific objects
1385
knit_index = KnitGraphIndex(
1386
self.repo._pack_collection.signature_index.combined_index,
1387
add_callback=add_callback, parents=False)
1388
self.repo._signature_knit = knit.KnitVersionedFile(
1389
'signatures', self.transport.clone('..'),
1390
self.repo.control_files._file_mode,
1391
create=False, access_mode=self.repo._access_mode(),
1392
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1393
access_method=self.repo._pack_collection.signature_index.knit_access)
1394
return self.repo._signature_knit
1397
class KnitPackTextStore(VersionedFileStore):
1398
"""Presents a TextStore abstraction on top of packs.
1400
This class works by replacing the original VersionedFileStore.
1401
We need to do this because the KnitPackRevisionStore is less
1402
isolated in its layering - it uses services from the repo and shares them
1403
with all the data written in a single write group.
1406
def __init__(self, repo, transport, weavestore):
1407
"""Create a KnitPackTextStore on repo with weavestore.
1409
This will store its state in the Repository, use the
1410
indices FileNames to provide a KnitGraphIndex,
1411
and at the end of transactions write new indices.
1413
# don't call base class constructor - it's not suitable.
1414
# no transient data stored in the transaction
1416
self._precious = False
1418
self.transport = transport
1419
self.weavestore = weavestore
1420
# XXX for check() which isn't updated yet
1421
self._transport = weavestore._transport
1423
def get_weave_or_empty(self, file_id, transaction):
1424
"""Get a 'Knit' backed by the .tix indices.
1426
The transaction parameter is ignored.
1428
self.repo._pack_collection.ensure_loaded()
1429
add_callback = self.repo._pack_collection.text_index.add_callback
1430
# setup knit specific objects
1431
file_id_index = GraphIndexPrefixAdapter(
1432
self.repo._pack_collection.text_index.combined_index,
1433
(file_id, ), 1, add_nodes_callback=add_callback)
1434
knit_index = KnitGraphIndex(file_id_index,
1435
add_callback=file_id_index.add_nodes,
1436
deltas=True, parents=True)
1437
return knit.KnitVersionedFile('text:' + file_id,
1438
self.transport.clone('..'),
1441
access_method=self.repo._pack_collection.text_index.knit_access,
1442
factory=knit.KnitPlainFactory())
1444
get_weave = get_weave_or_empty
1447
"""Generate a list of the fileids inserted, for use by check."""
1448
self.repo._pack_collection.ensure_loaded()
1450
for index, key, value, refs in \
1451
self.repo._pack_collection.text_index.combined_index.iter_all_entries():
1456
class InventoryKnitThunk(object):
1457
"""An object to manage thunking get_inventory_weave to pack based knits."""
1459
def __init__(self, repo, transport):
1460
"""Create an InventoryKnitThunk for repo at transport.
1462
This will store its state in the Repository, use the
1463
indices FileNames to provide a KnitGraphIndex,
1464
and at the end of transactions write a new index..
1467
self.transport = transport
1469
def get_weave(self):
1470
"""Get a 'Knit' that contains inventory data."""
1471
self.repo._pack_collection.ensure_loaded()
1472
add_callback = self.repo._pack_collection.inventory_index.add_callback
1473
# setup knit specific objects
1474
knit_index = KnitGraphIndex(
1475
self.repo._pack_collection.inventory_index.combined_index,
1476
add_callback=add_callback, deltas=True, parents=True)
1477
return knit.KnitVersionedFile(
1478
'inventory', self.transport.clone('..'),
1479
self.repo.control_files._file_mode,
1480
create=False, access_mode=self.repo._access_mode(),
1481
index=knit_index, delta=True, factory=knit.KnitPlainFactory(),
1482
access_method=self.repo._pack_collection.inventory_index.knit_access)
2198
1485
class KnitPackRepository(KnitRepository):
2199
"""Repository with knit objects stored inside pack containers.
