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
17
from __future__ import absolute_import
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22
17
from bzrlib.lazy_import import lazy_import
23
18
lazy_import(globals(), """
24
19
from itertools import izip
27
24
from bzrlib import (
39
30
from bzrlib.index import (
40
34
CombinedGraphIndex,
41
35
GraphIndexPrefixAdapter,
37
from bzrlib.knit import KnitGraphIndex, _PackAccess, _KnitData
38
from bzrlib.osutils import rand_chars
39
from bzrlib.pack import ContainerWriter
40
from bzrlib.store import revision
41
from bzrlib import tsort
44
43
from bzrlib import (
51
from bzrlib.decorators import (
56
from bzrlib.lock import LogicalLockResult
58
from bzrlib.decorators import needs_read_lock, needs_write_lock
59
from bzrlib.repofmt.knitrepo import KnitRepository
57
60
from bzrlib.repository import (
60
RepositoryFormatMetaDir,
61
RepositoryWriteLockResult,
63
from bzrlib.vf_repository import (
64
MetaDirVersionedFileRepository,
65
MetaDirVersionedFileRepositoryFormat,
66
VersionedFileCommitBuilder,
67
VersionedFileRootCommitBuilder,
69
from bzrlib.trace import (
76
class PackCommitBuilder(VersionedFileCommitBuilder):
77
"""Subclass of VersionedFileCommitBuilder to add texts with pack semantics.
63
MetaDirRepositoryFormat,
66
import bzrlib.revision as _mod_revision
67
from bzrlib.store.revision.knit import KnitRevisionStore
68
from bzrlib.store.versioned import VersionedFileStore
69
from bzrlib.trace import mutter, note, warning
72
class PackCommitBuilder(CommitBuilder):
73
"""A subclass of CommitBuilder to add texts with pack semantics.
79
75
Specifically this uses one knit object rather than one knit object per
80
76
added text, reducing memory and object pressure.
83
79
def __init__(self, repository, parents, config, timestamp=None,
84
80
timezone=None, committer=None, revprops=None,
85
revision_id=None, lossy=False):
86
VersionedFileCommitBuilder.__init__(self, repository, parents, config,
82
CommitBuilder.__init__(self, repository, parents, config,
87
83
timestamp=timestamp, timezone=timezone, committer=committer,
88
revprops=revprops, revision_id=revision_id, lossy=lossy)
84
revprops=revprops, revision_id=revision_id)
89
85
self._file_graph = graph.Graph(
90
86
repository._pack_collection.text_index.combined_index)
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,
92
93
def _heads(self, file_id, revision_ids):
93
94
keys = [(file_id, revision_id) for revision_id in revision_ids]
94
95
return set([key[1] for key in self._file_graph.heads(keys)])
97
class PackRootCommitBuilder(VersionedFileRootCommitBuilder):
98
class PackRootCommitBuilder(RootCommitBuilder):
98
99
"""A subclass of RootCommitBuilder to add texts with pack semantics.
100
101
Specifically this uses one knit object rather than one knit object per
101
102
added text, reducing memory and object pressure.
104
105
def __init__(self, repository, parents, config, timestamp=None,
105
106
timezone=None, committer=None, revprops=None,
106
revision_id=None, lossy=False):
107
super(PackRootCommitBuilder, self).__init__(repository, parents,
108
config, timestamp=timestamp, timezone=timezone,
109
committer=committer, revprops=revprops, revision_id=revision_id,
108
CommitBuilder.__init__(self, repository, parents, config,
109
timestamp=timestamp, timezone=timezone, committer=committer,
110
revprops=revprops, revision_id=revision_id)
111
111
self._file_graph = graph.Graph(
112
112
repository._pack_collection.text_index.combined_index)
114
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
115
return self.repository._pack_collection._add_text_to_weave(file_id,
116
self._new_revision_id, new_lines, parents, nostore_sha,
114
119
def _heads(self, file_id, revision_ids):
115
120
keys = [(file_id, revision_id) for revision_id in revision_ids]
116
121
return set([key[1] for key in self._file_graph.heads(keys)])
258
209
return not self.__eq__(other)
260
211
def __repr__(self):
261
return "<%s.%s object at 0x%x, %s, %s" % (
262
self.__class__.__module__, self.__class__.__name__, id(self),
263
self.pack_transport, self.name)
266
class ResumedPack(ExistingPack):
268
def __init__(self, name, revision_index, inventory_index, text_index,
269
signature_index, upload_transport, pack_transport, index_transport,
270
pack_collection, chk_index=None):
271
"""Create a ResumedPack object."""
272
ExistingPack.__init__(self, pack_transport, name, revision_index,
273
inventory_index, text_index, signature_index,
275
self.upload_transport = upload_transport
276
self.index_transport = index_transport
277
self.index_sizes = [None, None, None, None]
279
('revision', revision_index),
280
('inventory', inventory_index),
281
('text', text_index),
282
('signature', signature_index),
284
if chk_index is not None:
285
indices.append(('chk', chk_index))
286
self.index_sizes.append(None)
287
for index_type, index in indices:
288
offset = self.index_offset(index_type)
289
self.index_sizes[offset] = index._size
290
self.index_class = pack_collection._index_class
291
self._pack_collection = pack_collection
292
self._state = 'resumed'
293
# XXX: perhaps check that the .pack file exists?
295
def access_tuple(self):
296
if self._state == 'finished':
297
return Pack.access_tuple(self)
298
elif self._state == 'resumed':
299
return self.upload_transport, self.file_name()
301
raise AssertionError(self._state)
304
self.upload_transport.delete(self.file_name())
305
indices = [self.revision_index, self.inventory_index, self.text_index,
306
self.signature_index]
307
if self.chk_index is not None:
308
indices.append(self.chk_index)
309
for index in indices:
310
index._transport.delete(index._name)
313
self._check_references()
314
index_types = ['revision', 'inventory', 'text', 'signature']
315
if self.chk_index is not None:
316
index_types.append('chk')
317
for index_type in index_types:
318
old_name = self.index_name(index_type, self.name)
319
new_name = '../indices/' + old_name
320
self.upload_transport.move(old_name, new_name)
321
self._replace_index_with_readonly(index_type)
322
new_name = '../packs/' + self.file_name()
323
self.upload_transport.move(self.file_name(), new_name)
324
self._state = 'finished'
326
def _get_external_refs(self, index):
327
"""Return compression parents for this index that are not present.
329
This returns any compression parents that are referenced by this index,
330
which are not contained *in* this index. They may be present elsewhere.
332
return index.external_references(1)
212
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
213
id(self), self.transport, self.name)
335
216
class NewPack(Pack):
336
217
"""An in memory proxy for a pack which is being created."""
338
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
219
# A map of index 'type' to the file extension and position in the
221
index_definitions = {
222
'revision': ('.rix', 0),
223
'inventory': ('.iix', 1),
225
'signature': ('.six', 3),
228
def __init__(self, upload_transport, index_transport, pack_transport,
229
upload_suffix='', file_mode=None):
339
230
"""Create a NewPack instance.
341
:param pack_collection: A PackCollection into which this is being inserted.
232
:param upload_transport: A writable transport for the pack to be
233
incrementally uploaded to.
234
:param index_transport: A writable transport for the pack's indices to
235
be written to when the pack is finished.
236
:param pack_transport: A writable transport for the pack to be renamed
237
to when the upload is complete. This *must* be the same as
238
upload_transport.clone('../packs').
342
239
:param upload_suffix: An optional suffix to be given to any temporary
343
240
files created during the pack creation. e.g '.autopack'
344
:param file_mode: Unix permissions for newly created file.
241
:param file_mode: An optional file mode to create the new files with.
346
243
# The relative locations of the packs are constrained, but all are
347
244
# passed in because the caller has them, so as to avoid object churn.
348
index_builder_class = pack_collection._index_builder_class
349
if pack_collection.chk_index is not None:
350
chk_index = index_builder_class(reference_lists=0)
353
245
Pack.__init__(self,
354
246
# Revisions: parents list, no text compression.
355
index_builder_class(reference_lists=1),
247
InMemoryGraphIndex(reference_lists=1),
356
248
# Inventory: We want to map compression only, but currently the
357
249
# knit code hasn't been updated enough to understand that, so we
358
250
# have a regular 2-list index giving parents and compression
360
index_builder_class(reference_lists=2),
252
InMemoryGraphIndex(reference_lists=2),
361
253
# Texts: compression and per file graph, for all fileids - so two
362
254
# reference lists and two elements in the key tuple.
363
index_builder_class(reference_lists=2, key_elements=2),
255
InMemoryGraphIndex(reference_lists=2, key_elements=2),
364
256
# Signatures: Just blobs to store, no compression, no parents
366
index_builder_class(reference_lists=0),
367
# CHK based storage - just blobs, no compression or parents.
258
InMemoryGraphIndex(reference_lists=0),
370
self._pack_collection = pack_collection
371
# When we make readonly indices, we need this.
372
self.index_class = pack_collection._index_class
373
260
# where should the new pack be opened
374
self.upload_transport = pack_collection._upload_transport
261
self.upload_transport = upload_transport
375
262
# where are indices written out to
376
self.index_transport = pack_collection._index_transport
263
self.index_transport = index_transport
377
264
# where is the pack renamed to when it is finished?
378
self.pack_transport = pack_collection._pack_transport
265
self.pack_transport = pack_transport
379
266
# What file mode to upload the pack and indices with.
380
267
self._file_mode = file_mode
381
268
# tracks the content written to the .pack file.
382
self._hash = osutils.md5()
383
# a tuple with the length in bytes of the indices, once the pack
384
# is finalised. (rev, inv, text, sigs, chk_if_in_use)
269
self._hash = md5.new()
270
# a four-tuple with the length in bytes of the indices, once the pack
271
# is finalised. (rev, inv, text, sigs)
385
272
self.index_sizes = None
386
273
# How much data to cache when writing packs. Note that this is not
387
274
# synchronised with reads, because it's not in the transport layer, so
743
633
Sets self._text_filter appropriately.
