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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
from __future__ import absolute_import
17
22
from bzrlib.lazy_import import lazy_import
18
23
lazy_import(globals(), """
19
24
from itertools import izip
24
27
from bzrlib import (
30
39
from bzrlib.index import (
34
40
CombinedGraphIndex,
35
41
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
43
44
from bzrlib import (
57
from bzrlib.decorators import needs_read_lock, needs_write_lock
58
from bzrlib.repofmt.knitrepo import KnitRepository
51
from bzrlib.decorators import (
56
from bzrlib.lock import LogicalLockResult
59
57
from bzrlib.repository import (
62
MetaDirRepositoryFormat,
66
import bzrlib.revision as _mod_revision
67
from bzrlib.store.revision.knit import KnitRevisionStore
68
from bzrlib.store.versioned import VersionedFileStore
60
RepositoryFormatMetaDir,
61
RepositoryWriteLockResult,
63
from bzrlib.vf_repository import (
64
MetaDirVersionedFileRepository,
65
MetaDirVersionedFileRepositoryFormat,
66
VersionedFileCommitBuilder,
67
VersionedFileRootCommitBuilder,
69
69
from bzrlib.trace import (
77
class PackCommitBuilder(CommitBuilder):
78
"""A subclass of CommitBuilder to add texts with pack semantics.
76
class PackCommitBuilder(VersionedFileCommitBuilder):
77
"""Subclass of VersionedFileCommitBuilder to add texts with pack semantics.
80
79
Specifically this uses one knit object rather than one knit object per
81
80
added text, reducing memory and object pressure.
84
83
def __init__(self, repository, parents, config, timestamp=None,
85
84
timezone=None, committer=None, revprops=None,
87
CommitBuilder.__init__(self, repository, parents, config,
85
revision_id=None, lossy=False):
86
VersionedFileCommitBuilder.__init__(self, repository, parents, config,
88
87
timestamp=timestamp, timezone=timezone, committer=committer,
89
revprops=revprops, revision_id=revision_id)
88
revprops=revprops, revision_id=revision_id, lossy=lossy)
90
89
self._file_graph = graph.Graph(
91
90
repository._pack_collection.text_index.combined_index)
93
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
94
return self.repository._pack_collection._add_text_to_weave(file_id,
95
self._new_revision_id, new_lines, parents, nostore_sha,
98
92
def _heads(self, file_id, revision_ids):
99
93
keys = [(file_id, revision_id) for revision_id in revision_ids]
100
94
return set([key[1] for key in self._file_graph.heads(keys)])
103
class PackRootCommitBuilder(RootCommitBuilder):
97
class PackRootCommitBuilder(VersionedFileRootCommitBuilder):
104
98
"""A subclass of RootCommitBuilder to add texts with pack semantics.
106
100
Specifically this uses one knit object rather than one knit object per
107
101
added text, reducing memory and object pressure.
110
104
def __init__(self, repository, parents, config, timestamp=None,
111
105
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)
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,
116
111
self._file_graph = graph.Graph(
117
112
repository._pack_collection.text_index.combined_index)
119
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
120
return self.repository._pack_collection._add_text_to_weave(file_id,
121
self._new_revision_id, new_lines, parents, nostore_sha,
124
114
def _heads(self, file_id, revision_ids):
125
115
keys = [(file_id, revision_id) for revision_id in revision_ids]
126
116
return set([key[1] for key in self._file_graph.heads(keys)])
214
258
return not self.__eq__(other)
216
260
def __repr__(self):
217
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
218
id(self), self.transport, self.name)
261
return "<%s.%s object at 0x%x, %s, %s" % (
262
self.__class__.__module__, self.__class__.__name__, id(self),
263
self.pack_transport, self.name)
266
class ResumedPack(ExistingPack):
268
def __init__(self, name, revision_index, inventory_index, text_index,
269
signature_index, upload_transport, pack_transport, index_transport,
270
pack_collection, chk_index=None):
271
"""Create a ResumedPack object."""
272
ExistingPack.__init__(self, pack_transport, name, revision_index,
273
inventory_index, text_index, signature_index,
275
self.upload_transport = upload_transport
276
self.index_transport = index_transport
277
self.index_sizes = [None, None, None, None]
279
('revision', revision_index),
280
('inventory', inventory_index),
281
('text', text_index),
282
('signature', signature_index),
284
if chk_index is not None:
285
indices.append(('chk', chk_index))
286
self.index_sizes.append(None)
287
for index_type, index in indices:
288
offset = self.index_offset(index_type)
289
self.index_sizes[offset] = index._size
290
self.index_class = pack_collection._index_class
291
self._pack_collection = pack_collection
292
self._state = 'resumed'
293
# XXX: perhaps check that the .pack file exists?
295
def access_tuple(self):
296
if self._state == 'finished':
297
return Pack.access_tuple(self)
298
elif self._state == 'resumed':
299
return self.upload_transport, self.file_name()
301
raise AssertionError(self._state)
304
self.upload_transport.delete(self.file_name())
305
indices = [self.revision_index, self.inventory_index, self.text_index,
306
self.signature_index]
307
if self.chk_index is not None:
308
indices.append(self.chk_index)
309
for index in indices:
310
index._transport.delete(index._name)
313
self._check_references()
314
index_types = ['revision', 'inventory', 'text', 'signature']
315
if self.chk_index is not None:
316
index_types.append('chk')
317
for index_type in index_types:
318
old_name = self.index_name(index_type, self.name)
319
new_name = '../indices/' + old_name
320
self.upload_transport.move(old_name, new_name)
321
self._replace_index_with_readonly(index_type)
322
new_name = '../packs/' + self.file_name()
323
self.upload_transport.move(self.file_name(), new_name)
324
self._state = 'finished'
326
def _get_external_refs(self, index):
327
"""Return compression parents for this index that are not present.
329
This returns any compression parents that are referenced by this index,
330
which are not contained *in* this index. They may be present elsewhere.
332
return index.external_references(1)
221
335
class NewPack(Pack):
222
336
"""An in memory proxy for a pack which is being created."""
224
# A map of index 'type' to the file extension and position in the
226
index_definitions = {
227
'revision': ('.rix', 0),
228
'inventory': ('.iix', 1),
230
'signature': ('.six', 3),
233
def __init__(self, upload_transport, index_transport, pack_transport,
234
upload_suffix='', file_mode=None):
338
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
235
339
"""Create a NewPack instance.
237
:param upload_transport: A writable transport for the pack to be
238
incrementally uploaded to.
239
:param index_transport: A writable transport for the pack's indices to
240
be written to when the pack is finished.
241
:param pack_transport: A writable transport for the pack to be renamed
242
to when the upload is complete. This *must* be the same as
243
upload_transport.clone('../packs').
341
:param pack_collection: A PackCollection into which this is being inserted.
244
342
:param upload_suffix: An optional suffix to be given to any temporary
245
343
files created during the pack creation. e.g '.autopack'
246
:param file_mode: An optional file mode to create the new files with.
344
:param file_mode: Unix permissions for newly created file.
248
346
# The relative locations of the packs are constrained, but all are
249
347
# passed in because the caller has them, so as to avoid object churn.
348
index_builder_class = pack_collection._index_builder_class
349
if pack_collection.chk_index is not None:
350
chk_index = index_builder_class(reference_lists=0)
250
353
Pack.__init__(self,
251
354
# Revisions: parents list, no text compression.
252
InMemoryGraphIndex(reference_lists=1),
355
index_builder_class(reference_lists=1),
253
356
# Inventory: We want to map compression only, but currently the
254
357
# knit code hasn't been updated enough to understand that, so we
255
358
# have a regular 2-list index giving parents and compression
257
InMemoryGraphIndex(reference_lists=2),
360
index_builder_class(reference_lists=2),
258
361
# Texts: compression and per file graph, for all fileids - so two
259
362
# reference lists and two elements in the key tuple.
260
InMemoryGraphIndex(reference_lists=2, key_elements=2),
363
index_builder_class(reference_lists=2, key_elements=2),
261
364
# Signatures: Just blobs to store, no compression, no parents
263
InMemoryGraphIndex(reference_lists=0),
366
index_builder_class(reference_lists=0),
367
# CHK based storage - just blobs, no compression or parents.
370
self._pack_collection = pack_collection
371
# When we make readonly indices, we need this.
372
self.index_class = pack_collection._index_class
265
373
# where should the new pack be opened
266
self.upload_transport = upload_transport
374
self.upload_transport = pack_collection._upload_transport
267
375
# where are indices written out to
268
self.index_transport = index_transport
376
self.index_transport = pack_collection._index_transport
269
377
# where is the pack renamed to when it is finished?
270
self.pack_transport = pack_transport
378
self.pack_transport = pack_collection._pack_transport
271
379
# What file mode to upload the pack and indices with.
272
380
self._file_mode = file_mode
273
381
# tracks the content written to the .pack file.
274
self._hash = md5.new()
275
# a four-tuple with the length in bytes of the indices, once the pack
276
# is finalised. (rev, inv, text, sigs)
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)
277
385
self.index_sizes = None
278
386
# How much data to cache when writing packs. Note that this is not
279
387
# synchronised with reads, because it's not in the transport layer, so
638
743
Sets self._text_filter appropriately.
