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
23
from bzrlib import (
38
29
from bzrlib.index import (
39
30
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.pack import ContainerWriter
38
from bzrlib.store import revision
50
40
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 (
53
from bzrlib.decorators import needs_read_lock, needs_write_lock
67
54
from bzrlib.repofmt.knitrepo import KnitRepository
68
55
from bzrlib.repository import (
70
58
MetaDirRepositoryFormat,
75
from bzrlib.trace import (
61
import bzrlib.revision as _mod_revision
62
from bzrlib.store.revision.knit import KnitRevisionStore
63
from bzrlib.store.versioned import VersionedFileStore
64
from bzrlib.trace import mutter, note, warning
82
67
class PackCommitBuilder(CommitBuilder):
83
68
"""A subclass of CommitBuilder to add texts with pack semantics.
85
70
Specifically this uses one knit object rather than one knit object per
86
71
added text, reducing memory and object pressure.
89
def __init__(self, repository, parents, config, timestamp=None,
90
timezone=None, committer=None, revprops=None,
92
CommitBuilder.__init__(self, repository, parents, config,
93
timestamp=timestamp, timezone=timezone, committer=committer,
94
revprops=revprops, revision_id=revision_id)
95
self._file_graph = graph.Graph(
96
repository._pack_collection.text_index.combined_index)
98
def _heads(self, file_id, revision_ids):
99
keys = [(file_id, revision_id) for revision_id in revision_ids]
100
return set([key[1] for key in self._file_graph.heads(keys)])
74
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
75
return self.repository._pack_collection._add_text_to_weave(file_id,
76
self._new_revision_id, new_lines, parents, nostore_sha,
103
80
class PackRootCommitBuilder(RootCommitBuilder):
104
81
"""A subclass of RootCommitBuilder to add texts with pack semantics.
106
83
Specifically this uses one knit object rather than one knit object per
107
84
added text, reducing memory and object pressure.
110
def __init__(self, repository, parents, config, timestamp=None,
111
timezone=None, committer=None, revprops=None,
113
CommitBuilder.__init__(self, repository, parents, config,
114
timestamp=timestamp, timezone=timezone, committer=committer,
115
revprops=revprops, revision_id=revision_id)
116
self._file_graph = graph.Graph(
117
repository._pack_collection.text_index.combined_index)
119
def _heads(self, file_id, revision_ids):
120
keys = [(file_id, revision_id) for revision_id in revision_ids]
121
return set([key[1] for key in self._file_graph.heads(keys)])
87
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
88
return self.repository._pack_collection._add_text_to_weave(file_id,
89
self._new_revision_id, new_lines, parents, nostore_sha,
124
93
class Pack(object):
261
169
return not self.__eq__(other)
263
171
def __repr__(self):
264
return "<%s.%s object at 0x%x, %s, %s" % (
265
self.__class__.__module__, self.__class__.__name__, id(self),
266
self.pack_transport, self.name)
269
class ResumedPack(ExistingPack):
271
def __init__(self, name, revision_index, inventory_index, text_index,
272
signature_index, upload_transport, pack_transport, index_transport,
273
pack_collection, chk_index=None):
274
"""Create a ResumedPack object."""
275
ExistingPack.__init__(self, pack_transport, name, revision_index,
276
inventory_index, text_index, signature_index,
278
self.upload_transport = upload_transport
279
self.index_transport = index_transport
280
self.index_sizes = [None, None, None, None]
282
('revision', revision_index),
283
('inventory', inventory_index),
284
('text', text_index),
285
('signature', signature_index),
287
if chk_index is not None:
288
indices.append(('chk', chk_index))
289
self.index_sizes.append(None)
290
for index_type, index in indices:
291
offset = self.index_offset(index_type)
292
self.index_sizes[offset] = index._size
293
self.index_class = pack_collection._index_class
294
self._pack_collection = pack_collection
295
self._state = 'resumed'
296
# XXX: perhaps check that the .pack file exists?
298
def access_tuple(self):
299
if self._state == 'finished':
300
return Pack.access_tuple(self)
301
elif self._state == 'resumed':
302
return self.upload_transport, self.file_name()
304
raise AssertionError(self._state)
307
self.upload_transport.delete(self.file_name())
308
indices = [self.revision_index, self.inventory_index, self.text_index,
309
self.signature_index]
310
if self.chk_index is not None:
311
indices.append(self.chk_index)
312
for index in indices:
313
index._transport.delete(index._name)
316
self._check_references()
317
index_types = ['revision', 'inventory', 'text', 'signature']
318
if self.chk_index is not None:
319
index_types.append('chk')
320
for index_type in index_types:
321
old_name = self.index_name(index_type, self.name)
322
new_name = '../indices/' + old_name
323
self.upload_transport.rename(old_name, new_name)
324
self._replace_index_with_readonly(index_type)
325
new_name = '../packs/' + self.file_name()
326
self.upload_transport.rename(self.file_name(), new_name)
327
self._state = 'finished'
329
def _get_external_refs(self, index):
330
"""Return compression parents for this index that are not present.
332
This returns any compression parents that are referenced by this index,
333
which are not contained *in* this index. They may be present elsewhere.
335
return index.external_references(1)
172
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
173
id(self), self.transport, self.name)
338
176
class NewPack(Pack):
339
177
"""An in memory proxy for a pack which is being created."""
341
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
179
# A map of index 'type' to the file extension and position in the
181
index_definitions = {
182
'revision': ('.rix', 0),
183
'inventory': ('.iix', 1),
185
'signature': ('.six', 3),
188
def __init__(self, upload_transport, index_transport, pack_transport,
342
190
"""Create a NewPack instance.
344
:param pack_collection: A PackCollection into which this is being inserted.
192
:param upload_transport: A writable transport for the pack to be
193
incrementally uploaded to.
194
:param index_transport: A writable transport for the pack's indices to
195
be written to when the pack is finished.
196
:param pack_transport: A writable transport for the pack to be renamed
197
to when the upload is complete. This *must* be the same as
198
upload_transport.clone('../packs').
345
199
:param upload_suffix: An optional suffix to be given to any temporary
346
200
files created during the pack creation. e.g '.autopack'
347
:param file_mode: Unix permissions for newly created file.
349
202
# The relative locations of the packs are constrained, but all are
350
203
# passed in because the caller has them, so as to avoid object churn.
351
index_builder_class = pack_collection._index_builder_class
352
if pack_collection.chk_index is not None:
353
chk_index = index_builder_class(reference_lists=0)
356
204
Pack.__init__(self,
357
205
# Revisions: parents list, no text compression.
358
index_builder_class(reference_lists=1),
206
InMemoryGraphIndex(reference_lists=1),
359
207
# Inventory: We want to map compression only, but currently the
360
208
# knit code hasn't been updated enough to understand that, so we
361
209
# have a regular 2-list index giving parents and compression
363
index_builder_class(reference_lists=2),
211
InMemoryGraphIndex(reference_lists=2),
364
212
# Texts: compression and per file graph, for all fileids - so two
365
213
# reference lists and two elements in the key tuple.
366
index_builder_class(reference_lists=2, key_elements=2),
214
InMemoryGraphIndex(reference_lists=2, key_elements=2),
367
215
# Signatures: Just blobs to store, no compression, no parents
369
index_builder_class(reference_lists=0),
370
# CHK based storage - just blobs, no compression or parents.
217
InMemoryGraphIndex(reference_lists=0),
373
self._pack_collection = pack_collection
374
# When we make readonly indices, we need this.
375
self.index_class = pack_collection._index_class
376
219
# where should the new pack be opened
377
self.upload_transport = pack_collection._upload_transport
220
self.upload_transport = upload_transport
378
221
# where are indices written out to
379
self.index_transport = pack_collection._index_transport
222
self.index_transport = index_transport
380
223
# where is the pack renamed to when it is finished?
381
self.pack_transport = pack_collection._pack_transport
382
# What file mode to upload the pack and indices with.
383
self._file_mode = file_mode
224
self.pack_transport = pack_transport
384
225
# tracks the content written to the .pack file.
385
226
self._hash = osutils.md5()
386
# a tuple with the length in bytes of the indices, once the pack
387
# is finalised. (rev, inv, text, sigs, chk_if_in_use)
227
# a four-tuple with the length in bytes of the indices, once the pack
228
# is finalised. (rev, inv, text, sigs)
388
229
self.index_sizes = None
389
230
# How much data to cache when writing packs. Note that this is not
390
231
# synchronised with reads, because it's not in the transport layer, so
749
531
def open_pack(self):
750
532
"""Open a pack for the pack we are creating."""