2201
The layering for a KnitPackRepository is:
2203
Graph | HPSS | Repository public layer |
2204
===================================================
2205
Tuple based apis below, string based, and key based apis above
2206
---------------------------------------------------
2208
Provides .texts, .revisions etc
2209
This adapts the N-tuple keys to physical knit records which only have a
2210
single string identifier (for historical reasons), which in older formats
2211
was always the revision_id, and in the mapped code for packs is always
2212
the last element of key tuples.
2213
---------------------------------------------------
2215
A separate GraphIndex is used for each of the
2216
texts/inventories/revisions/signatures contained within each individual
2217
pack file. The GraphIndex layer works in N-tuples and is unaware of any
2219
===================================================
2223
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
1486
"""Experimental graph-knit using repository."""
1488
def __init__(self, _format, a_bzrdir, control_files, _revision_store,
1489
control_store, text_store, _commit_builder_class, _serializer):
2225
1490
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2226
_commit_builder_class, _serializer)
2227
index_transport = self._transport.clone('indices')
2228
self._pack_collection = RepositoryPackCollection(self, self._transport,
1491
_revision_store, control_store, text_store, _commit_builder_class,
1493
index_transport = control_files._transport.clone('indices')
1494
self._pack_collection = RepositoryPackCollection(self, control_files._transport,
2229
1495
index_transport,
2230
self._transport.clone('upload'),
2231
self._transport.clone('packs'),
2232
_format.index_builder_class,
2233
_format.index_class,
2234
use_chk_index=self._format.supports_chks,
2236
self.inventories = KnitVersionedFiles(
2237
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2238
add_callback=self._pack_collection.inventory_index.add_callback,
2239
deltas=True, parents=True, is_locked=self.is_locked),
2240
data_access=self._pack_collection.inventory_index.data_access,
2241
max_delta_chain=200)
2242
self.revisions = KnitVersionedFiles(
2243
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2244
add_callback=self._pack_collection.revision_index.add_callback,
2245
deltas=False, parents=True, is_locked=self.is_locked,
2246
track_external_parent_refs=True),
2247
data_access=self._pack_collection.revision_index.data_access,
2249
self.signatures = KnitVersionedFiles(
2250
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2251
add_callback=self._pack_collection.signature_index.add_callback,
2252
deltas=False, parents=False, is_locked=self.is_locked),
2253
data_access=self._pack_collection.signature_index.data_access,
2255
self.texts = KnitVersionedFiles(
2256
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2257
add_callback=self._pack_collection.text_index.add_callback,
2258
deltas=True, parents=True, is_locked=self.is_locked),
2259
data_access=self._pack_collection.text_index.data_access,
2260
max_delta_chain=200)
2261
if _format.supports_chks:
2262
# No graph, no compression:- references from chks are between
2263
# different objects not temporal versions of the same; and without
2264
# some sort of temporal structure knit compression will just fail.
2265
self.chk_bytes = KnitVersionedFiles(
2266
_KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2267
add_callback=self._pack_collection.chk_index.add_callback,
2268
deltas=False, parents=False, is_locked=self.is_locked),
2269
data_access=self._pack_collection.chk_index.data_access,
2272
self.chk_bytes = None
1496
control_files._transport.clone('upload'),
1497
control_files._transport.clone('packs'))
1498
self._revision_store = KnitPackRevisionStore(self, index_transport, self._revision_store)
1499
self.weave_store = KnitPackTextStore(self, index_transport, self.weave_store)
1500
self._inv_thunk = InventoryKnitThunk(self, index_transport)