745
raise NotImplementedError(self._copy_inventory_texts)
635
# select inventory keys
636
inv_keys = self._revision_keys # currently the same keyspace, and note that
637
# querying for keys here could introduce a bug where an inventory item
638
# is missed, so do not change it to query separately without cross
639
# checking like the text key check below.
640
inventory_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
641
self.packs, 'inventory_index')[0]
642
inv_nodes = self._pack_collection._index_contents(inventory_index_map, inv_keys)
643
# copy inventory keys and adjust values
644
# XXX: Should be a helper function to allow different inv representation
646
self.pb.update("Copying inventory texts", 2)
647
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
648
# Only grab the output lines if we will be processing them
649
output_lines = bool(self.revision_ids)
650
inv_lines = self._copy_nodes_graph(inventory_index_map,
651
self.new_pack._writer, self.new_pack.inventory_index,
652
readv_group_iter, total_items, output_lines=output_lines)
653
if self.revision_ids:
654
self._process_inventory_lines(inv_lines)
656
# eat the iterator to cause it to execute.
658
self._text_filter = None
659
if 'pack' in debug.debug_flags:
660
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
661
time.ctime(), self._pack_collection._upload_transport.base,
662
self.new_pack.random_name,
663
self.new_pack.inventory_index.key_count(),
664
time.time() - self.new_pack.start_time)
747
666
def _copy_text_texts(self):
748
raise NotImplementedError(self._copy_text_texts)
668
text_index_map, text_nodes = self._get_text_nodes()
669
if self._text_filter is not None:
670
# We could return the keys copied as part of the return value from
671
# _copy_nodes_graph but this doesn't work all that well with the
672
# need to get line output too, so we check separately, and as we're
673
# going to buffer everything anyway, we check beforehand, which
674
# saves reading knit data over the wire when we know there are
676
text_nodes = set(text_nodes)
677
present_text_keys = set(_node[1] for _node in text_nodes)
678
missing_text_keys = set(self._text_filter) - present_text_keys
679
if missing_text_keys:
680
# TODO: raise a specific error that can handle many missing
682
a_missing_key = missing_text_keys.pop()
683
raise errors.RevisionNotPresent(a_missing_key[1],
685
# copy text keys and adjust values
686
self.pb.update("Copying content texts", 3)
687
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
688
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
689
self.new_pack.text_index, readv_group_iter, total_items))
690
self._log_copied_texts()
692
def _check_references(self):
693
"""Make sure our external refereneces are present."""
694
external_refs = self.new_pack._external_compression_parents_of_texts()
696
index = self._pack_collection.text_index.combined_index
697
found_items = list(index.iter_entries(external_refs))
698
if len(found_items) != len(external_refs):
699
found_keys = set(k for idx, k, refs, value in found_items)
700
missing_items = external_refs - found_keys
701
missing_file_id, missing_revision_id = missing_items.pop()
702
raise errors.RevisionNotPresent(missing_revision_id,
750
705
def _create_pack_from_packs(self):
751
raise NotImplementedError(self._create_pack_from_packs)
706
self.pb.update("Opening pack", 0, 5)
707
self.new_pack = self.open_pack()
708
new_pack = self.new_pack
709
# buffer data - we won't be reading-back during the pack creation and
710
# this makes a significant difference on sftp pushes.
711
new_pack.set_write_cache_size(1024*1024)
712
if 'pack' in debug.debug_flags:
713
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
714
for a_pack in self.packs]
715
if self.revision_ids is not None:
716
rev_count = len(self.revision_ids)
719
mutter('%s: create_pack: creating pack from source packs: '
720
'%s%s %s revisions wanted %s t=0',
721
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
722
plain_pack_list, rev_count)
723
self._copy_revision_texts()
724
self._copy_inventory_texts()
725
self._copy_text_texts()
726
# select signature keys
727
signature_filter = self._revision_keys # same keyspace
728
signature_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
729
self.packs, 'signature_index')[0]
730
signature_nodes = self._pack_collection._index_contents(signature_index_map,
732
# copy signature keys and adjust values
733
self.pb.update("Copying signature texts", 4)
734
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
735
new_pack.signature_index)
736
if 'pack' in debug.debug_flags:
737
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
738
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
739
new_pack.signature_index.key_count(),
740
time.time() - new_pack.start_time)
741
self._check_references()
742
if not self._use_pack(new_pack):
745
self.pb.update("Finishing pack", 5)
747
self._pack_collection.allocate(new_pack)
750
def _copy_nodes(self, nodes, index_map, writer, write_index):
751
"""Copy knit nodes between packs with no graph references."""
752
pb = ui.ui_factory.nested_progress_bar()
754
return self._do_copy_nodes(nodes, index_map, writer,
759
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb):
760
# for record verification
761
knit_data = _KnitData(None)
762
# plan a readv on each source pack:
764
nodes = sorted(nodes)
765
# how to map this into knit.py - or knit.py into this?
766
# we don't want the typical knit logic, we want grouping by pack
767
# at this point - perhaps a helper library for the following code
768
# duplication points?
770
for index, key, value in nodes:
771
if index not in request_groups:
772
request_groups[index] = []
773
request_groups[index].append((key, value))
775
pb.update("Copied record", record_index, len(nodes))
776
for index, items in request_groups.iteritems():
777
pack_readv_requests = []
778
for key, value in items:
779
# ---- KnitGraphIndex.get_position
780
bits = value[1:].split(' ')
781
offset, length = int(bits[0]), int(bits[1])
782
pack_readv_requests.append((offset, length, (key, value[0])))
783
# linear scan up the pack
784
pack_readv_requests.sort()
786
transport, path = index_map[index]
787
reader = pack.make_readv_reader(transport, path,
788
[offset[0:2] for offset in pack_readv_requests])
789
for (names, read_func), (_1, _2, (key, eol_flag)) in \
790
izip(reader.iter_records(), pack_readv_requests):
791
raw_data = read_func(None)
792
# check the header only
793
df, _ = knit_data._parse_record_header(key[-1], raw_data)
795
pos, size = writer.add_bytes_record(raw_data, names)
796
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
797
pb.update("Copied record", record_index)
800
def _copy_nodes_graph(self, index_map, writer, write_index,
801
readv_group_iter, total_items, output_lines=False):
802
"""Copy knit nodes between packs.
804
:param output_lines: Return lines present in the copied data as
805
an iterator of line,version_id.
807
pb = ui.ui_factory.nested_progress_bar()
809
for result in self._do_copy_nodes_graph(index_map, writer,
810
write_index, output_lines, pb, readv_group_iter, total_items):
813
# Python 2.4 does not permit try:finally: in a generator.
819
def _do_copy_nodes_graph(self, index_map, writer, write_index,
820
output_lines, pb, readv_group_iter, total_items):
821
# for record verification
822
knit_data = _KnitData(None)
823
# for line extraction when requested (inventories only)
825
factory = knit.KnitPlainFactory()
827
pb.update("Copied record", record_index, total_items)
828
for index, readv_vector, node_vector in readv_group_iter:
830
transport, path = index_map[index]
831
reader = pack.make_readv_reader(transport, path, readv_vector)
832
for (names, read_func), (key, eol_flag, references) in \
833
izip(reader.iter_records(), node_vector):
834
raw_data = read_func(None)
837
# read the entire thing
838
content, _ = knit_data._parse_record(version_id, raw_data)
839
if len(references[-1]) == 0:
840
line_iterator = factory.get_fulltext_content(content)
842
line_iterator = factory.get_linedelta_content(content)
843
for line in line_iterator:
844
yield line, version_id
846
# check the header only
847
df, _ = knit_data._parse_record_header(version_id, raw_data)
849
pos, size = writer.add_bytes_record(raw_data, names)
850
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
851
pb.update("Copied record", record_index)
854
def _get_text_nodes(self):
855
text_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
856
self.packs, 'text_index')[0]
857
return text_index_map, self._pack_collection._index_contents(text_index_map,
860
def _least_readv_node_readv(self, nodes):
861
"""Generate request groups for nodes using the least readv's.
863
:param nodes: An iterable of graph index nodes.
864
:return: Total node count and an iterator of the data needed to perform
865
readvs to obtain the data for nodes. Each item yielded by the
866
iterator is a tuple with:
867
index, readv_vector, node_vector. readv_vector is a list ready to
868
hand to the transport readv method, and node_vector is a list of
869
(key, eol_flag, references) for the the node retrieved by the
870
matching readv_vector.
872
# group by pack so we do one readv per pack
873
nodes = sorted(nodes)
876
for index, key, value, references in nodes:
877
if index not in request_groups:
878
request_groups[index] = []
879
request_groups[index].append((key, value, references))
881
for index, items in request_groups.iteritems():
882
pack_readv_requests = []
883
for key, value, references in items:
884
# ---- KnitGraphIndex.get_position
885
bits = value[1:].split(' ')
886
offset, length = int(bits[0]), int(bits[1])
887
pack_readv_requests.append(
888
((offset, length), (key, value[0], references)))
889
# linear scan up the pack to maximum range combining.
890
pack_readv_requests.sort()
891
# split out the readv and the node data.
892
pack_readv = [readv for readv, node in pack_readv_requests]
893
node_vector = [node for readv, node in pack_readv_requests]
894
result.append((index, pack_readv, node_vector))
753
897
def _log_copied_texts(self):
754
898
if 'pack' in debug.debug_flags:
767
930
return new_pack.data_inserted()
933
class OptimisingPacker(Packer):
934
"""A packer which spends more time to create better disk layouts."""
936
def _revision_node_readv(self, revision_nodes):
937
"""Return the total revisions and the readv's to issue.
939
This sort places revisions in topological order with the ancestors
942
:param revision_nodes: The revision index contents for the packs being
943
incorporated into the new pack.
944
:return: As per _least_readv_node_readv.