751
new_pack = self._pack_collection.pack_factory(self._pack_collection,
752
upload_suffix=self.suffix,
753
file_mode=self._pack_collection.repo.bzrdir._get_file_mode())
754
# We know that we will process all nodes in order, and don't need to
755
# query, so don't combine any indices spilled to disk until we are done
756
new_pack.revision_index.set_optimize(combine_backing_indices=False)
757
new_pack.inventory_index.set_optimize(combine_backing_indices=False)
758
new_pack.text_index.set_optimize(combine_backing_indices=False)
759
new_pack.signature_index.set_optimize(combine_backing_indices=False)
762
def _update_pack_order(self, entries, index_to_pack_map):
763
"""Determine how we want our packs to be ordered.
765
This changes the sort order of the self.packs list so that packs unused
766
by 'entries' will be at the end of the list, so that future requests
767
can avoid probing them. Used packs will be at the front of the
768
self.packs list, in the order of their first use in 'entries'.
770
:param entries: A list of (index, ...) tuples
771
:param index_to_pack_map: A mapping from index objects to pack objects.
775
for entry in entries:
777
if index not in seen_indexes:
778
packs.append(index_to_pack_map[index])
779
seen_indexes.add(index)
780
if len(packs) == len(self.packs):
781
if 'pack' in debug.debug_flags:
782
mutter('Not changing pack list, all packs used.')
784
seen_packs = set(packs)
785
for pack in self.packs:
786
if pack not in seen_packs:
533
return NewPack(self._pack_collection._upload_transport,
534
self._pack_collection._index_transport,
535
self._pack_collection._pack_transport, upload_suffix=self.suffix)
537
def _create_pack_from_packs(self):
538
self.pb.update("Opening pack", 0, 5)
539
new_pack = self.open_pack()
540
# buffer data - we won't be reading-back during the pack creation and
541
# this makes a significant difference on sftp pushes.
542
new_pack.set_write_cache_size(1024*1024)
789
543
if 'pack' in debug.debug_flags:
790
old_names = [p.access_tuple()[1] for p in self.packs]
791
new_names = [p.access_tuple()[1] for p in packs]
792
mutter('Reordering packs\nfrom: %s\n to: %s',
793
old_names, new_names)
796
def _copy_revision_texts(self):
797
"""Copy revision data to the new pack."""
544
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
545
for a_pack in self.packs]
546
if self.revision_ids is not None:
547
rev_count = len(self.revision_ids)
550
mutter('%s: create_pack: creating pack from source packs: '
551
'%s%s %s revisions wanted %s t=0',
552
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
553
plain_pack_list, rev_count)
798
554
# select revisions
799
555
if self.revision_ids:
800
556
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
802
558
revision_keys = None
803
560
# select revision keys
804
revision_index_map, revision_indices = self._pack_map_and_index_list(
806
revision_nodes = self._index_contents(revision_indices, revision_keys)
807
revision_nodes = list(revision_nodes)
808
self._update_pack_order(revision_nodes, revision_index_map)
561
revision_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
562
self.packs, 'revision_index')[0]
563
revision_nodes = self._pack_collection._index_contents(revision_index_map, revision_keys)
809
564
# copy revision keys and adjust values
810
565
self.pb.update("Copying revision texts", 1)
811
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
812
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
813
self.new_pack.revision_index, readv_group_iter, total_items))
566
list(self._copy_nodes_graph(revision_nodes, revision_index_map,
567
new_pack._writer, new_pack.revision_index))
814
568
if 'pack' in debug.debug_flags:
815
569
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
816
time.ctime(), self._pack_collection._upload_transport.base,
817
self.new_pack.random_name,
818
self.new_pack.revision_index.key_count(),
819
time.time() - self.new_pack.start_time)
820
self._revision_keys = revision_keys
822
def _copy_inventory_texts(self):
823
"""Copy the inventory texts to the new pack.
825
self._revision_keys is used to determine what inventories to copy.
827
Sets self._text_filter appropriately.
570
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
571
new_pack.revision_index.key_count(),
572
time.time() - new_pack.start_time)
829
573
# select inventory keys
830
inv_keys = self._revision_keys # currently the same keyspace, and note that
574
inv_keys = revision_keys # currently the same keyspace, and note that
831
575
# querying for keys here could introduce a bug where an inventory item
832
576
# is missed, so do not change it to query separately without cross
833
577
# checking like the text key check below.
834
inventory_index_map, inventory_indices = self._pack_map_and_index_list(
836
inv_nodes = self._index_contents(inventory_indices, inv_keys)
578
inventory_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
579
self.packs, 'inventory_index')[0]
580
inv_nodes = self._pack_collection._index_contents(inventory_index_map, inv_keys)
837
581
# copy inventory keys and adjust values
838
582
# XXX: Should be a helper function to allow different inv representation
840
584
self.pb.update("Copying inventory texts", 2)
841
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
842
# Only grab the output lines if we will be processing them
843
output_lines = bool(self.revision_ids)
844
inv_lines = self._copy_nodes_graph(inventory_index_map,
845
self.new_pack._writer, self.new_pack.inventory_index,
846
readv_group_iter, total_items, output_lines=output_lines)
585
inv_lines = self._copy_nodes_graph(inv_nodes, inventory_index_map,
586
new_pack._writer, new_pack.inventory_index, output_lines=True)
847
587
if self.revision_ids:
848
self._process_inventory_lines(inv_lines)
588
fileid_revisions = self._pack_collection.repo._find_file_ids_from_xml_inventory_lines(
589
inv_lines, self.revision_ids)
591
for fileid, file_revids in fileid_revisions.iteritems():
593
[(fileid, file_revid) for file_revid in file_revids])
850
595
# eat the iterator to cause it to execute.
852
self._text_filter = None
853
598
if 'pack' in debug.debug_flags:
854
599
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
855
time.ctime(), self._pack_collection._upload_transport.base,
856
self.new_pack.random_name,
857
self.new_pack.inventory_index.key_count(),
858
time.time() - self.new_pack.start_time)
860
def _copy_text_texts(self):
600
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
601
new_pack.inventory_index.key_count(),
602
time.time() - new_pack.start_time)
861
603
# select text keys
862
text_index_map, text_nodes = self._get_text_nodes()
863
if self._text_filter is not None:
604
text_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
605
self.packs, 'text_index')[0]
606
text_nodes = self._pack_collection._index_contents(text_index_map, text_filter)
607
if text_filter is not None:
864
608
# We could return the keys copied as part of the return value from
865
609
# _copy_nodes_graph but this doesn't work all that well with the
866
610
# need to get line output too, so we check separately, and as we're
1036
712
pb = ui.ui_factory.nested_progress_bar()
1038
for result in self._do_copy_nodes_graph(index_map, writer,
1039
write_index, output_lines, pb, readv_group_iter, total_items):
