1
# Copyright (C) 2005, 2006, 2007, 2008 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
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
20
from bzrlib.lazy_import import lazy_import
21
lazy_import(globals(), """
22
from itertools import izip
37
from bzrlib.index import (
41
GraphIndexPrefixAdapter,
44
from bzrlib.knit import (
50
from bzrlib import tsort
57
revision as _mod_revision,
61
from bzrlib.decorators import needs_write_lock
62
from bzrlib.btree_index import (
66
from bzrlib.index import (
70
from bzrlib.repofmt.knitrepo import KnitRepository
71
from bzrlib.repository import (
73
MetaDirRepositoryFormat,
78
import bzrlib.revision as _mod_revision
79
from bzrlib.trace import (
85
class PackCommitBuilder(CommitBuilder):
86
"""A subclass of CommitBuilder to add texts with pack semantics.
88
Specifically this uses one knit object rather than one knit object per
89
added text, reducing memory and object pressure.
92
def __init__(self, repository, parents, config, timestamp=None,
93
timezone=None, committer=None, revprops=None,
95
CommitBuilder.__init__(self, repository, parents, config,
96
timestamp=timestamp, timezone=timezone, committer=committer,
97
revprops=revprops, revision_id=revision_id)
98
self._file_graph = graph.Graph(
99
repository._pack_collection.text_index.combined_index)
101
def _heads(self, file_id, revision_ids):
102
keys = [(file_id, revision_id) for revision_id in revision_ids]
103
return set([key[1] for key in self._file_graph.heads(keys)])
106
class PackRootCommitBuilder(RootCommitBuilder):
107
"""A subclass of RootCommitBuilder to add texts with pack semantics.
109
Specifically this uses one knit object rather than one knit object per
110
added text, reducing memory and object pressure.
113
def __init__(self, repository, parents, config, timestamp=None,
114
timezone=None, committer=None, revprops=None,
116
CommitBuilder.__init__(self, repository, parents, config,
117
timestamp=timestamp, timezone=timezone, committer=committer,
118
revprops=revprops, revision_id=revision_id)
119
self._file_graph = graph.Graph(
120
repository._pack_collection.text_index.combined_index)
122
def _heads(self, file_id, revision_ids):
123
keys = [(file_id, revision_id) for revision_id in revision_ids]
124
return set([key[1] for key in self._file_graph.heads(keys)])
128
"""An in memory proxy for a pack and its indices.
130
This is a base class that is not directly used, instead the classes
131
ExistingPack and NewPack are used.
134
# A map of index 'type' to the file extension and position in the
136
index_definitions = {
138
'revision': ('.rix', 0),
139
'inventory': ('.iix', 1),
141
'signature': ('.six', 3),
144
def __init__(self, revision_index, inventory_index, text_index,
145
signature_index, chk_index=None):
146
"""Create a pack instance.
148
:param revision_index: A GraphIndex for determining what revisions are
149
present in the Pack and accessing the locations of their texts.
150
:param inventory_index: A GraphIndex for determining what inventories are
151
present in the Pack and accessing the locations of their
153
:param text_index: A GraphIndex for determining what file texts
154
are present in the pack and accessing the locations of their
155
texts/deltas (via (fileid, revisionid) tuples).
156
:param signature_index: A GraphIndex for determining what signatures are
157
present in the Pack and accessing the locations of their texts.
158
:param chk_index: A GraphIndex for accessing content by CHK, if the
161
self.revision_index = revision_index
162
self.inventory_index = inventory_index
163
self.text_index = text_index
164
self.signature_index = signature_index
165
self.chk_index = chk_index
167
def access_tuple(self):
168
"""Return a tuple (transport, name) for the pack content."""
169
return self.pack_transport, self.file_name()
171
def _check_references(self):
172
"""Make sure our external references are present.
174
Packs are allowed to have deltas whose base is not in the pack, but it
175
must be present somewhere in this collection. It is not allowed to
176
have deltas based on a fallback repository.
177
(See <https://bugs.launchpad.net/bzr/+bug/288751>)
180
for (index_name, external_refs, index) in [
182
self._get_external_refs(self.text_index),
183
self._pack_collection.text_index.combined_index),
185
self._get_external_refs(self.inventory_index),
186
self._pack_collection.inventory_index.combined_index),
188
missing = external_refs.difference(
189
k for (idx, k, v, r) in
190
index.iter_entries(external_refs))
192
missing_items[index_name] = sorted(list(missing))
194
from pprint import pformat
195
raise errors.BzrCheckError(
196
"Newly created pack file %r has delta references to "
197
"items not in its repository:\n%s"
198
% (self, pformat(missing_items)))
201
"""Get the file name for the pack on disk."""
202
return self.name + '.pack'
204
def get_revision_count(self):
205
return self.revision_index.key_count()
207
def index_name(self, index_type, name):
208
"""Get the disk name of an index type for pack name 'name'."""
209
return name + Pack.index_definitions[index_type][0]
211
def index_offset(self, index_type):
212
"""Get the position in a index_size array for a given index type."""
213
return Pack.index_definitions[index_type][1]
215
def inventory_index_name(self, name):
216
"""The inv index is the name + .iix."""
217
return self.index_name('inventory', name)
219
def revision_index_name(self, name):
220
"""The revision index is the name + .rix."""
221
return self.index_name('revision', name)
223
def signature_index_name(self, name):
224
"""The signature index is the name + .six."""
225
return self.index_name('signature', name)
227
def text_index_name(self, name):
228
"""The text index is the name + .tix."""
229
return self.index_name('text', name)
231
def _replace_index_with_readonly(self, index_type):
232
setattr(self, index_type + '_index',
233
self.index_class(self.index_transport,
234
self.index_name(index_type, self.name),
235
self.index_sizes[self.index_offset(index_type)]))
238
class ExistingPack(Pack):
239
"""An in memory proxy for an existing .pack and its disk indices."""
241
def __init__(self, pack_transport, name, revision_index, inventory_index,
242
text_index, signature_index, chk_index=None):
243
"""Create an ExistingPack object.
245
:param pack_transport: The transport where the pack file resides.
246
:param name: The name of the pack on disk in the pack_transport.
248
Pack.__init__(self, revision_index, inventory_index, text_index,
249
signature_index, chk_index)
251
self.pack_transport = pack_transport
252
if None in (revision_index, inventory_index, text_index,
253
signature_index, name, pack_transport):
254
raise AssertionError()
256
def __eq__(self, other):
257
return self.__dict__ == other.__dict__
259
def __ne__(self, other):
260
return not self.__eq__(other)
263
return "<%s.%s object at 0x%x, %s, %s" % (
264
self.__class__.__module__, self.__class__.__name__, id(self),
265
self.pack_transport, self.name)
268
class ResumedPack(ExistingPack):
270
def __init__(self, name, revision_index, inventory_index, text_index,
271
signature_index, upload_transport, pack_transport, index_transport,
272
pack_collection, chk_index=None):
273
"""Create a ResumedPack object."""
274
ExistingPack.__init__(self, pack_transport, name, revision_index,
275
inventory_index, text_index, signature_index,
277
self.upload_transport = upload_transport
278
self.index_transport = index_transport
279
self.index_sizes = [None, None, None, None]
281
('revision', revision_index),
282
('inventory', inventory_index),
283
('text', text_index),
284
('signature', signature_index),
286
if chk_index is not None:
287
indices.append(('chk', chk_index))
288
self.index_sizes.append(None)
289
for index_type, index in indices:
290
offset = self.index_offset(index_type)
291
self.index_sizes[offset] = index._size
292
self.index_class = pack_collection._index_class
293
self._pack_collection = pack_collection
294
self._state = 'resumed'
295
# XXX: perhaps check that the .pack file exists?
297
def access_tuple(self):
298
if self._state == 'finished':
299
return Pack.access_tuple(self)
300
elif self._state == 'resumed':
301
return self.upload_transport, self.file_name()
303
raise AssertionError(self._state)
306
self.upload_transport.delete(self.file_name())
307
indices = [self.revision_index, self.inventory_index, self.text_index,
308
self.signature_index]
309
if self.chk_index is not None:
310
indices.append(self.chk_index)
311
for index in indices:
312
index._transport.delete(index._name)
315
self._check_references()
316
new_name = '../packs/' + self.file_name()
317
self.upload_transport.rename(self.file_name(), new_name)
318
index_types = ['revision', 'inventory', 'text', 'signature']
319
if self.chk_index is not None:
320
index_types.append('chk')
321
for index_type in index_types:
322
old_name = self.index_name(index_type, self.name)
323
new_name = '../indices/' + old_name
324
self.upload_transport.rename(old_name, new_name)
325
self._replace_index_with_readonly(index_type)
326
self._state = 'finished'
328
def _get_external_refs(self, index):
329
"""Return compression parents for this index that are not present.