946
# build an ancestors dict
949
for index, key, value, references in revision_nodes:
950
ancestors[key] = references[0]
951
by_key[key] = (index, value, references)
952
order = tsort.topo_sort(ancestors)
954
# Single IO is pathological, but it will work as a starting point.
956
for key in reversed(order):
957
index, value, references = by_key[key]
958
# ---- KnitGraphIndex.get_position
959
bits = value[1:].split(' ')
960
offset, length = int(bits[0]), int(bits[1])
962
(index, [(offset, length)], [(key, value[0], references)]))
963
# TODO: combine requests in the same index that are in ascending order.
964
return total, requests
967
class ReconcilePacker(Packer):
968
"""A packer which regenerates indices etc as it copies.
970
This is used by ``bzr reconcile`` to cause parent text pointers to be
974
def _extra_init(self):
975
self._data_changed = False
977
def _process_inventory_lines(self, inv_lines):
978
"""Generate a text key reference map rather for reconciling with."""
979
repo = self._pack_collection.repo
980
refs = repo._find_text_key_references_from_xml_inventory_lines(
982
self._text_refs = refs
983
# during reconcile we:
984
# - convert unreferenced texts to full texts
985
# - correct texts which reference a text not copied to be full texts
986
# - copy all others as-is but with corrected parents.
987
# - so at this point we don't know enough to decide what becomes a full
989
self._text_filter = None
991
def _copy_text_texts(self):
992
"""generate what texts we should have and then copy."""
993
self.pb.update("Copying content texts", 3)
994
# we have three major tasks here:
995
# 1) generate the ideal index
996
repo = self._pack_collection.repo
997
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
999
self.new_pack.revision_index.iter_all_entries()])
1000
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1001
# 2) generate a text_nodes list that contains all the deltas that can
1002
# be used as-is, with corrected parents.
1005
discarded_nodes = []
1006
NULL_REVISION = _mod_revision.NULL_REVISION
1007
text_index_map, text_nodes = self._get_text_nodes()
1008
for node in text_nodes:
1014
ideal_parents = tuple(ideal_index[node[1]])
1016
discarded_nodes.append(node)
1017
self._data_changed = True
1019
if ideal_parents == (NULL_REVISION,):
1021
if ideal_parents == node[3][0]:
1023
ok_nodes.append(node)
1024
elif ideal_parents[0:1] == node[3][0][0:1]:
1025
# the left most parent is the same, or there are no parents
1026
# today. Either way, we can preserve the representation as
1027
# long as we change the refs to be inserted.
1028
self._data_changed = True
1029
ok_nodes.append((node[0], node[1], node[2],
1030
(ideal_parents, node[3][1])))
1031
self._data_changed = True
1033
# Reinsert this text completely
1034
bad_texts.append((node[1], ideal_parents))
1035
self._data_changed = True
1036
# we're finished with some data.
1039
# 3) bulk copy the ok data
1040
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1041
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1042
self.new_pack.text_index, readv_group_iter, total_items))
1043
# 4) adhoc copy all the other texts.
1044
# We have to topologically insert all texts otherwise we can fail to
1045
# reconcile when parts of a single delta chain are preserved intact,
1046
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1047
# reinserted, and if d3 has incorrect parents it will also be
1048
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1049
# copied), so we will try to delta, but d2 is not currently able to be
1050
# extracted because it's basis d1 is not present. Topologically sorting
1051
# addresses this. The following generates a sort for all the texts that
1052
# are being inserted without having to reference the entire text key
1053
# space (we only topo sort the revisions, which is smaller).
1054
topo_order = tsort.topo_sort(ancestors)
1055
rev_order = dict(zip(topo_order, range(len(topo_order))))
1056
bad_texts.sort(key=lambda key:rev_order[key[0][1]])
1057
transaction = repo.get_transaction()
1058
file_id_index = GraphIndexPrefixAdapter(
1059
self.new_pack.text_index,
1061
add_nodes_callback=self.new_pack.text_index.add_nodes)
1062
knit_index = KnitGraphIndex(file_id_index,
1063
add_callback=file_id_index.add_nodes,
1064
deltas=True, parents=True)
1065
output_knit = knit.KnitVersionedFile('reconcile-texts',
1066
self._pack_collection.transport,
1068
access_method=_PackAccess(
1069
{self.new_pack.text_index:self.new_pack.access_tuple()},
1070
(self.new_pack._writer, self.new_pack.text_index)),
1071
factory=knit.KnitPlainFactory())
1072
for key, parent_keys in bad_texts:
1073
# We refer to the new pack to delta data being output.
1074
# A possible improvement would be to catch errors on short reads
1075
# and only flush then.
1076
self.new_pack.flush()
1078
for parent_key in parent_keys:
1079
if parent_key[0] != key[0]:
1080
# Graph parents must match the fileid
1081
raise errors.BzrError('Mismatched key parent %r:%r' %
1083
parents.append(parent_key[1])
1084
source_weave = repo.weave_store.get_weave(key[0], transaction)
1085
text_lines = source_weave.get_lines(key[1])
1086
# adapt the 'knit' to the current file_id.
1087
file_id_index = GraphIndexPrefixAdapter(
1088
self.new_pack.text_index,
1090
add_nodes_callback=self.new_pack.text_index.add_nodes)
1091
knit_index._graph_index = file_id_index
1092
knit_index._add_callback = file_id_index.add_nodes
1093
output_knit.add_lines_with_ghosts(
1094
key[1], parents, text_lines, random_id=True, check_content=False)
1095
# 5) check that nothing inserted has a reference outside the keyspace.
1096
missing_text_keys = self.new_pack._external_compression_parents_of_texts()
1097
if missing_text_keys:
1098
raise errors.BzrError('Reference to missing compression parents %r'
1100
self._log_copied_texts()
1102
def _use_pack(self, new_pack):
1103
"""Override _use_pack to check for reconcile having changed content."""
1104
# XXX: we might be better checking this at the copy time.
1105
original_inventory_keys = set()
1106
inv_index = self._pack_collection.inventory_index.combined_index
1107
for entry in inv_index.iter_all_entries():
1108
original_inventory_keys.add(entry[1])
1109
new_inventory_keys = set()
1110
for entry in new_pack.inventory_index.iter_all_entries():
1111
new_inventory_keys.add(entry[1])
1112
if new_inventory_keys != original_inventory_keys:
1113
self._data_changed = True
1114
return new_pack.data_inserted() and self._data_changed
770
1117
class RepositoryPackCollection(object):
771
"""Management of packs within a repository.
773
:ivar _names: map of {pack_name: (index_size,)}
777
resumed_pack_factory = None
778
normal_packer_class = None
779
optimising_packer_class = None
1118
"""Management of packs within a repository."""
781
1120
def __init__(self, repo, transport, index_transport, upload_transport,
782
pack_transport, index_builder_class, index_class,
784
1122
"""Create a new RepositoryPackCollection.
786
:param transport: Addresses the repository base directory
1124
:param transport: Addresses the repository base directory
787
1125
(typically .bzr/repository/).
788
1126
:param index_transport: Addresses the directory containing indices.
789
1127
:param upload_transport: Addresses the directory into which packs are written
790
1128
while they're being created.
791
1129
:param pack_transport: Addresses the directory of existing complete packs.
792
:param index_builder_class: The index builder class to use.
793
:param index_class: The index class to use.
794
:param use_chk_index: Whether to setup and manage a CHK index.
796
# XXX: This should call self.reset()
797
1131
self.repo = repo
798
1132
self.transport = transport
799
1133
self._index_transport = index_transport
800
1134
self._upload_transport = upload_transport
801
1135
self._pack_transport = pack_transport
802
self._index_builder_class = index_builder_class
803
self._index_class = index_class
804
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3,
1136
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
807
1138
# name:Pack mapping
809
1139
self._packs_by_name = {}
810
1140
# the previous pack-names content
811
1141
self._packs_at_load = None
812
1142
# when a pack is being created by this object, the state of that pack.
813
1143
self._new_pack = None
814
1144
# aggregated revision index data
815
flush = self._flush_new_pack
816
self.revision_index = AggregateIndex(self.reload_pack_names, flush)
817
self.inventory_index = AggregateIndex(self.reload_pack_names, flush)
818
self.text_index = AggregateIndex(self.reload_pack_names, flush)
819
self.signature_index = AggregateIndex(self.reload_pack_names, flush)
820
all_indices = [self.revision_index, self.inventory_index,
821
self.text_index, self.signature_index]
823
self.chk_index = AggregateIndex(self.reload_pack_names, flush)
824
all_indices.append(self.chk_index)
826
# used to determine if we're using a chk_index elsewhere.
827
self.chk_index = None
828
# Tell all the CombinedGraphIndex objects about each other, so they can
829
# share hints about which pack names to search first.
830
all_combined = [agg_idx.combined_index for agg_idx in all_indices]
831
for combined_idx in all_combined:
832
combined_idx.set_sibling_indices(
833
set(all_combined).difference([combined_idx]))
835
self._resumed_packs = []
836
self.config_stack = config.LocationStack(self.transport.base)
839
return '%s(%r)' % (self.__class__.__name__, self.repo)
1145
self.revision_index = AggregateIndex()
1146
self.inventory_index = AggregateIndex()
1147
self.text_index = AggregateIndex()
1148
self.signature_index = AggregateIndex()
841
1150
def add_pack_to_memory(self, pack):
842
1151
"""Make a Pack object available to the repository to satisfy queries.
844
1153
:param pack: A Pack object.