1042
# Python 2.4 does not permit try:finally: in a generator.
714
return self._do_copy_nodes_graph(nodes, index_map, writer,
715
write_index, output_lines, pb)
1048
def _do_copy_nodes_graph(self, index_map, writer, write_index,
1049
output_lines, pb, readv_group_iter, total_items):
719
def _do_copy_nodes_graph(self, nodes, index_map, writer, write_index,
1050
721
# for record verification
1051
knit = KnitVersionedFiles(None, None)
722
knit_data = _KnitData(None)
1052
723
# for line extraction when requested (inventories only)
1053
724
if output_lines:
1054
factory = KnitPlainFactory()
725
factory = knit.KnitPlainFactory()
726
# plan a readv on each source pack:
728
nodes = sorted(nodes)
729
# how to map this into knit.py - or knit.py into this?
730
# we don't want the typical knit logic, we want grouping by pack
731
# at this point - perhaps a helper library for the following code
732
# duplication points?
1055
734
record_index = 0
1056
pb.update("Copied record", record_index, total_items)
1057
for index, readv_vector, node_vector in readv_group_iter:
735
pb.update("Copied record", record_index, len(nodes))
736
for index, key, value, references in nodes:
737
if index not in request_groups:
738
request_groups[index] = []
739
request_groups[index].append((key, value, references))
740
for index, items in request_groups.iteritems():
741
pack_readv_requests = []
742
for key, value, references in items:
743
# ---- KnitGraphIndex.get_position
744
bits = value[1:].split(' ')
745
offset, length = int(bits[0]), int(bits[1])
746
pack_readv_requests.append((offset, length, (key, value[0], references)))
747
# linear scan up the pack
748
pack_readv_requests.sort()
1059
pack_obj = index_map[index]
1060
transport, path = pack_obj.access_tuple()
1062
reader = pack.make_readv_reader(transport, path, readv_vector)
1063
except errors.NoSuchFile:
1064
if self._reload_func is not None:
1067
for (names, read_func), (key, eol_flag, references) in \
1068
izip(reader.iter_records(), node_vector):
750
transport, path = index_map[index]
751
reader = pack.make_readv_reader(transport, path,
752
[offset[0:2] for offset in pack_readv_requests])
753
for (names, read_func), (_1, _2, (key, eol_flag, references)) in \
754
izip(reader.iter_records(), pack_readv_requests):
1069
755
raw_data = read_func(None)
1070
757
if output_lines:
1071
758
# read the entire thing
1072
content, _ = knit._parse_record(key[-1], raw_data)
759
content, _ = knit_data._parse_record(version_id, raw_data)
1073
760
if len(references[-1]) == 0:
1074
761
line_iterator = factory.get_fulltext_content(content)
1076
763
line_iterator = factory.get_linedelta_content(content)
1077
764
for line in line_iterator:
765
yield line, version_id
1080
767
# check the header only
1081
df, _ = knit._parse_record_header(key, raw_data)
768
df, _ = knit_data._parse_record_header(version_id, raw_data)
1083
770
pos, size = writer.add_bytes_record(raw_data, names)
1084
771
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1085
772
pb.update("Copied record", record_index)
1086
773
record_index += 1
1088
def _get_text_nodes(self):
1089
text_index_map, text_indices = self._pack_map_and_index_list(
1091
return text_index_map, self._index_contents(text_indices,
1094
def _least_readv_node_readv(self, nodes):
1095
"""Generate request groups for nodes using the least readv's.
1097
:param nodes: An iterable of graph index nodes.
1098
:return: Total node count and an iterator of the data needed to perform
1099
readvs to obtain the data for nodes. Each item yielded by the
1100
iterator is a tuple with:
1101
index, readv_vector, node_vector. readv_vector is a list ready to
1102
hand to the transport readv method, and node_vector is a list of
1103
(key, eol_flag, references) for the node retrieved by the
1104
matching readv_vector.
1106
# group by pack so we do one readv per pack
1107
nodes = sorted(nodes)
1110
for index, key, value, references in nodes:
1111
if index not in request_groups:
1112
request_groups[index] = []
1113
request_groups[index].append((key, value, references))
1115
for index, items in request_groups.iteritems():
1116
pack_readv_requests = []
1117
for key, value, references in items:
1118
# ---- KnitGraphIndex.get_position
1119
bits = value[1:].split(' ')
1120
offset, length = int(bits[0]), int(bits[1])
1121
pack_readv_requests.append(
1122
((offset, length), (key, value[0], references)))
1123
# linear scan up the pack to maximum range combining.
1124
pack_readv_requests.sort()
1125
# split out the readv and the node data.
1126
pack_readv = [readv for readv, node in pack_readv_requests]
1127
node_vector = [node for readv, node in pack_readv_requests]
1128
result.append((index, pack_readv, node_vector))
1129
return total, result
1131
def _log_copied_texts(self):
1132
if 'pack' in debug.debug_flags:
1133
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
1134
time.ctime(), self._pack_collection._upload_transport.base,
1135
self.new_pack.random_name,
1136
self.new_pack.text_index.key_count(),
1137
time.time() - self.new_pack.start_time)
1139
def _process_inventory_lines(self, inv_lines):
1140
"""Use up the inv_lines generator and setup a text key filter."""
1141
repo = self._pack_collection.repo
1142
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
1143
inv_lines, self.revision_keys)
1145
for fileid, file_revids in fileid_revisions.iteritems():
1146
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
1147
self._text_filter = text_filter
1149
def _revision_node_readv(self, revision_nodes):
1150
"""Return the total revisions and the readv's to issue.
1152
:param revision_nodes: The revision index contents for the packs being
1153
incorporated into the new pack.
1154
:return: As per _least_readv_node_readv.
1156
return self._least_readv_node_readv(revision_nodes)
1158
def _use_pack(self, new_pack):
1159
"""Return True if new_pack should be used.
1161
:param new_pack: The pack that has just been created.
1162
:return: True if the pack should be used.
1164
return new_pack.data_inserted()
1167
class OptimisingPacker(Packer):
1168
"""A packer which spends more time to create better disk layouts."""
1170
def _revision_node_readv(self, revision_nodes):
1171
"""Return the total revisions and the readv's to issue.
1173
This sort places revisions in topological order with the ancestors
1176
:param revision_nodes: The revision index contents for the packs being
1177
incorporated into the new pack.
1178
:return: As per _least_readv_node_readv.
1180
# build an ancestors dict
1183
for index, key, value, references in revision_nodes:
1184
ancestors[key] = references[0]
1185
by_key[key] = (index, value, references)
1186
order = tsort.topo_sort(ancestors)
1188
# Single IO is pathological, but it will work as a starting point.
1190
for key in reversed(order):
1191
index, value, references = by_key[key]
1192
# ---- KnitGraphIndex.get_position
1193
bits = value[1:].split(' ')
1194
offset, length = int(bits[0]), int(bits[1])
1196
(index, [(offset, length)], [(key, value[0], references)]))
1197
# TODO: combine requests in the same index that are in ascending order.
1198
return total, requests
1200
def open_pack(self):
1201
"""Open a pack for the pack we are creating."""