331
This returns any compression parents that are referenced by this index,
332
which are not contained *in* this index. They may be present elsewhere.
334
return index.external_references(1)
338
"""An in memory proxy for a pack which is being created."""
340
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
341
"""Create a NewPack instance.
343
:param pack_collection: A PackCollection into which this is being inserted.
344
:param upload_suffix: An optional suffix to be given to any temporary
345
files created during the pack creation. e.g '.autopack'
346
:param file_mode: Unix permissions for newly created file.
348
# The relative locations of the packs are constrained, but all are
349
# passed in because the caller has them, so as to avoid object churn.
350
index_builder_class = pack_collection._index_builder_class
351
if pack_collection.chk_index is not None:
352
chk_index = index_builder_class(reference_lists=0)
356
# Revisions: parents list, no text compression.
357
index_builder_class(reference_lists=1),
358
# Inventory: We want to map compression only, but currently the
359
# knit code hasn't been updated enough to understand that, so we
360
# have a regular 2-list index giving parents and compression
362
index_builder_class(reference_lists=2),
363
# Texts: compression and per file graph, for all fileids - so two
364
# reference lists and two elements in the key tuple.
365
index_builder_class(reference_lists=2, key_elements=2),
366
# Signatures: Just blobs to store, no compression, no parents
368
index_builder_class(reference_lists=0),
369
# CHK based storage - just blobs, no compression or parents.
372
self._pack_collection = pack_collection
373
# When we make readonly indices, we need this.
374
self.index_class = pack_collection._index_class
375
# where should the new pack be opened
376
self.upload_transport = pack_collection._upload_transport
377
# where are indices written out to
378
self.index_transport = pack_collection._index_transport
379
# where is the pack renamed to when it is finished?
380
self.pack_transport = pack_collection._pack_transport
381
# What file mode to upload the pack and indices with.
382
self._file_mode = file_mode
383
# tracks the content written to the .pack file.
384
self._hash = osutils.md5()
385
# a tuple with the length in bytes of the indices, once the pack
386
# is finalised. (rev, inv, text, sigs, chk_if_in_use)
387
self.index_sizes = None
388
# How much data to cache when writing packs. Note that this is not
389
# synchronised with reads, because it's not in the transport layer, so
390
# is not safe unless the client knows it won't be reading from the pack
392
self._cache_limit = 0
393
# the temporary pack file name.
394
self.random_name = osutils.rand_chars(20) + upload_suffix
395
# when was this pack started ?
396
self.start_time = time.time()
397
# open an output stream for the data added to the pack.
398
self.write_stream = self.upload_transport.open_write_stream(
399
self.random_name, mode=self._file_mode)
400
if 'pack' in debug.debug_flags:
401
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
402
time.ctime(), self.upload_transport.base, self.random_name,
403
time.time() - self.start_time)
404
# A list of byte sequences to be written to the new pack, and the
405
# aggregate size of them. Stored as a list rather than separate
406
# variables so that the _write_data closure below can update them.
407
self._buffer = [[], 0]
408
# create a callable for adding data
410
# robertc says- this is a closure rather than a method on the object
411
# so that the variables are locals, and faster than accessing object
413
def _write_data(bytes, flush=False, _buffer=self._buffer,
414
_write=self.write_stream.write, _update=self._hash.update):
415
_buffer[0].append(bytes)
416
_buffer[1] += len(bytes)
418
if _buffer[1] > self._cache_limit or flush:
419
bytes = ''.join(_buffer[0])
423
# expose this on self, for the occasion when clients want to add data.
424
self._write_data = _write_data
425
# a pack writer object to serialise pack records.
426
self._writer = pack.ContainerWriter(self._write_data)
428
# what state is the pack in? (open, finished, aborted)
432
"""Cancel creating this pack."""
433
self._state = 'aborted'
434
self.write_stream.close()
435
# Remove the temporary pack file.
436
self.upload_transport.delete(self.random_name)
437
# The indices have no state on disk.
439
def access_tuple(self):
440
"""Return a tuple (transport, name) for the pack content."""
441
if self._state == 'finished':
442
return Pack.access_tuple(self)
443
elif self._state == 'open':
444
return self.upload_transport, self.random_name
446
raise AssertionError(self._state)
448
def data_inserted(self):
449
"""True if data has been added to this pack."""
450
return bool(self.get_revision_count() or
451
self.inventory_index.key_count() or
452
self.text_index.key_count() or
453
self.signature_index.key_count() or
454
(self.chk_index is not None and self.chk_index.key_count()))
456
def finish(self, suspend=False):
457
"""Finish the new pack.
460
- finalises the content
461
- assigns a name (the md5 of the content, currently)
462
- writes out the associated indices
463
- renames the pack into place.
464
- stores the index size tuple for the pack in the index_sizes
469
self._write_data('', flush=True)
470
self.name = self._hash.hexdigest()
472
self._check_references()
474
# XXX: It'd be better to write them all to temporary names, then
475
# rename them all into place, so that the window when only some are
476
# visible is smaller. On the other hand none will be seen until
477
# they're in the names list.
478
self.index_sizes = [None, None, None, None]
479
self._write_index('revision', self.revision_index, 'revision', suspend)
480
self._write_index('inventory', self.inventory_index, 'inventory',
482
self._write_index('text', self.text_index, 'file texts', suspend)
483
self._write_index('signature', self.signature_index,
484
'revision signatures', suspend)
485
if self.chk_index is not None:
486
self.index_sizes.append(None)
487
self._write_index('chk', self.chk_index,
488
'content hash bytes', suspend)
489
self.write_stream.close()
490
# Note that this will clobber an existing pack with the same name,
491
# without checking for hash collisions. While this is undesirable this
492
# is something that can be rectified in a subsequent release. One way
493
# to rectify it may be to leave the pack at the original name, writing
494
# its pack-names entry as something like 'HASH: index-sizes
495
# temporary-name'. Allocate that and check for collisions, if it is
496
# collision free then rename it into place. If clients know this scheme
497
# they can handle missing-file errors by:
498
# - try for HASH.pack
499
# - try for temporary-name
500
# - refresh the pack-list to see if the pack is now absent
501
new_name = self.name + '.pack'
503
new_name = '../packs/' + new_name
504
self.upload_transport.rename(self.random_name, new_name)
505
self._state = 'finished'
506
if 'pack' in debug.debug_flags:
507
# XXX: size might be interesting?
508
mutter('%s: create_pack: pack finished: %s%s->%s t+%6.3fs',
509
time.ctime(), self.upload_transport.base, self.random_name,
510
new_name, time.time() - self.start_time)
513
"""Flush any current data."""
515
bytes = ''.join(self._buffer[0])
516
self.write_stream.write(bytes)
517
self._hash.update(bytes)
518
self._buffer[:] = [[], 0]
520
def _get_external_refs(self, index):
521
return index._external_references()
523
def set_write_cache_size(self, size):
524
self._cache_limit = size
526
def _write_index(self, index_type, index, label, suspend=False):
527
"""Write out an index.
529
:param index_type: The type of index to write - e.g. 'revision'.
530
:param index: The index object to serialise.
531
:param label: What label to give the index e.g. 'revision'.
533
index_name = self.index_name(index_type, self.name)
535
transport = self.upload_transport
537
transport = self.index_transport
538
self.index_sizes[self.index_offset(index_type)] = transport.put_file(
539
index_name, index.finish(), mode=self._file_mode)
540
if 'pack' in debug.debug_flags:
541
# XXX: size might be interesting?
542
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
543
time.ctime(), label, self.upload_transport.base,
544
self.random_name, time.time() - self.start_time)
545
# Replace the writable index on this object with a readonly,
546
# presently unloaded index. We should alter
547
# the index layer to make its finish() error if add_node is
548
# subsequently used. RBC
549
self._replace_index_with_readonly(index_type)
552
class AggregateIndex(object):
553
"""An aggregated index for the RepositoryPackCollection.
555
AggregateIndex is reponsible for managing the PackAccess object,
556
Index-To-Pack mapping, and all indices list for a specific type of index
557
such as 'revision index'.
559
A CombinedIndex provides an index on a single key space built up
560
from several on-disk indices. The AggregateIndex builds on this
561
to provide a knit access layer, and allows having up to one writable
562
index within the collection.
564
# XXX: Probably 'can be written to' could/should be separated from 'acts
565
# like a knit index' -- mbp 20071024
567
def __init__(self, reload_func=None, flush_func=None):
568
"""Create an AggregateIndex.
570
:param reload_func: A function to call if we find we are missing an
571
index. Should have the form reload_func() => True if the list of
572
active pack files has changed.