2273
1501
# True when the repository object is 'write locked' (as opposed to the
2274
# physical lock only taken out around changes to the pack-names list.)
1502
# physical lock only taken out around changes to the pack-names list.)
2275
1503
# Another way to represent this would be a decorator around the control
2276
1504
# files object that presents logical locks as physical ones - if this
2277
1505
# gets ugly consider that alternative design. RBC 20071011
2279
1507
self._transaction = None
2281
1509
self._reconcile_does_inventory_gc = True
2282
self._reconcile_fixes_text_parents = True
1510
self._reconcile_fixes_text_parents = False
2283
1511
self._reconcile_backsup_inventory = False
2285
def _warn_if_deprecated(self, branch=None):
2286
# This class isn't deprecated, but one sub-format is
2287
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2288
super(KnitPackRepository, self)._warn_if_deprecated(branch)
2290
1513
def _abort_write_group(self):
2291
self.revisions._index._key_dependencies.clear()
2292
1514
self._pack_collection._abort_write_group()
2294
def _get_source(self, to_format):
2295
if to_format.network_name() == self._format.network_name():
2296
return KnitPackStreamSource(self, to_format)
2297
return super(KnitPackRepository, self)._get_source(to_format)
1516
def _access_mode(self):
1517
"""Return 'w' or 'r' for depending on whether a write lock is active.
1519
This method is a helper for the Knit-thunking support objects.
1521
if self.is_write_locked():
1525
def _find_inconsistent_revision_parents(self):
1526
"""Find revisions with incorrectly cached parents.
1528
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1529
parents-in-revision).
1531
assert self.is_locked()
1532
pb = ui.ui_factory.nested_progress_bar()
1535
revision_nodes = self._pack_collection.revision_index \
1536
.combined_index.iter_all_entries()
1537
index_positions = []
1538
# Get the cached index values for all revisions, and also the location
1539
# in each index of the revision text so we can perform linear IO.
1540
for index, key, value, refs in revision_nodes:
1541
pos, length = value[1:].split(' ')
1542
index_positions.append((index, int(pos), key[0],
1543
tuple(parent[0] for parent in refs[0])))
1544
pb.update("Reading revision index.", 0, 0)
1545
index_positions.sort()
1546
batch_count = len(index_positions) / 1000 + 1
1547
pb.update("Checking cached revision graph.", 0, batch_count)
1548
for offset in xrange(batch_count):
1549
pb.update("Checking cached revision graph.", offset)
1550
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1553
rev_ids = [item[2] for item in to_query]
1554
revs = self.get_revisions(rev_ids)
1555
for revision, item in zip(revs, to_query):
1556
index_parents = item[3]
1557
rev_parents = tuple(revision.parent_ids)
1558
if index_parents != rev_parents:
1559
result.append((revision.revision_id, index_parents, rev_parents))
1564
def get_parents(self, revision_ids):
1565
"""See StackedParentsProvider.get_parents.
1567
This implementation accesses the combined revision index to provide
1570
self._pack_collection.ensure_loaded()
1571
index = self._pack_collection.revision_index.combined_index
1573
for revision_id in revision_ids:
1574
if revision_id != _mod_revision.NULL_REVISION:
1575
search_keys.add((revision_id,))
1576
found_parents = {_mod_revision.NULL_REVISION:[]}
1577
for index, key, value, refs in index.iter_entries(search_keys):
1580
parents = (_mod_revision.NULL_REVISION,)
1582
parents = tuple(parent[0] for parent in parents)
1583
found_parents[key[0]] = parents
1585
for revision_id in revision_ids:
1587
result.append(found_parents[revision_id])
2299
1592
def _make_parents_provider(self):
2300
return graph.CachingParentsProvider(self)
2302
1595
def _refresh_data(self):
2303
if not self.is_locked():
2305
self._pack_collection.reload_pack_names()
1596
if self._write_lock_count == 1 or (
1597
self.control_files._lock_count == 1 and
1598
self.control_files._lock_mode == 'r'):
1599
# forget what names there are
1600
self._pack_collection.reset()
1601
# XXX: Better to do an in-memory merge when acquiring a new lock -
1602
# factor out code from _save_pack_names.
1603
self._pack_collection.ensure_loaded()
2307
1605
def _start_write_group(self):
2308
1606
self._pack_collection._start_write_group()
2310
1608
def _commit_write_group(self):
2311
hint = self._pack_collection._commit_write_group()
2312
self.revisions._index._key_dependencies.clear()
2315
def suspend_write_group(self):