1202
new_pack = super(OptimisingPacker, self).open_pack()
1203
# Turn on the optimization flags for all the index builders.
1204
new_pack.revision_index.set_optimize(for_size=True)
1205
new_pack.inventory_index.set_optimize(for_size=True)
1206
new_pack.text_index.set_optimize(for_size=True)
1207
new_pack.signature_index.set_optimize(for_size=True)
1211
776
class ReconcilePacker(Packer):
1212
777
"""A packer which regenerates indices etc as it copies.
1214
779
This is used by ``bzr reconcile`` to cause parent text pointers to be
1218
def _extra_init(self):
1219
self._data_changed = False
1221
def _process_inventory_lines(self, inv_lines):
1222
"""Generate a text key reference map rather for reconciling with."""
1223
repo = self._pack_collection.repo
1224
refs = repo._find_text_key_references_from_xml_inventory_lines(
1226
self._text_refs = refs
1227
# during reconcile we:
1228
# - convert unreferenced texts to full texts
1229
# - correct texts which reference a text not copied to be full texts
1230
# - copy all others as-is but with corrected parents.
1231
# - so at this point we don't know enough to decide what becomes a full
1233
self._text_filter = None
1235
def _copy_text_texts(self):
1236
"""generate what texts we should have and then copy."""
1237
self.pb.update("Copying content texts", 3)
1238
# we have three major tasks here:
1239
# 1) generate the ideal index
1240
repo = self._pack_collection.repo
1241
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1242
_1, key, _2, refs in
1243
self.new_pack.revision_index.iter_all_entries()])
1244
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1245
# 2) generate a text_nodes list that contains all the deltas that can
1246
# be used as-is, with corrected parents.
1249
discarded_nodes = []
1250
NULL_REVISION = _mod_revision.NULL_REVISION
1251
text_index_map, text_nodes = self._get_text_nodes()
1252
for node in text_nodes:
1258
ideal_parents = tuple(ideal_index[node[1]])
1260
discarded_nodes.append(node)
1261
self._data_changed = True
1263
if ideal_parents == (NULL_REVISION,):
1265
if ideal_parents == node[3][0]:
1267
ok_nodes.append(node)
1268
elif ideal_parents[0:1] == node[3][0][0:1]:
1269
# the left most parent is the same, or there are no parents
1270
# today. Either way, we can preserve the representation as
1271
# long as we change the refs to be inserted.
1272
self._data_changed = True
1273
ok_nodes.append((node[0], node[1], node[2],
1274
(ideal_parents, node[3][1])))
1275
self._data_changed = True
1277
# Reinsert this text completely
1278
bad_texts.append((node[1], ideal_parents))
1279
self._data_changed = True
1280
# we're finished with some data.
1283
# 3) bulk copy the ok data
1284
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1285
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1286
self.new_pack.text_index, readv_group_iter, total_items))
1287
# 4) adhoc copy all the other texts.
1288
# We have to topologically insert all texts otherwise we can fail to
1289
# reconcile when parts of a single delta chain are preserved intact,
1290
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1291
# reinserted, and if d3 has incorrect parents it will also be
1292
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1293
# copied), so we will try to delta, but d2 is not currently able to be
1294
# extracted because it's basis d1 is not present. Topologically sorting
1295
# addresses this. The following generates a sort for all the texts that
1296
# are being inserted without having to reference the entire text key
1297
# space (we only topo sort the revisions, which is smaller).
1298
topo_order = tsort.topo_sort(ancestors)
1299
rev_order = dict(zip(topo_order, range(len(topo_order))))
1300
bad_texts.sort(key=lambda key:rev_order.get(key[0][1], 0))
1301
transaction = repo.get_transaction()
1302
file_id_index = GraphIndexPrefixAdapter(
1303
self.new_pack.text_index,
1305
add_nodes_callback=self.new_pack.text_index.add_nodes)
1306
data_access = _DirectPackAccess(
1307
{self.new_pack.text_index:self.new_pack.access_tuple()})
1308
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1309
self.new_pack.access_tuple())
1310
output_texts = KnitVersionedFiles(
1311
_KnitGraphIndex(self.new_pack.text_index,
1312
add_callback=self.new_pack.text_index.add_nodes,
1313
deltas=True, parents=True, is_locked=repo.is_locked),
1314
data_access=data_access, max_delta_chain=200)
1315
for key, parent_keys in bad_texts:
1316
# We refer to the new pack to delta data being output.
1317
# A possible improvement would be to catch errors on short reads
1318
# and only flush then.
1319
self.new_pack.flush()
1321
for parent_key in parent_keys:
1322
if parent_key[0] != key[0]:
1323
# Graph parents must match the fileid
1324
raise errors.BzrError('Mismatched key parent %r:%r' %
1326
parents.append(parent_key[1])
1327
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1328
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1329
output_texts.add_lines(key, parent_keys, text_lines,
1330
random_id=True, check_content=False)
1331
# 5) check that nothing inserted has a reference outside the keyspace.
1332
missing_text_keys = self.new_pack.text_index._external_references()
1333
if missing_text_keys:
1334
raise errors.BzrCheckError('Reference to missing compression parents %r'
1335
% (missing_text_keys,))
1336
self._log_copied_texts()
1338
def _use_pack(self, new_pack):
1339
"""Override _use_pack to check for reconcile having changed content."""
1340
# XXX: we might be better checking this at the copy time.
1341
original_inventory_keys = set()
1342
inv_index = self._pack_collection.inventory_index.combined_index
1343
for entry in inv_index.iter_all_entries():
1344
original_inventory_keys.add(entry[1])
1345
new_inventory_keys = set()
1346
for entry in new_pack.inventory_index.iter_all_entries():
1347
new_inventory_keys.add(entry[1])
1348
if new_inventory_keys != original_inventory_keys:
1349
self._data_changed = True
1350
return new_pack.data_inserted() and self._data_changed
1353
784
class RepositoryPackCollection(object):
1354
"""Management of packs within a repository.
1356
:ivar _names: map of {pack_name: (index_size,)}
1359
pack_factory = NewPack
1360
resumed_pack_factory = ResumedPack
785
"""Management of packs within a repository."""
1362
787
def __init__(self, repo, transport, index_transport, upload_transport,
1363
pack_transport, index_builder_class, index_class,
1365
789
"""Create a new RepositoryPackCollection.
1367
:param transport: Addresses the repository base directory
791
:param transport: Addresses the repository base directory
1368
792
(typically .bzr/repository/).
1369
793
:param index_transport: Addresses the directory containing indices.
1370
794
:param upload_transport: Addresses the directory into which packs are written
1371
795
while they're being created.
1372
796
:param pack_transport: Addresses the directory of existing complete packs.
1373
:param index_builder_class: The index builder class to use.
1374
:param index_class: The index class to use.
1375
:param use_chk_index: Whether to setup and manage a CHK index.
