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 (
54
from bzrlib.decorators import needs_read_lock, needs_write_lock
67
55
from bzrlib.repofmt.knitrepo import KnitRepository
68
56
from bzrlib.repository import (
70
59
MetaDirRepositoryFormat,
72
RepositoryWriteLockResult,
76
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
83
68
class PackCommitBuilder(CommitBuilder):
84
69
"""A subclass of CommitBuilder to add texts with pack semantics.
86
71
Specifically this uses one knit object rather than one knit object per
87
72
added text, reducing memory and object pressure.
90
def __init__(self, repository, parents, config, timestamp=None,
91
timezone=None, committer=None, revprops=None,
93
CommitBuilder.__init__(self, repository, parents, config,
94
timestamp=timestamp, timezone=timezone, committer=committer,
95
revprops=revprops, revision_id=revision_id)
96
self._file_graph = graph.Graph(
97
repository._pack_collection.text_index.combined_index)
99
def _heads(self, file_id, revision_ids):
100
keys = [(file_id, revision_id) for revision_id in revision_ids]
101
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,
104
81
class PackRootCommitBuilder(RootCommitBuilder):
105
82
"""A subclass of RootCommitBuilder to add texts with pack semantics.
107
84
Specifically this uses one knit object rather than one knit object per
108
85
added text, reducing memory and object pressure.
111
def __init__(self, repository, parents, config, timestamp=None,
112
timezone=None, committer=None, revprops=None,
114
CommitBuilder.__init__(self, repository, parents, config,
115
timestamp=timestamp, timezone=timezone, committer=committer,
116
revprops=revprops, revision_id=revision_id)
117
self._file_graph = graph.Graph(
118
repository._pack_collection.text_index.combined_index)
120
def _heads(self, file_id, revision_ids):
121
keys = [(file_id, revision_id) for revision_id in revision_ids]
122
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,
125
94
class Pack(object):
262
170
return not self.__eq__(other)
264
172
def __repr__(self):
265
return "<%s.%s object at 0x%x, %s, %s" % (
266
self.__class__.__module__, self.__class__.__name__, id(self),
267
self.pack_transport, self.name)
270
class ResumedPack(ExistingPack):
272
def __init__(self, name, revision_index, inventory_index, text_index,
273
signature_index, upload_transport, pack_transport, index_transport,
274
pack_collection, chk_index=None):
275
"""Create a ResumedPack object."""
276
ExistingPack.__init__(self, pack_transport, name, revision_index,
277
inventory_index, text_index, signature_index,
279
self.upload_transport = upload_transport
280
self.index_transport = index_transport
281
self.index_sizes = [None, None, None, None]
283
('revision', revision_index),
284
('inventory', inventory_index),
285
('text', text_index),
286
('signature', signature_index),
288
if chk_index is not None:
289
indices.append(('chk', chk_index))
290
self.index_sizes.append(None)
291
for index_type, index in indices:
292
offset = self.index_offset(index_type)
293
self.index_sizes[offset] = index._size
294
self.index_class = pack_collection._index_class
295
self._pack_collection = pack_collection
296
self._state = 'resumed'
297
# XXX: perhaps check that the .pack file exists?
299
def access_tuple(self):
300
if self._state == 'finished':
301
return Pack.access_tuple(self)
302
elif self._state == 'resumed':
303
return self.upload_transport, self.file_name()
305
raise AssertionError(self._state)
308
self.upload_transport.delete(self.file_name())
309
indices = [self.revision_index, self.inventory_index, self.text_index,
310
self.signature_index]
311
if self.chk_index is not None:
312
indices.append(self.chk_index)
313
for index in indices:
314
index._transport.delete(index._name)
317
self._check_references()
318
index_types = ['revision', 'inventory', 'text', 'signature']
319
if self.chk_index is not None:
320
index_types.append('chk')
321
for index_type in index_types:
322
old_name = self.index_name(index_type, self.name)
323
new_name = '../indices/' + old_name
324
self.upload_transport.rename(old_name, new_name)
325
self._replace_index_with_readonly(index_type)
326
new_name = '../packs/' + self.file_name()
327
self.upload_transport.rename(self.file_name(), new_name)
328
self._state = 'finished'
330
def _get_external_refs(self, index):
331
"""Return compression parents for this index that are not present.
333
This returns any compression parents that are referenced by this index,
334
which are not contained *in* this index. They may be present elsewhere.
336
return index.external_references(1)
173
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
174
id(self), self.transport, self.name)
339
177
class NewPack(Pack):
340
178
"""An in memory proxy for a pack which is being created."""
342
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,
343
191
"""Create a NewPack instance.
345
: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').
346
200
:param upload_suffix: An optional suffix to be given to any temporary
347
201
files created during the pack creation. e.g '.autopack'
348
:param file_mode: Unix permissions for newly created file.
350
203
# The relative locations of the packs are constrained, but all are
351
204
# passed in because the caller has them, so as to avoid object churn.
352
index_builder_class = pack_collection._index_builder_class
353
if pack_collection.chk_index is not None:
354
chk_index = index_builder_class(reference_lists=0)
357
205
Pack.__init__(self,
358
206
# Revisions: parents list, no text compression.
359
index_builder_class(reference_lists=1),
207
InMemoryGraphIndex(reference_lists=1),
360
208
# Inventory: We want to map compression only, but currently the
361
209
# knit code hasn't been updated enough to understand that, so we
362
210
# have a regular 2-list index giving parents and compression
364
index_builder_class(reference_lists=2),
212
InMemoryGraphIndex(reference_lists=2),
365
213
# Texts: compression and per file graph, for all fileids - so two
366
214
# reference lists and two elements in the key tuple.
367
index_builder_class(reference_lists=2, key_elements=2),
215
InMemoryGraphIndex(reference_lists=2, key_elements=2),
368
216
# Signatures: Just blobs to store, no compression, no parents
370
index_builder_class(reference_lists=0),
371
# CHK based storage - just blobs, no compression or parents.
218
InMemoryGraphIndex(reference_lists=0),
374
self._pack_collection = pack_collection
375
# When we make readonly indices, we need this.
376
self.index_class = pack_collection._index_class
377
220
# where should the new pack be opened
378
self.upload_transport = pack_collection._upload_transport
221
self.upload_transport = upload_transport
379
222
# where are indices written out to
380
self.index_transport = pack_collection._index_transport
223
self.index_transport = index_transport
381
224
# where is the pack renamed to when it is finished?
382
self.pack_transport = pack_collection._pack_transport
383
# What file mode to upload the pack and indices with.
384
self._file_mode = file_mode
225
self.pack_transport = pack_transport
385
226
# tracks the content written to the .pack file.
386
self._hash = osutils.md5()
387
# a tuple with the length in bytes of the indices, once the pack
388
# is finalised. (rev, inv, text, sigs, chk_if_in_use)
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)
389
230
self.index_sizes = None
390
231
# How much data to cache when writing packs. Note that this is not
391
232
# synchronised with reads, because it's not in the transport layer, so
750
532
def open_pack(self):
751
533
"""Open a pack for the pack we are creating."""
752
new_pack = self._pack_collection.pack_factory(self._pack_collection,
753
upload_suffix=self.suffix,
754
file_mode=self._pack_collection.repo.bzrdir._get_file_mode())
755
# We know that we will process all nodes in order, and don't need to
756
# query, so don't combine any indices spilled to disk until we are done
757
new_pack.revision_index.set_optimize(combine_backing_indices=False)
758
new_pack.inventory_index.set_optimize(combine_backing_indices=False)
759
new_pack.text_index.set_optimize(combine_backing_indices=False)
760
new_pack.signature_index.set_optimize(combine_backing_indices=False)
763
def _update_pack_order(self, entries, index_to_pack_map):
764
"""Determine how we want our packs to be ordered.
766
This changes the sort order of the self.packs list so that packs unused
767
by 'entries' will be at the end of the list, so that future requests
768
can avoid probing them. Used packs will be at the front of the
769
self.packs list, in the order of their first use in 'entries'.
771
:param entries: A list of (index, ...) tuples
772
:param index_to_pack_map: A mapping from index objects to pack objects.
776
for entry in entries:
778
if index not in seen_indexes:
779
packs.append(index_to_pack_map[index])
780
seen_indexes.add(index)
781
if len(packs) == len(self.packs):
782
if 'pack' in debug.debug_flags:
783
mutter('Not changing pack list, all packs used.')