2316
# XXX check self._write_group is self.get_transaction()?
2317
tokens = self._pack_collection._suspend_write_group()
2318
self.revisions._index._key_dependencies.clear()
2319
self._write_group = None
2322
def _resume_write_group(self, tokens):
2323
self._start_write_group()
2325
self._pack_collection._resume_write_group(tokens)
2326
except errors.UnresumableWriteGroup:
2327
self._abort_write_group()
2329
for pack in self._pack_collection._resumed_packs:
2330
self.revisions._index.scan_unvalidated_index(pack.revision_index)
1609
return self._pack_collection._commit_write_group()
1611
def get_inventory_weave(self):
1612
return self._inv_thunk.get_weave()
2332
1614
def get_transaction(self):
2333
1615
if self._write_lock_count:
2654
1862
def get_format_description(self):
2655
1863
"""See RepositoryFormat.get_format_description()."""
2656
1864
return "Packs containing knits with subtree support\n"
2659
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2660
"""A rich-root, no subtrees parameterized Pack repository.
2662
This repository format uses the xml6 serializer to get:
2663
- support for recording full info about the tree root
2665
This format was introduced in 1.0.
2668
repository_class = KnitPackRepository
2669
_commit_builder_class = PackRootCommitBuilder
2670
rich_root_data = True
2671
supports_tree_reference = False
2673
def _serializer(self):
2674
return xml6.serializer_v6
2675
# What index classes to use
2676
index_builder_class = InMemoryGraphIndex
2677
index_class = GraphIndex
2679
def _get_matching_bzrdir(self):
2680
return bzrdir.format_registry.make_bzrdir(
2683
def _ignore_setting_bzrdir(self, format):
2686
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2688
def get_format_string(self):
2689
"""See RepositoryFormat.get_format_string()."""
2690
return ("Bazaar pack repository format 1 with rich root"
2691
" (needs bzr 1.0)\n")
2693
def get_format_description(self):
2694
"""See RepositoryFormat.get_format_description()."""
2695
return "Packs containing knits with rich root support\n"
2698
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2699
"""Repository that supports external references to allow stacking.
2703
Supports external lookups, which results in non-truncated ghosts after
2704
reconcile compared to pack-0.92 formats.
2707
repository_class = KnitPackRepository
2708
_commit_builder_class = PackCommitBuilder
2709
supports_external_lookups = True
2710
# What index classes to use
2711
index_builder_class = InMemoryGraphIndex
2712
index_class = GraphIndex
2715
def _serializer(self):
2716
return xml5.serializer_v5
2718
def _get_matching_bzrdir(self):
2719
return bzrdir.format_registry.make_bzrdir('1.6')
2721
def _ignore_setting_bzrdir(self, format):
2724
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2726
def get_format_string(self):
2727
"""See RepositoryFormat.get_format_string()."""
2728
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2730
def get_format_description(self):
2731
"""See RepositoryFormat.get_format_description()."""
2732
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2735
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2736
"""A repository with rich roots and stacking.
2738
New in release 1.6.1.
2740
Supports stacking on other repositories, allowing data to be accessed
2741
without being stored locally.
2744
repository_class = KnitPackRepository
2745
_commit_builder_class = PackRootCommitBuilder
2746
rich_root_data = True
2747
supports_tree_reference = False # no subtrees
2748
supports_external_lookups = True
2749
# What index classes to use
2750
index_builder_class = InMemoryGraphIndex
2751
index_class = GraphIndex
2754
def _serializer(self):
2755
return xml6.serializer_v6
2757
def _get_matching_bzrdir(self):
2758
return bzrdir.format_registry.make_bzrdir(
2761
def _ignore_setting_bzrdir(self, format):
2764
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2766
def get_format_string(self):
2767
"""See RepositoryFormat.get_format_string()."""
2768
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2770
def get_format_description(self):
2771
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2774
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2775
"""A repository with rich roots and external references.
2779
Supports external lookups, which results in non-truncated ghosts after
2780
reconcile compared to pack-0.92 formats.