640
# select inventory keys
641
inv_keys = self._revision_keys # currently the same keyspace, and note that
642
# querying for keys here could introduce a bug where an inventory item
643
# is missed, so do not change it to query separately without cross
644
# checking like the text key check below.
645
inventory_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
646
self.packs, 'inventory_index')[0]
647
inv_nodes = self._pack_collection._index_contents(inventory_index_map, inv_keys)
648
# copy inventory keys and adjust values
649
# XXX: Should be a helper function to allow different inv representation
651
self.pb.update("Copying inventory texts", 2)
652
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
653
# Only grab the output lines if we will be processing them
654
output_lines = bool(self.revision_ids)
655
inv_lines = self._copy_nodes_graph(inventory_index_map,
656
self.new_pack._writer, self.new_pack.inventory_index,
657
readv_group_iter, total_items, output_lines=output_lines)
658
if self.revision_ids:
659
self._process_inventory_lines(inv_lines)
661
# eat the iterator to cause it to execute.
663
self._text_filter = None
664
if 'pack' in debug.debug_flags:
665
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
666
time.ctime(), self._pack_collection._upload_transport.base,
667
self.new_pack.random_name,
668
self.new_pack.inventory_index.key_count(),
669
time.time() - self.new_pack.start_time)
745
raise NotImplementedError(self._copy_inventory_texts)
671
747
def _copy_text_texts(self):
673
text_index_map, text_nodes = self._get_text_nodes()
674
if self._text_filter is not None:
675
# We could return the keys copied as part of the return value from
676
# _copy_nodes_graph but this doesn't work all that well with the
677
# need to get line output too, so we check separately, and as we're
678
# going to buffer everything anyway, we check beforehand, which
679
# saves reading knit data over the wire when we know there are
681
text_nodes = set(text_nodes)
682
present_text_keys = set(_node[1] for _node in text_nodes)
683
missing_text_keys = set(self._text_filter) - present_text_keys
684
if missing_text_keys:
685
# TODO: raise a specific error that can handle many missing
687
a_missing_key = missing_text_keys.pop()
688
raise errors.RevisionNotPresent(a_missing_key[1],
690
# copy text keys and adjust values
691
self.pb.update("Copying content texts", 3)
692
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
693
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
694
self.new_pack.text_index, readv_group_iter, total_items))
695
self._log_copied_texts()
697
def _check_references(self):
698
"""Make sure our external refereneces are present."""
699
external_refs = self.new_pack._external_compression_parents_of_texts()
701
index = self._pack_collection.text_index.combined_index
702
found_items = list(index.iter_entries(external_refs))
703
if len(found_items) != len(external_refs):
704
found_keys = set(k for idx, k, refs, value in found_items)
705
missing_items = external_refs - found_keys
706
missing_file_id, missing_revision_id = missing_items.pop()
707
raise errors.RevisionNotPresent(missing_revision_id,
748
raise NotImplementedError(self._copy_text_texts)
710
750
def _create_pack_from_packs(self):
711
self.pb.update("Opening pack", 0, 5)
712
self.new_pack = self.open_pack()
713
new_pack = self.new_pack
714
# buffer data - we won't be reading-back during the pack creation and
715
# this makes a significant difference on sftp pushes.
716
new_pack.set_write_cache_size(1024*1024)
717
if 'pack' in debug.debug_flags:
718
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
719
for a_pack in self.packs]
720
if self.revision_ids is not None:
721
rev_count = len(self.revision_ids)
724
mutter('%s: create_pack: creating pack from source packs: '
725
'%s%s %s revisions wanted %s t=0',
726
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
727
plain_pack_list, rev_count)
728
self._copy_revision_texts()
729
self._copy_inventory_texts()
730
self._copy_text_texts()
731
# select signature keys
732
signature_filter = self._revision_keys # same keyspace
733
signature_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
734
self.packs, 'signature_index')[0]
735
signature_nodes = self._pack_collection._index_contents(signature_index_map,
737
# copy signature keys and adjust values
738
self.pb.update("Copying signature texts", 4)
739
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
740
new_pack.signature_index)
741
if 'pack' in debug.debug_flags:
742
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
743
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
744
new_pack.signature_index.key_count(),
745
time.time() - new_pack.start_time)
746
self._check_references()
747
if not self._use_pack(new_pack):
750
self.pb.update("Finishing pack", 5)
752
self._pack_collection.allocate(new_pack)
755
def _copy_nodes(self, nodes, index_map, writer, write_index):
756
"""Copy knit nodes between packs with no graph references."""
757
pb = ui.ui_factory.nested_progress_bar()
759
return self._do_copy_nodes(nodes, index_map, writer,
764
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb):
765
# for record verification
766
knit_data = _KnitData(None)
767
# plan a readv on each source pack:
769
nodes = sorted(nodes)
770
# how to map this into knit.py - or knit.py into this?
771
# we don't want the typical knit logic, we want grouping by pack
772
# at this point - perhaps a helper library for the following code
773
# duplication points?
775
for index, key, value in nodes:
776
if index not in request_groups:
777
request_groups[index] = []
778
request_groups[index].append((key, value))
780
pb.update("Copied record", record_index, len(nodes))
781
for index, items in request_groups.iteritems():
782
pack_readv_requests = []
783
for key, value in items:
784
# ---- KnitGraphIndex.get_position
785
bits = value[1:].split(' ')
786
offset, length = int(bits[0]), int(bits[1])
787
pack_readv_requests.append((offset, length, (key, value[0])))
788
# linear scan up the pack
789
pack_readv_requests.sort()
791
transport, path = index_map[index]
792
reader = pack.make_readv_reader(transport, path,
793
[offset[0:2] for offset in pack_readv_requests])
794
for (names, read_func), (_1, _2, (key, eol_flag)) in \
795
izip(reader.iter_records(), pack_readv_requests):
796
raw_data = read_func(None)
797
# check the header only
798
df, _ = knit_data._parse_record_header(key[-1], raw_data)
800
pos, size = writer.add_bytes_record(raw_data, names)
801
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
802
pb.update("Copied record", record_index)
805
def _copy_nodes_graph(self, index_map, writer, write_index,
806
readv_group_iter, total_items, output_lines=False):
807
"""Copy knit nodes between packs.
809
:param output_lines: Return lines present in the copied data as
810
an iterator of line,version_id.
812
pb = ui.ui_factory.nested_progress_bar()
814
for result in self._do_copy_nodes_graph(index_map, writer,
815
write_index, output_lines, pb, readv_group_iter, total_items):
818
# Python 2.4 does not permit try:finally: in a generator.
824
def _do_copy_nodes_graph(self, index_map, writer, write_index,
825
output_lines, pb, readv_group_iter, total_items):
826
# for record verification
827
knit_data = _KnitData(None)
828
# for line extraction when requested (inventories only)
830
factory = knit.KnitPlainFactory()
832
pb.update("Copied record", record_index, total_items)
833
for index, readv_vector, node_vector in readv_group_iter:
835
transport, path = index_map[index]
836
reader = pack.make_readv_reader(transport, path, readv_vector)
837
for (names, read_func), (key, eol_flag, references) in \
838
izip(reader.iter_records(), node_vector):
839
raw_data = read_func(None)
842
# read the entire thing
843
content, _ = knit_data._parse_record(version_id, raw_data)
844
if len(references[-1]) == 0:
845
line_iterator = factory.get_fulltext_content(content)
847
line_iterator = factory.get_linedelta_content(content)
848
for line in line_iterator:
849
yield line, version_id
851
# check the header only
852
df, _ = knit_data._parse_record_header(version_id, raw_data)
854
pos, size = writer.add_bytes_record(raw_data, names)
855
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
856
pb.update("Copied record", record_index)
859
def _get_text_nodes(self):
860
text_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
861
self.packs, 'text_index')[0]
862
return text_index_map, self._pack_collection._index_contents(text_index_map,
865
def _least_readv_node_readv(self, nodes):
866
"""Generate request groups for nodes using the least readv's.