785
seen_packs = set(packs)
786
for pack in self.packs:
787
if pack not in seen_packs:
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)
790
544
if 'pack' in debug.debug_flags:
791
old_names = [p.access_tuple()[1] for p in self.packs]
792
new_names = [p.access_tuple()[1] for p in packs]
793
mutter('Reordering packs\nfrom: %s\n to: %s',
794
old_names, new_names)
797
def _copy_revision_texts(self):
798
"""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)
799
555
# select revisions
800
556
if self.revision_ids:
801
557
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
803
559
revision_keys = None
804
561
# select revision keys
805
revision_index_map, revision_indices = self._pack_map_and_index_list(
807
revision_nodes = self._index_contents(revision_indices, revision_keys)
808
revision_nodes = list(revision_nodes)
809
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)
810
565
# copy revision keys and adjust values
811
566
self.pb.update("Copying revision texts", 1)
812
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
813
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
814
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))
815
569
if 'pack' in debug.debug_flags:
816
570
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
817
time.ctime(), self._pack_collection._upload_transport.base,
818
self.new_pack.random_name,
819
self.new_pack.revision_index.key_count(),
820
time.time() - self.new_pack.start_time)
821
self._revision_keys = revision_keys
823
def _copy_inventory_texts(self):
824
"""Copy the inventory texts to the new pack.
826
self._revision_keys is used to determine what inventories to copy.
828
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)
830
574
# select inventory keys
831
inv_keys = self._revision_keys # currently the same keyspace, and note that
575
inv_keys = revision_keys # currently the same keyspace, and note that
832
576
# querying for keys here could introduce a bug where an inventory item
833
577
# is missed, so do not change it to query separately without cross
834
578
# checking like the text key check below.
835
inventory_index_map, inventory_indices = self._pack_map_and_index_list(
837
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)
838
582
# copy inventory keys and adjust values
839
583
# XXX: Should be a helper function to allow different inv representation
841
585
self.pb.update("Copying inventory texts", 2)
842
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
843
# Only grab the output lines if we will be processing them
844
output_lines = bool(self.revision_ids)
845
inv_lines = self._copy_nodes_graph(inventory_index_map,
846
self.new_pack._writer, self.new_pack.inventory_index,
847
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)
848
588
if self.revision_ids:
849
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])
851
596
# eat the iterator to cause it to execute.
853
self._text_filter = None
854
599
if 'pack' in debug.debug_flags:
855
600
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
856
time.ctime(), self._pack_collection._upload_transport.base,
857
self.new_pack.random_name,
858
self.new_pack.inventory_index.key_count(),
859
time.time() - self.new_pack.start_time)
861
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)
862
604
# select text keys
863
text_index_map, text_nodes = self._get_text_nodes()
864
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:
865
609
# We could return the keys copied as part of the return value from
866
610
# _copy_nodes_graph but this doesn't work all that well with the
867
611
# need to get line output too, so we check separately, and as we're
1037
713
pb = ui.ui_factory.nested_progress_bar()
1039
for result in self._do_copy_nodes_graph(index_map, writer,
1040
write_index, output_lines, pb, readv_group_iter, total_items):
1043
# 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)
1049
def _do_copy_nodes_graph(self, index_map, writer, write_index,
1050
output_lines, pb, readv_group_iter, total_items):
720
def _do_copy_nodes_graph(self, nodes, index_map, writer, write_index,
1051
722
# for record verification
1052
knit = KnitVersionedFiles(None, None)
723
knit_data = _KnitData(None)
1053
724
# for line extraction when requested (inventories only)
1054
725
if output_lines:
1055
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?
1056
735
record_index = 0
1057
pb.update("Copied record", record_index, total_items)
1058
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()
1060
pack_obj = index_map[index]
1061
transport, path = pack_obj.access_tuple()
1063
reader = pack.make_readv_reader(transport, path, readv_vector)
1064
except errors.NoSuchFile:
1065
if self._reload_func is not None:
1068
for (names, read_func), (key, eol_flag, references) in \
1069
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):
1070
756
raw_data = read_func(None)
1071
758
if output_lines:
1072
759
# read the entire thing
1073
content, _ = knit._parse_record(key[-1], raw_data)
760
content, _ = knit_data._parse_record(version_id, raw_data)
1074
761
if len(references[-1]) == 0:
1075
762
line_iterator = factory.get_fulltext_content(content)
1077
764
line_iterator = factory.get_linedelta_content(content)
1078
765
for line in line_iterator:
766
yield line, version_id
1081
768
# check the header only
1082
df, _ = knit._parse_record_header(key, raw_data)
769
df, _ = knit_data._parse_record_header(version_id, raw_data)
1084
771
pos, size = writer.add_bytes_record(raw_data, names)
1085
772
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1086
773
pb.update("Copied record", record_index)
1087
774
record_index += 1
1089
def _get_text_nodes(self):
1090
text_index_map, text_indices = self._pack_map_and_index_list(
1092
return text_index_map, self._index_contents(text_indices,
1095
def _least_readv_node_readv(self, nodes):
1096
"""Generate request groups for nodes using the least readv's.
1098
:param nodes: An iterable of graph index nodes.
1099
:return: Total node count and an iterator of the data needed to perform
1100
readvs to obtain the data for nodes. Each item yielded by the
1101
iterator is a tuple with:
1102
index, readv_vector, node_vector. readv_vector is a list ready to
1103
hand to the transport readv method, and node_vector is a list of
1104
(key, eol_flag, references) for the node retrieved by the
1105
matching readv_vector.
1107
# group by pack so we do one readv per pack
1108
nodes = sorted(nodes)
1111
for index, key, value, references in nodes:
1112
if index not in request_groups:
1113
request_groups[index] = []
1114
request_groups[index].append((key, value, references))
1116
for index, items in request_groups.iteritems():
1117
pack_readv_requests = []
1118
for key, value, references in items:
1119
# ---- KnitGraphIndex.get_position
1120
bits = value[1:].split(' ')
1121
offset, length = int(bits[0]), int(bits[1])
1122
pack_readv_requests.append(
1123
((offset, length), (key, value[0], references)))
1124
# linear scan up the pack to maximum range combining.
1125
pack_readv_requests.sort()
1126
# split out the readv and the node data.
1127
pack_readv = [readv for readv, node in pack_readv_requests]
1128
node_vector = [node for readv, node in pack_readv_requests]
1129
result.append((index, pack_readv, node_vector))
1130
return total, result
1132
def _log_copied_texts(self):
1133
if 'pack' in debug.debug_flags:
1134
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
1135
time.ctime(), self._pack_collection._upload_transport.base,
1136
self.new_pack.random_name,
1137
self.new_pack.text_index.key_count(),
1138
time.time() - self.new_pack.start_time)
1140
def _process_inventory_lines(self, inv_lines):
1141
"""Use up the inv_lines generator and setup a text key filter."""
1142
repo = self._pack_collection.repo
1143
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
1144
inv_lines, self.revision_keys)
1146
for fileid, file_revids in fileid_revisions.iteritems():
1147
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
1148
self._text_filter = text_filter
1150
def _revision_node_readv(self, revision_nodes):
1151
"""Return the total revisions and the readv's to issue.
1153
:param revision_nodes: The revision index contents for the packs being
1154
incorporated into the new pack.
1155
:return: As per _least_readv_node_readv.
1157
return self._least_readv_node_readv(revision_nodes)
1159
def _use_pack(self, new_pack):
1160
"""Return True if new_pack should be used.
1162
:param new_pack: The pack that has just been created.
1163
:return: True if the pack should be used.
1165
return new_pack.data_inserted()
1168
class OptimisingPacker(Packer):
1169
"""A packer which spends more time to create better disk layouts."""
1171
def _revision_node_readv(self, revision_nodes):
1172
"""Return the total revisions and the readv's to issue.
1174
This sort places revisions in topological order with the ancestors
1177
:param revision_nodes: The revision index contents for the packs being
1178
incorporated into the new pack.
1179
:return: As per _least_readv_node_readv.
1181
# build an ancestors dict
1184
for index, key, value, references in revision_nodes:
1185
ancestors[key] = references[0]
1186
by_key[key] = (index, value, references)
1187
order = tsort.topo_sort(ancestors)
1189
# Single IO is pathological, but it will work as a starting point.
1191
for key in reversed(order):
1192
index, value, references = by_key[key]
1193
# ---- KnitGraphIndex.get_position
1194
bits = value[1:].split(' ')
1195
offset, length = int(bits[0]), int(bits[1])
1197
(index, [(offset, length)], [(key, value[0], references)]))
1198
# TODO: combine requests in the same index that are in ascending order.
1199
return total, requests
1201
def open_pack(self):
1202
"""Open a pack for the pack we are creating."""