574
self._reload_func = reload_func
575
self.index_to_pack = {}
576
self.combined_index = CombinedGraphIndex([], reload_func=reload_func)
577
self.data_access = _DirectPackAccess(self.index_to_pack,
578
reload_func=reload_func,
579
flush_func=flush_func)
580
self.add_callback = None
582
def replace_indices(self, index_to_pack, indices):
583
"""Replace the current mappings with fresh ones.
585
This should probably not be used eventually, rather incremental add and
586
removal of indices. It has been added during refactoring of existing
589
:param index_to_pack: A mapping from index objects to
590
(transport, name) tuples for the pack file data.
591
:param indices: A list of indices.
593
# refresh the revision pack map dict without replacing the instance.
594
self.index_to_pack.clear()
595
self.index_to_pack.update(index_to_pack)
596
# XXX: API break - clearly a 'replace' method would be good?
597
self.combined_index._indices[:] = indices
598
# the current add nodes callback for the current writable index if
600
self.add_callback = None
602
def add_index(self, index, pack):
603
"""Add index to the aggregate, which is an index for Pack pack.
605
Future searches on the aggregate index will seach this new index
606
before all previously inserted indices.
608
:param index: An Index for the pack.
609
:param pack: A Pack instance.
611
# expose it to the index map
612
self.index_to_pack[index] = pack.access_tuple()
613
# put it at the front of the linear index list
614
self.combined_index.insert_index(0, index)
616
def add_writable_index(self, index, pack):
617
"""Add an index which is able to have data added to it.
619
There can be at most one writable index at any time. Any
620
modifications made to the knit are put into this index.
622
:param index: An index from the pack parameter.
623
:param pack: A Pack instance.
625
if self.add_callback is not None:
626
raise AssertionError(
627
"%s already has a writable index through %s" % \
628
(self, self.add_callback))
629
# allow writing: queue writes to a new index
630
self.add_index(index, pack)
631
# Updates the index to packs mapping as a side effect,
632
self.data_access.set_writer(pack._writer, index, pack.access_tuple())
633
self.add_callback = index.add_nodes
636
"""Reset all the aggregate data to nothing."""
637
self.data_access.set_writer(None, None, (None, None))
638
self.index_to_pack.clear()
639
del self.combined_index._indices[:]
640
self.add_callback = None
642
def remove_index(self, index, pack):
643
"""Remove index from the indices used to answer queries.
645
:param index: An index from the pack parameter.
646
:param pack: A Pack instance.
648
del self.index_to_pack[index]
649
self.combined_index._indices.remove(index)
650
if (self.add_callback is not None and
651
getattr(index, 'add_nodes', None) == self.add_callback):
652
self.add_callback = None
653
self.data_access.set_writer(None, None, (None, None))
656
class Packer(object):
657
"""Create a pack from packs."""
659
def __init__(self, pack_collection, packs, suffix, revision_ids=None,
663
:param pack_collection: A RepositoryPackCollection object where the
664
new pack is being written to.
665
:param packs: The packs to combine.
666
:param suffix: The suffix to use on the temporary files for the pack.
667
:param revision_ids: Revision ids to limit the pack to.
668
:param reload_func: A function to call if a pack file/index goes
669
missing. The side effect of calling this function should be to
670
update self.packs. See also AggregateIndex
674
self.revision_ids = revision_ids
675
# The pack object we are creating.
677
self._pack_collection = pack_collection
678
self._reload_func = reload_func
679
# The index layer keys for the revisions being copied. None for 'all
681
self._revision_keys = None
682
# What text keys to copy. None for 'all texts'. This is set by
683
# _copy_inventory_texts
684
self._text_filter = None
687
def _extra_init(self):
688
"""A template hook to allow extending the constructor trivially."""
690
def _pack_map_and_index_list(self, index_attribute):
691
"""Convert a list of packs to an index pack map and index list.
693
:param index_attribute: The attribute that the desired index is found
695
:return: A tuple (map, list) where map contains the dict from
696
index:pack_tuple, and list contains the indices in the preferred
701
for pack_obj in self.packs:
702
index = getattr(pack_obj, index_attribute)
703
indices.append(index)
704
pack_map[index] = pack_obj
705
return pack_map, indices
707
def _index_contents(self, indices, key_filter=None):
708
"""Get an iterable of the index contents from a pack_map.
710
:param indices: The list of indices to query
711
:param key_filter: An optional filter to limit the keys returned.
713
all_index = CombinedGraphIndex(indices)
714
if key_filter is None:
715
return all_index.iter_all_entries()
717
return all_index.iter_entries(key_filter)
719
def pack(self, pb=None):
720
"""Create a new pack by reading data from other packs.
722
This does little more than a bulk copy of data. One key difference
723
is that data with the same item key across multiple packs is elided
724
from the output. The new pack is written into the current pack store
725
along with its indices, and the name added to the pack names. The
726
source packs are not altered and are not required to be in the current
729
:param pb: An optional progress bar to use. A nested bar is created if
731
:return: A Pack object, or None if nothing was copied.
733
# open a pack - using the same name as the last temporary file
734
# - which has already been flushed, so its safe.
735
# XXX: - duplicate code warning with start_write_group; fix before
736
# considering 'done'.
737
if self._pack_collection._new_pack is not None:
738
raise errors.BzrError('call to %s.pack() while another pack is'
740
% (self.__class__.__name__,))
741
if self.revision_ids is not None:
742
if len(self.revision_ids) == 0:
743
# silly fetch request.
746
self.revision_ids = frozenset(self.revision_ids)
747
self.revision_keys = frozenset((revid,) for revid in
750
self.pb = ui.ui_factory.nested_progress_bar()
754
return self._create_pack_from_packs()
760
"""Open a pack for the pack we are creating."""
761
new_pack = self._pack_collection.pack_factory(self._pack_collection,
762
upload_suffix=self.suffix,
763
file_mode=self._pack_collection.repo.bzrdir._get_file_mode())
764
# We know that we will process all nodes in order, and don't need to
765
# query, so don't combine any indices spilled to disk until we are done
766
new_pack.revision_index.set_optimize(combine_backing_indices=False)
767
new_pack.inventory_index.set_optimize(combine_backing_indices=False)
768
new_pack.text_index.set_optimize(combine_backing_indices=False)
769
new_pack.signature_index.set_optimize(combine_backing_indices=False)
772
def _update_pack_order(self, entries, index_to_pack_map):
773
"""Determine how we want our packs to be ordered.
775
This changes the sort order of the self.packs list so that packs unused
776
by 'entries' will be at the end of the list, so that future requests
777
can avoid probing them. Used packs will be at the front of the
778
self.packs list, in the order of their first use in 'entries'.
780
:param entries: A list of (index, ...) tuples
781
:param index_to_pack_map: A mapping from index objects to pack objects.
785
for entry in entries:
787
if index not in seen_indexes:
788
packs.append(index_to_pack_map[index])
789
seen_indexes.add(index)
790
if len(packs) == len(self.packs):
791
if 'pack' in debug.debug_flags:
792
mutter('Not changing pack list, all packs used.')
794
seen_packs = set(packs)
795
for pack in self.packs:
796
if pack not in seen_packs:
799
if 'pack' in debug.debug_flags:
800
old_names = [p.access_tuple()[1] for p in self.packs]
801
new_names = [p.access_tuple()[1] for p in packs]
802
mutter('Reordering packs\nfrom: %s\n to: %s',
803
old_names, new_names)
806
def _copy_revision_texts(self):
807
"""Copy revision data to the new pack."""
809
if self.revision_ids:
810
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
813
# select revision keys
814
revision_index_map, revision_indices = self._pack_map_and_index_list(
816
revision_nodes = self._index_contents(revision_indices, revision_keys)
817
revision_nodes = list(revision_nodes)
818
self._update_pack_order(revision_nodes, revision_index_map)
819
# copy revision keys and adjust values
820
self.pb.update("Copying revision texts", 1)
821
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
822
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
823
self.new_pack.revision_index, readv_group_iter, total_items))
824
if 'pack' in debug.debug_flags:
825
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
826
time.ctime(), self._pack_collection._upload_transport.base,
827
self.new_pack.random_name,
828
self.new_pack.revision_index.key_count(),
829
time.time() - self.new_pack.start_time)
830
self._revision_keys = revision_keys
832
def _copy_inventory_texts(self):
833
"""Copy the inventory texts to the new pack.
835
self._revision_keys is used to determine what inventories to copy.
837
Sets self._text_filter appropriately.