868
:param nodes: An iterable of graph index nodes.
869
:return: Total node count and an iterator of the data needed to perform
870
readvs to obtain the data for nodes. Each item yielded by the
871
iterator is a tuple with:
872
index, readv_vector, node_vector. readv_vector is a list ready to
873
hand to the transport readv method, and node_vector is a list of
874
(key, eol_flag, references) for the the node retrieved by the
875
matching readv_vector.
877
# group by pack so we do one readv per pack
878
nodes = sorted(nodes)
881
for index, key, value, references in nodes:
882
if index not in request_groups:
883
request_groups[index] = []
884
request_groups[index].append((key, value, references))
886
for index, items in request_groups.iteritems():
887
pack_readv_requests = []
888
for key, value, references in items:
889
# ---- KnitGraphIndex.get_position
890
bits = value[1:].split(' ')
891
offset, length = int(bits[0]), int(bits[1])
892
pack_readv_requests.append(
893
((offset, length), (key, value[0], references)))
894
# linear scan up the pack to maximum range combining.
895
pack_readv_requests.sort()
896
# split out the readv and the node data.
897
pack_readv = [readv for readv, node in pack_readv_requests]
898
node_vector = [node for readv, node in pack_readv_requests]
899
result.append((index, pack_readv, node_vector))
751
raise NotImplementedError(self._create_pack_from_packs)
902
753
def _log_copied_texts(self):
903
754
if 'pack' in debug.debug_flags:
935
767
return new_pack.data_inserted()
938
class OptimisingPacker(Packer):
939
"""A packer which spends more time to create better disk layouts."""
941
def _revision_node_readv(self, revision_nodes):
942
"""Return the total revisions and the readv's to issue.
944
This sort places revisions in topological order with the ancestors
947
:param revision_nodes: The revision index contents for the packs being
948
incorporated into the new pack.
949
:return: As per _least_readv_node_readv.
951
# build an ancestors dict
954
for index, key, value, references in revision_nodes:
955
ancestors[key] = references[0]
956
by_key[key] = (index, value, references)
957
order = tsort.topo_sort(ancestors)
959
# Single IO is pathological, but it will work as a starting point.
961
for key in reversed(order):
962
index, value, references = by_key[key]
963
# ---- KnitGraphIndex.get_position
964
bits = value[1:].split(' ')
965
offset, length = int(bits[0]), int(bits[1])
967
(index, [(offset, length)], [(key, value[0], references)]))
968
# TODO: combine requests in the same index that are in ascending order.
969
return total, requests
972
class ReconcilePacker(Packer):
973
"""A packer which regenerates indices etc as it copies.
975
This is used by ``bzr reconcile`` to cause parent text pointers to be
979
def _extra_init(self):
980
self._data_changed = False
982
def _process_inventory_lines(self, inv_lines):
983
"""Generate a text key reference map rather for reconciling with."""
984
repo = self._pack_collection.repo
985
refs = repo._find_text_key_references_from_xml_inventory_lines(
987
self._text_refs = refs
988
# during reconcile we:
989
# - convert unreferenced texts to full texts
990
# - correct texts which reference a text not copied to be full texts
991
# - copy all others as-is but with corrected parents.
992
# - so at this point we don't know enough to decide what becomes a full
994
self._text_filter = None
996
def _copy_text_texts(self):
997
"""generate what texts we should have and then copy."""
998
self.pb.update("Copying content texts", 3)
999
# we have three major tasks here:
1000
# 1) generate the ideal index
1001
repo = self._pack_collection.repo
1002
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1003
_1, key, _2, refs in
1004
self.new_pack.revision_index.iter_all_entries()])
1005
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1006
# 2) generate a text_nodes list that contains all the deltas that can
1007
# be used as-is, with corrected parents.
1010
discarded_nodes = []
1011
NULL_REVISION = _mod_revision.NULL_REVISION
1012
text_index_map, text_nodes = self._get_text_nodes()
1013
for node in text_nodes:
1019
ideal_parents = tuple(ideal_index[node[1]])
1021
discarded_nodes.append(node)
1022
self._data_changed = True
1024
if ideal_parents == (NULL_REVISION,):
1026
if ideal_parents == node[3][0]:
1028
ok_nodes.append(node)
1029
elif ideal_parents[0:1] == node[3][0][0:1]:
1030
# the left most parent is the same, or there are no parents
1031
# today. Either way, we can preserve the representation as
1032
# long as we change the refs to be inserted.
1033
self._data_changed = True
1034
ok_nodes.append((node[0], node[1], node[2],
1035
(ideal_parents, node[3][1])))
1036
self._data_changed = True
1038
# Reinsert this text completely
1039
bad_texts.append((node[1], ideal_parents))
1040
self._data_changed = True
1041
# we're finished with some data.
1044
# 3) bulk copy the ok data
1045
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1046
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1047
self.new_pack.text_index, readv_group_iter, total_items))
1048
# 4) adhoc copy all the other texts.
1049
# We have to topologically insert all texts otherwise we can fail to
1050
# reconcile when parts of a single delta chain are preserved intact,
1051
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1052
# reinserted, and if d3 has incorrect parents it will also be
1053
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1054
# copied), so we will try to delta, but d2 is not currently able to be
1055
# extracted because it's basis d1 is not present. Topologically sorting
1056
# addresses this. The following generates a sort for all the texts that
1057
# are being inserted without having to reference the entire text key
1058
# space (we only topo sort the revisions, which is smaller).
1059
topo_order = tsort.topo_sort(ancestors)
1060
rev_order = dict(zip(topo_order, range(len(topo_order))))
1061
bad_texts.sort(key=lambda key:rev_order[key[0][1]])
1062
transaction = repo.get_transaction()
1063
file_id_index = GraphIndexPrefixAdapter(
1064
self.new_pack.text_index,
1066
add_nodes_callback=self.new_pack.text_index.add_nodes)
1067
knit_index = KnitGraphIndex(file_id_index,
1068
add_callback=file_id_index.add_nodes,
1069
deltas=True, parents=True)
1070
output_knit = knit.KnitVersionedFile('reconcile-texts',
1071
self._pack_collection.transport,
1073
access_method=_PackAccess(
1074
{self.new_pack.text_index:self.new_pack.access_tuple()},
1075
(self.new_pack._writer, self.new_pack.text_index)),
1076
factory=knit.KnitPlainFactory())
1077
for key, parent_keys in bad_texts:
1078
# We refer to the new pack to delta data being output.
1079
# A possible improvement would be to catch errors on short reads
1080
# and only flush then.
1081
self.new_pack.flush()
1083
for parent_key in parent_keys:
1084
if parent_key[0] != key[0]:
1085
# Graph parents must match the fileid
1086
raise errors.BzrError('Mismatched key parent %r:%r' %
1088
parents.append(parent_key[1])
1089
source_weave = repo.weave_store.get_weave(key[0], transaction)
1090
text_lines = source_weave.get_lines(key[1])
1091
# adapt the 'knit' to the current file_id.
1092
file_id_index = GraphIndexPrefixAdapter(
1093
self.new_pack.text_index,
1095
add_nodes_callback=self.new_pack.text_index.add_nodes)
1096
knit_index._graph_index = file_id_index
1097
knit_index._add_callback = file_id_index.add_nodes
1098
output_knit.add_lines_with_ghosts(
1099
key[1], parents, text_lines, random_id=True, check_content=False)
1100
# 5) check that nothing inserted has a reference outside the keyspace.
1101
missing_text_keys = self.new_pack._external_compression_parents_of_texts()
1102
if missing_text_keys:
1103
raise errors.BzrError('Reference to missing compression parents %r'
1104
% (missing_text_keys,))
1105
self._log_copied_texts()
1107
def _use_pack(self, new_pack):
1108
"""Override _use_pack to check for reconcile having changed content."""
1109
# XXX: we might be better checking this at the copy time.
1110
original_inventory_keys = set()
1111
inv_index = self._pack_collection.inventory_index.combined_index
1112
for entry in inv_index.iter_all_entries():
1113
original_inventory_keys.add(entry[1])
1114
new_inventory_keys = set()
1115
for entry in new_pack.inventory_index.iter_all_entries():
1116
new_inventory_keys.add(entry[1])
1117
if new_inventory_keys != original_inventory_keys:
1118
self._data_changed = True
1119
return new_pack.data_inserted() and self._data_changed
1122
770
class RepositoryPackCollection(object):
1123
"""Management of packs within a repository."""