1203
new_pack = super(OptimisingPacker, self).open_pack()
1204
# Turn on the optimization flags for all the index builders.
1205
new_pack.revision_index.set_optimize(for_size=True)
1206
new_pack.inventory_index.set_optimize(for_size=True)
1207
new_pack.text_index.set_optimize(for_size=True)
1208
new_pack.signature_index.set_optimize(for_size=True)
1212
777
class ReconcilePacker(Packer):
1213
778
"""A packer which regenerates indices etc as it copies.
1215
780
This is used by ``bzr reconcile`` to cause parent text pointers to be
1219
def _extra_init(self):
1220
self._data_changed = False
1222
def _process_inventory_lines(self, inv_lines):
1223
"""Generate a text key reference map rather for reconciling with."""
1224
repo = self._pack_collection.repo
1225
refs = repo._find_text_key_references_from_xml_inventory_lines(
1227
self._text_refs = refs
1228
# during reconcile we:
1229
# - convert unreferenced texts to full texts
1230
# - correct texts which reference a text not copied to be full texts
1231
# - copy all others as-is but with corrected parents.
1232
# - so at this point we don't know enough to decide what becomes a full
1234
self._text_filter = None
1236
def _copy_text_texts(self):
1237
"""generate what texts we should have and then copy."""
1238
self.pb.update("Copying content texts", 3)
1239
# we have three major tasks here:
1240
# 1) generate the ideal index
1241
repo = self._pack_collection.repo
1242
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1243
_1, key, _2, refs in
1244
self.new_pack.revision_index.iter_all_entries()])
1245
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1246
# 2) generate a text_nodes list that contains all the deltas that can
1247
# be used as-is, with corrected parents.
1250
discarded_nodes = []
1251
NULL_REVISION = _mod_revision.NULL_REVISION
1252
text_index_map, text_nodes = self._get_text_nodes()
1253
for node in text_nodes:
1259
ideal_parents = tuple(ideal_index[node[1]])
1261
discarded_nodes.append(node)
1262
self._data_changed = True
1264
if ideal_parents == (NULL_REVISION,):
1266
if ideal_parents == node[3][0]:
1268
ok_nodes.append(node)
1269
elif ideal_parents[0:1] == node[3][0][0:1]:
1270
# the left most parent is the same, or there are no parents
1271
# today. Either way, we can preserve the representation as
1272
# long as we change the refs to be inserted.
1273
self._data_changed = True
1274
ok_nodes.append((node[0], node[1], node[2],
1275
(ideal_parents, node[3][1])))
1276
self._data_changed = True
1278
# Reinsert this text completely
1279
bad_texts.append((node[1], ideal_parents))
1280
self._data_changed = True
1281
# we're finished with some data.
1284
# 3) bulk copy the ok data
1285
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1286
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1287
self.new_pack.text_index, readv_group_iter, total_items))
1288
# 4) adhoc copy all the other texts.
1289
# We have to topologically insert all texts otherwise we can fail to
1290
# reconcile when parts of a single delta chain are preserved intact,
1291
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1292
# reinserted, and if d3 has incorrect parents it will also be
1293
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1294
# copied), so we will try to delta, but d2 is not currently able to be
1295
# extracted because it's basis d1 is not present. Topologically sorting
1296
# addresses this. The following generates a sort for all the texts that
1297
# are being inserted without having to reference the entire text key
1298
# space (we only topo sort the revisions, which is smaller).
1299
topo_order = tsort.topo_sort(ancestors)
1300
rev_order = dict(zip(topo_order, range(len(topo_order))))
1301
bad_texts.sort(key=lambda key:rev_order.get(key[0][1], 0))
1302
transaction = repo.get_transaction()
1303
file_id_index = GraphIndexPrefixAdapter(
1304
self.new_pack.text_index,
1306
add_nodes_callback=self.new_pack.text_index.add_nodes)
1307
data_access = _DirectPackAccess(
1308
{self.new_pack.text_index:self.new_pack.access_tuple()})
1309
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1310
self.new_pack.access_tuple())
1311
output_texts = KnitVersionedFiles(
1312
_KnitGraphIndex(self.new_pack.text_index,
1313
add_callback=self.new_pack.text_index.add_nodes,
1314
deltas=True, parents=True, is_locked=repo.is_locked),
1315
data_access=data_access, max_delta_chain=200)
1316
for key, parent_keys in bad_texts:
1317
# We refer to the new pack to delta data being output.
1318
# A possible improvement would be to catch errors on short reads
1319
# and only flush then.
1320
self.new_pack.flush()
1322
for parent_key in parent_keys:
1323
if parent_key[0] != key[0]:
1324
# Graph parents must match the fileid
1325
raise errors.BzrError('Mismatched key parent %r:%r' %
1327
parents.append(parent_key[1])
1328
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1329
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1330
output_texts.add_lines(key, parent_keys, text_lines,
1331
random_id=True, check_content=False)
1332
# 5) check that nothing inserted has a reference outside the keyspace.
1333
missing_text_keys = self.new_pack.text_index._external_references()
1334
if missing_text_keys:
1335
raise errors.BzrCheckError('Reference to missing compression parents %r'
1336
% (missing_text_keys,))
1337
self._log_copied_texts()
1339
def _use_pack(self, new_pack):
1340
"""Override _use_pack to check for reconcile having changed content."""
1341
# XXX: we might be better checking this at the copy time.
1342
original_inventory_keys = set()
1343
inv_index = self._pack_collection.inventory_index.combined_index
1344
for entry in inv_index.iter_all_entries():
1345
original_inventory_keys.add(entry[1])
1346
new_inventory_keys = set()
1347
for entry in new_pack.inventory_index.iter_all_entries():
1348
new_inventory_keys.add(entry[1])
1349
if new_inventory_keys != original_inventory_keys:
1350
self._data_changed = True
1351
return new_pack.data_inserted() and self._data_changed
1354
785
class RepositoryPackCollection(object):
1355
"""Management of packs within a repository.
1357
:ivar _names: map of {pack_name: (index_size,)}
1360
pack_factory = NewPack
1361
resumed_pack_factory = ResumedPack
786
"""Management of packs within a repository."""
1363
788
def __init__(self, repo, transport, index_transport, upload_transport,
1364
pack_transport, index_builder_class, index_class,
1366
790
"""Create a new RepositoryPackCollection.
1368
:param transport: Addresses the repository base directory
792
:param transport: Addresses the repository base directory
1369
793
(typically .bzr/repository/).
1370
794
:param index_transport: Addresses the directory containing indices.
1371
795
:param upload_transport: Addresses the directory into which packs are written
1372
796
while they're being created.
1373
797
:param pack_transport: Addresses the directory of existing complete packs.
1374
:param index_builder_class: The index builder class to use.
1375
:param index_class: The index class to use.
1376
:param use_chk_index: Whether to setup and manage a CHK index.
1378
# XXX: This should call self.reset()
1379
799
self.repo = repo
1380
800
self.transport = transport
1381
801
self._index_transport = index_transport
1382
802
self._upload_transport = upload_transport
1383
803
self._pack_transport = pack_transport
1384
self._index_builder_class = index_builder_class
1385
self._index_class = index_class
1386
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3,
804
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
1389
806
# name:Pack mapping
1391
807
self._packs_by_name = {}
1392
808
# the previous pack-names content
1393
809
self._packs_at_load = None
1394
810
# when a pack is being created by this object, the state of that pack.
1395
811
self._new_pack = None
1396
812
# aggregated revision index data
1397
flush = self._flush_new_pack
1398
self.revision_index = AggregateIndex(self.reload_pack_names, flush)
1399
self.inventory_index = AggregateIndex(self.reload_pack_names, flush)
1400
self.text_index = AggregateIndex(self.reload_pack_names, flush)
1401
self.signature_index = AggregateIndex(self.reload_pack_names, flush)
1402
all_indices = [self.revision_index, self.inventory_index,
1403
self.text_index, self.signature_index]
1405
self.chk_index = AggregateIndex(self.reload_pack_names, flush)
1406
all_indices.append(self.chk_index)
1408
# used to determine if we're using a chk_index elsewhere.
1409
self.chk_index = None
1410
# Tell all the CombinedGraphIndex objects about each other, so they can
1411
# share hints about which pack names to search first.
1412
all_combined = [agg_idx.combined_index for agg_idx in all_indices]
1413
for combined_idx in all_combined:
1414
combined_idx.set_sibling_indices(
1415
set(all_combined).difference([combined_idx]))
1417
self._resumed_packs = []
1420
return '%s(%r)' % (self.__class__.__name__, self.repo)
813
self.revision_index = AggregateIndex()
814
self.inventory_index = AggregateIndex()
815
self.text_index = AggregateIndex()
816
self.signature_index = AggregateIndex()
1422
818
def add_pack_to_memory(self, pack):
1423
819
"""Make a Pack object available to the repository to satisfy queries.