846
1155
if pack.name in self._packs_by_name:
847
raise AssertionError(
848
'pack %s already in _packs_by_name' % (pack.name,))
1156
raise AssertionError()
849
1157
self.packs.append(pack)
850
1158
self._packs_by_name[pack.name] = pack
851
1159
self.revision_index.add_index(pack.revision_index, pack)
852
1160
self.inventory_index.add_index(pack.inventory_index, pack)
853
1161
self.text_index.add_index(pack.text_index, pack)
854
1162
self.signature_index.add_index(pack.signature_index, pack)
855
if self.chk_index is not None:
856
self.chk_index.add_index(pack.chk_index, pack)
1164
def _add_text_to_weave(self, file_id, revision_id, new_lines, parents,
1165
nostore_sha, random_revid):
1166
file_id_index = GraphIndexPrefixAdapter(
1167
self.text_index.combined_index,
1169
add_nodes_callback=self.text_index.add_callback)
1170
self.repo._text_knit._index._graph_index = file_id_index
1171
self.repo._text_knit._index._add_callback = file_id_index.add_nodes
1172
return self.repo._text_knit.add_lines_with_ghosts(
1173
revision_id, parents, new_lines, nostore_sha=nostore_sha,
1174
random_id=random_revid, check_content=False)[0:2]
858
1176
def all_packs(self):
859
1177
"""Return a list of all the Pack objects this repository has.
910
1224
# group their data with the relevant commit, and that may
911
1225
# involve rewriting ancient history - which autopack tries to
912
1226
# avoid. Alternatively we could not group the data but treat
913
# each of these as having a single revision, and thus add
1227
# each of these as having a single revision, and thus add
914
1228
# one revision for each to the total revision count, to get
915
1229
# a matching distribution.
917
1231
existing_packs.append((revision_count, pack))
918
1232
pack_operations = self.plan_autopack_combinations(
919
1233
existing_packs, pack_distribution)
920
num_new_packs = len(pack_operations)
921
num_old_packs = sum([len(po[1]) for po in pack_operations])
922
num_revs_affected = sum([po[0] for po in pack_operations])
923
mutter('Auto-packing repository %s, which has %d pack files, '
924
'containing %d revisions. Packing %d files into %d affecting %d'
925
' revisions', self, total_packs, total_revisions, num_old_packs,
926
num_new_packs, num_revs_affected)
927
result = self._execute_pack_operations(pack_operations, packer_class=self.normal_packer_class,
928
reload_func=self._restart_autopack)
929
mutter('Auto-packing repository %s completed', self)
1234
self._execute_pack_operations(pack_operations)
932
def _execute_pack_operations(self, pack_operations, packer_class,
1237
def _execute_pack_operations(self, pack_operations, _packer_class=Packer):
934
1238
"""Execute a series of pack operations.
936
1240
:param pack_operations: A list of [revision_count, packs_to_combine].
937
:param packer_class: The class of packer to use
938
:return: The new pack names.
1241
:param _packer_class: The class of packer to use (default: Packer).
940
1244
for revision_count, packs in pack_operations:
941
1245
# we may have no-ops from the setup logic
942
1246
if len(packs) == 0:
944
packer = packer_class(self, packs, '.autopack',
945
reload_func=reload_func)
947
result = packer.pack()
948
except errors.RetryWithNewPacks:
949
# An exception is propagating out of this context, make sure
950
# this packer has cleaned up. Packer() doesn't set its new_pack
951
# state into the RepositoryPackCollection object, so we only
952
# have access to it directly here.
953
if packer.new_pack is not None:
954
packer.new_pack.abort()
1248
_packer_class(self, packs, '.autopack').pack()
958
1249
for pack in packs:
959
1250
self._remove_pack_from_memory(pack)
960
1251
# record the newly available packs and stop advertising the old
963
for _, packs in pack_operations:
964
to_be_obsoleted.extend(packs)
965
result = self._save_pack_names(clear_obsolete_packs=True,
966
obsolete_packs=to_be_obsoleted)
969
def _flush_new_pack(self):
970
if self._new_pack is not None:
971
self._new_pack.flush()
1253
self._save_pack_names(clear_obsolete_packs=True)
1254
# Move the old packs out of the way now they are no longer referenced.
1255
for revision_count, packs in pack_operations:
1256
self._obsolete_packs(packs)
973
1258
def lock_names(self):
974
1259
"""Acquire the mutex around the pack-names index.
976
1261
This cannot be used in the middle of a read-only transaction on the
979
1264
self.repo.control_files.lock_write()
981
def _already_packed(self):
982
"""Is the collection already packed?"""
983
return not (self.repo._format.pack_compresses or (len(self._names) > 1))
985
def pack(self, hint=None, clean_obsolete_packs=False):
986
1267
"""Pack the pack collection totally."""
987
1268
self.ensure_loaded()
988
1269
total_packs = len(self._names)
989
if self._already_packed():
1271
# This is arguably wrong because we might not be optimal, but for
1272
# now lets leave it in. (e.g. reconcile -> one pack. But not
991
1275
total_revisions = self.revision_index.combined_index.key_count()
992
1276
# XXX: the following may want to be a class, to pack with a given
994
1278
mutter('Packing repository %s, which has %d pack files, '
995
'containing %d revisions with hint %r.', self, total_packs,
996
total_revisions, hint)
999
self._try_pack_operations(hint)
1000
except RetryPackOperations:
1004
if clean_obsolete_packs:
1005
self._clear_obsolete_packs()
1007
def _try_pack_operations(self, hint):
1008
"""Calculate the pack operations based on the hint (if any), and
1279
'containing %d revisions into 1 packs.', self, total_packs,
1011
1281
# determine which packs need changing
1282
pack_distribution = [1]
1012
1283
pack_operations = [[0, []]]
1013
1284
for pack in self.all_packs():
1014
if hint is None or pack.name in hint:
1015
# Either no hint was provided (so we are packing everything),
1016
# or this pack was included in the hint.
1017
pack_operations[-1][0] += pack.get_revision_count()
1018
pack_operations[-1][1].append(pack)
1019
self._execute_pack_operations(pack_operations,
1020
packer_class=self.optimising_packer_class,
1021
reload_func=self._restart_pack_operations)
1285
pack_operations[-1][0] += pack.get_revision_count()
1286
pack_operations[-1][1].append(pack)
1287
self._execute_pack_operations(pack_operations, OptimisingPacker)
1023
1289
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1024
1290
"""Plan a pack operation.
1321
1493
self._packs_by_name = {}
1322
1494
self._packs_at_load = None
1496
def _make_index_map(self, index_suffix):
1497
"""Return information on existing indices.
1499
:param suffix: Index suffix added to pack name.
1501
:returns: (pack_map, indices) where indices is a list of GraphIndex
1502
objects, and pack_map is a mapping from those objects to the
1503
pack tuple they describe.
1505
# TODO: stop using this; it creates new indices unnecessarily.
1506
self.ensure_loaded()
1507
suffix_map = {'.rix': 'revision_index',
1508
'.six': 'signature_index',
1509
'.iix': 'inventory_index',
1510
'.tix': 'text_index',
1512
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1513
suffix_map[index_suffix])
1515
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1516
"""Convert a list of packs to an index pack map and index list.
1518
:param packs: The packs list to process.
1519
:param index_attribute: The attribute that the desired index is found
1521
:return: A tuple (map, list) where map contains the dict from
1522
index:pack_tuple, and lsit contains the indices in the same order
1528
index = getattr(pack, index_attribute)
1529
indices.append(index)
1530
pack_map[index] = (pack.pack_transport, pack.file_name())
1531
return pack_map, indices
1533
def _index_contents(self, pack_map, key_filter=None):
1534
"""Get an iterable of the index contents from a pack_map.
1536
:param pack_map: A map from indices to pack details.
1537
:param key_filter: An optional filter to limit the
1540
indices = [index for index in pack_map.iterkeys()]
1541
all_index = CombinedGraphIndex(indices)
1542
if key_filter is None:
1543
return all_index.iter_all_entries()
1545
return all_index.iter_entries(key_filter)
1324
1547
def _unlock_names(self):
1325
1548
"""Release the mutex around the pack-names index."""
1326
1549
self.repo.control_files.unlock()
1328
def _diff_pack_names(self):
1329
"""Read the pack names from disk, and compare it to the one in memory.
1331
:return: (disk_nodes, deleted_nodes, new_nodes)
1332
disk_nodes The final set of nodes that should be referenced
1333
deleted_nodes Nodes which have been removed from when we started
1334
new_nodes Nodes that are newly introduced
1336
# load the disk nodes across
1338
for index, key, value in self._iter_disk_pack_index():
1339
disk_nodes.add((key, value))
1340
orig_disk_nodes = set(disk_nodes)
1342
# do a two-way diff against our original content
1343
current_nodes = set()
1344
for name, sizes in self._names.iteritems():
1346
((name, ), ' '.join(str(size) for size in sizes)))
1348
# Packs no longer present in the repository, which were present when we
1349
# locked the repository
1350
deleted_nodes = self._packs_at_load - current_nodes
1351
# Packs which this process is adding
1352
new_nodes = current_nodes - self._packs_at_load
1354
# Update the disk_nodes set to include the ones we are adding, and
1355
# remove the ones which were removed by someone else
1356
disk_nodes.difference_update(deleted_nodes)
1357
disk_nodes.update(new_nodes)
1359
return disk_nodes, deleted_nodes, new_nodes, orig_disk_nodes
1361
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1362
"""Given the correct set of pack files, update our saved info.
1364
:return: (removed, added, modified)
1365
removed pack names removed from self._names
1366
added pack names added to self._names
1367
modified pack names that had changed value
1372
## self._packs_at_load = disk_nodes
1551
def _save_pack_names(self, clear_obsolete_packs=False):
1552
"""Save the list of packs.
1554
This will take out the mutex around the pack names list for the
1555
duration of the method call. If concurrent updates have been made, a
1556
three-way merge between the current list and the current in memory list
1559
:param clear_obsolete_packs: If True, clear out the contents of the
1560
obsolete_packs directory.