839
# select inventory keys
840
inv_keys = self._revision_keys # currently the same keyspace, and note that
841
# querying for keys here could introduce a bug where an inventory item
842
# is missed, so do not change it to query separately without cross
843
# checking like the text key check below.
844
inventory_index_map, inventory_indices = self._pack_map_and_index_list(
846
inv_nodes = self._index_contents(inventory_indices, inv_keys)
847
# copy inventory keys and adjust values
848
# XXX: Should be a helper function to allow different inv representation
850
self.pb.update("Copying inventory texts", 2)
851
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
852
# Only grab the output lines if we will be processing them
853
output_lines = bool(self.revision_ids)
854
inv_lines = self._copy_nodes_graph(inventory_index_map,
855
self.new_pack._writer, self.new_pack.inventory_index,
856
readv_group_iter, total_items, output_lines=output_lines)
857
if self.revision_ids:
858
self._process_inventory_lines(inv_lines)
860
# eat the iterator to cause it to execute.
862
self._text_filter = None
863
if 'pack' in debug.debug_flags:
864
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
865
time.ctime(), self._pack_collection._upload_transport.base,
866
self.new_pack.random_name,
867
self.new_pack.inventory_index.key_count(),
868
time.time() - self.new_pack.start_time)
870
def _copy_text_texts(self):
872
text_index_map, text_nodes = self._get_text_nodes()
873
if self._text_filter is not None:
874
# We could return the keys copied as part of the return value from
875
# _copy_nodes_graph but this doesn't work all that well with the
876
# need to get line output too, so we check separately, and as we're
877
# going to buffer everything anyway, we check beforehand, which
878
# saves reading knit data over the wire when we know there are
880
text_nodes = set(text_nodes)
881
present_text_keys = set(_node[1] for _node in text_nodes)
882
missing_text_keys = set(self._text_filter) - present_text_keys
883
if missing_text_keys:
884
# TODO: raise a specific error that can handle many missing
886
mutter("missing keys during fetch: %r", missing_text_keys)
887
a_missing_key = missing_text_keys.pop()
888
raise errors.RevisionNotPresent(a_missing_key[1],
890
# copy text keys and adjust values
891
self.pb.update("Copying content texts", 3)
892
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
893
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
894
self.new_pack.text_index, readv_group_iter, total_items))
895
self._log_copied_texts()
897
def _create_pack_from_packs(self):
898
self.pb.update("Opening pack", 0, 5)
899
self.new_pack = self.open_pack()
900
new_pack = self.new_pack
901
# buffer data - we won't be reading-back during the pack creation and
902
# this makes a significant difference on sftp pushes.
903
new_pack.set_write_cache_size(1024*1024)
904
if 'pack' in debug.debug_flags:
905
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
906
for a_pack in self.packs]
907
if self.revision_ids is not None:
908
rev_count = len(self.revision_ids)
911
mutter('%s: create_pack: creating pack from source packs: '
912
'%s%s %s revisions wanted %s t=0',
913
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
914
plain_pack_list, rev_count)
915
self._copy_revision_texts()
916
self._copy_inventory_texts()
917
self._copy_text_texts()
918
# select signature keys
919
signature_filter = self._revision_keys # same keyspace
920
signature_index_map, signature_indices = self._pack_map_and_index_list(
922
signature_nodes = self._index_contents(signature_indices,
924
# copy signature keys and adjust values
925
self.pb.update("Copying signature texts", 4)
926
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
927
new_pack.signature_index)
928
if 'pack' in debug.debug_flags:
929
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
930
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
931
new_pack.signature_index.key_count(),
932
time.time() - new_pack.start_time)
934
# NB XXX: how to check CHK references are present? perhaps by yielding
935
# the items? How should that interact with stacked repos?
936
if new_pack.chk_index is not None:
938
if 'pack' in debug.debug_flags:
939
mutter('%s: create_pack: chk content copied: %s%s %d items t+%6.3fs',
940
time.ctime(), self._pack_collection._upload_transport.base,
941
new_pack.random_name,
942
new_pack.chk_index.key_count(),
943
time.time() - new_pack.start_time)
944
new_pack._check_references()
945
if not self._use_pack(new_pack):
948
self.pb.update("Finishing pack", 5)
950
self._pack_collection.allocate(new_pack)
953
def _copy_chks(self, refs=None):
954
# XXX: Todo, recursive follow-pointers facility when fetching some
956
chk_index_map, chk_indices = self._pack_map_and_index_list(
958
chk_nodes = self._index_contents(chk_indices, refs)
960
# TODO: This isn't strictly tasteful as we are accessing some private
961
# variables (_serializer). Perhaps a better way would be to have
962
# Repository._deserialise_chk_node()
963
search_key_func = chk_map.search_key_registry.get(
964
self._pack_collection.repo._serializer.search_key_name)
965
def accumlate_refs(lines):
966
# XXX: move to a generic location
968
bytes = ''.join(lines)
969
node = chk_map._deserialise(bytes, ("unknown",), search_key_func)
970
new_refs.update(node.refs())
971
self._copy_nodes(chk_nodes, chk_index_map, self.new_pack._writer,
972
self.new_pack.chk_index, output_lines=accumlate_refs)
975
def _copy_nodes(self, nodes, index_map, writer, write_index,
977
"""Copy knit nodes between packs with no graph references.
979
:param output_lines: Output full texts of copied items.
981
pb = ui.ui_factory.nested_progress_bar()
983
return self._do_copy_nodes(nodes, index_map, writer,
984
write_index, pb, output_lines=output_lines)
988
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb,
990
# for record verification
991
knit = KnitVersionedFiles(None, None)
992
# plan a readv on each source pack:
994
nodes = sorted(nodes)
995
# how to map this into knit.py - or knit.py into this?
996
# we don't want the typical knit logic, we want grouping by pack
997
# at this point - perhaps a helper library for the following code
998
# duplication points?
1000
for index, key, value in nodes:
1001
if index not in request_groups:
1002
request_groups[index] = []
1003
request_groups[index].append((key, value))
1005
pb.update("Copied record", record_index, len(nodes))
1006
for index, items in request_groups.iteritems():
1007
pack_readv_requests = []
1008
for key, value in items:
1009
# ---- KnitGraphIndex.get_position
1010
bits = value[1:].split(' ')
1011
offset, length = int(bits[0]), int(bits[1])
1012
pack_readv_requests.append((offset, length, (key, value[0])))
1013
# linear scan up the pack
1014
pack_readv_requests.sort()
1016
pack_obj = index_map[index]
1017
transport, path = pack_obj.access_tuple()
1019
reader = pack.make_readv_reader(transport, path,
1020
[offset[0:2] for offset in pack_readv_requests])
1021
except errors.NoSuchFile:
1022
if self._reload_func is not None:
1025
for (names, read_func), (_1, _2, (key, eol_flag)) in \
1026
izip(reader.iter_records(), pack_readv_requests):
1027
raw_data = read_func(None)
1028
# check the header only
1029
if output_lines is not None:
1030
output_lines(knit._parse_record(key[-1], raw_data)[0])
1032
df, _ = knit._parse_record_header(key, raw_data)
1034
pos, size = writer.add_bytes_record(raw_data, names)
1035
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
1036
pb.update("Copied record", record_index)
1039
def _copy_nodes_graph(self, index_map, writer, write_index,
1040
readv_group_iter, total_items, output_lines=False):
1041
"""Copy knit nodes between packs.
1043
:param output_lines: Return lines present in the copied data as
1044
an iterator of line,version_id.
1046
pb = ui.ui_factory.nested_progress_bar()
1048
for result in self._do_copy_nodes_graph(index_map, writer,
1049
write_index, output_lines, pb, readv_group_iter, total_items):
1052
# Python 2.4 does not permit try:finally: in a generator.
1058
def _do_copy_nodes_graph(self, index_map, writer, write_index,
1059
output_lines, pb, readv_group_iter, total_items):
1060
# for record verification
1061
knit = KnitVersionedFiles(None, None)
1062
# for line extraction when requested (inventories only)
1064
factory = KnitPlainFactory()
1066
pb.update("Copied record", record_index, total_items)
1067
for index, readv_vector, node_vector in readv_group_iter:
1069
pack_obj = index_map[index]
1070
transport, path = pack_obj.access_tuple()
1072
reader = pack.make_readv_reader(transport, path, readv_vector)
1073
except errors.NoSuchFile:
1074
if self._reload_func is not None:
1077
for (names, read_func), (key, eol_flag, references) in \
1078
izip(reader.iter_records(), node_vector):
1079
raw_data = read_func(None)
1081
# read the entire thing
1082
content, _ = knit._parse_record(key[-1], raw_data)
1083
if len(references[-1]) == 0:
1084
line_iterator = factory.get_fulltext_content(content)
1086
line_iterator = factory.get_linedelta_content(content)
1087
for line in line_iterator:
1090
# check the header only
1091
df, _ = knit._parse_record_header(key, raw_data)
1093
pos, size = writer.add_bytes_record(raw_data, names)
1094
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1095
pb.update("Copied record", record_index)
1098
def _get_text_nodes(self):
1099
text_index_map, text_indices = self._pack_map_and_index_list(
1101
return text_index_map, self._index_contents(text_indices,
1104
def _least_readv_node_readv(self, nodes):
1105
"""Generate request groups for nodes using the least readv's.