1425
821
:param pack: A Pack object.
1427
if pack.name in self._packs_by_name:
1428
raise AssertionError(
1429
'pack %s already in _packs_by_name' % (pack.name,))
823
assert pack.name not in self._packs_by_name
1430
824
self.packs.append(pack)
1431
825
self._packs_by_name[pack.name] = pack
1432
826
self.revision_index.add_index(pack.revision_index, pack)
1433
827
self.inventory_index.add_index(pack.inventory_index, pack)
1434
828
self.text_index.add_index(pack.text_index, pack)
1435
829
self.signature_index.add_index(pack.signature_index, pack)
1436
if self.chk_index is not None:
1437
self.chk_index.add_index(pack.chk_index, pack)
831
def _add_text_to_weave(self, file_id, revision_id, new_lines, parents,
832
nostore_sha, random_revid):
833
file_id_index = GraphIndexPrefixAdapter(
834
self.text_index.combined_index,
836
add_nodes_callback=self.text_index.add_callback)
837
self.repo._text_knit._index._graph_index = file_id_index
838
self.repo._text_knit._index._add_callback = file_id_index.add_nodes
839
return self.repo._text_knit.add_lines_with_ghosts(
840
revision_id, parents, new_lines, nostore_sha=nostore_sha,
841
random_id=random_revid, check_content=False)[0:2]
1439
843
def all_packs(self):
1440
844
"""Return a list of all the Pack objects this repository has.
1491
891
# group their data with the relevant commit, and that may
1492
892
# involve rewriting ancient history - which autopack tries to
1493
893
# avoid. Alternatively we could not group the data but treat
1494
# each of these as having a single revision, and thus add
894
# each of these as having a single revision, and thus add
1495
895
# one revision for each to the total revision count, to get
1496
896
# a matching distribution.
1498
898
existing_packs.append((revision_count, pack))
1499
899
pack_operations = self.plan_autopack_combinations(
1500
900
existing_packs, pack_distribution)
1501
num_new_packs = len(pack_operations)
1502
num_old_packs = sum([len(po[1]) for po in pack_operations])
1503
num_revs_affected = sum([po[0] for po in pack_operations])
1504
mutter('Auto-packing repository %s, which has %d pack files, '
1505
'containing %d revisions. Packing %d files into %d affecting %d'
1506
' revisions', self, total_packs, total_revisions, num_old_packs,
1507
num_new_packs, num_revs_affected)
1508
result = self._execute_pack_operations(pack_operations,
1509
reload_func=self._restart_autopack)
1510
mutter('Auto-packing repository %s completed', self)
901
self._execute_pack_operations(pack_operations)
1513
def _execute_pack_operations(self, pack_operations, _packer_class=Packer,
904
def _execute_pack_operations(self, pack_operations):
1515
905
"""Execute a series of pack operations.
1517
907
:param pack_operations: A list of [revision_count, packs_to_combine].
1518
:param _packer_class: The class of packer to use (default: Packer).
1519
:return: The new pack names.
1521
910
for revision_count, packs in pack_operations:
1522
911
# we may have no-ops from the setup logic
1523
912
if len(packs) == 0:
1525
packer = _packer_class(self, packs, '.autopack',
1526
reload_func=reload_func)
1529
except errors.RetryWithNewPacks:
1530
# An exception is propagating out of this context, make sure
1531
# this packer has cleaned up. Packer() doesn't set its new_pack
1532
# state into the RepositoryPackCollection object, so we only
1533
# have access to it directly here.
1534
if packer.new_pack is not None:
1535
packer.new_pack.abort()
914
Packer(self, packs, '.autopack').pack()
1537
915
for pack in packs:
1538
916
self._remove_pack_from_memory(pack)
1539
917
# record the newly available packs and stop advertising the old
1541
to_be_obsoleted = []
1542
for _, packs in pack_operations:
1543
to_be_obsoleted.extend(packs)
1544
result = self._save_pack_names(clear_obsolete_packs=True,
1545
obsolete_packs=to_be_obsoleted)
1548
def _flush_new_pack(self):
1549
if self._new_pack is not None:
1550
self._new_pack.flush()
919
self._save_pack_names(clear_obsolete_packs=True)
920
# Move the old packs out of the way now they are no longer referenced.
921
for revision_count, packs in pack_operations:
922
self._obsolete_packs(packs)
1552
924
def lock_names(self):
1553
925
"""Acquire the mutex around the pack-names index.
1555
927
This cannot be used in the middle of a read-only transaction on the
1558
930
self.repo.control_files.lock_write()
1560
def _already_packed(self):
1561
"""Is the collection already packed?"""
1562
return not (self.repo._format.pack_compresses or (len(self._names) > 1))
1564
def pack(self, hint=None, clean_obsolete_packs=False):
1565
933
"""Pack the pack collection totally."""
1566
934
self.ensure_loaded()
1567
935
total_packs = len(self._names)
1568
if self._already_packed():
1570
938
total_revisions = self.revision_index.combined_index.key_count()
1571
939
# XXX: the following may want to be a class, to pack with a given
1573
941
mutter('Packing repository %s, which has %d pack files, '
1574
'containing %d revisions with hint %r.', self, total_packs,
1575
total_revisions, hint)
942
'containing %d revisions into 1 packs.', self, total_packs,
1576
944
# determine which packs need changing
945
pack_distribution = [1]
1577
946
pack_operations = [[0, []]]
1578
947
for pack in self.all_packs():
1579
if hint is None or pack.name in hint:
1580
# Either no hint was provided (so we are packing everything),
1581
# or this pack was included in the hint.
1582
pack_operations[-1][0] += pack.get_revision_count()
1583
pack_operations[-1][1].append(pack)
1584
self._execute_pack_operations(pack_operations, OptimisingPacker)
1586
if clean_obsolete_packs:
1587
self._clear_obsolete_packs()
948
revision_count = pack.get_revision_count()
949
pack_operations[-1][0] += revision_count
950
pack_operations[-1][1].append(pack)
951
self._execute_pack_operations(pack_operations)
1589
953
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1590
954
"""Plan a pack operation.
1874
1158
self._packs_by_name = {}
1875
1159
self._packs_at_load = None
1161
def _make_index_map(self, index_suffix):
1162
"""Return information on existing indices.
1164
:param suffix: Index suffix added to pack name.
1166
:returns: (pack_map, indices) where indices is a list of GraphIndex
1167
objects, and pack_map is a mapping from those objects to the
1168
pack tuple they describe.
1170
# TODO: stop using this; it creates new indices unnecessarily.
1171
self.ensure_loaded()
1172
suffix_map = {'.rix': 'revision_index',
1173
'.six': 'signature_index',
1174
'.iix': 'inventory_index',
1175
'.tix': 'text_index',
1177
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1178
suffix_map[index_suffix])
1180
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1181
"""Convert a list of packs to an index pack map and index list.
1183
:param packs: The packs list to process.
1184
:param index_attribute: The attribute that the desired index is found
1186
:return: A tuple (map, list) where map contains the dict from
1187
index:pack_tuple, and lsit contains the indices in the same order
1193
index = getattr(pack, index_attribute)
1194
indices.append(index)
1195
pack_map[index] = (pack.pack_transport, pack.file_name())
1196
return pack_map, indices
1198
def _index_contents(self, pack_map, key_filter=None):
1199
"""Get an iterable of the index contents from a pack_map.
1201
:param pack_map: A map from indices to pack details.
1202
:param key_filter: An optional filter to limit the
1205
indices = [index for index in pack_map.iterkeys()]
1206
all_index = CombinedGraphIndex(indices)
1207
if key_filter is None:
1208
return all_index.iter_all_entries()
1210
return all_index.iter_entries(key_filter)
1877
1212
def _unlock_names(self):
1878
1213
"""Release the mutex around the pack-names index."""
1879
1214
self.repo.control_files.unlock()
1881
def _diff_pack_names(self):
1882
"""Read the pack names from disk, and compare it to the one in memory.