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
1125
781
def __init__(self, repo, transport, index_transport, upload_transport,
782
pack_transport, index_builder_class, index_class,
1127
784
"""Create a new RepositoryPackCollection.
1129
:param transport: Addresses the repository base directory
786
:param transport: Addresses the repository base directory
1130
787
(typically .bzr/repository/).
1131
788
:param index_transport: Addresses the directory containing indices.
1132
789
:param upload_transport: Addresses the directory into which packs are written
1133
790
while they're being created.
1134
791
: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()
1136
797
self.repo = repo
1137
798
self.transport = transport
1138
799
self._index_transport = index_transport
1139
800
self._upload_transport = upload_transport
1140
801
self._pack_transport = pack_transport
1141
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
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,
1143
807
# name:Pack mapping
1144
809
self._packs_by_name = {}
1145
810
# the previous pack-names content
1146
811
self._packs_at_load = None
1147
812
# when a pack is being created by this object, the state of that pack.
1148
813
self._new_pack = None
1149
814
# aggregated revision index data
1150
self.revision_index = AggregateIndex()
1151
self.inventory_index = AggregateIndex()
1152
self.text_index = AggregateIndex()
1153
self.signature_index = AggregateIndex()
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)
1155
841
def add_pack_to_memory(self, pack):
1156
842
"""Make a Pack object available to the repository to satisfy queries.
1158
844
:param pack: A Pack object.
1160
846
if pack.name in self._packs_by_name:
1161
raise AssertionError()
847
raise AssertionError(
848
'pack %s already in _packs_by_name' % (pack.name,))
1162
849
self.packs.append(pack)
1163
850
self._packs_by_name[pack.name] = pack
1164
851
self.revision_index.add_index(pack.revision_index, pack)
1165
852
self.inventory_index.add_index(pack.inventory_index, pack)
1166
853
self.text_index.add_index(pack.text_index, pack)
1167
854
self.signature_index.add_index(pack.signature_index, pack)
1169
def _add_text_to_weave(self, file_id, revision_id, new_lines, parents,
1170
nostore_sha, random_revid):
1171
file_id_index = GraphIndexPrefixAdapter(
1172
self.text_index.combined_index,
1174
add_nodes_callback=self.text_index.add_callback)
1175
self.repo._text_knit._index._graph_index = file_id_index
1176
self.repo._text_knit._index._add_callback = file_id_index.add_nodes
1177
return self.repo._text_knit.add_lines_with_ghosts(
1178
revision_id, parents, new_lines, nostore_sha=nostore_sha,
1179
random_id=random_revid, check_content=False)[0:2]
855
if self.chk_index is not None:
856
self.chk_index.add_index(pack.chk_index, pack)
1181
858
def all_packs(self):
1182
859
"""Return a list of all the Pack objects this repository has.
1229
910
# group their data with the relevant commit, and that may
1230
911
# involve rewriting ancient history - which autopack tries to
1231
912
# avoid. Alternatively we could not group the data but treat
1232
# each of these as having a single revision, and thus add
913
# each of these as having a single revision, and thus add
1233
914
# one revision for each to the total revision count, to get
1234
915
# a matching distribution.
1236
917
existing_packs.append((revision_count, pack))
1237
918
pack_operations = self.plan_autopack_combinations(
1238
919
existing_packs, pack_distribution)
1239
self._execute_pack_operations(pack_operations)
920
num_new_packs = len(pack_operations)
921
num_old_packs = sum([len(po[1]) for po in pack_operations])
922
num_revs_affected = sum([po[0] for po in pack_operations])
923
mutter('Auto-packing repository %s, which has %d pack files, '
924
'containing %d revisions. Packing %d files into %d affecting %d'
925
' revisions', self, total_packs, total_revisions, num_old_packs,
926
num_new_packs, num_revs_affected)
927
result = self._execute_pack_operations(pack_operations, packer_class=self.normal_packer_class,
928
reload_func=self._restart_autopack)
929
mutter('Auto-packing repository %s completed', self)
1242
def _execute_pack_operations(self, pack_operations, _packer_class=Packer):
932
def _execute_pack_operations(self, pack_operations, packer_class,
1243
934
"""Execute a series of pack operations.
1245
936
:param pack_operations: A list of [revision_count, packs_to_combine].
1246
:param _packer_class: The class of packer to use (default: Packer).
937
:param packer_class: The class of packer to use
938
:return: The new pack names.
1249
940
for revision_count, packs in pack_operations:
1250
941
# we may have no-ops from the setup logic
1251
942
if len(packs) == 0:
1253
_packer_class(self, packs, '.autopack').pack()
944
packer = packer_class(self, packs, '.autopack',
945
reload_func=reload_func)
947
result = packer.pack()
948
except errors.RetryWithNewPacks:
949
# An exception is propagating out of this context, make sure
950
# this packer has cleaned up. Packer() doesn't set its new_pack
951
# state into the RepositoryPackCollection object, so we only
952
# have access to it directly here.
953
if packer.new_pack is not None:
954
packer.new_pack.abort()
1254
958
for pack in packs:
1255
959
self._remove_pack_from_memory(pack)
1256
960
# record the newly available packs and stop advertising the old
1258
self._save_pack_names(clear_obsolete_packs=True)
1259
# Move the old packs out of the way now they are no longer referenced.
1260
for revision_count, packs in pack_operations:
1261
self._obsolete_packs(packs)
963
for _, packs in pack_operations:
964
to_be_obsoleted.extend(packs)
965
result = self._save_pack_names(clear_obsolete_packs=True,
966
obsolete_packs=to_be_obsoleted)
969
def _flush_new_pack(self):
970
if self._new_pack is not None:
971
self._new_pack.flush()
1263
973
def lock_names(self):
1264
974
"""Acquire the mutex around the pack-names index.
1266
976
This cannot be used in the middle of a read-only transaction on the
1269
979
self.repo.control_files.lock_write()
981
def _already_packed(self):
982
"""Is the collection already packed?"""
983
return not (self.repo._format.pack_compresses or (len(self._names) > 1))
985
def pack(self, hint=None, clean_obsolete_packs=False):
1272
986
"""Pack the pack collection totally."""
1273
987
self.ensure_loaded()
1274
988
total_packs = len(self._names)
1276
# This is arguably wrong because we might not be optimal, but for
1277
# now lets leave it in. (e.g. reconcile -> one pack. But not
989
if self._already_packed():
1280
991
total_revisions = self.revision_index.combined_index.key_count()
1281
992
# XXX: the following may want to be a class, to pack with a given
1283
994
mutter('Packing repository %s, which has %d pack files, '
1284
'containing %d revisions into 1 packs.', self, total_packs,
995
'containing %d revisions with hint %r.', self, total_packs,
996
total_revisions, hint)
999
self._try_pack_operations(hint)
1000
except RetryPackOperations:
1004
if clean_obsolete_packs:
1005
self._clear_obsolete_packs()
1007
def _try_pack_operations(self, hint):
1008
"""Calculate the pack operations based on the hint (if any), and
1286
1011
# determine which packs need changing
1287
pack_distribution = [1]
1288
1012
pack_operations = [[0, []]]
1289
1013
for pack in self.all_packs():
1290
pack_operations[-1][0] += pack.get_revision_count()
1291
pack_operations[-1][1].append(pack)
1292
self._execute_pack_operations(pack_operations, OptimisingPacker)
1014
if hint is None or pack.name in hint:
1015
# Either no hint was provided (so we are packing everything),
1016
# or this pack was included in the hint.
1017
pack_operations[-1][0] += pack.get_revision_count()
1018
pack_operations[-1][1].append(pack)
1019
self._execute_pack_operations(pack_operations,
1020
packer_class=self.optimising_packer_class,
1021
reload_func=self._restart_pack_operations)
1294
1023
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1295
1024
"""Plan a pack operation.
1498
1321
self._packs_by_name = {}
1499
1322
self._packs_at_load = None
1501
def _make_index_map(self, index_suffix):
1502
"""Return information on existing indices.
1504
:param suffix: Index suffix added to pack name.
1506
:returns: (pack_map, indices) where indices is a list of GraphIndex
1507
objects, and pack_map is a mapping from those objects to the
1508
pack tuple they describe.
1510
# TODO: stop using this; it creates new indices unnecessarily.
1511
self.ensure_loaded()
1512
suffix_map = {'.rix': 'revision_index',
1513
'.six': 'signature_index',
1514
'.iix': 'inventory_index',
1515
'.tix': 'text_index',
1517
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1518
suffix_map[index_suffix])
1520
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1521
"""Convert a list of packs to an index pack map and index list.