1564
builder = GraphIndexBuilder()
1565
# load the disk nodes across
1567
for index, key, value in self._iter_disk_pack_index():
1568
disk_nodes.add((key, value))
1569
# do a two-way diff against our original content
1570
current_nodes = set()
1571
for name, sizes in self._names.iteritems():
1573
((name, ), ' '.join(str(size) for size in sizes)))
1574
deleted_nodes = self._packs_at_load - current_nodes
1575
new_nodes = current_nodes - self._packs_at_load
1576
disk_nodes.difference_update(deleted_nodes)
1577
disk_nodes.update(new_nodes)
1578
# TODO: handle same-name, index-size-changes here -
1579
# e.g. use the value from disk, not ours, *unless* we're the one
1581
for key, value in disk_nodes:
1582
builder.add_node(key, value)
1583
self.transport.put_file('pack-names', builder.finish(),
1584
mode=self.repo.control_files._file_mode)
1585
# move the baseline forward
1586
self._packs_at_load = disk_nodes
1587
# now clear out the obsolete packs directory
1588
if clear_obsolete_packs:
1589
self.transport.clone('obsolete_packs').delete_multi(
1590
self.transport.list_dir('obsolete_packs'))
1592
self._unlock_names()
1593
# synchronise the memory packs list with what we just wrote:
1373
1594
new_names = dict(disk_nodes)
1374
1595
# drop no longer present nodes
1375
1596
for pack in self.all_packs():
1376
1597
if (pack.name,) not in new_names:
1377
removed.append(pack.name)
1378
1598
self._remove_pack_from_memory(pack)
1379
1599
# add new nodes/refresh existing ones
1380
1600
for key, value in disk_nodes:
1390
1610
# disk index because the set values are the same, unless
1391
1611
# the only index shows up as deleted by the set difference
1392
1612
# - which it may. Until there is a specific test for this,
1393
# assume it's broken. RBC 20071017.
1613
# assume its broken. RBC 20071017.
1394
1614
self._remove_pack_from_memory(self.get_pack_by_name(name))
1395
1615
self._names[name] = sizes
1396
1616
self.get_pack_by_name(name)
1397
modified.append(name)
1400
1619
self._names[name] = sizes
1401
1620
self.get_pack_by_name(name)
1403
return removed, added, modified
1405
def _save_pack_names(self, clear_obsolete_packs=False, obsolete_packs=None):
1406
"""Save the list of packs.
1408
This will take out the mutex around the pack names list for the
1409
duration of the method call. If concurrent updates have been made, a
1410
three-way merge between the current list and the current in memory list
1413
:param clear_obsolete_packs: If True, clear out the contents of the
1414
obsolete_packs directory.
1415
:param obsolete_packs: Packs that are obsolete once the new pack-names
1416
file has been written.
1417
:return: A list of the names saved that were not previously on disk.
1419
already_obsolete = []
1422
builder = self._index_builder_class()
1423
(disk_nodes, deleted_nodes, new_nodes,
1424
orig_disk_nodes) = self._diff_pack_names()
1425
# TODO: handle same-name, index-size-changes here -
1426
# e.g. use the value from disk, not ours, *unless* we're the one
1428
for key, value in disk_nodes:
1429
builder.add_node(key, value)
1430
self.transport.put_file('pack-names', builder.finish(),
1431
mode=self.repo.bzrdir._get_file_mode())
1432
self._packs_at_load = disk_nodes
1433
if clear_obsolete_packs:
1436
to_preserve = set([o.name for o in obsolete_packs])
1437
already_obsolete = self._clear_obsolete_packs(to_preserve)
1439
self._unlock_names()
1440
# synchronise the memory packs list with what we just wrote:
1441
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1443
# TODO: We could add one more condition here. "if o.name not in
1444
# orig_disk_nodes and o != the new_pack we haven't written to
1445
# disk yet. However, the new pack object is not easily
1446
# accessible here (it would have to be passed through the
1447
# autopacking code, etc.)
1448
obsolete_packs = [o for o in obsolete_packs
1449
if o.name not in already_obsolete]
1450
self._obsolete_packs(obsolete_packs)
1451
return [new_node[0][0] for new_node in new_nodes]
1453
def reload_pack_names(self):
1454
"""Sync our pack listing with what is present in the repository.
1456
This should be called when we find out that something we thought was
1457
present is now missing. This happens when another process re-packs the
1460
:return: True if the in-memory list of packs has been altered at all.
1462
# The ensure_loaded call is to handle the case where the first call
1463
# made involving the collection was to reload_pack_names, where we
1464
# don't have a view of disk contents. It's a bit of a bandaid, and
1465
# causes two reads of pack-names, but it's a rare corner case not
1466
# struck with regular push/pull etc.
1467
first_read = self.ensure_loaded()
1470
# out the new value.
1471
(disk_nodes, deleted_nodes, new_nodes,
1472
orig_disk_nodes) = self._diff_pack_names()
1473
# _packs_at_load is meant to be the explicit list of names in
1474
# 'pack-names' at then start. As such, it should not contain any
1475
# pending names that haven't been written out yet.
1476
self._packs_at_load = orig_disk_nodes
1478
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1479
if removed or added or modified:
1483
def _restart_autopack(self):
1484
"""Reload the pack names list, and restart the autopack code."""
1485
if not self.reload_pack_names():
1486
# Re-raise the original exception, because something went missing
1487
# and a restart didn't find it
1489
raise errors.RetryAutopack(self.repo, False, sys.exc_info())
1491
def _restart_pack_operations(self):
1492
"""Reload the pack names list, and restart the autopack code."""
1493
if not self.reload_pack_names():
1494
# Re-raise the original exception, because something went missing
1495
# and a restart didn't find it
1497
raise RetryPackOperations(self.repo, False, sys.exc_info())
1499
def _clear_obsolete_packs(self, preserve=None):
1500
"""Delete everything from the obsolete-packs directory.
1502
:return: A list of pack identifiers (the filename without '.pack') that
1503
were found in obsolete_packs.
1506
obsolete_pack_transport = self.transport.clone('obsolete_packs')
1507
if preserve is None:
1510
obsolete_pack_files = obsolete_pack_transport.list_dir('.')
1511
except errors.NoSuchFile:
1513
for filename in obsolete_pack_files:
1514
name, ext = osutils.splitext(filename)
1517
if name in preserve:
1520
obsolete_pack_transport.delete(filename)
1521
except (errors.PathError, errors.TransportError), e:
1522
warning("couldn't delete obsolete pack, skipping it:\n%s"
1526
1622
def _start_write_group(self):
1527
1623
# Do not permit preparation for writing if we're not in a 'write lock'.
1528
1624
if not self.repo.is_write_locked():
1529
1625
raise errors.NotWriteLocked(self)
1530
self._new_pack = self.pack_factory(self, upload_suffix='.pack',
1531
file_mode=self.repo.bzrdir._get_file_mode())
1626
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1627
self._pack_transport, upload_suffix='.pack',
1628
file_mode=self.repo.control_files._file_mode)
1532
1629
# allow writing: queue writes to a new index
1533
1630
self.revision_index.add_writable_index(self._new_pack.revision_index,
1534
1631
self._new_pack)
1536
1633
self._new_pack)
1537
1634
self.text_index.add_writable_index(self._new_pack.text_index,
1538
1635
self._new_pack)
1539
self._new_pack.text_index.set_optimize(combine_backing_indices=False)
1540
1636
self.signature_index.add_writable_index(self._new_pack.signature_index,
1541
1637
self._new_pack)
1542
if self.chk_index is not None:
1543
self.chk_index.add_writable_index(self._new_pack.chk_index,
1545
self.repo.chk_bytes._index._add_callback = self.chk_index.add_callback
1546
self._new_pack.chk_index.set_optimize(combine_backing_indices=False)
1548
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
1549
self.repo.revisions._index._add_callback = self.revision_index.add_callback
1550
self.repo.signatures._index._add_callback = self.signature_index.add_callback
1551
self.repo.texts._index._add_callback = self.text_index.add_callback
1639
# reused revision and signature knits may need updating
1641
# "Hysterical raisins. client code in bzrlib grabs those knits outside
1642
# of write groups and then mutates it inside the write group."
1643
if self.repo._revision_knit is not None:
1644
self.repo._revision_knit._index._add_callback = \
1645
self.revision_index.add_callback
1646
if self.repo._signature_knit is not None:
1647
self.repo._signature_knit._index._add_callback = \
1648
self.signature_index.add_callback
1649
# create a reused knit object for text addition in commit.
1650
self.repo._text_knit = self.repo.weave_store.get_weave_or_empty(
1553
1653
def _abort_write_group(self):
1554
1654
# FIXME: just drop the transient index.
1555
1655
# forget what names there are
1556
1656
if self._new_pack is not None:
1557
operation = cleanup.OperationWithCleanups(self._new_pack.abort)
1558
operation.add_cleanup(setattr, self, '_new_pack', None)
1559
# If we aborted while in the middle of finishing the write
1560
# group, _remove_pack_indices could fail because the indexes are
1561
# already gone. But they're not there we shouldn't fail in this
1562
# case, so we pass ignore_missing=True.
1563
operation.add_cleanup(self._remove_pack_indices, self._new_pack,
1564
ignore_missing=True)
1565
operation.run_simple()
1566
for resumed_pack in self._resumed_packs:
1567
operation = cleanup.OperationWithCleanups(resumed_pack.abort)
1568
# See comment in previous finally block.
1569
operation.add_cleanup(self._remove_pack_indices, resumed_pack,
1570
ignore_missing=True)
1571
operation.run_simple()
1572
del self._resumed_packs[:]
1574
def _remove_resumed_pack_indices(self):
1575
for resumed_pack in self._resumed_packs:
1576
self._remove_pack_indices(resumed_pack)
1577
del self._resumed_packs[:]
1579
def _check_new_inventories(self):
1580
"""Detect missing inventories in this write group.
1582
:returns: list of strs, summarising any problems found. If the list is
1583
empty no problems were found.