1884
:return: (disk_nodes, deleted_nodes, new_nodes)
1885
disk_nodes The final set of nodes that should be referenced
1886
deleted_nodes Nodes which have been removed from when we started
1887
new_nodes Nodes that are newly introduced
1889
# load the disk nodes across
1891
for index, key, value in self._iter_disk_pack_index():
1892
disk_nodes.add((key, value))
1893
orig_disk_nodes = set(disk_nodes)
1895
# do a two-way diff against our original content
1896
current_nodes = set()
1897
for name, sizes in self._names.iteritems():
1899
((name, ), ' '.join(str(size) for size in sizes)))
1901
# Packs no longer present in the repository, which were present when we
1902
# locked the repository
1903
deleted_nodes = self._packs_at_load - current_nodes
1904
# Packs which this process is adding
1905
new_nodes = current_nodes - self._packs_at_load
1907
# Update the disk_nodes set to include the ones we are adding, and
1908
# remove the ones which were removed by someone else
1909
disk_nodes.difference_update(deleted_nodes)
1910
disk_nodes.update(new_nodes)
1912
return disk_nodes, deleted_nodes, new_nodes, orig_disk_nodes
1914
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1915
"""Given the correct set of pack files, update our saved info.
1917
:return: (removed, added, modified)
1918
removed pack names removed from self._names
1919
added pack names added to self._names
1920
modified pack names that had changed value
1925
## self._packs_at_load = disk_nodes
1216
def _save_pack_names(self, clear_obsolete_packs=False):
1217
"""Save the list of packs.
1219
This will take out the mutex around the pack names list for the
1220
duration of the method call. If concurrent updates have been made, a
1221
three-way merge between the current list and the current in memory list
1224
:param clear_obsolete_packs: If True, clear out the contents of the
1225
obsolete_packs directory.
1229
builder = GraphIndexBuilder()
1230
# load the disk nodes across
1232
for index, key, value in self._iter_disk_pack_index():
1233
disk_nodes.add((key, value))
1234
# do a two-way diff against our original content
1235
current_nodes = set()
1236
for name, sizes in self._names.iteritems():
1238
((name, ), ' '.join(str(size) for size in sizes)))
1239
deleted_nodes = self._packs_at_load - current_nodes
1240
new_nodes = current_nodes - self._packs_at_load
1241
disk_nodes.difference_update(deleted_nodes)
1242
disk_nodes.update(new_nodes)
1243
# TODO: handle same-name, index-size-changes here -
1244
# e.g. use the value from disk, not ours, *unless* we're the one
1246
for key, value in disk_nodes:
1247
builder.add_node(key, value)
1248
self.transport.put_file('pack-names', builder.finish())
1249
# move the baseline forward
1250
self._packs_at_load = disk_nodes
1251
# now clear out the obsolete packs directory
1252
if clear_obsolete_packs:
1253
self.transport.clone('obsolete_packs').delete_multi(
1254
self.transport.list_dir('obsolete_packs'))
1256
self._unlock_names()
1257
# synchronise the memory packs list with what we just wrote:
1926
1258
new_names = dict(disk_nodes)
1927
1259
# drop no longer present nodes
1928
1260
for pack in self.all_packs():
1929
1261
if (pack.name,) not in new_names:
1930
removed.append(pack.name)
1931
1262
self._remove_pack_from_memory(pack)
1932
1263
# add new nodes/refresh existing ones
1933
1264
for key, value in disk_nodes:
1947
1278
self._remove_pack_from_memory(self.get_pack_by_name(name))
1948
1279
self._names[name] = sizes
1949
1280
self.get_pack_by_name(name)
1950
modified.append(name)
1953
1283
self._names[name] = sizes
1954
1284
self.get_pack_by_name(name)
1956
return removed, added, modified
1958
def _save_pack_names(self, clear_obsolete_packs=False, obsolete_packs=None):
1959
"""Save the list of packs.
1961
This will take out the mutex around the pack names list for the
1962
duration of the method call. If concurrent updates have been made, a
1963
three-way merge between the current list and the current in memory list
1966
:param clear_obsolete_packs: If True, clear out the contents of the
1967
obsolete_packs directory.
1968
:param obsolete_packs: Packs that are obsolete once the new pack-names
1969
file has been written.
1970
:return: A list of the names saved that were not previously on disk.
1972
already_obsolete = []
1975
builder = self._index_builder_class()
1976
(disk_nodes, deleted_nodes, new_nodes,
1977
orig_disk_nodes) = self._diff_pack_names()
1978
# TODO: handle same-name, index-size-changes here -
1979
# e.g. use the value from disk, not ours, *unless* we're the one
1981
for key, value in disk_nodes:
1982
builder.add_node(key, value)
1983
self.transport.put_file('pack-names', builder.finish(),
1984
mode=self.repo.bzrdir._get_file_mode())
1985
self._packs_at_load = disk_nodes
1986
if clear_obsolete_packs:
1989
to_preserve = set([o.name for o in obsolete_packs])
1990
already_obsolete = self._clear_obsolete_packs(to_preserve)
1992
self._unlock_names()
1993
# synchronise the memory packs list with what we just wrote:
1994
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1996
# TODO: We could add one more condition here. "if o.name not in
1997
# orig_disk_nodes and o != the new_pack we haven't written to
1998
# disk yet. However, the new pack object is not easily
1999
# accessible here (it would have to be passed through the
2000
# autopacking code, etc.)
2001
obsolete_packs = [o for o in obsolete_packs
2002
if o.name not in already_obsolete]
2003
self._obsolete_packs(obsolete_packs)
2004
return [new_node[0][0] for new_node in new_nodes]
2006
def reload_pack_names(self):
2007
"""Sync our pack listing with what is present in the repository.
2009
This should be called when we find out that something we thought was
2010
present is now missing. This happens when another process re-packs the
2013
:return: True if the in-memory list of packs has been altered at all.
2015
# The ensure_loaded call is to handle the case where the first call
2016
# made involving the collection was to reload_pack_names, where we
2017
# don't have a view of disk contents. Its a bit of a bandaid, and
2018
# causes two reads of pack-names, but its a rare corner case not struck
2019
# with regular push/pull etc.
2020
first_read = self.ensure_loaded()
2023
# out the new value.
2024
(disk_nodes, deleted_nodes, new_nodes,
2025
orig_disk_nodes) = self._diff_pack_names()
2026
# _packs_at_load is meant to be the explicit list of names in
2027
# 'pack-names' at then start. As such, it should not contain any
2028
# pending names that haven't been written out yet.
2029
self._packs_at_load = orig_disk_nodes
2031
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
2032
if removed or added or modified:
2036
def _restart_autopack(self):
2037
"""Reload the pack names list, and restart the autopack code."""
2038
if not self.reload_pack_names():
2039
# Re-raise the original exception, because something went missing
2040
# and a restart didn't find it
2042
raise errors.RetryAutopack(self.repo, False, sys.exc_info())
2044
def _clear_obsolete_packs(self, preserve=None):
2045
"""Delete everything from the obsolete-packs directory.
2047
:return: A list of pack identifiers (the filename without '.pack') that
2048
were found in obsolete_packs.
2051
obsolete_pack_transport = self.transport.clone('obsolete_packs')
2052
if preserve is None:
2054
for filename in obsolete_pack_transport.list_dir('.'):
2055
name, ext = osutils.splitext(filename)
2058
if name in preserve:
2061
obsolete_pack_transport.delete(filename)
2062
except (errors.PathError, errors.TransportError), e:
2063
warning("couldn't delete obsolete pack, skipping it:\n%s"
2067
1286
def _start_write_group(self):
2068
1287
# Do not permit preparation for writing if we're not in a 'write lock'.
2069
1288
if not self.repo.is_write_locked():
2070
1289
raise errors.NotWriteLocked(self)
2071
self._new_pack = self.pack_factory(self, upload_suffix='.pack',
2072
file_mode=self.repo.bzrdir._get_file_mode())
1290
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1291
self._pack_transport, upload_suffix='.pack')
2073
1292
# allow writing: queue writes to a new index
2074
1293
self.revision_index.add_writable_index(self._new_pack.revision_index,
2075
1294
self._new_pack)
2077
1296
self._new_pack)
2078
1297
self.text_index.add_writable_index(self._new_pack.text_index,
2079
1298
self._new_pack)
2080
self._new_pack.text_index.set_optimize(combine_backing_indices=False)
2081
1299
self.signature_index.add_writable_index(self._new_pack.signature_index,
2082
1300
self._new_pack)
2083
if self.chk_index is not None:
2084
self.chk_index.add_writable_index(self._new_pack.chk_index,
2086
self.repo.chk_bytes._index._add_callback = self.chk_index.add_callback
2087
self._new_pack.chk_index.set_optimize(combine_backing_indices=False)