1523
:param packs: The packs list to process.
1524
:param index_attribute: The attribute that the desired index is found
1526
:return: A tuple (map, list) where map contains the dict from
1527
index:pack_tuple, and lsit contains the indices in the same order
1533
index = getattr(pack, index_attribute)
1534
indices.append(index)
1535
pack_map[index] = (pack.pack_transport, pack.file_name())
1536
return pack_map, indices
1538
def _index_contents(self, pack_map, key_filter=None):
1539
"""Get an iterable of the index contents from a pack_map.
1541
:param pack_map: A map from indices to pack details.
1542
:param key_filter: An optional filter to limit the
1545
indices = [index for index in pack_map.iterkeys()]
1546
all_index = CombinedGraphIndex(indices)
1547
if key_filter is None:
1548
return all_index.iter_all_entries()
1550
return all_index.iter_entries(key_filter)
1552
1324
def _unlock_names(self):
1553
1325
"""Release the mutex around the pack-names index."""
1554
1326
self.repo.control_files.unlock()
1556
def _save_pack_names(self, clear_obsolete_packs=False):
1557
"""Save the list of packs.
1559
This will take out the mutex around the pack names list for the
1560
duration of the method call. If concurrent updates have been made, a
1561
three-way merge between the current list and the current in memory list
1564
:param clear_obsolete_packs: If True, clear out the contents of the
1565
obsolete_packs directory.
1569
builder = GraphIndexBuilder()
1570
# load the disk nodes across
1572
for index, key, value in self._iter_disk_pack_index():
1573
disk_nodes.add((key, value))
1574
# do a two-way diff against our original content
1575
current_nodes = set()
1576
for name, sizes in self._names.iteritems():
1578
((name, ), ' '.join(str(size) for size in sizes)))
1579
deleted_nodes = self._packs_at_load - current_nodes
1580
new_nodes = current_nodes - self._packs_at_load
1581
disk_nodes.difference_update(deleted_nodes)
1582
disk_nodes.update(new_nodes)
1583
# TODO: handle same-name, index-size-changes here -
1584
# e.g. use the value from disk, not ours, *unless* we're the one
1586
for key, value in disk_nodes:
1587
builder.add_node(key, value)
1588
self.transport.put_file('pack-names', builder.finish(),
1589
mode=self.repo.control_files._file_mode)
1590
# move the baseline forward
1591
self._packs_at_load = disk_nodes
1592
# now clear out the obsolete packs directory
1593
if clear_obsolete_packs:
1594
self.transport.clone('obsolete_packs').delete_multi(
1595
self.transport.list_dir('obsolete_packs'))
1597
self._unlock_names()
1598
# synchronise the memory packs list with what we just wrote:
1328
def _diff_pack_names(self):
1329
"""Read the pack names from disk, and compare it to the one in memory.
1331
:return: (disk_nodes, deleted_nodes, new_nodes)
1332
disk_nodes The final set of nodes that should be referenced
1333
deleted_nodes Nodes which have been removed from when we started
1334
new_nodes Nodes that are newly introduced
1336
# load the disk nodes across
1338
for index, key, value in self._iter_disk_pack_index():
1339
disk_nodes.add((key, value))
1340
orig_disk_nodes = set(disk_nodes)
1342
# do a two-way diff against our original content
1343
current_nodes = set()
1344
for name, sizes in self._names.iteritems():
1346
((name, ), ' '.join(str(size) for size in sizes)))
1348
# Packs no longer present in the repository, which were present when we
1349
# locked the repository
1350
deleted_nodes = self._packs_at_load - current_nodes
1351
# Packs which this process is adding
1352
new_nodes = current_nodes - self._packs_at_load
1354
# Update the disk_nodes set to include the ones we are adding, and
1355
# remove the ones which were removed by someone else
1356
disk_nodes.difference_update(deleted_nodes)
1357
disk_nodes.update(new_nodes)
1359
return disk_nodes, deleted_nodes, new_nodes, orig_disk_nodes
1361
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1362
"""Given the correct set of pack files, update our saved info.
1364
:return: (removed, added, modified)
1365
removed pack names removed from self._names
1366
added pack names added to self._names
1367
modified pack names that had changed value
1372
## self._packs_at_load = disk_nodes
1599
1373
new_names = dict(disk_nodes)
1600
1374
# drop no longer present nodes
1601
1375
for pack in self.all_packs():
1602
1376
if (pack.name,) not in new_names:
1377
removed.append(pack.name)
1603
1378
self._remove_pack_from_memory(pack)
1604
1379
# add new nodes/refresh existing ones
1605
1380
for key, value in disk_nodes:
1615
1390
# disk index because the set values are the same, unless
1616
1391
# the only index shows up as deleted by the set difference
1617
1392
# - which it may. Until there is a specific test for this,
1618
# assume its broken. RBC 20071017.
1393
# assume it's broken. RBC 20071017.
1619
1394
self._remove_pack_from_memory(self.get_pack_by_name(name))
1620
1395
self._names[name] = sizes
1621
1396
self.get_pack_by_name(name)
1397
modified.append(name)
1624
1400
self._names[name] = sizes
1625
1401
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"
1627
1526
def _start_write_group(self):
1628
1527
# Do not permit preparation for writing if we're not in a 'write lock'.
1629
1528
if not self.repo.is_write_locked():
1630
1529
raise errors.NotWriteLocked(self)
1631
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1632
self._pack_transport, upload_suffix='.pack',
1633
file_mode=self.repo.control_files._file_mode)
1530
self._new_pack = self.pack_factory(self, upload_suffix='.pack',
1531
file_mode=self.repo.bzrdir._get_file_mode())
1634
1532
# allow writing: queue writes to a new index
1635
1533
self.revision_index.add_writable_index(self._new_pack.revision_index,
1636
1534
self._new_pack)
1638
1536
self._new_pack)
1639
1537
self.text_index.add_writable_index(self._new_pack.text_index,
1640
1538
self._new_pack)
1539
self._new_pack.text_index.set_optimize(combine_backing_indices=False)
1641
1540
self.signature_index.add_writable_index(self._new_pack.signature_index,
1642
1541
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)
1644
# reused revision and signature knits may need updating
1646
# "Hysterical raisins. client code in bzrlib grabs those knits outside
1647
# of write groups and then mutates it inside the write group."
1648
if self.repo._revision_knit is not None:
1649
self.repo._revision_knit._index._add_callback = \
1650
self.revision_index.add_callback
1651
if self.repo._signature_knit is not None:
1652
self.repo._signature_knit._index._add_callback = \
1653
self.signature_index.add_callback
1654
# create a reused knit object for text addition in commit.
1655
self.repo._text_knit = self.repo.weave_store.get_weave_or_empty(
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
1658
1553
def _abort_write_group(self):
1659
1554
# FIXME: just drop the transient index.
1660
1555
# forget what names there are
1661
1556
if self._new_pack is not None:
1662
self._new_pack.abort()
1663
self._remove_pack_indices(self._new_pack)
1664
self._new_pack = None
1665
self.repo._text_knit = None
1557
operation = cleanup.OperationWithCleanups(self._new_pack.abort)
1558
operation.add_cleanup(setattr, self, '_new_pack', None)
1559
# If we aborted while in the middle of finishing the write
1560
# group, _remove_pack_indices could fail because the indexes are
1561
# already gone. But they're not there we shouldn't fail in this
1562
# case, so we pass ignore_missing=True.
1563
operation.add_cleanup(self._remove_pack_indices, self._new_pack,
1564
ignore_missing=True)
1565
operation.run_simple()
1566
for resumed_pack in self._resumed_packs:
1567
operation = cleanup.OperationWithCleanups(resumed_pack.abort)
1568
# See comment in previous finally block.
1569
operation.add_cleanup(self._remove_pack_indices, resumed_pack,
1570
ignore_missing=True)
1571
operation.run_simple()
1572
del self._resumed_packs[:]
1574
def _remove_resumed_pack_indices(self):
1575
for resumed_pack in self._resumed_packs:
1576
self._remove_pack_indices(resumed_pack)
1577
del self._resumed_packs[:]
1579
def _check_new_inventories(self):
1580
"""Detect missing inventories in this write group.
1582
:returns: list of strs, summarising any problems found. If the list is
1583
empty no problems were found.