1107
:param nodes: An iterable of graph index nodes.
1108
:return: Total node count and an iterator of the data needed to perform
1109
readvs to obtain the data for nodes. Each item yielded by the
1110
iterator is a tuple with:
1111
index, readv_vector, node_vector. readv_vector is a list ready to
1112
hand to the transport readv method, and node_vector is a list of
1113
(key, eol_flag, references) for the the node retrieved by the
1114
matching readv_vector.
1116
# group by pack so we do one readv per pack
1117
nodes = sorted(nodes)
1120
for index, key, value, references in nodes:
1121
if index not in request_groups:
1122
request_groups[index] = []
1123
request_groups[index].append((key, value, references))
1125
for index, items in request_groups.iteritems():
1126
pack_readv_requests = []
1127
for key, value, references in items:
1128
# ---- KnitGraphIndex.get_position
1129
bits = value[1:].split(' ')
1130
offset, length = int(bits[0]), int(bits[1])
1131
pack_readv_requests.append(
1132
((offset, length), (key, value[0], references)))
1133
# linear scan up the pack to maximum range combining.
1134
pack_readv_requests.sort()
1135
# split out the readv and the node data.
1136
pack_readv = [readv for readv, node in pack_readv_requests]
1137
node_vector = [node for readv, node in pack_readv_requests]
1138
result.append((index, pack_readv, node_vector))
1139
return total, result
1141
def _log_copied_texts(self):
1142
if 'pack' in debug.debug_flags:
1143
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
1144
time.ctime(), self._pack_collection._upload_transport.base,
1145
self.new_pack.random_name,
1146
self.new_pack.text_index.key_count(),
1147
time.time() - self.new_pack.start_time)
1149
def _process_inventory_lines(self, inv_lines):
1150
"""Use up the inv_lines generator and setup a text key filter."""
1151
repo = self._pack_collection.repo
1152
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
1153
inv_lines, self.revision_keys)
1155
for fileid, file_revids in fileid_revisions.iteritems():
1156
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
1157
self._text_filter = text_filter
1159
def _revision_node_readv(self, revision_nodes):
1160
"""Return the total revisions and the readv's to issue.
1162
:param revision_nodes: The revision index contents for the packs being
1163
incorporated into the new pack.
1164
:return: As per _least_readv_node_readv.
1166
return self._least_readv_node_readv(revision_nodes)
1168
def _use_pack(self, new_pack):
1169
"""Return True if new_pack should be used.
1171
:param new_pack: The pack that has just been created.
1172
:return: True if the pack should be used.
1174
return new_pack.data_inserted()
1177
class OptimisingPacker(Packer):
1178
"""A packer which spends more time to create better disk layouts."""
1180
def _revision_node_readv(self, revision_nodes):
1181
"""Return the total revisions and the readv's to issue.
1183
This sort places revisions in topological order with the ancestors
1186
:param revision_nodes: The revision index contents for the packs being
1187
incorporated into the new pack.
1188
:return: As per _least_readv_node_readv.
1190
# build an ancestors dict
1193
for index, key, value, references in revision_nodes:
1194
ancestors[key] = references[0]
1195
by_key[key] = (index, value, references)
1196
order = tsort.topo_sort(ancestors)
1198
# Single IO is pathological, but it will work as a starting point.
1200
for key in reversed(order):
1201
index, value, references = by_key[key]
1202
# ---- KnitGraphIndex.get_position
1203
bits = value[1:].split(' ')
1204
offset, length = int(bits[0]), int(bits[1])
1206
(index, [(offset, length)], [(key, value[0], references)]))
1207
# TODO: combine requests in the same index that are in ascending order.
1208
return total, requests
1210
def open_pack(self):
1211
"""Open a pack for the pack we are creating."""
1212
new_pack = super(OptimisingPacker, self).open_pack()
1213
# Turn on the optimization flags for all the index builders.
1214
new_pack.revision_index.set_optimize(for_size=True)
1215
new_pack.inventory_index.set_optimize(for_size=True)
1216
new_pack.text_index.set_optimize(for_size=True)
1217
new_pack.signature_index.set_optimize(for_size=True)
1221
class ReconcilePacker(Packer):
1222
"""A packer which regenerates indices etc as it copies.
1224
This is used by ``bzr reconcile`` to cause parent text pointers to be
1228
def _extra_init(self):
1229
self._data_changed = False
1231
def _process_inventory_lines(self, inv_lines):
1232
"""Generate a text key reference map rather for reconciling with."""
1233
repo = self._pack_collection.repo
1234
refs = repo._find_text_key_references_from_xml_inventory_lines(
1236
self._text_refs = refs
1237
# during reconcile we:
1238
# - convert unreferenced texts to full texts
1239
# - correct texts which reference a text not copied to be full texts
1240
# - copy all others as-is but with corrected parents.
1241
# - so at this point we don't know enough to decide what becomes a full
1243
self._text_filter = None
1245
def _copy_text_texts(self):
1246
"""generate what texts we should have and then copy."""
1247
self.pb.update("Copying content texts", 3)
1248
# we have three major tasks here:
1249
# 1) generate the ideal index
1250
repo = self._pack_collection.repo
1251
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1252
_1, key, _2, refs in
1253
self.new_pack.revision_index.iter_all_entries()])
1254
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1255
# 2) generate a text_nodes list that contains all the deltas that can
1256
# be used as-is, with corrected parents.
1259
discarded_nodes = []
1260
NULL_REVISION = _mod_revision.NULL_REVISION
1261
text_index_map, text_nodes = self._get_text_nodes()
1262
for node in text_nodes:
1268
ideal_parents = tuple(ideal_index[node[1]])
1270
discarded_nodes.append(node)
1271
self._data_changed = True
1273
if ideal_parents == (NULL_REVISION,):
1275
if ideal_parents == node[3][0]:
1277
ok_nodes.append(node)
1278
elif ideal_parents[0:1] == node[3][0][0:1]:
1279
# the left most parent is the same, or there are no parents
1280
# today. Either way, we can preserve the representation as
1281
# long as we change the refs to be inserted.
1282
self._data_changed = True
1283
ok_nodes.append((node[0], node[1], node[2],
1284
(ideal_parents, node[3][1])))
1285
self._data_changed = True
1287
# Reinsert this text completely
1288
bad_texts.append((node[1], ideal_parents))
1289
self._data_changed = True
1290
# we're finished with some data.
1293
# 3) bulk copy the ok data
1294
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1295
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1296
self.new_pack.text_index, readv_group_iter, total_items))
1297
# 4) adhoc copy all the other texts.
1298
# We have to topologically insert all texts otherwise we can fail to
1299
# reconcile when parts of a single delta chain are preserved intact,
1300
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1301
# reinserted, and if d3 has incorrect parents it will also be
1302
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1303
# copied), so we will try to delta, but d2 is not currently able to be
1304
# extracted because it's basis d1 is not present. Topologically sorting
1305
# addresses this. The following generates a sort for all the texts that
1306
# are being inserted without having to reference the entire text key
1307
# space (we only topo sort the revisions, which is smaller).
1308
topo_order = tsort.topo_sort(ancestors)
1309
rev_order = dict(zip(topo_order, range(len(topo_order))))
1310
bad_texts.sort(key=lambda key:rev_order.get(key[0][1], 0))
1311
transaction = repo.get_transaction()
1312
file_id_index = GraphIndexPrefixAdapter(
1313
self.new_pack.text_index,
1315
add_nodes_callback=self.new_pack.text_index.add_nodes)
1316
data_access = _DirectPackAccess(
1317
{self.new_pack.text_index:self.new_pack.access_tuple()})
1318
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1319
self.new_pack.access_tuple())
1320
output_texts = KnitVersionedFiles(
1321
_KnitGraphIndex(self.new_pack.text_index,
1322
add_callback=self.new_pack.text_index.add_nodes,
1323
deltas=True, parents=True, is_locked=repo.is_locked),
1324
data_access=data_access, max_delta_chain=200)
1325
for key, parent_keys in bad_texts:
1326
# We refer to the new pack to delta data being output.
1327
# A possible improvement would be to catch errors on short reads
1328
# and only flush then.
1329
self.new_pack.flush()
1331
for parent_key in parent_keys:
1332
if parent_key[0] != key[0]:
1333
# Graph parents must match the fileid
1334
raise errors.BzrError('Mismatched key parent %r:%r' %
1336
parents.append(parent_key[1])
1337
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1338
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1339
output_texts.add_lines(key, parent_keys, text_lines,
1340
random_id=True, check_content=False)
1341
# 5) check that nothing inserted has a reference outside the keyspace.
1342
missing_text_keys = self.new_pack.text_index._external_references()
1343
if missing_text_keys:
1344
raise errors.BzrCheckError('Reference to missing compression parents %r'
1345
% (missing_text_keys,))
1346
self._log_copied_texts()
1348
def _use_pack(self, new_pack):
1349
"""Override _use_pack to check for reconcile having changed content."""