2089
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
2090
self.repo.revisions._index._add_callback = self.revision_index.add_callback
2091
self.repo.signatures._index._add_callback = self.signature_index.add_callback
2092
self.repo.texts._index._add_callback = self.text_index.add_callback
1302
# reused revision and signature knits may need updating
1304
# "Hysterical raisins. client code in bzrlib grabs those knits outside
1305
# of write groups and then mutates it inside the write group."
1306
if self.repo._revision_knit is not None:
1307
self.repo._revision_knit._index._add_callback = \
1308
self.revision_index.add_callback
1309
if self.repo._signature_knit is not None:
1310
self.repo._signature_knit._index._add_callback = \
1311
self.signature_index.add_callback
1312
# create a reused knit object for text addition in commit.
1313
self.repo._text_knit = self.repo.weave_store.get_weave_or_empty(
2094
1316
def _abort_write_group(self):
2095
1317
# FIXME: just drop the transient index.
2096
1318
# forget what names there are
2097
if self._new_pack is not None:
2098
operation = cleanup.OperationWithCleanups(self._new_pack.abort)
2099
operation.add_cleanup(setattr, self, '_new_pack', None)
2100
# If we aborted while in the middle of finishing the write
2101
# group, _remove_pack_indices could fail because the indexes are
2102
# already gone. But they're not there we shouldn't fail in this
2103
# case, so we pass ignore_missing=True.
2104
operation.add_cleanup(self._remove_pack_indices, self._new_pack,
2105
ignore_missing=True)
2106
operation.run_simple()
2107
for resumed_pack in self._resumed_packs:
2108
operation = cleanup.OperationWithCleanups(resumed_pack.abort)
2109
# See comment in previous finally block.
2110
operation.add_cleanup(self._remove_pack_indices, resumed_pack,
2111
ignore_missing=True)
2112
operation.run_simple()
2113
del self._resumed_packs[:]
2115
def _remove_resumed_pack_indices(self):
2116
for resumed_pack in self._resumed_packs:
2117
self._remove_pack_indices(resumed_pack)
2118
del self._resumed_packs[:]
2120
def _check_new_inventories(self):
2121
"""Detect missing inventories in this write group.
2123
:returns: list of strs, summarising any problems found. If the list is
2124
empty no problems were found.
2126
# The base implementation does no checks. GCRepositoryPackCollection
1319
self._new_pack.abort()
1320
self._remove_pack_indices(self._new_pack)
1321
self._new_pack = None
1322
self.repo._text_knit = None
2130
1324
def _commit_write_group(self):
2132
for prefix, versioned_file in (
2133
('revisions', self.repo.revisions),
2134
('inventories', self.repo.inventories),
2135
('texts', self.repo.texts),
2136
('signatures', self.repo.signatures),
2138
missing = versioned_file.get_missing_compression_parent_keys()
2139
all_missing.update([(prefix,) + key for key in missing])
2141
raise errors.BzrCheckError(
2142
"Repository %s has missing compression parent(s) %r "
2143
% (self.repo, sorted(all_missing)))
2144
problems = self._check_new_inventories()
2146
problems_summary = '\n'.join(problems)
2147
raise errors.BzrCheckError(
2148
"Cannot add revision(s) to repository: " + problems_summary)
2149
1325
self._remove_pack_indices(self._new_pack)
2150
any_new_content = False
2151
1326
if self._new_pack.data_inserted():
2152
1327
# get all the data to disk and read to use
2153
1328
self._new_pack.finish()
2154
1329
self.allocate(self._new_pack)
2155
1330
self._new_pack = None
2156
any_new_content = True
2158
self._new_pack.abort()
2159
self._new_pack = None
2160
for resumed_pack in self._resumed_packs:
2161
# XXX: this is a pretty ugly way to turn the resumed pack into a
2162
# properly committed pack.
2163
self._names[resumed_pack.name] = None
2164
self._remove_pack_from_memory(resumed_pack)
2165
resumed_pack.finish()
2166
self.allocate(resumed_pack)
2167
any_new_content = True
2168
del self._resumed_packs[:]
2170
result = self.autopack()
1331
if not self.autopack():
2172
1332
# when autopack takes no steps, the names list is still
2174
return self._save_pack_names()
2178
def _suspend_write_group(self):
2179
tokens = [pack.name for pack in self._resumed_packs]
2180
self._remove_pack_indices(self._new_pack)
2181
if self._new_pack.data_inserted():
2182
# get all the data to disk and read to use
2183
self._new_pack.finish(suspend=True)
2184
tokens.append(self._new_pack.name)
2185
self._new_pack = None
1334
self._save_pack_names()
2187
1336
self._new_pack.abort()
2188
1337
self._new_pack = None
2189
self._remove_resumed_pack_indices()
2192
def _resume_write_group(self, tokens):
2193
for token in tokens:
2194
self._resume_pack(token)
1338
self.repo._text_knit = None
1341
class KnitPackRevisionStore(KnitRevisionStore):
1342
"""An object to adapt access from RevisionStore's to use KnitPacks.
1344
This class works by replacing the original RevisionStore.
1345
We need to do this because the KnitPackRevisionStore is less
1346
isolated in its layering - it uses services from the repo.
1349
def __init__(self, repo, transport, revisionstore):
1350
"""Create a KnitPackRevisionStore on repo with revisionstore.
1352
This will store its state in the Repository, use the
1353
indices to provide a KnitGraphIndex,
1354
and at the end of transactions write new indices.
1356
KnitRevisionStore.__init__(self, revisionstore.versioned_file_store)
1358
self._serializer = revisionstore._serializer
1359
self.transport = transport
1361
def get_revision_file(self, transaction):
1362
"""Get the revision versioned file object."""
1363
if getattr(self.repo, '_revision_knit', None) is not None:
1364
return self.repo._revision_knit
1365
self.repo._pack_collection.ensure_loaded()
1366
add_callback = self.repo._pack_collection.revision_index.add_callback
1367
# setup knit specific objects
1368
knit_index = KnitGraphIndex(
1369
self.repo._pack_collection.revision_index.combined_index,
1370
add_callback=add_callback)
1371
self.repo._revision_knit = knit.KnitVersionedFile(
1372
'revisions', self.transport.clone('..'),
1373
self.repo.control_files._file_mode,
1374
create=False, access_mode=self.repo._access_mode(),
1375
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1376
access_method=self.repo._pack_collection.revision_index.knit_access)
1377
return self.repo._revision_knit
1379
def get_signature_file(self, transaction):
1380
"""Get the signature versioned file object."""
1381
if getattr(self.repo, '_signature_knit', None) is not None:
1382
return self.repo._signature_knit
1383
self.repo._pack_collection.ensure_loaded()
1384
add_callback = self.repo._pack_collection.signature_index.add_callback
1385
# setup knit specific objects
1386
knit_index = KnitGraphIndex(
1387
self.repo._pack_collection.signature_index.combined_index,
1388
add_callback=add_callback, parents=False)
1389
self.repo._signature_knit = knit.KnitVersionedFile(
1390
'signatures', self.transport.clone('..'),
1391
self.repo.control_files._file_mode,
1392
create=False, access_mode=self.repo._access_mode(),
1393
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1394
access_method=self.repo._pack_collection.signature_index.knit_access)
1395
return self.repo._signature_knit
1398
class KnitPackTextStore(VersionedFileStore):
1399
"""Presents a TextStore abstraction on top of packs.
1401
This class works by replacing the original VersionedFileStore.
1402
We need to do this because the KnitPackRevisionStore is less
1403
isolated in its layering - it uses services from the repo and shares them
1404
with all the data written in a single write group.
1407
def __init__(self, repo, transport, weavestore):
1408
"""Create a KnitPackTextStore on repo with weavestore.
1410
This will store its state in the Repository, use the
1411
indices FileNames to provide a KnitGraphIndex,
1412
and at the end of transactions write new indices.
1414
# don't call base class constructor - it's not suitable.
1415
# no transient data stored in the transaction
1417
self._precious = False
1419
self.transport = transport
1420
self.weavestore = weavestore
1421
# XXX for check() which isn't updated yet
1422
self._transport = weavestore._transport
1424
def get_weave_or_empty(self, file_id, transaction):
1425
"""Get a 'Knit' backed by the .tix indices.
1427
The transaction parameter is ignored.
1429
self.repo._pack_collection.ensure_loaded()
1430
add_callback = self.repo._pack_collection.text_index.add_callback
1431
# setup knit specific objects
1432
file_id_index = GraphIndexPrefixAdapter(
1433
self.repo._pack_collection.text_index.combined_index,
1434
(file_id, ), 1, add_nodes_callback=add_callback)
1435
knit_index = KnitGraphIndex(file_id_index,
1436
add_callback=file_id_index.add_nodes,
1437
deltas=True, parents=True)
1438
return knit.KnitVersionedFile('text:' + file_id,
1439
self.transport.clone('..'),
1442
access_method=self.repo._pack_collection.text_index.knit_access,
1443
factory=knit.KnitPlainFactory())
1445
get_weave = get_weave_or_empty
1448
"""Generate a list of the fileids inserted, for use by check."""