1585
# The base implementation does no checks. GCRepositoryPackCollection
1667
1589
def _commit_write_group(self):
1591
for prefix, versioned_file in (
1592
('revisions', self.repo.revisions),
1593
('inventories', self.repo.inventories),
1594
('texts', self.repo.texts),
1595
('signatures', self.repo.signatures),
1597
missing = versioned_file.get_missing_compression_parent_keys()
1598
all_missing.update([(prefix,) + key for key in missing])
1600
raise errors.BzrCheckError(
1601
"Repository %s has missing compression parent(s) %r "
1602
% (self.repo, sorted(all_missing)))
1603
problems = self._check_new_inventories()
1605
problems_summary = '\n'.join(problems)
1606
raise errors.BzrCheckError(
1607
"Cannot add revision(s) to repository: " + problems_summary)
1668
1608
self._remove_pack_indices(self._new_pack)
1609
any_new_content = False
1669
1610
if self._new_pack.data_inserted():
1670
1611
# get all the data to disk and read to use
1671
1612
self._new_pack.finish()
1672
1613
self.allocate(self._new_pack)
1673
1614
self._new_pack = None
1674
if not self.autopack():
1615
any_new_content = True
1617
self._new_pack.abort()
1618
self._new_pack = None
1619
for resumed_pack in self._resumed_packs:
1620
# XXX: this is a pretty ugly way to turn the resumed pack into a
1621
# properly committed pack.
1622
self._names[resumed_pack.name] = None
1623
self._remove_pack_from_memory(resumed_pack)
1624
resumed_pack.finish()
1625
self.allocate(resumed_pack)
1626
any_new_content = True
1627
del self._resumed_packs[:]
1629
result = self.autopack()
1675
1631
# when autopack takes no steps, the names list is still
1677
self._save_pack_names()
1633
return self._save_pack_names()
1637
def _suspend_write_group(self):
1638
tokens = [pack.name for pack in self._resumed_packs]
1639
self._remove_pack_indices(self._new_pack)
1640
if self._new_pack.data_inserted():
1641
# get all the data to disk and read to use
1642
self._new_pack.finish(suspend=True)
1643
tokens.append(self._new_pack.name)
1644
self._new_pack = None
1679
1646
self._new_pack.abort()
1680
1647
self._new_pack = None
1681
self.repo._text_knit = None
1684
class KnitPackRevisionStore(KnitRevisionStore):
1685
"""An object to adapt access from RevisionStore's to use KnitPacks.
1687
This class works by replacing the original RevisionStore.
1688
We need to do this because the KnitPackRevisionStore is less
1689
isolated in its layering - it uses services from the repo.
1692
def __init__(self, repo, transport, revisionstore):
1693
"""Create a KnitPackRevisionStore on repo with revisionstore.
1695
This will store its state in the Repository, use the
1696
indices to provide a KnitGraphIndex,
1697
and at the end of transactions write new indices.
1699
KnitRevisionStore.__init__(self, revisionstore.versioned_file_store)
1701
self._serializer = revisionstore._serializer
1702
self.transport = transport
1704
def get_revision_file(self, transaction):
1705
"""Get the revision versioned file object."""
1706
if getattr(self.repo, '_revision_knit', None) is not None:
1707
return self.repo._revision_knit
1708
self.repo._pack_collection.ensure_loaded()
1709
add_callback = self.repo._pack_collection.revision_index.add_callback
1710
# setup knit specific objects
1711
knit_index = KnitGraphIndex(
1712
self.repo._pack_collection.revision_index.combined_index,
1713
add_callback=add_callback)
1714
self.repo._revision_knit = knit.KnitVersionedFile(
1715
'revisions', self.transport.clone('..'),
1716
self.repo.control_files._file_mode,
1718
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1719
access_method=self.repo._pack_collection.revision_index.knit_access)
1720
return self.repo._revision_knit
1722
def get_signature_file(self, transaction):
1723
"""Get the signature versioned file object."""
1724
if getattr(self.repo, '_signature_knit', None) is not None:
1725
return self.repo._signature_knit
1726
self.repo._pack_collection.ensure_loaded()
1727
add_callback = self.repo._pack_collection.signature_index.add_callback
1728
# setup knit specific objects
1729
knit_index = KnitGraphIndex(
1730
self.repo._pack_collection.signature_index.combined_index,
1731
add_callback=add_callback, parents=False)
1732
self.repo._signature_knit = knit.KnitVersionedFile(
1733
'signatures', self.transport.clone('..'),
1734
self.repo.control_files._file_mode,
1736
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1737
access_method=self.repo._pack_collection.signature_index.knit_access)
1738
return self.repo._signature_knit
1741
class KnitPackTextStore(VersionedFileStore):
1742
"""Presents a TextStore abstraction on top of packs.
1744
This class works by replacing the original VersionedFileStore.
1745
We need to do this because the KnitPackRevisionStore is less
1746
isolated in its layering - it uses services from the repo and shares them
1747
with all the data written in a single write group.
1750
def __init__(self, repo, transport, weavestore):
1751
"""Create a KnitPackTextStore on repo with weavestore.
1753
This will store its state in the Repository, use the
1754
indices FileNames to provide a KnitGraphIndex,
1755
and at the end of transactions write new indices.
1757
# don't call base class constructor - it's not suitable.
1758
# no transient data stored in the transaction
1760
self._precious = False
1762
self.transport = transport
1763
self.weavestore = weavestore
1764
# XXX for check() which isn't updated yet
1765
self._transport = weavestore._transport
1767
def get_weave_or_empty(self, file_id, transaction):
1768
"""Get a 'Knit' backed by the .tix indices.
1770
The transaction parameter is ignored.
1772
self.repo._pack_collection.ensure_loaded()
1773
add_callback = self.repo._pack_collection.text_index.add_callback
1774
# setup knit specific objects
1775
file_id_index = GraphIndexPrefixAdapter(
1776
self.repo._pack_collection.text_index.combined_index,
1777
(file_id, ), 1, add_nodes_callback=add_callback)
1778
knit_index = KnitGraphIndex(file_id_index,
1779
add_callback=file_id_index.add_nodes,
1780
deltas=True, parents=True)
1781
return knit.KnitVersionedFile('text:' + file_id,
1782
self.transport.clone('..'),
1785
access_method=self.repo._pack_collection.text_index.knit_access,
1786
factory=knit.KnitPlainFactory())
1788
get_weave = get_weave_or_empty
1791
"""Generate a list of the fileids inserted, for use by check."""
1792
self.repo._pack_collection.ensure_loaded()
1794
for index, key, value, refs in \
1795
self.repo._pack_collection.text_index.combined_index.iter_all_entries():
1800
class InventoryKnitThunk(object):
1801
"""An object to manage thunking get_inventory_weave to pack based knits."""
1803
def __init__(self, repo, transport):
1804
"""Create an InventoryKnitThunk for repo at transport.
1806
This will store its state in the Repository, use the
1807
indices FileNames to provide a KnitGraphIndex,
1808
and at the end of transactions write a new index..
1811
self.transport = transport
1813
def get_weave(self):
1814
"""Get a 'Knit' that contains inventory data."""
1815
self.repo._pack_collection.ensure_loaded()
1816
add_callback = self.repo._pack_collection.inventory_index.add_callback
1817
# setup knit specific objects
1818
knit_index = KnitGraphIndex(
1819
self.repo._pack_collection.inventory_index.combined_index,
1820
add_callback=add_callback, deltas=True, parents=True)
1821
return knit.KnitVersionedFile(
1822
'inventory', self.transport.clone('..'),
1823
self.repo.control_files._file_mode,
1825
index=knit_index, delta=True, factory=knit.KnitPlainFactory(),
1826
access_method=self.repo._pack_collection.inventory_index.knit_access)
1829
class KnitPackRepository(KnitRepository):
1830
"""Experimental graph-knit using repository."""
1832
def __init__(self, _format, a_bzrdir, control_files, _revision_store,
1833
control_store, text_store, _commit_builder_class, _serializer):
1834
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
1835
_revision_store, control_store, text_store, _commit_builder_class,
1837
index_transport = control_files._transport.clone('indices')
1838
self._pack_collection = RepositoryPackCollection(self, control_files._transport,
1840
control_files._transport.clone('upload'),
1841
control_files._transport.clone('packs'))
1842
self._revision_store = KnitPackRevisionStore(self, index_transport, self._revision_store)
1843
self.weave_store = KnitPackTextStore(self, index_transport, self.weave_store)
1844
self._inv_thunk = InventoryKnitThunk(self, index_transport)
1845
# True when the repository object is 'write locked' (as opposed to the
1846
# physical lock only taken out around changes to the pack-names list.)
1847
# Another way to represent this would be a decorator around the control
1848
# files object that presents logical locks as physical ones - if this
1849
# gets ugly consider that alternative design. RBC 20071011
1850
self._write_lock_count = 0
1851
self._transaction = None
1853
self._reconcile_does_inventory_gc = True
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
1854
1693
self._reconcile_fixes_text_parents = True
1855
self._reconcile_backsup_inventory = False
1694
if self._format.supports_external_lookups:
1695
self._unstacked_provider = graph.CachingParentsProvider(
1696
self._make_parents_provider_unstacked())
1698
self._unstacked_provider = graph.CachingParentsProvider(self)
1699
self._unstacked_provider.disable_cache()
1702
def _all_revision_ids(self):
1703
"""See Repository.all_revision_ids()."""