1350
# XXX: we might be better checking this at the copy time.
1351
original_inventory_keys = set()
1352
inv_index = self._pack_collection.inventory_index.combined_index
1353
for entry in inv_index.iter_all_entries():
1354
original_inventory_keys.add(entry[1])
1355
new_inventory_keys = set()
1356
for entry in new_pack.inventory_index.iter_all_entries():
1357
new_inventory_keys.add(entry[1])
1358
if new_inventory_keys != original_inventory_keys:
1359
self._data_changed = True
1360
return new_pack.data_inserted() and self._data_changed
1363
class RepositoryPackCollection(object):
1364
"""Management of packs within a repository.
1366
:ivar _names: map of {pack_name: (index_size,)}
1369
pack_factory = NewPack
1370
resumed_pack_factory = ResumedPack
1372
def __init__(self, repo, transport, index_transport, upload_transport,
1373
pack_transport, index_builder_class, index_class,
1375
"""Create a new RepositoryPackCollection.
1377
:param transport: Addresses the repository base directory
1378
(typically .bzr/repository/).
1379
:param index_transport: Addresses the directory containing indices.
1380
:param upload_transport: Addresses the directory into which packs are written
1381
while they're being created.
1382
:param pack_transport: Addresses the directory of existing complete packs.
1383
:param index_builder_class: The index builder class to use.
1384
:param index_class: The index class to use.
1385
:param use_chk_index: Whether to setup and manage a CHK index.
1387
# XXX: This should call self.reset()
1389
self.transport = transport
1390
self._index_transport = index_transport
1391
self._upload_transport = upload_transport
1392
self._pack_transport = pack_transport
1393
self._index_builder_class = index_builder_class
1394
self._index_class = index_class
1395
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3,
1400
self._packs_by_name = {}
1401
# the previous pack-names content
1402
self._packs_at_load = None
1403
# when a pack is being created by this object, the state of that pack.
1404
self._new_pack = None
1405
# aggregated revision index data
1406
flush = self._flush_new_pack
1407
self.revision_index = AggregateIndex(self.reload_pack_names, flush)
1408
self.inventory_index = AggregateIndex(self.reload_pack_names, flush)
1409
self.text_index = AggregateIndex(self.reload_pack_names, flush)
1410
self.signature_index = AggregateIndex(self.reload_pack_names, flush)
1412
self.chk_index = AggregateIndex(self.reload_pack_names, flush)
1414
# used to determine if we're using a chk_index elsewhere.
1415
self.chk_index = None
1417
self._resumed_packs = []
1419
def add_pack_to_memory(self, pack):
1420
"""Make a Pack object available to the repository to satisfy queries.
1422
:param pack: A Pack object.
1424
if pack.name in self._packs_by_name:
1425
raise AssertionError(
1426
'pack %s already in _packs_by_name' % (pack.name,))
1427
self.packs.append(pack)
1428
self._packs_by_name[pack.name] = pack
1429
self.revision_index.add_index(pack.revision_index, pack)
1430
self.inventory_index.add_index(pack.inventory_index, pack)
1431
self.text_index.add_index(pack.text_index, pack)
1432
self.signature_index.add_index(pack.signature_index, pack)
1433
if self.chk_index is not None:
1434
self.chk_index.add_index(pack.chk_index, pack)
1436
def all_packs(self):
1437
"""Return a list of all the Pack objects this repository has.
1439
Note that an in-progress pack being created is not returned.
1441
:return: A list of Pack objects for all the packs in the repository.
1444
for name in self.names():
1445
result.append(self.get_pack_by_name(name))
1449
"""Pack the pack collection incrementally.
1451
This will not attempt global reorganisation or recompression,
1452
rather it will just ensure that the total number of packs does
1453
not grow without bound. It uses the _max_pack_count method to
1454
determine if autopacking is needed, and the pack_distribution
1455
method to determine the number of revisions in each pack.
1457
If autopacking takes place then the packs name collection will have
1458
been flushed to disk - packing requires updating the name collection
1459
in synchronisation with certain steps. Otherwise the names collection
1462
:return: Something evaluating true if packing took place.
1466
return self._do_autopack()
1467
except errors.RetryAutopack:
1468
# If we get a RetryAutopack exception, we should abort the
1469
# current action, and retry.
1472
def _do_autopack(self):
1473
# XXX: Should not be needed when the management of indices is sane.
1474
total_revisions = self.revision_index.combined_index.key_count()
1475
total_packs = len(self._names)
1476
if self._max_pack_count(total_revisions) >= total_packs:
1478
# determine which packs need changing
1479
pack_distribution = self.pack_distribution(total_revisions)
1481
for pack in self.all_packs():
1482
revision_count = pack.get_revision_count()
1483
if revision_count == 0:
1484
# revision less packs are not generated by normal operation,
1485
# only by operations like sign-my-commits, and thus will not
1486
# tend to grow rapdily or without bound like commit containing
1487
# packs do - leave them alone as packing them really should
1488
# group their data with the relevant commit, and that may
1489
# involve rewriting ancient history - which autopack tries to
1490
# avoid. Alternatively we could not group the data but treat
1491
# each of these as having a single revision, and thus add
1492
# one revision for each to the total revision count, to get
1493
# a matching distribution.
1495
existing_packs.append((revision_count, pack))
1496
pack_operations = self.plan_autopack_combinations(
1497
existing_packs, pack_distribution)
1498
num_new_packs = len(pack_operations)
1499
num_old_packs = sum([len(po[1]) for po in pack_operations])
1500
num_revs_affected = sum([po[0] for po in pack_operations])
1501
mutter('Auto-packing repository %s, which has %d pack files, '
1502
'containing %d revisions. Packing %d files into %d affecting %d'
1503
' revisions', self, total_packs, total_revisions, num_old_packs,
1504
num_new_packs, num_revs_affected)
1505
result = self._execute_pack_operations(pack_operations,
1506
reload_func=self._restart_autopack)
1507
mutter('Auto-packing repository %s completed', self)
1510
def _execute_pack_operations(self, pack_operations, _packer_class=Packer,
1512
"""Execute a series of pack operations.
1514
:param pack_operations: A list of [revision_count, packs_to_combine].
1515
:param _packer_class: The class of packer to use (default: Packer).
1516
:return: The new pack names.
1518
for revision_count, packs in pack_operations:
1519
# we may have no-ops from the setup logic
1522
packer = _packer_class(self, packs, '.autopack',
1523
reload_func=reload_func)
1526
except errors.RetryWithNewPacks:
1527
# An exception is propagating out of this context, make sure
1528
# this packer has cleaned up. Packer() doesn't set its new_pack
1529
# state into the RepositoryPackCollection object, so we only
1530
# have access to it directly here.
1531
if packer.new_pack is not None:
1532
packer.new_pack.abort()
1535
self._remove_pack_from_memory(pack)
1536
# record the newly available packs and stop advertising the old
1538
result = self._save_pack_names(clear_obsolete_packs=True)
1539
# Move the old packs out of the way now they are no longer referenced.
1540
for revision_count, packs in pack_operations:
1541
self._obsolete_packs(packs)
1544
def _flush_new_pack(self):
1545
if self._new_pack is not None:
1546
self._new_pack.flush()
1548
def lock_names(self):
1549
"""Acquire the mutex around the pack-names index.
1551
This cannot be used in the middle of a read-only transaction on the
1554
self.repo.control_files.lock_write()
1556
def _already_packed(self):
1557
"""Is the collection already packed?"""
1558
return not (self.repo._format.pack_compresses or (len(self._names) > 1))
1560
def pack(self, hint=None):
1561
"""Pack the pack collection totally."""