2782
This format was deprecated because the serializer it uses accidentally
2783
supported subtrees, when the format was not intended to. This meant that
2784
someone could accidentally fetch from an incorrect repository.
2787
repository_class = KnitPackRepository
2788
_commit_builder_class = PackRootCommitBuilder
2789
rich_root_data = True
2790
supports_tree_reference = False # no subtrees
2792
supports_external_lookups = True
2793
# What index classes to use
2794
index_builder_class = InMemoryGraphIndex
2795
index_class = GraphIndex
2798
def _serializer(self):
2799
return xml7.serializer_v7
2801
def _get_matching_bzrdir(self):
2802
matching = bzrdir.format_registry.make_bzrdir(
2804
matching.repository_format = self
2807
def _ignore_setting_bzrdir(self, format):
2810
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2812
def get_format_string(self):
2813
"""See RepositoryFormat.get_format_string()."""
2814
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2816
def get_format_description(self):
2817
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2821
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2822
"""A repository with stacking and btree indexes,
2823
without rich roots or subtrees.
2825
This is equivalent to pack-1.6 with B+Tree indices.
2828
repository_class = KnitPackRepository
2829
_commit_builder_class = PackCommitBuilder
2830
supports_external_lookups = True
2831
# What index classes to use
2832
index_builder_class = BTreeBuilder
2833
index_class = BTreeGraphIndex
2836
def _serializer(self):
2837
return xml5.serializer_v5
2839
def _get_matching_bzrdir(self):
2840
return bzrdir.format_registry.make_bzrdir('1.9')
2842
def _ignore_setting_bzrdir(self, format):
2845
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2847
def get_format_string(self):
2848
"""See RepositoryFormat.get_format_string()."""
2849
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2851
def get_format_description(self):
2852
"""See RepositoryFormat.get_format_description()."""
2853
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2856
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2857
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2859
1.6-rich-root with B+Tree indices.
2862
repository_class = KnitPackRepository
2863
_commit_builder_class = PackRootCommitBuilder
2864
rich_root_data = True
2865
supports_tree_reference = False # no subtrees
2866
supports_external_lookups = True
2867
# What index classes to use
2868
index_builder_class = BTreeBuilder
2869
index_class = BTreeGraphIndex
2872
def _serializer(self):
2873
return xml6.serializer_v6
2875
def _get_matching_bzrdir(self):
2876
return bzrdir.format_registry.make_bzrdir(
2879
def _ignore_setting_bzrdir(self, format):
2882
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2884
def get_format_string(self):
2885
"""See RepositoryFormat.get_format_string()."""
2886
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2888
def get_format_description(self):
2889
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2892
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2893
"""A subtrees development repository.
2895
This format should be retained until the second release after bzr 1.7.
2897
1.6.1-subtree[as it might have been] with B+Tree indices.
2899
This is [now] retained until we have a CHK based subtree format in
2903
repository_class = KnitPackRepository
2904
_commit_builder_class = PackRootCommitBuilder
2905
rich_root_data = True
2907
supports_tree_reference = True
2908
supports_external_lookups = True
2909
# What index classes to use
2910
index_builder_class = BTreeBuilder
2911
index_class = BTreeGraphIndex
2914
def _serializer(self):
2915
return xml7.serializer_v7
2917
def _get_matching_bzrdir(self):
2918
return bzrdir.format_registry.make_bzrdir(
2919
'development-subtree')
2921
def _ignore_setting_bzrdir(self, format):
2924
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2926
def get_format_string(self):
2927
"""See RepositoryFormat.get_format_string()."""
2928
return ("Bazaar development format 2 with subtree support "
2929
"(needs bzr.dev from before 1.8)\n")
2931
def get_format_description(self):
2932
"""See RepositoryFormat.get_format_description()."""
2933
return ("Development repository format, currently the same as "
2934
"1.6.1-subtree with B+Tree indices.\n")
added
removed
bzrlib/repofmt/pack_repo.py