1704
return [key[0] for key in self.revisions.keys()]
1857
1706
def _abort_write_group(self):
1707
self.revisions._index._key_dependencies.clear()
1858
1708
self._pack_collection._abort_write_group()
1860
def _find_inconsistent_revision_parents(self):
1861
"""Find revisions with incorrectly cached parents.
1863
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1864
parents-in-revision).
1866
if not self.is_locked():
1867
raise errors.ObjectNotLocked(self)
1868
pb = ui.ui_factory.nested_progress_bar()
1871
revision_nodes = self._pack_collection.revision_index \
1872
.combined_index.iter_all_entries()
1873
index_positions = []
1874
# Get the cached index values for all revisions, and also the location
1875
# in each index of the revision text so we can perform linear IO.
1876
for index, key, value, refs in revision_nodes:
1877
pos, length = value[1:].split(' ')
1878
index_positions.append((index, int(pos), key[0],
1879
tuple(parent[0] for parent in refs[0])))
1880
pb.update("Reading revision index.", 0, 0)
1881
index_positions.sort()
1882
batch_count = len(index_positions) / 1000 + 1
1883
pb.update("Checking cached revision graph.", 0, batch_count)
1884
for offset in xrange(batch_count):
1885
pb.update("Checking cached revision graph.", offset)
1886
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1889
rev_ids = [item[2] for item in to_query]
1890
revs = self.get_revisions(rev_ids)
1891
for revision, item in zip(revs, to_query):
1892
index_parents = item[3]
1893
rev_parents = tuple(revision.parent_ids)
1894
if index_parents != rev_parents:
1895
result.append((revision.revision_id, index_parents, rev_parents))
1900
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
1901
def get_parents(self, revision_ids):
1902
"""See graph._StackedParentsProvider.get_parents."""
1903
parent_map = self.get_parent_map(revision_ids)
1904
return [parent_map.get(r, None) for r in revision_ids]
1906
def get_parent_map(self, keys):
1907
"""See graph._StackedParentsProvider.get_parent_map
1909
This implementation accesses the combined revision index to provide
1912
self._pack_collection.ensure_loaded()
1913
index = self._pack_collection.revision_index.combined_index
1916
raise ValueError('get_parent_map(None) is not valid')
1917
if _mod_revision.NULL_REVISION in keys:
1918
keys.discard(_mod_revision.NULL_REVISION)
1919
found_parents = {_mod_revision.NULL_REVISION:()}
1922
search_keys = set((revision_id,) for revision_id in keys)
1923
for index, key, value, refs in index.iter_entries(search_keys):
1926
parents = (_mod_revision.NULL_REVISION,)
1928
parents = tuple(parent[0] for parent in parents)
1929
found_parents[key[0]] = parents
1930
return found_parents
1932
def has_revisions(self, revision_ids):
1933
"""See Repository.has_revisions()."""
1934
revision_ids = set(revision_ids)
1935
result = revision_ids.intersection(
1936
set([None, _mod_revision.NULL_REVISION]))
1937
revision_ids.difference_update(result)
1938
index = self._pack_collection.revision_index.combined_index
1939
keys = [(revision_id,) for revision_id in revision_ids]
1940
result.update(node[1][0] for node in index.iter_entries(keys))
1943
1710
def _make_parents_provider(self):
1944
return graph.CachingParentsProvider(self)
1711
if not self._format.supports_external_lookups:
1712
return self._unstacked_provider
1713
return graph.StackedParentsProvider(_LazyListJoin(
1714
[self._unstacked_provider], self._fallback_repositories))
1946
1716
def _refresh_data(self):
1947
if self._write_lock_count == 1 or (
1948
self.control_files._lock_count == 1 and
1949
self.control_files._lock_mode == 'r'):
1950
# forget what names there are
1951
self._pack_collection.reset()
1952
# XXX: Better to do an in-memory merge when acquiring a new lock -
1953
# factor out code from _save_pack_names.
1954
self._pack_collection.ensure_loaded()
1717
if not self.is_locked():
1719
self._pack_collection.reload_pack_names()
1720
self._unstacked_provider.disable_cache()
1721
self._unstacked_provider.enable_cache()
1956
1723
def _start_write_group(self):
1957
1724
self._pack_collection._start_write_group()
1959
1726
def _commit_write_group(self):
1960
return self._pack_collection._commit_write_group()
1962
def get_inventory_weave(self):
1963
return self._inv_thunk.get_weave()
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)
1965
1752
def get_transaction(self):
1966
1753
if self._write_lock_count:
2110
1906
mutter('creating repository in %s.', a_bzrdir.transport.base)
2111
1907
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
2112
builder = GraphIndexBuilder()
1908
builder = self.index_builder_class()
2113
1909
files = [('pack-names', builder.finish())]
2114
1910
utf8_files = [('format', self.get_format_string())]
2116
1912
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
2117
return self.open(a_bzrdir=a_bzrdir, _found=True)
1913
repository = self.open(a_bzrdir=a_bzrdir, _found=True)
1914
self._run_post_repo_init_hooks(repository, a_bzrdir, shared)
2119
1917
def open(self, a_bzrdir, _found=False, _override_transport=None):
2120
1918
"""See RepositoryFormat.open().
2122
1920
:param _override_transport: INTERNAL USE ONLY. Allows opening the
2123
1921
repository at a slightly different url
2124
1922
than normal. I.e. during 'upgrade'.
2127
format = RepositoryFormat.find_format(a_bzrdir)
1925
format = RepositoryFormatMetaDir.find_format(a_bzrdir)
2128
1926
if _override_transport is not None:
2129
1927
repo_transport = _override_transport
2131
1929
repo_transport = a_bzrdir.get_repository_transport(None)
2132
1930
control_files = lockable_files.LockableFiles(repo_transport,
2133
1931
'lock', lockdir.LockDir)
2134
text_store = self._get_text_store(repo_transport, control_files)
2135
control_store = self._get_control_store(repo_transport, control_files)
2136
_revision_store = self._get_revision_store(repo_transport, control_files)
2137
1932
return self.repository_class(_format=self,
2138
1933
a_bzrdir=a_bzrdir,
2139
1934
control_files=control_files,
2140
_revision_store=_revision_store,
2141
control_store=control_store,
2142
text_store=text_store,
2143
1935
_commit_builder_class=self._commit_builder_class,
2144
1936
_serializer=self._serializer)
2147
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2148
"""A no-subtrees parameterized Pack repository.
2150
This format was introduced in 0.92.
2153
repository_class = KnitPackRepository
2154
_commit_builder_class = PackCommitBuilder
2155
_serializer = xml5.serializer_v5
2157
def _get_matching_bzrdir(self):
2158
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2160
def _ignore_setting_bzrdir(self, format):
2163
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2165
def get_format_string(self):
2166
"""See RepositoryFormat.get_format_string()."""
2167
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2169
def get_format_description(self):
2170
"""See RepositoryFormat.get_format_description()."""
2171
return "Packs containing knits without subtree support"
2173
def check_conversion_target(self, target_format):
2177
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2178
"""A subtrees parameterized Pack repository.
2180
This repository format uses the xml7 serializer to get:
2181
- support for recording full info about the tree root
2182
- support for recording tree-references
2184
This format was introduced in 0.92.
2187
repository_class = KnitPackRepository
2188
_commit_builder_class = PackRootCommitBuilder
2189
rich_root_data = True
2190
supports_tree_reference = True
2191
_serializer = xml7.serializer_v7
2193
def _get_matching_bzrdir(self):
2194
return bzrdir.format_registry.make_bzrdir(
2195
'pack-0.92-subtree')
2197
def _ignore_setting_bzrdir(self, format):
2200
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2202
def check_conversion_target(self, target_format):
2203
if not target_format.rich_root_data:
2204
raise errors.BadConversionTarget(
2205
'Does not support rich root data.', target_format)
2206
if not getattr(target_format, 'supports_tree_reference', False):
2207
raise errors.BadConversionTarget(
2208
'Does not support nested trees', target_format)
2210
def get_format_string(self):
2211
"""See RepositoryFormat.get_format_string()."""