1562
self.ensure_loaded()
1563
total_packs = len(self._names)
1564
if self._already_packed():
1565
# This is arguably wrong because we might not be optimal, but for
1566
# now lets leave it in. (e.g. reconcile -> one pack. But not
1569
total_revisions = self.revision_index.combined_index.key_count()
1570
# XXX: the following may want to be a class, to pack with a given
1572
mutter('Packing repository %s, which has %d pack files, '
1573
'containing %d revisions into 1 packs.', self, total_packs,
1575
# determine which packs need changing
1576
pack_distribution = [1]
1577
pack_operations = [[0, []]]
1578
for pack in self.all_packs():
1579
pack_operations[-1][0] += pack.get_revision_count()
1580
pack_operations[-1][1].append(pack)
1581
self._execute_pack_operations(pack_operations, OptimisingPacker)
1583
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1584
"""Plan a pack operation.
1586
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1588
:param pack_distribution: A list with the number of revisions desired
1591
if len(existing_packs) <= len(pack_distribution):
1593
existing_packs.sort(reverse=True)
1594
pack_operations = [[0, []]]
1595
# plan out what packs to keep, and what to reorganise
1596
while len(existing_packs):
1597
# take the largest pack, and if its less than the head of the
1598
# distribution chart we will include its contents in the new pack
1599
# for that position. If its larger, we remove its size from the
1600
# distribution chart
1601
next_pack_rev_count, next_pack = existing_packs.pop(0)
1602
if next_pack_rev_count >= pack_distribution[0]:
1603
# this is already packed 'better' than this, so we can
1604
# not waste time packing it.
1605
while next_pack_rev_count > 0:
1606
next_pack_rev_count -= pack_distribution[0]
1607
if next_pack_rev_count >= 0:
1609
del pack_distribution[0]
1611
# didn't use that entire bucket up
1612
pack_distribution[0] = -next_pack_rev_count
1614
# add the revisions we're going to add to the next output pack
1615
pack_operations[-1][0] += next_pack_rev_count
1616
# allocate this pack to the next pack sub operation
1617
pack_operations[-1][1].append(next_pack)
1618
if pack_operations[-1][0] >= pack_distribution[0]:
1619
# this pack is used up, shift left.
1620
del pack_distribution[0]
1621
pack_operations.append([0, []])
1622
# Now that we know which pack files we want to move, shove them all
1623
# into a single pack file.
1625
final_pack_list = []
1626
for num_revs, pack_files in pack_operations:
1627
final_rev_count += num_revs
1628
final_pack_list.extend(pack_files)
1629
if len(final_pack_list) == 1:
1630
raise AssertionError('We somehow generated an autopack with a'
1631
' single pack file being moved.')
1633
return [[final_rev_count, final_pack_list]]
1635
def ensure_loaded(self):
1636
"""Ensure we have read names from disk.
1638
:return: True if the disk names had not been previously read.
1640
# NB: if you see an assertion error here, its probably access against
1641
# an unlocked repo. Naughty.
1642
if not self.repo.is_locked():
1643
raise errors.ObjectNotLocked(self.repo)
1644
if self._names is None:
1646
self._packs_at_load = set()
1647
for index, key, value in self._iter_disk_pack_index():
1649
self._names[name] = self._parse_index_sizes(value)
1650
self._packs_at_load.add((key, value))
1654
# populate all the metadata.
1658
def _parse_index_sizes(self, value):
1659
"""Parse a string of index sizes."""
1660
return tuple([int(digits) for digits in value.split(' ')])
1662
def get_pack_by_name(self, name):
1663
"""Get a Pack object by name.
1665
:param name: The name of the pack - e.g. '123456'
1666
:return: A Pack object.
1669
return self._packs_by_name[name]
1671
rev_index = self._make_index(name, '.rix')
1672
inv_index = self._make_index(name, '.iix')
1673
txt_index = self._make_index(name, '.tix')
1674
sig_index = self._make_index(name, '.six')
1675
if self.chk_index is not None:
1676
chk_index = self._make_index(name, '.cix')
1679
result = ExistingPack(self._pack_transport, name, rev_index,
1680
inv_index, txt_index, sig_index, chk_index)
1681
self.add_pack_to_memory(result)
1684
def _resume_pack(self, name):
1685
"""Get a suspended Pack object by name.
1687
:param name: The name of the pack - e.g. '123456'
1688
:return: A Pack object.
1690
if not re.match('[a-f0-9]{32}', name):
1691
# Tokens should be md5sums of the suspended pack file, i.e. 32 hex
1693
raise errors.UnresumableWriteGroup(
1694
self.repo, [name], 'Malformed write group token')
1696
rev_index = self._make_index(name, '.rix', resume=True)
1697
inv_index = self._make_index(name, '.iix', resume=True)
1698
txt_index = self._make_index(name, '.tix', resume=True)
1699
sig_index = self._make_index(name, '.six', resume=True)
1700
if self.chk_index is not None:
1701
chk_index = self._make_index(name, '.cix', resume=True)
1704
result = self.resumed_pack_factory(name, rev_index, inv_index,
1705
txt_index, sig_index, self._upload_transport,
1706
self._pack_transport, self._index_transport, self,
1707
chk_index=chk_index)
1708
except errors.NoSuchFile, e:
1709
raise errors.UnresumableWriteGroup(self.repo, [name], str(e))
1710
self.add_pack_to_memory(result)
1711
self._resumed_packs.append(result)
1714
def allocate(self, a_new_pack):
1715
"""Allocate name in the list of packs.
1717
:param a_new_pack: A NewPack instance to be added to the collection of
1718
packs for this repository.
1720
self.ensure_loaded()
1721
if a_new_pack.name in self._names:
1722
raise errors.BzrError(
1723
'Pack %r already exists in %s' % (a_new_pack.name, self))
1724
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1725
self.add_pack_to_memory(a_new_pack)
1727
def _iter_disk_pack_index(self):
1728
"""Iterate over the contents of the pack-names index.
1730
This is used when loading the list from disk, and before writing to
1731
detect updates from others during our write operation.
1732
:return: An iterator of the index contents.
1734
return self._index_class(self.transport, 'pack-names', None
1735
).iter_all_entries()
1737
def _make_index(self, name, suffix, resume=False):
1738
size_offset = self._suffix_offsets[suffix]
1739
index_name = name + suffix
1741
transport = self._upload_transport
1742
index_size = transport.stat(index_name).st_size
1744
transport = self._index_transport
1745
index_size = self._names[name][size_offset]
1746
return self._index_class(transport, index_name, index_size)
1748
def _max_pack_count(self, total_revisions):
1749
"""Return the maximum number of packs to use for total revisions.
1751
:param total_revisions: The total number of revisions in the
1754
if not total_revisions:
1756
digits = str(total_revisions)
1758
for digit in digits:
1759
result += int(digit)
1763
"""Provide an order to the underlying names."""
1764
return sorted(self._names.keys())
1766
def _obsolete_packs(self, packs):
1767
"""Move a number of packs which have been obsoleted out of the way.
1769
Each pack and its associated indices are moved out of the way.
1771
Note: for correctness this function should only be called after a new
1772
pack names index has been written without these pack names, and with
1773
the names of packs that contain the data previously available via these
1776
:param packs: The packs to obsolete.
1777
:param return: None.
1780
pack.pack_transport.rename(pack.file_name(),
1781
'../obsolete_packs/' + pack.file_name())
1782
# TODO: Probably needs to know all possible indices for this pack
1783
# - or maybe list the directory and move all indices matching this
1784
# name whether we recognize it or not?
1785
suffixes = ['.iix', '.six', '.tix', '.rix']
1786
if self.chk_index is not None:
1787
suffixes.append('.cix')
1788
for suffix in suffixes:
1789
self._index_transport.rename(pack.name + suffix,
1790
'../obsolete_packs/' + pack.name + suffix)
1792
def pack_distribution(self, total_revisions):
1793
"""Generate a list of the number of revisions to put in each pack.
1795
:param total_revisions: The total number of revisions in the
1798
if total_revisions == 0:
1800
digits = reversed(str(total_revisions))
1802
for exponent, count in enumerate(digits):
1803
size = 10 ** exponent
1804
for pos in range(int(count)):
1806
return list(reversed(result))
1808
def _pack_tuple(self, name):
1809
"""Return a tuple with the transport and file name for a pack name."""
1810
return self._pack_transport, name + '.pack'
1812
def _remove_pack_from_memory(self, pack):
1813
"""Remove pack from the packs accessed by this repository.
1815
Only affects memory state, until self._save_pack_names() is invoked.
1817
self._names.pop(pack.name)
1818
self._packs_by_name.pop(pack.name)
1819
self._remove_pack_indices(pack)
1820
self.packs.remove(pack)
1822
def _remove_pack_indices(self, pack):
1823
"""Remove the indices for pack from the aggregated indices."""