1585
# The base implementation does no checks. GCRepositoryPackCollection
1657
self._new_pack.abort()
1658
self._remove_pack_indices(self._new_pack)
1659
self._new_pack = None
1660
self.repo._text_knit = None
1589
1662
def _commit_write_group(self):
1591
for prefix, versioned_file in (
1592
('revisions', self.repo.revisions),
1593
('inventories', self.repo.inventories),
1594
('texts', self.repo.texts),
1595
('signatures', self.repo.signatures),
1597
missing = versioned_file.get_missing_compression_parent_keys()
1598
all_missing.update([(prefix,) + key for key in missing])
1600
raise errors.BzrCheckError(
1601
"Repository %s has missing compression parent(s) %r "
1602
% (self.repo, sorted(all_missing)))
1603
problems = self._check_new_inventories()
1605
problems_summary = '\n'.join(problems)
1606
raise errors.BzrCheckError(
1607
"Cannot add revision(s) to repository: " + problems_summary)
1608
1663
self._remove_pack_indices(self._new_pack)
1609
any_new_content = False
1610
1664
if self._new_pack.data_inserted():
1611
1665
# get all the data to disk and read to use
1612
1666
self._new_pack.finish()
1613
1667
self.allocate(self._new_pack)
1614
1668
self._new_pack = None
1615
any_new_content = True
1617
self._new_pack.abort()
1618
self._new_pack = None
1619
for resumed_pack in self._resumed_packs:
1620
# XXX: this is a pretty ugly way to turn the resumed pack into a
1621
# properly committed pack.
1622
self._names[resumed_pack.name] = None
1623
self._remove_pack_from_memory(resumed_pack)
1624
resumed_pack.finish()
1625
self.allocate(resumed_pack)
1626
any_new_content = True
1627
del self._resumed_packs[:]
1629
result = self.autopack()
1669
if not self.autopack():
1631
1670
# when autopack takes no steps, the names list is still
1633
return self._save_pack_names()
1637
def _suspend_write_group(self):
1638
tokens = [pack.name for pack in self._resumed_packs]
1639
self._remove_pack_indices(self._new_pack)
1640
if self._new_pack.data_inserted():
1641
# get all the data to disk and read to use
1642
self._new_pack.finish(suspend=True)
1643
tokens.append(self._new_pack.name)
1644
self._new_pack = None
1672
self._save_pack_names()
1646
1674
self._new_pack.abort()
1647
1675
self._new_pack = None
1648
self._remove_resumed_pack_indices()
1651
def _resume_write_group(self, tokens):
1652
for token in tokens:
1653
self._resume_pack(token)
1656
class PackRepository(MetaDirVersionedFileRepository):
1657
"""Repository with knit objects stored inside pack containers.
1659
The layering for a KnitPackRepository is:
1661
Graph | HPSS | Repository public layer |
1662
===================================================
1663
Tuple based apis below, string based, and key based apis above
1664
---------------------------------------------------
1666
Provides .texts, .revisions etc
1667
This adapts the N-tuple keys to physical knit records which only have a
1668
single string identifier (for historical reasons), which in older formats
1669
was always the revision_id, and in the mapped code for packs is always
1670
the last element of key tuples.
1671
---------------------------------------------------
1673
A separate GraphIndex is used for each of the
1674
texts/inventories/revisions/signatures contained within each individual
1675
pack file. The GraphIndex layer works in N-tuples and is unaware of any
1677
===================================================
1681
# These attributes are inherited from the Repository base class. Setting
1682
# them to None ensures that if the constructor is changed to not initialize
1683
# them, or a subclass fails to call the constructor, that an error will
1684
# occur rather than the system working but generating incorrect data.
1685
_commit_builder_class = None
1688
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
1690
MetaDirRepository.__init__(self, _format, a_bzrdir, control_files)
1691
self._commit_builder_class = _commit_builder_class
1692
self._serializer = _serializer
1676
self.repo._text_knit = None
1679
class KnitPackRevisionStore(KnitRevisionStore):
1680
"""An object to adapt access from RevisionStore's to use KnitPacks.
1682
This class works by replacing the original RevisionStore.
1683
We need to do this because the KnitPackRevisionStore is less
1684
isolated in its layering - it uses services from the repo.
1687
def __init__(self, repo, transport, revisionstore):
1688
"""Create a KnitPackRevisionStore on repo with revisionstore.
1690
This will store its state in the Repository, use the
1691
indices to provide a KnitGraphIndex,
1692
and at the end of transactions write new indices.
1694
KnitRevisionStore.__init__(self, revisionstore.versioned_file_store)
1696
self._serializer = revisionstore._serializer
1697
self.transport = transport
1699
def get_revision_file(self, transaction):
1700
"""Get the revision versioned file object."""
1701
if getattr(self.repo, '_revision_knit', None) is not None:
1702
return self.repo._revision_knit
1703
self.repo._pack_collection.ensure_loaded()
1704
add_callback = self.repo._pack_collection.revision_index.add_callback
1705
# setup knit specific objects
1706
knit_index = KnitGraphIndex(
1707
self.repo._pack_collection.revision_index.combined_index,
1708
add_callback=add_callback)
1709
self.repo._revision_knit = knit.KnitVersionedFile(
1710
'revisions', self.transport.clone('..'),
1711
self.repo.control_files._file_mode,
1713
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1714
access_method=self.repo._pack_collection.revision_index.knit_access)
1715
return self.repo._revision_knit
1717
def get_signature_file(self, transaction):
1718
"""Get the signature versioned file object."""
1719
if getattr(self.repo, '_signature_knit', None) is not None:
1720
return self.repo._signature_knit
1721
self.repo._pack_collection.ensure_loaded()
1722
add_callback = self.repo._pack_collection.signature_index.add_callback
1723
# setup knit specific objects
1724
knit_index = KnitGraphIndex(
1725
self.repo._pack_collection.signature_index.combined_index,
1726
add_callback=add_callback, parents=False)
1727
self.repo._signature_knit = knit.KnitVersionedFile(
1728
'signatures', self.transport.clone('..'),
1729
self.repo.control_files._file_mode,
1731
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1732
access_method=self.repo._pack_collection.signature_index.knit_access)
1733
return self.repo._signature_knit
1736
class KnitPackTextStore(VersionedFileStore):
1737
"""Presents a TextStore abstraction on top of packs.
1739
This class works by replacing the original VersionedFileStore.
1740
We need to do this because the KnitPackRevisionStore is less
1741
isolated in its layering - it uses services from the repo and shares them
1742
with all the data written in a single write group.
1745
def __init__(self, repo, transport, weavestore):
1746
"""Create a KnitPackTextStore on repo with weavestore.
1748
This will store its state in the Repository, use the
1749
indices FileNames to provide a KnitGraphIndex,
1750
and at the end of transactions write new indices.
1752
# don't call base class constructor - it's not suitable.
1753
# no transient data stored in the transaction
1755
self._precious = False
1757
self.transport = transport
1758
self.weavestore = weavestore
1759
# XXX for check() which isn't updated yet
1760
self._transport = weavestore._transport
1762
def get_weave_or_empty(self, file_id, transaction):
1763
"""Get a 'Knit' backed by the .tix indices.
1765
The transaction parameter is ignored.
1767
self.repo._pack_collection.ensure_loaded()
1768
add_callback = self.repo._pack_collection.text_index.add_callback
1769
# setup knit specific objects
1770
file_id_index = GraphIndexPrefixAdapter(
1771
self.repo._pack_collection.text_index.combined_index,
1772
(file_id, ), 1, add_nodes_callback=add_callback)
1773
knit_index = KnitGraphIndex(file_id_index,
1774
add_callback=file_id_index.add_nodes,
1775
deltas=True, parents=True)
1776
return knit.KnitVersionedFile('text:' + file_id,
1777
self.transport.clone('..'),
1780
access_method=self.repo._pack_collection.text_index.knit_access,
1781
factory=knit.KnitPlainFactory())
1783
get_weave = get_weave_or_empty
1786
"""Generate a list of the fileids inserted, for use by check."""
1787
self.repo._pack_collection.ensure_loaded()
1789
for index, key, value, refs in \
1790
self.repo._pack_collection.text_index.combined_index.iter_all_entries():
1795
class InventoryKnitThunk(object):
1796
"""An object to manage thunking get_inventory_weave to pack based knits."""
1798
def __init__(self, repo, transport):
1799
"""Create an InventoryKnitThunk for repo at transport.
1801
This will store its state in the Repository, use the
1802
indices FileNames to provide a KnitGraphIndex,
1803
and at the end of transactions write a new index..
1806
self.transport = transport
1808
def get_weave(self):
1809
"""Get a 'Knit' that contains inventory data."""
1810
self.repo._pack_collection.ensure_loaded()
1811
add_callback = self.repo._pack_collection.inventory_index.add_callback
1812
# setup knit specific objects
1813
knit_index = KnitGraphIndex(
1814
self.repo._pack_collection.inventory_index.combined_index,
1815
add_callback=add_callback, deltas=True, parents=True)
1816
return knit.KnitVersionedFile(
1817
'inventory', self.transport.clone('..'),
1818
self.repo.control_files._file_mode,
1820
index=knit_index, delta=True, factory=knit.KnitPlainFactory(),
1821
access_method=self.repo._pack_collection.inventory_index.knit_access)
1824
class KnitPackRepository(KnitRepository):
1825
"""Experimental graph-knit using repository."""
1827
def __init__(self, _format, a_bzrdir, control_files, _revision_store,
1828
control_store, text_store, _commit_builder_class, _serializer):
1829
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
1830
_revision_store, control_store, text_store, _commit_builder_class,
1832
index_transport = control_files._transport.clone('indices')
1833
self._pack_collection = RepositoryPackCollection(self, control_files._transport,
1835
control_files._transport.clone('upload'),
1836
control_files._transport.clone('packs'))
1837
self._revision_store = KnitPackRevisionStore(self, index_transport, self._revision_store)
1838
self.weave_store = KnitPackTextStore(self, index_transport, self.weave_store)
1839
self._inv_thunk = InventoryKnitThunk(self, index_transport)
1840
# True when the repository object is 'write locked' (as opposed to the
1841
# physical lock only taken out around changes to the pack-names list.)
1842
# Another way to represent this would be a decorator around the control
1843
# files object that presents logical locks as physical ones - if this
1844
# gets ugly consider that alternative design. RBC 20071011
1845
self._write_lock_count = 0
1846
self._transaction = None
1848
self._reconcile_does_inventory_gc = True
1693
1849
self._reconcile_fixes_text_parents = True
1694
if self._format.supports_external_lookups:
1695
self._unstacked_provider = graph.CachingParentsProvider(
1696
self._make_parents_provider_unstacked())
1850
self._reconcile_backsup_inventory = False
1852
def _abort_write_group(self):
1853
self._pack_collection._abort_write_group()
1855
def _find_inconsistent_revision_parents(self):
1856
"""Find revisions with incorrectly cached parents.