2212
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2214
def get_format_description(self):
2215
"""See RepositoryFormat.get_format_description()."""
2216
return "Packs containing knits with subtree support\n"
2219
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2220
"""A rich-root, no subtrees parameterized Pack repository.
2222
This repository format uses the xml6 serializer to get:
2223
- support for recording full info about the tree root
2225
This format was introduced in 1.0.
2228
repository_class = KnitPackRepository
2229
_commit_builder_class = PackRootCommitBuilder
2230
rich_root_data = True
2231
supports_tree_reference = False
2232
_serializer = xml6.serializer_v6
2234
def _get_matching_bzrdir(self):
2235
return bzrdir.format_registry.make_bzrdir(
2238
def _ignore_setting_bzrdir(self, format):
2241
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2243
def check_conversion_target(self, target_format):
2244
if not target_format.rich_root_data:
2245
raise errors.BadConversionTarget(
2246
'Does not support rich root data.', target_format)
2248
def get_format_string(self):
2249
"""See RepositoryFormat.get_format_string()."""
2250
return ("Bazaar pack repository format 1 with rich root"
2251
" (needs bzr 1.0)\n")
2253
def get_format_description(self):
2254
"""See RepositoryFormat.get_format_description()."""
2255
return "Packs containing knits with rich root support\n"
2258
class RepositoryFormatPackDevelopment0(RepositoryFormatPack):
2259
"""A no-subtrees development repository.
2261
This format should be retained until the second release after bzr 1.0.
2263
No changes to the disk behaviour from pack-0.92.
2266
repository_class = KnitPackRepository
2267
_commit_builder_class = PackCommitBuilder
2268
_serializer = xml5.serializer_v5
2270
def _get_matching_bzrdir(self):
2271
return bzrdir.format_registry.make_bzrdir('development0')
2273
def _ignore_setting_bzrdir(self, format):
2276
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2278
def get_format_string(self):
2279
"""See RepositoryFormat.get_format_string()."""
2280
return "Bazaar development format 0 (needs bzr.dev from before 1.3)\n"
2282
def get_format_description(self):
2283
"""See RepositoryFormat.get_format_description()."""
2284
return ("Development repository format, currently the same as "
2287
def check_conversion_target(self, target_format):
2291
class RepositoryFormatPackDevelopment0Subtree(RepositoryFormatPack):
2292
"""A subtrees development repository.
2294
This format should be retained until the second release after bzr 1.0.
2296
No changes to the disk behaviour from pack-0.92-subtree.
2299
repository_class = KnitPackRepository
2300
_commit_builder_class = PackRootCommitBuilder
2301
rich_root_data = True
2302
supports_tree_reference = True
2303
_serializer = xml7.serializer_v7
2305
def _get_matching_bzrdir(self):
2306
return bzrdir.format_registry.make_bzrdir(
2307
'development0-subtree')
2309
def _ignore_setting_bzrdir(self, format):
2312
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2314
def check_conversion_target(self, target_format):
2315
if not target_format.rich_root_data:
2316
raise errors.BadConversionTarget(
2317
'Does not support rich root data.', target_format)
2318
if not getattr(target_format, 'supports_tree_reference', False):
2319
raise errors.BadConversionTarget(
2320
'Does not support nested trees', target_format)
2322
def get_format_string(self):
2323
"""See RepositoryFormat.get_format_string()."""
2324
return ("Bazaar development format 0 with subtree support "
2325
"(needs bzr.dev from before 1.3)\n")
2327
def get_format_description(self):
2328
"""See RepositoryFormat.get_format_description()."""
2329
return ("Development repository format, currently the same as "
2330
"pack-0.92-subtree\n")
1939
class RetryPackOperations(errors.RetryWithNewPacks):
1940
"""Raised when we are packing and we find a missing file.
1942
Meant as a signaling exception, to tell the RepositoryPackCollection.pack
1943
code it should try again.
1946
internal_error = True
1948
_fmt = ("Pack files have changed, reload and try pack again."
1949
" context: %(context)s %(orig_error)s")
1952
class _DirectPackAccess(object):
1953
"""Access to data in one or more packs with less translation."""
1955
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
1956
"""Create a _DirectPackAccess object.
1958
:param index_to_packs: A dict mapping index objects to the transport
1959
and file names for obtaining data.
1960
:param reload_func: A function to call if we determine that the pack
1961
files have moved and we need to reload our caches. See
1962
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
1964
self._container_writer = None
1965
self._write_index = None
1966
self._indices = index_to_packs
1967
self._reload_func = reload_func
1968
self._flush_func = flush_func
1970
def add_raw_records(self, key_sizes, raw_data):
1971
"""Add raw knit bytes to a storage area.
1973
The data is spooled to the container writer in one bytes-record per
1976
:param sizes: An iterable of tuples containing the key and size of each
1978
:param raw_data: A bytestring containing the data.
1979
:return: A list of memos to retrieve the record later. Each memo is an
1980
opaque index memo. For _DirectPackAccess the memo is (index, pos,
1981
length), where the index field is the write_index object supplied
1982
to the PackAccess object.
1984
if type(raw_data) is not str:
1985
raise AssertionError(
1986
'data must be plain bytes was %s' % type(raw_data))
1989
for key, size in key_sizes:
1990
p_offset, p_length = self._container_writer.add_bytes_record(
1991
raw_data[offset:offset+size], [])
1993
result.append((self._write_index, p_offset, p_length))
1997
"""Flush pending writes on this access object.
1999
This will flush any buffered writes to a NewPack.
2001
if self._flush_func is not None:
2004
def get_raw_records(self, memos_for_retrieval):
2005
"""Get the raw bytes for a records.
2007
:param memos_for_retrieval: An iterable containing the (index, pos,
2008
length) memo for retrieving the bytes. The Pack access method
2009
looks up the pack to use for a given record in its index_to_pack
2011
:return: An iterator over the bytes of the records.
2013
# first pass, group into same-index requests
2015
current_index = None
2016
for (index, offset, length) in memos_for_retrieval:
2017
if current_index == index:
2018
current_list.append((offset, length))
2020
if current_index is not None:
2021
request_lists.append((current_index, current_list))
2022
current_index = index
2023
current_list = [(offset, length)]
2024
# handle the last entry
2025
if current_index is not None:
2026
request_lists.append((current_index, current_list))
2027
for index, offsets in request_lists:
2029
transport, path = self._indices[index]
2031
# A KeyError here indicates that someone has triggered an index
2032
# reload, and this index has gone missing, we need to start
2034
if self._reload_func is None:
2035
# If we don't have a _reload_func there is nothing that can
2038
raise errors.RetryWithNewPacks(index,
2039
reload_occurred=True,
2040
exc_info=sys.exc_info())
2042
reader = pack.make_readv_reader(transport, path, offsets)
2043
for names, read_func in reader.iter_records():
2044
yield read_func(None)
2045
except errors.NoSuchFile:
2046
# A NoSuchFile error indicates that a pack file has gone
2047
# missing on disk, we need to trigger a reload, and start over.
2048
if self._reload_func is None:
2050
raise errors.RetryWithNewPacks(transport.abspath(path),
2051
reload_occurred=False,
2052
exc_info=sys.exc_info())
2054
def set_writer(self, writer, index, transport_packname):
2055
"""Set a writer to use for adding data."""
2056
if index is not None:
2057
self._indices[index] = transport_packname
2058
self._container_writer = writer
2059
self._write_index = index
2061
def reload_or_raise(self, retry_exc):
2062
"""Try calling the reload function, or re-raise the original exception.
2064
This should be called after _DirectPackAccess raises a
2065
RetryWithNewPacks exception. This function will handle the common logic
2066
of determining when the error is fatal versus being temporary.
2067
It will also make sure that the original exception is raised, rather
2068
than the RetryWithNewPacks exception.
2070
If this function returns, then the calling function should retry
2071
whatever operation was being performed. Otherwise an exception will
2074
:param retry_exc: A RetryWithNewPacks exception.
2077
if self._reload_func is None:
2079
elif not self._reload_func():
2080
# The reload claimed that nothing changed
2081
if not retry_exc.reload_occurred:
2082
# If there wasn't an earlier reload, then we really were
2083
# expecting to find changes. We didn't find them, so this is a
2087
exc_class, exc_value, exc_traceback = retry_exc.exc_info
2088
raise exc_class, exc_value, exc_traceback
added
removed
bzrlib/repofmt/pack_repo.py