1377
# XXX: This should call self.reset()
1378
798
self.repo = repo
1379
799
self.transport = transport
1380
800
self._index_transport = index_transport
1381
801
self._upload_transport = upload_transport
1382
802
self._pack_transport = pack_transport
1383
self._index_builder_class = index_builder_class
1384
self._index_class = index_class
1385
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3,
803
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
1388
805
# name:Pack mapping
1390
806
self._packs_by_name = {}
1391
807
# the previous pack-names content
1392
808
self._packs_at_load = None
1393
809
# when a pack is being created by this object, the state of that pack.
1394
810
self._new_pack = None
1395
811
# aggregated revision index data
1396
flush = self._flush_new_pack
1397
self.revision_index = AggregateIndex(self.reload_pack_names, flush)
1398
self.inventory_index = AggregateIndex(self.reload_pack_names, flush)
1399
self.text_index = AggregateIndex(self.reload_pack_names, flush)
1400
self.signature_index = AggregateIndex(self.reload_pack_names, flush)
1401
all_indices = [self.revision_index, self.inventory_index,
1402
self.text_index, self.signature_index]
1404
self.chk_index = AggregateIndex(self.reload_pack_names, flush)
1405
all_indices.append(self.chk_index)
1407
# used to determine if we're using a chk_index elsewhere.
1408
self.chk_index = None
1409
# Tell all the CombinedGraphIndex objects about each other, so they can
1410
# share hints about which pack names to search first.
1411
all_combined = [agg_idx.combined_index for agg_idx in all_indices]
1412
for combined_idx in all_combined:
1413
combined_idx.set_sibling_indices(
1414
set(all_combined).difference([combined_idx]))
1416
self._resumed_packs = []
1419
return '%s(%r)' % (self.__class__.__name__, self.repo)
812
self.revision_index = AggregateIndex()
813
self.inventory_index = AggregateIndex()
814
self.text_index = AggregateIndex()
815
self.signature_index = AggregateIndex()
1421
817
def add_pack_to_memory(self, pack):
1422
818
"""Make a Pack object available to the repository to satisfy queries.
1424
820
:param pack: A Pack object.
1426
if pack.name in self._packs_by_name:
1427
raise AssertionError(
1428
'pack %s already in _packs_by_name' % (pack.name,))
822
assert pack.name not in self._packs_by_name
1429
823
self.packs.append(pack)
1430
824
self._packs_by_name[pack.name] = pack
1431
825
self.revision_index.add_index(pack.revision_index, pack)
1432
826
self.inventory_index.add_index(pack.inventory_index, pack)
1433
827
self.text_index.add_index(pack.text_index, pack)
1434
828
self.signature_index.add_index(pack.signature_index, pack)
1435
if self.chk_index is not None:
1436
self.chk_index.add_index(pack.chk_index, pack)
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]
1438
842
def all_packs(self):
1439
843
"""Return a list of all the Pack objects this repository has.
1490
890
# group their data with the relevant commit, and that may
1491
891
# involve rewriting ancient history - which autopack tries to
1492
892
# avoid. Alternatively we could not group the data but treat
1493
# each of these as having a single revision, and thus add
893
# each of these as having a single revision, and thus add
1494
894
# one revision for each to the total revision count, to get
1495
895
# a matching distribution.
1497
897
existing_packs.append((revision_count, pack))
1498
898
pack_operations = self.plan_autopack_combinations(
1499
899
existing_packs, pack_distribution)
1500
num_new_packs = len(pack_operations)
1501
num_old_packs = sum([len(po[1]) for po in pack_operations])
1502
num_revs_affected = sum([po[0] for po in pack_operations])
1503
mutter('Auto-packing repository %s, which has %d pack files, '
1504
'containing %d revisions. Packing %d files into %d affecting %d'
1505
' revisions', self, total_packs, total_revisions, num_old_packs,
1506
num_new_packs, num_revs_affected)
1507
result = self._execute_pack_operations(pack_operations,
1508
reload_func=self._restart_autopack)
1509
mutter('Auto-packing repository %s completed', self)
900
self._execute_pack_operations(pack_operations)
1512
def _execute_pack_operations(self, pack_operations, _packer_class=Packer,
903
def _execute_pack_operations(self, pack_operations):
1514
904
"""Execute a series of pack operations.
1516
906
:param pack_operations: A list of [revision_count, packs_to_combine].
1517
:param _packer_class: The class of packer to use (default: Packer).
1518
:return: The new pack names.
1520
909
for revision_count, packs in pack_operations:
1521
910
# we may have no-ops from the setup logic
1522
911
if len(packs) == 0:
1524
packer = _packer_class(self, packs, '.autopack',
1525
reload_func=reload_func)
1528
except errors.RetryWithNewPacks:
1529
# An exception is propagating out of this context, make sure
1530
# this packer has cleaned up. Packer() doesn't set its new_pack
1531
# state into the RepositoryPackCollection object, so we only
1532
# have access to it directly here.
1533
if packer.new_pack is not None:
1534
packer.new_pack.abort()
913
Packer(self, packs, '.autopack').pack()
1536
914
for pack in packs:
1537
915
self._remove_pack_from_memory(pack)
1538
916
# record the newly available packs and stop advertising the old
1540
to_be_obsoleted = []
1541
for _, packs in pack_operations:
1542
to_be_obsoleted.extend(packs)
1543
result = self._save_pack_names(clear_obsolete_packs=True,
1544
obsolete_packs=to_be_obsoleted)
1547
def _flush_new_pack(self):
1548
if self._new_pack is not None:
1549
self._new_pack.flush()
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)
1551
923
def lock_names(self):
1552
924
"""Acquire the mutex around the pack-names index.
1554
926
This cannot be used in the middle of a read-only transaction on the
1557
929
self.repo.control_files.lock_write()
1559
def _already_packed(self):
1560
"""Is the collection already packed?"""
1561
return not (self.repo._format.pack_compresses or (len(self._names) > 1))
1563
def pack(self, hint=None, clean_obsolete_packs=False):
1564
932
"""Pack the pack collection totally."""
1565
933
self.ensure_loaded()
1566
934
total_packs = len(self._names)
1567
if self._already_packed():
1569
937
total_revisions = self.revision_index.combined_index.key_count()
1570
938
# XXX: the following may want to be a class, to pack with a given
1572
940
mutter('Packing repository %s, which has %d pack files, '
1573
'containing %d revisions with hint %r.', self, total_packs,
1574
total_revisions, hint)
941
'containing %d revisions into 1 packs.', self, total_packs,
1575
943
# determine which packs need changing
944
pack_distribution = [1]
1576
945
pack_operations = [[0, []]]
1577
946
for pack in self.all_packs():
1578
if hint is None or pack.name in hint:
1579
# Either no hint was provided (so we are packing everything),
1580
# or this pack was included in the hint.
1581
pack_operations[-1][0] += pack.get_revision_count()
1582
pack_operations[-1][1].append(pack)
1583
self._execute_pack_operations(pack_operations, OptimisingPacker)
1585
if clean_obsolete_packs:
1586
self._clear_obsolete_packs()
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)
1588
952
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1589
953
"""Plan a pack operation.
1873
1157
self._packs_by_name = {}
1874
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)
1876
1211
def _unlock_names(self):
1877
1212
"""Release the mutex around the pack-names index."""
1878
1213
self.repo.control_files.unlock()
1880
def _diff_pack_names(self):
1881
"""Read the pack names from disk, and compare it to the one in memory.
1883
:return: (disk_nodes, deleted_nodes, new_nodes)
1884
disk_nodes The final set of nodes that should be referenced
1885
deleted_nodes Nodes which have been removed from when we started
1886
new_nodes Nodes that are newly introduced
1888
# load the disk nodes across
1890
for index, key, value in self._iter_disk_pack_index():
1891
disk_nodes.add((key, value))
1892
orig_disk_nodes = set(disk_nodes)
1894
# do a two-way diff against our original content
1895
current_nodes = set()
1896
for name, sizes in self._names.iteritems():
1898
((name, ), ' '.join(str(size) for size in sizes)))
1900
# Packs no longer present in the repository, which were present when we
1901
# locked the repository
1902
deleted_nodes = self._packs_at_load - current_nodes
1903
# Packs which this process is adding
1904
new_nodes = current_nodes - self._packs_at_load
1906
# Update the disk_nodes set to include the ones we are adding, and
1907
# remove the ones which were removed by someone else
1908
disk_nodes.difference_update(deleted_nodes)
1909
disk_nodes.update(new_nodes)
1911
return disk_nodes, deleted_nodes, new_nodes, orig_disk_nodes
1913
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1914
"""Given the correct set of pack files, update our saved info.