1858
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1859
parents-in-revision).
1861
if not self.is_locked():
1862
raise errors.ObjectNotLocked(self)
1863
pb = ui.ui_factory.nested_progress_bar()
1866
revision_nodes = self._pack_collection.revision_index \
1867
.combined_index.iter_all_entries()
1868
index_positions = []
1869
# Get the cached index values for all revisions, and also the location
1870
# in each index of the revision text so we can perform linear IO.
1871
for index, key, value, refs in revision_nodes:
1872
pos, length = value[1:].split(' ')
1873
index_positions.append((index, int(pos), key[0],
1874
tuple(parent[0] for parent in refs[0])))
1875
pb.update("Reading revision index.", 0, 0)
1876
index_positions.sort()
1877
batch_count = len(index_positions) / 1000 + 1
1878
pb.update("Checking cached revision graph.", 0, batch_count)
1879
for offset in xrange(batch_count):
1880
pb.update("Checking cached revision graph.", offset)
1881
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1884
rev_ids = [item[2] for item in to_query]
1885
revs = self.get_revisions(rev_ids)
1886
for revision, item in zip(revs, to_query):
1887
index_parents = item[3]
1888
rev_parents = tuple(revision.parent_ids)
1889
if index_parents != rev_parents:
1890
result.append((revision.revision_id, index_parents, rev_parents))
1895
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
1896
def get_parents(self, revision_ids):
1897
"""See graph._StackedParentsProvider.get_parents."""
1898
parent_map = self.get_parent_map(revision_ids)
1899
return [parent_map.get(r, None) for r in revision_ids]
1901
def get_parent_map(self, keys):
1902
"""See graph._StackedParentsProvider.get_parent_map
1904
This implementation accesses the combined revision index to provide
1907
self._pack_collection.ensure_loaded()
1908
index = self._pack_collection.revision_index.combined_index
1910
if _mod_revision.NULL_REVISION in keys:
1911
keys.discard(_mod_revision.NULL_REVISION)
1912
found_parents = {_mod_revision.NULL_REVISION:()}
1698
self._unstacked_provider = graph.CachingParentsProvider(self)
1699
self._unstacked_provider.disable_cache()
1915
search_keys = set((revision_id,) for revision_id in keys)
1916
for index, key, value, refs in index.iter_entries(search_keys):
1919
parents = (_mod_revision.NULL_REVISION,)
1921
parents = tuple(parent[0] for parent in parents)
1922
found_parents[key[0]] = parents
1923
return found_parents
1925
@symbol_versioning.deprecated_method(symbol_versioning.one_four)
1701
1926
@needs_read_lock
1702
def _all_revision_ids(self):
1703
"""See Repository.all_revision_ids()."""
1704
return [key[0] for key in self.revisions.keys()]
1706
def _abort_write_group(self):
1707
self.revisions._index._key_dependencies.clear()
1708
self._pack_collection._abort_write_group()
1927
def get_revision_graph(self, revision_id=None):
1928
"""Return a dictionary containing the revision graph.
1930
:param revision_id: The revision_id to get a graph from. If None, then
1931
the entire revision graph is returned. This is a deprecated mode of
1932
operation and will be removed in the future.
1933
:return: a dictionary of revision_id->revision_parents_list.
1935
if 'evil' in debug.debug_flags:
1937
"get_revision_graph scales with size of history.")
1938
# special case NULL_REVISION
1939
if revision_id == _mod_revision.NULL_REVISION:
1941
if revision_id is None:
1942
revision_vf = self._get_revision_vf()
1943
return revision_vf.get_graph()
1944
g = self.get_graph()
1945
first = g.get_parent_map([revision_id])
1946
if revision_id not in first:
1947
raise errors.NoSuchRevision(self, revision_id)
1951
NULL_REVISION = _mod_revision.NULL_REVISION
1952
ghosts = set([NULL_REVISION])
1953
for rev_id, parent_ids in g.iter_ancestry([revision_id]):
1954
if parent_ids is None: # This is a ghost
1957
ancestry[rev_id] = parent_ids
1958
for p in parent_ids:
1960
children[p].append(rev_id)
1962
children[p] = [rev_id]
1964
if NULL_REVISION in ancestry:
1965
del ancestry[NULL_REVISION]
1967
# Find all nodes that reference a ghost, and filter the ghosts out
1968
# of their parent lists. To preserve the order of parents, and
1969
# avoid double filtering nodes, we just find all children first,
1971
children_of_ghosts = set()
1972
for ghost in ghosts:
1973
children_of_ghosts.update(children[ghost])
1975
for child in children_of_ghosts:
1976
ancestry[child] = tuple(p for p in ancestry[child]
1980
def has_revisions(self, revision_ids):
1981
"""See Repository.has_revisions()."""
1982
revision_ids = set(revision_ids)
1983
result = revision_ids.intersection(
1984
set([None, _mod_revision.NULL_REVISION]))
1985
revision_ids.difference_update(result)
1986
index = self._pack_collection.revision_index.combined_index
1987
keys = [(revision_id,) for revision_id in revision_ids]
1988
result.update(node[1][0] for node in index.iter_entries(keys))
1710
1991
def _make_parents_provider(self):
1711
if not self._format.supports_external_lookups:
1712
return self._unstacked_provider
1713
return graph.StackedParentsProvider(_LazyListJoin(
1714
[self._unstacked_provider], self._fallback_repositories))
1992
return graph.CachingParentsProvider(self)
1716
1994
def _refresh_data(self):
1717
if not self.is_locked():
1719
self._pack_collection.reload_pack_names()
1720
self._unstacked_provider.disable_cache()
1721
self._unstacked_provider.enable_cache()
1995
if self._write_lock_count == 1 or (
1996
self.control_files._lock_count == 1 and
1997
self.control_files._lock_mode == 'r'):
1998
# forget what names there are
1999
self._pack_collection.reset()
2000
# XXX: Better to do an in-memory merge when acquiring a new lock -
2001
# factor out code from _save_pack_names.
2002
self._pack_collection.ensure_loaded()
1723
2004
def _start_write_group(self):
1724
2005
self._pack_collection._start_write_group()
1726
2007
def _commit_write_group(self):
1727
hint = self._pack_collection._commit_write_group()
1728
self.revisions._index._key_dependencies.clear()
1729
# The commit may have added keys that were previously cached as
1730
# missing, so reset the cache.
1731
self._unstacked_provider.disable_cache()
1732
self._unstacked_provider.enable_cache()
1735
def suspend_write_group(self):
1736
# XXX check self._write_group is self.get_transaction()?
1737
tokens = self._pack_collection._suspend_write_group()
1738
self.revisions._index._key_dependencies.clear()
1739
self._write_group = None
1742
def _resume_write_group(self, tokens):
1743
self._start_write_group()
1745
self._pack_collection._resume_write_group(tokens)
1746
except errors.UnresumableWriteGroup:
1747
self._abort_write_group()
1749
for pack in self._pack_collection._resumed_packs:
1750
self.revisions._index.scan_unvalidated_index(pack.revision_index)
2008
return self._pack_collection._commit_write_group()
2010
def get_inventory_weave(self):
2011
return self._inv_thunk.get_weave()
1752
2013
def get_transaction(self):
1753
2014
if self._write_lock_count:
1906
2158
mutter('creating repository in %s.', a_bzrdir.transport.base)
1907
2159
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
1908
builder = self.index_builder_class()
2160
builder = GraphIndexBuilder()
1909
2161
files = [('pack-names', builder.finish())]
1910
2162
utf8_files = [('format', self.get_format_string())]
1912
2164
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
1913
repository = self.open(a_bzrdir=a_bzrdir, _found=True)
1914
self._run_post_repo_init_hooks(repository, a_bzrdir, shared)
2165
return self.open(a_bzrdir=a_bzrdir, _found=True)
1917
2167
def open(self, a_bzrdir, _found=False, _override_transport=None):
1918
2168
"""See RepositoryFormat.open().
1920
2170
:param _override_transport: INTERNAL USE ONLY. Allows opening the
1921
2171
repository at a slightly different url
1922
2172
than normal. I.e. during 'upgrade'.
1925
format = RepositoryFormatMetaDir.find_format(a_bzrdir)
2175
format = RepositoryFormat.find_format(a_bzrdir)
1926
2176
if _override_transport is not None:
1927
2177
repo_transport = _override_transport
1929
2179
repo_transport = a_bzrdir.get_repository_transport(None)
1930
2180
control_files = lockable_files.LockableFiles(repo_transport,
1931
2181
'lock', lockdir.LockDir)
2182
text_store = self._get_text_store(repo_transport, control_files)
2183
control_store = self._get_control_store(repo_transport, control_files)
2184
_revision_store = self._get_revision_store(repo_transport, control_files)
1932
2185
return self.repository_class(_format=self,
1933
2186
a_bzrdir=a_bzrdir,
1934
2187
control_files=control_files,
2188
_revision_store=_revision_store,
2189
control_store=control_store,
2190
text_store=text_store,
1935
2191
_commit_builder_class=self._commit_builder_class,
1936
2192
_serializer=self._serializer)
1939
class RetryPackOperations(errors.RetryWithNewPacks):
1940
"""Raised when we are packing and we find a missing file.
1942
Meant as a signaling exception, to tell the RepositoryPackCollection.pack
1943
code it should try again.