1449
self.repo._pack_collection.ensure_loaded()
1451
for index, key, value, refs in \
1452
self.repo._pack_collection.text_index.combined_index.iter_all_entries():
1457
class InventoryKnitThunk(object):
1458
"""An object to manage thunking get_inventory_weave to pack based knits."""
1460
def __init__(self, repo, transport):
1461
"""Create an InventoryKnitThunk for repo at transport.
1463
This will store its state in the Repository, use the
1464
indices FileNames to provide a KnitGraphIndex,
1465
and at the end of transactions write a new index..
1468
self.transport = transport
1470
def get_weave(self):
1471
"""Get a 'Knit' that contains inventory data."""
1472
self.repo._pack_collection.ensure_loaded()
1473
add_callback = self.repo._pack_collection.inventory_index.add_callback
1474
# setup knit specific objects
1475
knit_index = KnitGraphIndex(
1476
self.repo._pack_collection.inventory_index.combined_index,
1477
add_callback=add_callback, deltas=True, parents=True)
1478
return knit.KnitVersionedFile(
1479
'inventory', self.transport.clone('..'),
1480
self.repo.control_files._file_mode,
1481
create=False, access_mode=self.repo._access_mode(),
1482
index=knit_index, delta=True, factory=knit.KnitPlainFactory(),
1483
access_method=self.repo._pack_collection.inventory_index.knit_access)
2197
1486
class KnitPackRepository(KnitRepository):
2198
"""Repository with knit objects stored inside pack containers.
2200
The layering for a KnitPackRepository is:
2202
Graph | HPSS | Repository public layer |
2203
===================================================
2204
Tuple based apis below, string based, and key based apis above
2205
---------------------------------------------------
2207
Provides .texts, .revisions etc
2208
This adapts the N-tuple keys to physical knit records which only have a
2209
single string identifier (for historical reasons), which in older formats
2210
was always the revision_id, and in the mapped code for packs is always
2211
the last element of key tuples.
2212
---------------------------------------------------
2214
A separate GraphIndex is used for each of the
2215
texts/inventories/revisions/signatures contained within each individual
2216
pack file. The GraphIndex layer works in N-tuples and is unaware of any
2218
===================================================
2222
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
1487
"""Experimental graph-knit using repository."""
1489
def __init__(self, _format, a_bzrdir, control_files, _revision_store,
1490
control_store, text_store, _commit_builder_class, _serializer):
2224
1491
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2225
_commit_builder_class, _serializer)
2226
index_transport = self._transport.clone('indices')
2227
self._pack_collection = RepositoryPackCollection(self, self._transport,
1492
_revision_store, control_store, text_store, _commit_builder_class,
1494
index_transport = control_files._transport.clone('indices')
1495
self._pack_collection = RepositoryPackCollection(self, control_files._transport,
2228
1496
index_transport,
2229
self._transport.clone('upload'),
2230
self._transport.clone('packs'),
2231
_format.index_builder_class,
2232
_format.index_class,
2233
use_chk_index=self._format.supports_chks,
2235
self.inventories = KnitVersionedFiles(
2236
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2237
add_callback=self._pack_collection.inventory_index.add_callback,
2238
deltas=True, parents=True, is_locked=self.is_locked),
2239
data_access=self._pack_collection.inventory_index.data_access,
2240
max_delta_chain=200)
2241
self.revisions = KnitVersionedFiles(
2242
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2243
add_callback=self._pack_collection.revision_index.add_callback,
2244
deltas=False, parents=True, is_locked=self.is_locked,
2245
track_external_parent_refs=True),
2246
data_access=self._pack_collection.revision_index.data_access,
2248
self.signatures = KnitVersionedFiles(
2249
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2250
add_callback=self._pack_collection.signature_index.add_callback,
2251
deltas=False, parents=False, is_locked=self.is_locked),
2252
data_access=self._pack_collection.signature_index.data_access,
2254
self.texts = KnitVersionedFiles(
2255
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2256
add_callback=self._pack_collection.text_index.add_callback,
2257
deltas=True, parents=True, is_locked=self.is_locked),
2258
data_access=self._pack_collection.text_index.data_access,
2259
max_delta_chain=200)
2260
if _format.supports_chks:
2261
# No graph, no compression:- references from chks are between
2262
# different objects not temporal versions of the same; and without
2263
# some sort of temporal structure knit compression will just fail.
2264
self.chk_bytes = KnitVersionedFiles(
2265
_KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2266
add_callback=self._pack_collection.chk_index.add_callback,
2267
deltas=False, parents=False, is_locked=self.is_locked),
2268
data_access=self._pack_collection.chk_index.data_access,
2271
self.chk_bytes = None
1497
control_files._transport.clone('upload'),
1498
control_files._transport.clone('packs'))
1499
self._revision_store = KnitPackRevisionStore(self, index_transport, self._revision_store)
1500
self.weave_store = KnitPackTextStore(self, index_transport, self.weave_store)
1501
self._inv_thunk = InventoryKnitThunk(self, index_transport)
2272
1502
# True when the repository object is 'write locked' (as opposed to the
2273
# physical lock only taken out around changes to the pack-names list.)
1503
# physical lock only taken out around changes to the pack-names list.)
2274
1504
# Another way to represent this would be a decorator around the control
2275
1505
# files object that presents logical locks as physical ones - if this
2276
1506
# gets ugly consider that alternative design. RBC 20071011
2278
1508
self._transaction = None
2280
1510
self._reconcile_does_inventory_gc = True
2281
self._reconcile_fixes_text_parents = True
1511
self._reconcile_fixes_text_parents = False
2282
1512
self._reconcile_backsup_inventory = False
2284
def _warn_if_deprecated(self, branch=None):
2285
# This class isn't deprecated, but one sub-format is
2286
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2287
super(KnitPackRepository, self)._warn_if_deprecated(branch)
2289
1514
def _abort_write_group(self):
2290
self.revisions._index._key_dependencies.clear()
2291
1515
self._pack_collection._abort_write_group()
2293
def _get_source(self, to_format):
2294
if to_format.network_name() == self._format.network_name():
2295
return KnitPackStreamSource(self, to_format)
2296
return super(KnitPackRepository, self)._get_source(to_format)
1517
def _access_mode(self):
1518
"""Return 'w' or 'r' for depending on whether a write lock is active.
1520
This method is a helper for the Knit-thunking support objects.
1522
if self.is_write_locked():
1526
def _find_inconsistent_revision_parents(self):
1527
"""Find revisions with incorrectly cached parents.
1529
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1530
parents-in-revision).
1532
assert self.is_locked()
1533
pb = ui.ui_factory.nested_progress_bar()
1536
revision_nodes = self._pack_collection.revision_index \
1537
.combined_index.iter_all_entries()
1538
index_positions = []
1539
# Get the cached index values for all revisions, and also the location
1540
# in each index of the revision text so we can perform linear IO.
1541
for index, key, value, refs in revision_nodes:
1542
pos, length = value[1:].split(' ')
1543
index_positions.append((index, int(pos), key[0],
1544
tuple(parent[0] for parent in refs[0])))
1545
pb.update("Reading revision index.", 0, 0)
1546
index_positions.sort()
1547
batch_count = len(index_positions) / 1000 + 1
1548
pb.update("Checking cached revision graph.", 0, batch_count)
1549
for offset in xrange(batch_count):
1550
pb.update("Checking cached revision graph.", offset)
1551
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1554
rev_ids = [item[2] for item in to_query]
1555
revs = self.get_revisions(rev_ids)
1556
for revision, item in zip(revs, to_query):
1557
index_parents = item[3]
1558
rev_parents = tuple(revision.parent_ids)
1559
if index_parents != rev_parents:
1560
result.append((revision.revision_id, index_parents, rev_parents))
1565
def get_parents(self, revision_ids):
1566
"""See StackedParentsProvider.get_parents.