1916
:return: (removed, added, modified)
1917
removed pack names removed from self._names
1918
added pack names added to self._names
1919
modified pack names that had changed value
1924
## self._packs_at_load = disk_nodes
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:
1925
1257
new_names = dict(disk_nodes)
1926
1258
# drop no longer present nodes
1927
1259
for pack in self.all_packs():
1928
1260
if (pack.name,) not in new_names:
1929
removed.append(pack.name)
1930
1261
self._remove_pack_from_memory(pack)
1931
1262
# add new nodes/refresh existing ones
1932
1263
for key, value in disk_nodes:
1946
1277
self._remove_pack_from_memory(self.get_pack_by_name(name))
1947
1278
self._names[name] = sizes
1948
1279
self.get_pack_by_name(name)
1949
modified.append(name)
1952
1282
self._names[name] = sizes
1953
1283
self.get_pack_by_name(name)
1955
return removed, added, modified
1957
def _save_pack_names(self, clear_obsolete_packs=False, obsolete_packs=None):
1958
"""Save the list of packs.
1960
This will take out the mutex around the pack names list for the
1961
duration of the method call. If concurrent updates have been made, a
1962
three-way merge between the current list and the current in memory list
1965
:param clear_obsolete_packs: If True, clear out the contents of the
1966
obsolete_packs directory.
1967
:param obsolete_packs: Packs that are obsolete once the new pack-names
1968
file has been written.
1969
:return: A list of the names saved that were not previously on disk.
1971
already_obsolete = []
1974
builder = self._index_builder_class()
1975
(disk_nodes, deleted_nodes, new_nodes,
1976
orig_disk_nodes) = self._diff_pack_names()
1977
# TODO: handle same-name, index-size-changes here -
1978
# e.g. use the value from disk, not ours, *unless* we're the one
1980
for key, value in disk_nodes:
1981
builder.add_node(key, value)
1982
self.transport.put_file('pack-names', builder.finish(),
1983
mode=self.repo.bzrdir._get_file_mode())
1984
self._packs_at_load = disk_nodes
1985
if clear_obsolete_packs:
1988
to_preserve = set([o.name for o in obsolete_packs])
1989
already_obsolete = self._clear_obsolete_packs(to_preserve)
1991
self._unlock_names()
1992
# synchronise the memory packs list with what we just wrote:
1993
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1995
# TODO: We could add one more condition here. "if o.name not in
1996
# orig_disk_nodes and o != the new_pack we haven't written to
1997
# disk yet. However, the new pack object is not easily
1998
# accessible here (it would have to be passed through the
1999
# autopacking code, etc.)
2000
obsolete_packs = [o for o in obsolete_packs
2001
if o.name not in already_obsolete]
2002
self._obsolete_packs(obsolete_packs)
2003
return [new_node[0][0] for new_node in new_nodes]
2005
def reload_pack_names(self):
2006
"""Sync our pack listing with what is present in the repository.
2008
This should be called when we find out that something we thought was
2009
present is now missing. This happens when another process re-packs the
2012
:return: True if the in-memory list of packs has been altered at all.
2014
# The ensure_loaded call is to handle the case where the first call
2015
# made involving the collection was to reload_pack_names, where we
2016
# don't have a view of disk contents. Its a bit of a bandaid, and
2017
# causes two reads of pack-names, but its a rare corner case not struck
2018
# with regular push/pull etc.
2019
first_read = self.ensure_loaded()
2022
# out the new value.
2023
(disk_nodes, deleted_nodes, new_nodes,
2024
orig_disk_nodes) = self._diff_pack_names()
2025
# _packs_at_load is meant to be the explicit list of names in
2026
# 'pack-names' at then start. As such, it should not contain any
2027
# pending names that haven't been written out yet.
2028
self._packs_at_load = orig_disk_nodes
2030
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
2031
if removed or added or modified:
2035
def _restart_autopack(self):
2036
"""Reload the pack names list, and restart the autopack code."""
2037
if not self.reload_pack_names():
2038
# Re-raise the original exception, because something went missing
2039
# and a restart didn't find it
2041
raise errors.RetryAutopack(self.repo, False, sys.exc_info())
2043
def _clear_obsolete_packs(self, preserve=None):
2044
"""Delete everything from the obsolete-packs directory.
2046
:return: A list of pack identifiers (the filename without '.pack') that
2047
were found in obsolete_packs.
2050
obsolete_pack_transport = self.transport.clone('obsolete_packs')
2051
if preserve is None:
2053
for filename in obsolete_pack_transport.list_dir('.'):
2054
name, ext = osutils.splitext(filename)
2057
if name in preserve:
2060
obsolete_pack_transport.delete(filename)
2061
except (errors.PathError, errors.TransportError), e:
2062
warning("couldn't delete obsolete pack, skipping it:\n%s"
2066
1285
def _start_write_group(self):
2067
1286
# Do not permit preparation for writing if we're not in a 'write lock'.
2068
1287
if not self.repo.is_write_locked():
2069
1288
raise errors.NotWriteLocked(self)
2070
self._new_pack = self.pack_factory(self, upload_suffix='.pack',
2071
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')
2072
1291
# allow writing: queue writes to a new index
2073
1292
self.revision_index.add_writable_index(self._new_pack.revision_index,
2074
1293
self._new_pack)
2076
1295
self._new_pack)
2077
1296
self.text_index.add_writable_index(self._new_pack.text_index,
2078
1297
self._new_pack)
2079
self._new_pack.text_index.set_optimize(combine_backing_indices=False)
2080
1298
self.signature_index.add_writable_index(self._new_pack.signature_index,
2081
1299
self._new_pack)
2082
if self.chk_index is not None:
2083
self.chk_index.add_writable_index(self._new_pack.chk_index,
2085
self.repo.chk_bytes._index._add_callback = self.chk_index.add_callback
2086
self._new_pack.chk_index.set_optimize(combine_backing_indices=False)
2088
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
2089
self.repo.revisions._index._add_callback = self.revision_index.add_callback
2090
self.repo.signatures._index._add_callback = self.signature_index.add_callback
2091
self.repo.texts._index._add_callback = self.text_index.add_callback
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(
2093
1315
def _abort_write_group(self):
2094
1316
# FIXME: just drop the transient index.
2095
1317
# forget what names there are
2096
if self._new_pack is not None:
2097
operation = cleanup.OperationWithCleanups(self._new_pack.abort)
2098
operation.add_cleanup(setattr, self, '_new_pack', None)
2099
# If we aborted while in the middle of finishing the write
2100
# group, _remove_pack_indices could fail because the indexes are
2101
# already gone. But they're not there we shouldn't fail in this
2102
# case, so we pass ignore_missing=True.
2103
operation.add_cleanup(self._remove_pack_indices, self._new_pack,
2104
ignore_missing=True)
2105
operation.run_simple()
2106
for resumed_pack in self._resumed_packs:
2107
operation = cleanup.OperationWithCleanups(resumed_pack.abort)
2108
# See comment in previous finally block.
2109
operation.add_cleanup(self._remove_pack_indices, resumed_pack,
2110
ignore_missing=True)
2111
operation.run_simple()
2112
del self._resumed_packs[:]
2114
def _remove_resumed_pack_indices(self):
2115
for resumed_pack in self._resumed_packs:
2116
self._remove_pack_indices(resumed_pack)
2117
del self._resumed_packs[:]
2119
def _check_new_inventories(self):
2120
"""Detect missing inventories in this write group.