1946
internal_error = True
1948
_fmt = ("Pack files have changed, reload and try pack again."
1949
" context: %(context)s %(orig_error)s")
1952
class _DirectPackAccess(object):
1953
"""Access to data in one or more packs with less translation."""
1955
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
1956
"""Create a _DirectPackAccess object.
1958
:param index_to_packs: A dict mapping index objects to the transport
1959
and file names for obtaining data.
1960
:param reload_func: A function to call if we determine that the pack
1961
files have moved and we need to reload our caches. See
1962
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
1964
self._container_writer = None
1965
self._write_index = None
1966
self._indices = index_to_packs
1967
self._reload_func = reload_func
1968
self._flush_func = flush_func
1970
def add_raw_records(self, key_sizes, raw_data):
1971
"""Add raw knit bytes to a storage area.
1973
The data is spooled to the container writer in one bytes-record per
1976
:param sizes: An iterable of tuples containing the key and size of each
1978
:param raw_data: A bytestring containing the data.
1979
:return: A list of memos to retrieve the record later. Each memo is an
1980
opaque index memo. For _DirectPackAccess the memo is (index, pos,
1981
length), where the index field is the write_index object supplied
1982
to the PackAccess object.
1984
if type(raw_data) is not str:
1985
raise AssertionError(
1986
'data must be plain bytes was %s' % type(raw_data))
1989
for key, size in key_sizes:
1990
p_offset, p_length = self._container_writer.add_bytes_record(
1991
raw_data[offset:offset+size], [])
1993
result.append((self._write_index, p_offset, p_length))
1997
"""Flush pending writes on this access object.
1999
This will flush any buffered writes to a NewPack.
2001
if self._flush_func is not None:
2004
def get_raw_records(self, memos_for_retrieval):
2005
"""Get the raw bytes for a records.
2007
:param memos_for_retrieval: An iterable containing the (index, pos,
2008
length) memo for retrieving the bytes. The Pack access method
2009
looks up the pack to use for a given record in its index_to_pack
2011
:return: An iterator over the bytes of the records.
2013
# first pass, group into same-index requests
2015
current_index = None
2016
for (index, offset, length) in memos_for_retrieval:
2017
if current_index == index:
2018
current_list.append((offset, length))
2020
if current_index is not None:
2021
request_lists.append((current_index, current_list))
2022
current_index = index
2023
current_list = [(offset, length)]
2024
# handle the last entry
2025
if current_index is not None:
2026
request_lists.append((current_index, current_list))
2027
for index, offsets in request_lists:
2029
transport, path = self._indices[index]
2031
# A KeyError here indicates that someone has triggered an index
2032
# reload, and this index has gone missing, we need to start
2034
if self._reload_func is None:
2035
# If we don't have a _reload_func there is nothing that can
2038
raise errors.RetryWithNewPacks(index,
2039
reload_occurred=True,
2040
exc_info=sys.exc_info())
2042
reader = pack.make_readv_reader(transport, path, offsets)
2043
for names, read_func in reader.iter_records():
2044
yield read_func(None)
2045
except errors.NoSuchFile:
2046
# A NoSuchFile error indicates that a pack file has gone
2047
# missing on disk, we need to trigger a reload, and start over.
2048
if self._reload_func is None:
2050
raise errors.RetryWithNewPacks(transport.abspath(path),
2051
reload_occurred=False,
2052
exc_info=sys.exc_info())
2054
def set_writer(self, writer, index, transport_packname):
2055
"""Set a writer to use for adding data."""
2056
if index is not None:
2057
self._indices[index] = transport_packname
2058
self._container_writer = writer
2059
self._write_index = index
2061
def reload_or_raise(self, retry_exc):
2062
"""Try calling the reload function, or re-raise the original exception.
2064
This should be called after _DirectPackAccess raises a
2065
RetryWithNewPacks exception. This function will handle the common logic
2066
of determining when the error is fatal versus being temporary.
2067
It will also make sure that the original exception is raised, rather
2068
than the RetryWithNewPacks exception.
2070
If this function returns, then the calling function should retry
2071
whatever operation was being performed. Otherwise an exception will
2074
:param retry_exc: A RetryWithNewPacks exception.
2077
if self._reload_func is None:
2079
elif not self._reload_func():
2080
# The reload claimed that nothing changed
2081
if not retry_exc.reload_occurred:
2082
# If there wasn't an earlier reload, then we really were
2083
# expecting to find changes. We didn't find them, so this is a
2087
exc_class, exc_value, exc_traceback = retry_exc.exc_info
2088
raise exc_class, exc_value, exc_traceback
2195
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2196
"""A no-subtrees parameterized Pack repository.
2198
This format was introduced in 0.92.
2201
repository_class = KnitPackRepository
2202
_commit_builder_class = PackCommitBuilder
2203
_serializer = xml5.serializer_v5
2205
def _get_matching_bzrdir(self):
2206
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2208
def _ignore_setting_bzrdir(self, format):
2211
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2213
def get_format_string(self):
2214
"""See RepositoryFormat.get_format_string()."""
2215
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2217
def get_format_description(self):
2218
"""See RepositoryFormat.get_format_description()."""
2219
return "Packs containing knits without subtree support"
2221
def check_conversion_target(self, target_format):
2225
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2226
"""A subtrees parameterized Pack repository.
2228
This repository format uses the xml7 serializer to get:
2229
- support for recording full info about the tree root
2230
- support for recording tree-references
2232
This format was introduced in 0.92.
2235
repository_class = KnitPackRepository
2236
_commit_builder_class = PackRootCommitBuilder
2237
rich_root_data = True
2238
supports_tree_reference = True
2239
_serializer = xml7.serializer_v7
2241
def _get_matching_bzrdir(self):
2242
return bzrdir.format_registry.make_bzrdir(
2243
'pack-0.92-subtree')
2245
def _ignore_setting_bzrdir(self, format):
2248
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2250
def check_conversion_target(self, target_format):
2251
if not target_format.rich_root_data:
2252
raise errors.BadConversionTarget(
2253
'Does not support rich root data.', target_format)
2254
if not getattr(target_format, 'supports_tree_reference', False):
2255
raise errors.BadConversionTarget(
2256
'Does not support nested trees', target_format)
2258
def get_format_string(self):
2259
"""See RepositoryFormat.get_format_string()."""
2260
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2262
def get_format_description(self):
2263
"""See RepositoryFormat.get_format_description()."""
2264
return "Packs containing knits with subtree support\n"
2267
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2268
"""A rich-root, no subtrees parameterized Pack repository.
2270
This repository format uses the xml6 serializer to get:
2271
- support for recording full info about the tree root
2273
This format was introduced in 1.0.
2276
repository_class = KnitPackRepository
2277
_commit_builder_class = PackRootCommitBuilder
2278
rich_root_data = True
2279
supports_tree_reference = False
2280
_serializer = xml6.serializer_v6
2282
def _get_matching_bzrdir(self):
2283
return bzrdir.format_registry.make_bzrdir(
2286
def _ignore_setting_bzrdir(self, format):
2289
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2291
def check_conversion_target(self, target_format):
2292
if not target_format.rich_root_data:
2293
raise errors.BadConversionTarget(
2294
'Does not support rich root data.', target_format)
2296
def get_format_string(self):
2297
"""See RepositoryFormat.get_format_string()."""
2298
return ("Bazaar pack repository format 1 with rich root"
2299
" (needs bzr 1.0)\n")
2301
def get_format_description(self):
2302
"""See RepositoryFormat.get_format_description()."""
2303
return "Packs containing knits with rich root support\n"
2306
class RepositoryFormatPackDevelopment0(RepositoryFormatPack):
2307
"""A no-subtrees development repository.
2309
This format should be retained until the second release after bzr 1.0.
2311
No changes to the disk behaviour from pack-0.92.
2314
repository_class = KnitPackRepository
2315
_commit_builder_class = PackCommitBuilder
2316
_serializer = xml5.serializer_v5
2318
def _get_matching_bzrdir(self):
2319
return bzrdir.format_registry.make_bzrdir('development0')
2321
def _ignore_setting_bzrdir(self, format):
2324
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2326
def get_format_string(self):
2327
"""See RepositoryFormat.get_format_string()."""
2328
return "Bazaar development format 0 (needs bzr.dev from before 1.3)\n"
2330
def get_format_description(self):
2331
"""See RepositoryFormat.get_format_description()."""
2332
return ("Development repository format, currently the same as "
2335
def check_conversion_target(self, target_format):
2339
class RepositoryFormatPackDevelopment0Subtree(RepositoryFormatPack):
2340
"""A subtrees development repository.
2342
This format should be retained until the second release after bzr 1.0.
2344
No changes to the disk behaviour from pack-0.92-subtree.
2347
repository_class = KnitPackRepository
2348
_commit_builder_class = PackRootCommitBuilder
2349
rich_root_data = True
2350
supports_tree_reference = True
2351
_serializer = xml7.serializer_v7
2353
def _get_matching_bzrdir(self):
2354
return bzrdir.format_registry.make_bzrdir(
2355
'development0-subtree')
2357
def _ignore_setting_bzrdir(self, format):
2360
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2362
def check_conversion_target(self, target_format):
2363
if not target_format.rich_root_data:
2364
raise errors.BadConversionTarget(
2365
'Does not support rich root data.', target_format)
2366
if not getattr(target_format, 'supports_tree_reference', False):
2367
raise errors.BadConversionTarget(
2368
'Does not support nested trees', target_format)
2370
def get_format_string(self):
2371
"""See RepositoryFormat.get_format_string()."""
2372
return ("Bazaar development format 0 with subtree support "
2373
"(needs bzr.dev from before 1.3)\n")
2375
def get_format_description(self):
2376
"""See RepositoryFormat.get_format_description()."""
2377
return ("Development repository format, currently the same as "
2378
"pack-0.92-subtree\n")
added
removed
bzrlib/repofmt/pack_repo.py