1568
This implementation accesses the combined revision index to provide
1571
self._pack_collection.ensure_loaded()
1572
index = self._pack_collection.revision_index.combined_index
1574
for revision_id in revision_ids:
1575
if revision_id != _mod_revision.NULL_REVISION:
1576
search_keys.add((revision_id,))
1577
found_parents = {_mod_revision.NULL_REVISION:[]}
1578
for index, key, value, refs in index.iter_entries(search_keys):
1581
parents = (_mod_revision.NULL_REVISION,)
1583
parents = tuple(parent[0] for parent in parents)
1584
found_parents[key[0]] = parents
1586
for revision_id in revision_ids:
1588
result.append(found_parents[revision_id])
2298
1593
def _make_parents_provider(self):
2299
return graph.CachingParentsProvider(self)
2301
1596
def _refresh_data(self):
2302
if not self.is_locked():
2304
self._pack_collection.reload_pack_names()
1597
if self._write_lock_count == 1 or (
1598
self.control_files._lock_count == 1 and
1599
self.control_files._lock_mode == 'r'):
1600
# forget what names there are
1601
self._pack_collection.reset()
1602
# XXX: Better to do an in-memory merge when acquiring a new lock -
1603
# factor out code from _save_pack_names.
1604
self._pack_collection.ensure_loaded()
2306
1606
def _start_write_group(self):
2307
1607
self._pack_collection._start_write_group()
2309
1609
def _commit_write_group(self):
2310
hint = self._pack_collection._commit_write_group()
2311
self.revisions._index._key_dependencies.clear()
2314
def suspend_write_group(self):
2315
# XXX check self._write_group is self.get_transaction()?
2316
tokens = self._pack_collection._suspend_write_group()
2317
self.revisions._index._key_dependencies.clear()
2318
self._write_group = None
2321
def _resume_write_group(self, tokens):
2322
self._start_write_group()
2324
self._pack_collection._resume_write_group(tokens)
2325
except errors.UnresumableWriteGroup:
2326
self._abort_write_group()
2328
for pack in self._pack_collection._resumed_packs:
2329
self.revisions._index.scan_unvalidated_index(pack.revision_index)
1610
return self._pack_collection._commit_write_group()
1612
def get_inventory_weave(self):
1613
return self._inv_thunk.get_weave()
2331
1615
def get_transaction(self):
2332
1616
if self._write_lock_count:
2645
1863
def get_format_description(self):
2646
1864
"""See RepositoryFormat.get_format_description()."""
2647
1865
return "Packs containing knits with subtree support\n"
2650
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2651
"""A rich-root, no subtrees parameterized Pack repository.
2653
This repository format uses the xml6 serializer to get:
2654
- support for recording full info about the tree root
2656
This format was introduced in 1.0.
2659
repository_class = KnitPackRepository
2660
_commit_builder_class = PackRootCommitBuilder
2661
rich_root_data = True
2662
supports_tree_reference = False
2664
def _serializer(self):
2665
return xml6.serializer_v6
2666
# What index classes to use
2667
index_builder_class = InMemoryGraphIndex
2668
index_class = GraphIndex
2670
def _get_matching_bzrdir(self):
2671
return bzrdir.format_registry.make_bzrdir(
2674
def _ignore_setting_bzrdir(self, format):
2677
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2679
def get_format_string(self):
2680
"""See RepositoryFormat.get_format_string()."""
2681
return ("Bazaar pack repository format 1 with rich root"
2682
" (needs bzr 1.0)\n")
2684
def get_format_description(self):
2685
"""See RepositoryFormat.get_format_description()."""
2686
return "Packs containing knits with rich root support\n"
2689
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2690
"""Repository that supports external references to allow stacking.
2694
Supports external lookups, which results in non-truncated ghosts after
2695
reconcile compared to pack-0.92 formats.
2698
repository_class = KnitPackRepository
2699
_commit_builder_class = PackCommitBuilder
2700
supports_external_lookups = True
2701
# What index classes to use
2702
index_builder_class = InMemoryGraphIndex
2703
index_class = GraphIndex
2706
def _serializer(self):
2707
return xml5.serializer_v5
2709
def _get_matching_bzrdir(self):
2710
return bzrdir.format_registry.make_bzrdir('1.6')
2712
def _ignore_setting_bzrdir(self, format):
2715
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2717
def get_format_string(self):
2718
"""See RepositoryFormat.get_format_string()."""
2719
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2721
def get_format_description(self):
2722
"""See RepositoryFormat.get_format_description()."""
2723
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2726
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2727
"""A repository with rich roots and stacking.
2729
New in release 1.6.1.
2731
Supports stacking on other repositories, allowing data to be accessed
2732
without being stored locally.
2735
repository_class = KnitPackRepository
2736
_commit_builder_class = PackRootCommitBuilder
2737
rich_root_data = True
2738
supports_tree_reference = False # no subtrees
2739
supports_external_lookups = True
2740
# What index classes to use
2741
index_builder_class = InMemoryGraphIndex
2742
index_class = GraphIndex
2745
def _serializer(self):
2746
return xml6.serializer_v6
2748
def _get_matching_bzrdir(self):
2749
return bzrdir.format_registry.make_bzrdir(
2752
def _ignore_setting_bzrdir(self, format):
2755
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2757
def get_format_string(self):
2758
"""See RepositoryFormat.get_format_string()."""
2759
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2761
def get_format_description(self):
2762
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2765
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2766
"""A repository with rich roots and external references.
2770
Supports external lookups, which results in non-truncated ghosts after
2771
reconcile compared to pack-0.92 formats.
2773
This format was deprecated because the serializer it uses accidentally
2774
supported subtrees, when the format was not intended to. This meant that
2775
someone could accidentally fetch from an incorrect repository.
2778
repository_class = KnitPackRepository
2779
_commit_builder_class = PackRootCommitBuilder
2780
rich_root_data = True
2781
supports_tree_reference = False # no subtrees
2783
supports_external_lookups = True
2784
# What index classes to use
2785
index_builder_class = InMemoryGraphIndex
2786
index_class = GraphIndex
2789
def _serializer(self):
2790
return xml7.serializer_v7
2792
def _get_matching_bzrdir(self):
2793
matching = bzrdir.format_registry.make_bzrdir(
2795
matching.repository_format = self
2798
def _ignore_setting_bzrdir(self, format):
2801
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2803
def get_format_string(self):
2804
"""See RepositoryFormat.get_format_string()."""
2805
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2807
def get_format_description(self):
2808
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2812
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2813
"""A repository with stacking and btree indexes,
2814
without rich roots or subtrees.
2816
This is equivalent to pack-1.6 with B+Tree indices.
2819
repository_class = KnitPackRepository
2820
_commit_builder_class = PackCommitBuilder
2821
supports_external_lookups = True
2822
# What index classes to use
2823
index_builder_class = BTreeBuilder
2824
index_class = BTreeGraphIndex
2827
def _serializer(self):
2828
return xml5.serializer_v5
2830
def _get_matching_bzrdir(self):
2831
return bzrdir.format_registry.make_bzrdir('1.9')
2833
def _ignore_setting_bzrdir(self, format):
2836
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2838
def get_format_string(self):
2839
"""See RepositoryFormat.get_format_string()."""
2840
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2842
def get_format_description(self):
2843
"""See RepositoryFormat.get_format_description()."""
2844
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2847
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2848
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2850
1.6-rich-root with B+Tree indices.
2853
repository_class = KnitPackRepository
2854
_commit_builder_class = PackRootCommitBuilder
2855
rich_root_data = True
2856
supports_tree_reference = False # no subtrees
2857
supports_external_lookups = True
2858
# What index classes to use
2859
index_builder_class = BTreeBuilder
2860
index_class = BTreeGraphIndex
2863
def _serializer(self):
2864
return xml6.serializer_v6
2866
def _get_matching_bzrdir(self):
2867
return bzrdir.format_registry.make_bzrdir(
2870
def _ignore_setting_bzrdir(self, format):
2873
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2875
def get_format_string(self):
2876
"""See RepositoryFormat.get_format_string()."""
2877
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2879
def get_format_description(self):
2880
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2883
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2884
"""A subtrees development repository.
2886
This format should be retained until the second release after bzr 1.7.
2888
1.6.1-subtree[as it might have been] with B+Tree indices.
2890
This is [now] retained until we have a CHK based subtree format in
2894
repository_class = KnitPackRepository
2895
_commit_builder_class = PackRootCommitBuilder
2896
rich_root_data = True
2898
supports_tree_reference = True
2899
supports_external_lookups = True
2900
# What index classes to use
2901
index_builder_class = BTreeBuilder
2902
index_class = BTreeGraphIndex
2905
def _serializer(self):
2906
return xml7.serializer_v7
2908
def _get_matching_bzrdir(self):
2909
return bzrdir.format_registry.make_bzrdir(
2910
'development-subtree')
2912
def _ignore_setting_bzrdir(self, format):
2915
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2917
def get_format_string(self):
2918
"""See RepositoryFormat.get_format_string()."""
2919
return ("Bazaar development format 2 with subtree support "
2920
"(needs bzr.dev from before 1.8)\n")
2922
def get_format_description(self):
2923
"""See RepositoryFormat.get_format_description()."""
2924
return ("Development repository format, currently the same as "
2925
"1.6.1-subtree with B+Tree indices.\n")
added
removed
bzrlib/repofmt/pack_repo.py