2122
:returns: list of strs, summarising any problems found. If the list is
2123
empty no problems were found.
2125
# The base implementation does no checks. GCRepositoryPackCollection
1318
self._new_pack.abort()
1319
self._remove_pack_indices(self._new_pack)
1320
self._new_pack = None
1321
self.repo._text_knit = None
2129
1323
def _commit_write_group(self):
2131
for prefix, versioned_file in (
2132
('revisions', self.repo.revisions),
2133
('inventories', self.repo.inventories),
2134
('texts', self.repo.texts),
2135
('signatures', self.repo.signatures),
2137
missing = versioned_file.get_missing_compression_parent_keys()
2138
all_missing.update([(prefix,) + key for key in missing])
2140
raise errors.BzrCheckError(
2141
"Repository %s has missing compression parent(s) %r "
2142
% (self.repo, sorted(all_missing)))
2143
problems = self._check_new_inventories()
2145
problems_summary = '\n'.join(problems)
2146
raise errors.BzrCheckError(
2147
"Cannot add revision(s) to repository: " + problems_summary)
2148
1324
self._remove_pack_indices(self._new_pack)
2149
any_new_content = False
2150
1325
if self._new_pack.data_inserted():
2151
1326
# get all the data to disk and read to use
2152
1327
self._new_pack.finish()
2153
1328
self.allocate(self._new_pack)
2154
1329
self._new_pack = None
2155
any_new_content = True
2157
self._new_pack.abort()
2158
self._new_pack = None
2159
for resumed_pack in self._resumed_packs:
2160
# XXX: this is a pretty ugly way to turn the resumed pack into a
2161
# properly committed pack.
2162
self._names[resumed_pack.name] = None
2163
self._remove_pack_from_memory(resumed_pack)
2164
resumed_pack.finish()
2165
self.allocate(resumed_pack)
2166
any_new_content = True
2167
del self._resumed_packs[:]
2169
result = self.autopack()
1330
if not self.autopack():
2171
1331
# when autopack takes no steps, the names list is still
2173
return self._save_pack_names()
2177
def _suspend_write_group(self):
2178
tokens = [pack.name for pack in self._resumed_packs]
2179
self._remove_pack_indices(self._new_pack)
2180
if self._new_pack.data_inserted():
2181
# get all the data to disk and read to use
2182
self._new_pack.finish(suspend=True)
2183
tokens.append(self._new_pack.name)
2184
self._new_pack = None
1333
self._save_pack_names()
2186
1335
self._new_pack.abort()
2187
1336
self._new_pack = None
2188
self._remove_resumed_pack_indices()
2191
def _resume_write_group(self, tokens):
2192
for token in tokens:
2193
self._resume_pack(token)
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)
2196
1485
class KnitPackRepository(KnitRepository):
2197
"""Repository with knit objects stored inside pack containers.
2199
The layering for a KnitPackRepository is:
2201
Graph | HPSS | Repository public layer |
2202
===================================================
2203
Tuple based apis below, string based, and key based apis above
2204
---------------------------------------------------
2206
Provides .texts, .revisions etc
2207
This adapts the N-tuple keys to physical knit records which only have a
2208
single string identifier (for historical reasons), which in older formats
2209
was always the revision_id, and in the mapped code for packs is always
2210
the last element of key tuples.
2211
---------------------------------------------------
2213
A separate GraphIndex is used for each of the
2214
texts/inventories/revisions/signatures contained within each individual
2215
pack file. The GraphIndex layer works in N-tuples and is unaware of any
2217
===================================================
2221
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
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):
2223
1490
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2224
_commit_builder_class, _serializer)
2225
index_transport = self._transport.clone('indices')
2226
self._pack_collection = RepositoryPackCollection(self, self._transport,
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,
2227
1495
index_transport,
2228
self._transport.clone('upload'),
2229
self._transport.clone('packs'),
2230
_format.index_builder_class,
2231
_format.index_class,
2232
use_chk_index=self._format.supports_chks,
2234
self.inventories = KnitVersionedFiles(
2235
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2236
add_callback=self._pack_collection.inventory_index.add_callback,
2237
deltas=True, parents=True, is_locked=self.is_locked),
2238
data_access=self._pack_collection.inventory_index.data_access,
2239
max_delta_chain=200)
2240
self.revisions = KnitVersionedFiles(
2241
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2242
add_callback=self._pack_collection.revision_index.add_callback,
2243
deltas=False, parents=True, is_locked=self.is_locked,
2244
track_external_parent_refs=True),
2245
data_access=self._pack_collection.revision_index.data_access,
2247
self.signatures = KnitVersionedFiles(
2248
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2249
add_callback=self._pack_collection.signature_index.add_callback,
2250
deltas=False, parents=False, is_locked=self.is_locked),
2251
data_access=self._pack_collection.signature_index.data_access,
2253
self.texts = KnitVersionedFiles(
2254
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2255
add_callback=self._pack_collection.text_index.add_callback,
2256
deltas=True, parents=True, is_locked=self.is_locked),
2257
data_access=self._pack_collection.text_index.data_access,
2258
max_delta_chain=200)
2259
if _format.supports_chks:
2260
# No graph, no compression:- references from chks are between
2261
# different objects not temporal versions of the same; and without
2262
# some sort of temporal structure knit compression will just fail.
2263
self.chk_bytes = KnitVersionedFiles(
2264
_KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2265
add_callback=self._pack_collection.chk_index.add_callback,
2266
deltas=False, parents=False, is_locked=self.is_locked),
2267
data_access=self._pack_collection.chk_index.data_access,
2270
self.chk_bytes = None
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)
2271
1501
# True when the repository object is 'write locked' (as opposed to the
2272
# physical lock only taken out around changes to the pack-names list.)
1502
# physical lock only taken out around changes to the pack-names list.)
2273
1503
# Another way to represent this would be a decorator around the control
2274
1504
# files object that presents logical locks as physical ones - if this
2275
1505
# gets ugly consider that alternative design. RBC 20071011
2277
1507
self._transaction = None
2279
1509
self._reconcile_does_inventory_gc = True
2280
self._reconcile_fixes_text_parents = True
1510
self._reconcile_fixes_text_parents = False
2281
1511
self._reconcile_backsup_inventory = False
2283
def _warn_if_deprecated(self, branch=None):
2284
# This class isn't deprecated, but one sub-format is
2285
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2286
super(KnitPackRepository, self)._warn_if_deprecated(branch)
2288
1513
def _abort_write_group(self):
2289
self.revisions._index._key_dependencies.clear()
2290
1514
self._pack_collection._abort_write_group()
2292
def _get_source(self, to_format):
2293
if to_format.network_name() == self._format.network_name():
2294
return KnitPackStreamSource(self, to_format)
2295
return super(KnitPackRepository, self)._get_source(to_format)
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])
2297
1592
def _make_parents_provider(self):
2298
return graph.CachingParentsProvider(self)
2300
1595
def _refresh_data(self):
2301
if not self.is_locked():
2303
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()
2305
1605
def _start_write_group(self):
2306
1606
self._pack_collection._start_write_group()
2308
1608
def _commit_write_group(self):
2309
hint = self._pack_collection._commit_write_group()
2310
self.revisions._index._key_dependencies.clear()
2313
def suspend_write_group(self):
2314
# XXX check self._write_group is self.get_transaction()?
2315
tokens = self._pack_collection._suspend_write_group()
2316
self.revisions._index._key_dependencies.clear()
2317
self._write_group = None
2320
def _resume_write_group(self, tokens):
2321
self._start_write_group()
2323
self._pack_collection._resume_write_group(tokens)
2324
except errors.UnresumableWriteGroup:
2325
self._abort_write_group()
2327
for pack in self._pack_collection._resumed_packs:
2328
self.revisions._index.scan_unvalidated_index(pack.revision_index)
1609
return self._pack_collection._commit_write_group()
1611
def get_inventory_weave(self):
1612
return self._inv_thunk.get_weave()
2330
1614
def get_transaction(self):
2331
1615
if self._write_lock_count:
2642
1862
def get_format_description(self):
2643
1863
"""See RepositoryFormat.get_format_description()."""
2644
1864
return "Packs containing knits with subtree support\n"
2647
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2648
"""A rich-root, no subtrees parameterized Pack repository.
2650
This repository format uses the xml6 serializer to get:
2651
- support for recording full info about the tree root
2653
This format was introduced in 1.0.
2656
repository_class = KnitPackRepository
2657
_commit_builder_class = PackRootCommitBuilder
2658
rich_root_data = True
2659
supports_tree_reference = False
2661
def _serializer(self):
2662
return xml6.serializer_v6
2663
# What index classes to use
2664
index_builder_class = InMemoryGraphIndex
2665
index_class = GraphIndex
2667
def _get_matching_bzrdir(self):
2668
return bzrdir.format_registry.make_bzrdir(
2671
def _ignore_setting_bzrdir(self, format):
2674
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2676
def get_format_string(self):
2677
"""See RepositoryFormat.get_format_string()."""
2678
return ("Bazaar pack repository format 1 with rich root"
2679
" (needs bzr 1.0)\n")
2681
def get_format_description(self):
2682
"""See RepositoryFormat.get_format_description()."""
2683
return "Packs containing knits with rich root support\n"
2686
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2687
"""Repository that supports external references to allow stacking.
2691
Supports external lookups, which results in non-truncated ghosts after
2692
reconcile compared to pack-0.92 formats.
2695
repository_class = KnitPackRepository
2696
_commit_builder_class = PackCommitBuilder
2697
supports_external_lookups = True
2698
# What index classes to use
2699
index_builder_class = InMemoryGraphIndex
2700
index_class = GraphIndex
2703
def _serializer(self):
2704
return xml5.serializer_v5
2706
def _get_matching_bzrdir(self):
2707
return bzrdir.format_registry.make_bzrdir('1.6')
2709
def _ignore_setting_bzrdir(self, format):
2712
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2714
def get_format_string(self):
2715
"""See RepositoryFormat.get_format_string()."""
2716
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2718
def get_format_description(self):
2719
"""See RepositoryFormat.get_format_description()."""
2720
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2723
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2724
"""A repository with rich roots and stacking.
2726
New in release 1.6.1.
2728
Supports stacking on other repositories, allowing data to be accessed
2729
without being stored locally.
2732
repository_class = KnitPackRepository
2733
_commit_builder_class = PackRootCommitBuilder
2734
rich_root_data = True
2735
supports_tree_reference = False # no subtrees
2736
supports_external_lookups = True
2737
# What index classes to use
2738
index_builder_class = InMemoryGraphIndex
2739
index_class = GraphIndex
2742
def _serializer(self):
2743
return xml6.serializer_v6
2745
def _get_matching_bzrdir(self):
2746
return bzrdir.format_registry.make_bzrdir(
2749
def _ignore_setting_bzrdir(self, format):
2752
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2754
def get_format_string(self):
2755
"""See RepositoryFormat.get_format_string()."""
2756
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2758
def get_format_description(self):
2759
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2762
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2763
"""A repository with rich roots and external references.
2767
Supports external lookups, which results in non-truncated ghosts after
2768
reconcile compared to pack-0.92 formats.
2770
This format was deprecated because the serializer it uses accidentally
2771
supported subtrees, when the format was not intended to. This meant that
2772
someone could accidentally fetch from an incorrect repository.
2775
repository_class = KnitPackRepository
2776
_commit_builder_class = PackRootCommitBuilder
2777
rich_root_data = True
2778
supports_tree_reference = False # no subtrees
2780
supports_external_lookups = True
2781
# What index classes to use
2782
index_builder_class = InMemoryGraphIndex
2783
index_class = GraphIndex
2786
def _serializer(self):
2787
return xml7.serializer_v7
2789
def _get_matching_bzrdir(self):
2790
matching = bzrdir.format_registry.make_bzrdir(
2792
matching.repository_format = self
2795
def _ignore_setting_bzrdir(self, format):
2798
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2800
def get_format_string(self):
2801
"""See RepositoryFormat.get_format_string()."""
2802
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2804
def get_format_description(self):
2805
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2809
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2810
"""A repository with stacking and btree indexes,
2811
without rich roots or subtrees.
2813
This is equivalent to pack-1.6 with B+Tree indices.
2816
repository_class = KnitPackRepository
2817
_commit_builder_class = PackCommitBuilder
2818
supports_external_lookups = True
2819
# What index classes to use
2820
index_builder_class = BTreeBuilder
2821
index_class = BTreeGraphIndex
2824
def _serializer(self):
2825
return xml5.serializer_v5
2827
def _get_matching_bzrdir(self):
2828
return bzrdir.format_registry.make_bzrdir('1.9')
2830
def _ignore_setting_bzrdir(self, format):
2833
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2835
def get_format_string(self):
2836
"""See RepositoryFormat.get_format_string()."""
2837
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2839
def get_format_description(self):
2840
"""See RepositoryFormat.get_format_description()."""
2841
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2844
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2845
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2847
1.6-rich-root with B+Tree indices.
2850
repository_class = KnitPackRepository
2851
_commit_builder_class = PackRootCommitBuilder
2852
rich_root_data = True
2853
supports_tree_reference = False # no subtrees
2854
supports_external_lookups = True
2855
# What index classes to use
2856
index_builder_class = BTreeBuilder
2857
index_class = BTreeGraphIndex
2860
def _serializer(self):
2861
return xml6.serializer_v6
2863
def _get_matching_bzrdir(self):
2864
return bzrdir.format_registry.make_bzrdir(
2867
def _ignore_setting_bzrdir(self, format):
2870
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2872
def get_format_string(self):
2873
"""See RepositoryFormat.get_format_string()."""
2874
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2876
def get_format_description(self):
2877
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2880
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2881
"""A subtrees development repository.
2883
This format should be retained until the second release after bzr 1.7.
2885
1.6.1-subtree[as it might have been] with B+Tree indices.
2887
This is [now] retained until we have a CHK based subtree format in
2891
repository_class = KnitPackRepository
2892
_commit_builder_class = PackRootCommitBuilder
2893
rich_root_data = True
2895
supports_tree_reference = True
2896
supports_external_lookups = True
2897
# What index classes to use
2898
index_builder_class = BTreeBuilder
2899
index_class = BTreeGraphIndex
2902
def _serializer(self):
2903
return xml7.serializer_v7
2905
def _get_matching_bzrdir(self):
2906
return bzrdir.format_registry.make_bzrdir(
2907
'development-subtree')
2909
def _ignore_setting_bzrdir(self, format):
2912
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2914
def get_format_string(self):
2915
"""See RepositoryFormat.get_format_string()."""
2916
return ("Bazaar development format 2 with subtree support "
2917
"(needs bzr.dev from before 1.8)\n")
2919
def get_format_description(self):
2920
"""See RepositoryFormat.get_format_description()."""
2921
return ("Development repository format, currently the same as "
2922
"1.6.1-subtree with B+Tree indices.\n")
added
removed
bzrlib/repofmt/pack_repo.py