1824
self.revision_index.remove_index(pack.revision_index, pack)
1825
self.inventory_index.remove_index(pack.inventory_index, pack)
1826
self.text_index.remove_index(pack.text_index, pack)
1827
self.signature_index.remove_index(pack.signature_index, pack)
1828
if self.chk_index is not None:
1829
self.chk_index.remove_index(pack.chk_index, pack)
1832
"""Clear all cached data."""
1833
# cached revision data
1834
self.revision_index.clear()
1835
# cached signature data
1836
self.signature_index.clear()
1837
# cached file text data
1838
self.text_index.clear()
1839
# cached inventory data
1840
self.inventory_index.clear()
1842
if self.chk_index is not None:
1843
self.chk_index.clear()
1844
# remove the open pack
1845
self._new_pack = None
1846
# information about packs.
1849
self._packs_by_name = {}
1850
self._packs_at_load = None
1852
def _unlock_names(self):
1853
"""Release the mutex around the pack-names index."""
1854
self.repo.control_files.unlock()
1856
def _diff_pack_names(self):
1857
"""Read the pack names from disk, and compare it to the one in memory.
1859
:return: (disk_nodes, deleted_nodes, new_nodes)
1860
disk_nodes The final set of nodes that should be referenced
1861
deleted_nodes Nodes which have been removed from when we started
1862
new_nodes Nodes that are newly introduced
1864
# load the disk nodes across
1866
for index, key, value in self._iter_disk_pack_index():
1867
disk_nodes.add((key, value))
1869
# do a two-way diff against our original content
1870
current_nodes = set()
1871
for name, sizes in self._names.iteritems():
1873
((name, ), ' '.join(str(size) for size in sizes)))
1875
# Packs no longer present in the repository, which were present when we
1876
# locked the repository
1877
deleted_nodes = self._packs_at_load - current_nodes
1878
# Packs which this process is adding
1879
new_nodes = current_nodes - self._packs_at_load
1881
# Update the disk_nodes set to include the ones we are adding, and
1882
# remove the ones which were removed by someone else
1883
disk_nodes.difference_update(deleted_nodes)
1884
disk_nodes.update(new_nodes)
1886
return disk_nodes, deleted_nodes, new_nodes
1888
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1889
"""Given the correct set of pack files, update our saved info.
1891
:return: (removed, added, modified)
1892
removed pack names removed from self._names
1893
added pack names added to self._names
1894
modified pack names that had changed value
1899
## self._packs_at_load = disk_nodes
1900
new_names = dict(disk_nodes)
1901
# drop no longer present nodes
1902
for pack in self.all_packs():
1903
if (pack.name,) not in new_names:
1904
removed.append(pack.name)
1905
self._remove_pack_from_memory(pack)
1906
# add new nodes/refresh existing ones
1907
for key, value in disk_nodes:
1909
sizes = self._parse_index_sizes(value)
1910
if name in self._names:
1912
if sizes != self._names[name]:
1913
# the pack for name has had its indices replaced - rare but
1914
# important to handle. XXX: probably can never happen today
1915
# because the three-way merge code above does not handle it
1916
# - you may end up adding the same key twice to the new
1917
# disk index because the set values are the same, unless
1918
# the only index shows up as deleted by the set difference
1919
# - which it may. Until there is a specific test for this,
1920
# assume its broken. RBC 20071017.
1921
self._remove_pack_from_memory(self.get_pack_by_name(name))
1922
self._names[name] = sizes
1923
self.get_pack_by_name(name)
1924
modified.append(name)
1927
self._names[name] = sizes
1928
self.get_pack_by_name(name)
1930
return removed, added, modified
1932
def _save_pack_names(self, clear_obsolete_packs=False):
1933
"""Save the list of packs.
1935
This will take out the mutex around the pack names list for the
1936
duration of the method call. If concurrent updates have been made, a
1937
three-way merge between the current list and the current in memory list
1940
:param clear_obsolete_packs: If True, clear out the contents of the
1941
obsolete_packs directory.
1942
:return: A list of the names saved that were not previously on disk.
1946
builder = self._index_builder_class()
1947
disk_nodes, deleted_nodes, new_nodes = self._diff_pack_names()
1948
# TODO: handle same-name, index-size-changes here -
1949
# e.g. use the value from disk, not ours, *unless* we're the one
1951
for key, value in disk_nodes:
1952
builder.add_node(key, value)
1953
self.transport.put_file('pack-names', builder.finish(),
1954
mode=self.repo.bzrdir._get_file_mode())
1955
# move the baseline forward
1956
self._packs_at_load = disk_nodes
1957
if clear_obsolete_packs:
1958
self._clear_obsolete_packs()
1960
self._unlock_names()
1961
# synchronise the memory packs list with what we just wrote:
1962
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1963
return [new_node[0][0] for new_node in new_nodes]
1965
def reload_pack_names(self):
1966
"""Sync our pack listing with what is present in the repository.
1968
This should be called when we find out that something we thought was
1969
present is now missing. This happens when another process re-packs the
1972
:return: True if the in-memory list of packs has been altered at all.
1974
# The ensure_loaded call is to handle the case where the first call
1975
# made involving the collection was to reload_pack_names, where we
1976
# don't have a view of disk contents. Its a bit of a bandaid, and
1977
# causes two reads of pack-names, but its a rare corner case not struck
1978
# with regular push/pull etc.
1979
first_read = self.ensure_loaded()
1982
# out the new value.
1983
disk_nodes, _, _ = self._diff_pack_names()
1984
self._packs_at_load = disk_nodes
1986
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1987
if removed or added or modified:
1991
def _restart_autopack(self):
1992
"""Reload the pack names list, and restart the autopack code."""
1993
if not self.reload_pack_names():
1994
# Re-raise the original exception, because something went missing
1995
# and a restart didn't find it
1997
raise errors.RetryAutopack(self.repo, False, sys.exc_info())
1999
def _clear_obsolete_packs(self):
2000
"""Delete everything from the obsolete-packs directory.
2002
obsolete_pack_transport = self.transport.clone('obsolete_packs')
2003
for filename in obsolete_pack_transport.list_dir('.'):
2005
obsolete_pack_transport.delete(filename)
2006
except (errors.PathError, errors.TransportError), e:
2007
warning("couldn't delete obsolete pack, skipping it:\n%s" % (e,))
2009
def _start_write_group(self):
2010
# Do not permit preparation for writing if we're not in a 'write lock'.
2011
if not self.repo.is_write_locked():
2012
raise errors.NotWriteLocked(self)
2013
self._new_pack = self.pack_factory(self, upload_suffix='.pack',
2014
file_mode=self.repo.bzrdir._get_file_mode())
2015
# allow writing: queue writes to a new index
2016
self.revision_index.add_writable_index(self._new_pack.revision_index,
2018
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
2020
self.text_index.add_writable_index(self._new_pack.text_index,
2022
self._new_pack.text_index.set_optimize(combine_backing_indices=False)
2023
self.signature_index.add_writable_index(self._new_pack.signature_index,
2025
if self.chk_index is not None:
2026
self.chk_index.add_writable_index(self._new_pack.chk_index,
2028
self.repo.chk_bytes._index._add_callback = self.chk_index.add_callback
2029
self._new_pack.chk_index.set_optimize(combine_backing_indices=False)
2031
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
2032
self.repo.revisions._index._add_callback = self.revision_index.add_callback
2033
self.repo.signatures._index._add_callback = self.signature_index.add_callback
2034
self.repo.texts._index._add_callback = self.text_index.add_callback
2036
def _abort_write_group(self):
2037
# FIXME: just drop the transient index.
2038
# forget what names there are
2039
if self._new_pack is not None:
2041
self._new_pack.abort()
2043
# XXX: If we aborted while in the middle of finishing the write
2044
# group, _remove_pack_indices can fail because the indexes are
2045
# already gone. If they're not there we shouldn't fail in this
2046
# case. -- mbp 20081113
2047
self._remove_pack_indices(self._new_pack)
2048
self._new_pack = None
2049
for resumed_pack in self._resumed_packs:
2051
resumed_pack.abort()
2053
# See comment in previous finally block.
2055
self._remove_pack_indices(resumed_pack)
2058
del self._resumed_packs[:]
2060
def _remove_resumed_pack_indices(self):
2061
for resumed_pack in self._resumed_packs:
2062
self._remove_pack_indices(resumed_pack)
2063
del self._resumed_packs[:]
2065
def _commit_write_group(self):
2067
for prefix, versioned_file in (
2068
('revisions', self.repo.revisions),
2069
('inventories', self.repo.inventories),
2070
('texts', self.repo.texts),
2071
('signatures', self.repo.signatures),
2073
missing = versioned_file.get_missing_compression_parent_keys()
2074
all_missing.update([(prefix,) + key for key in missing])
2076
raise errors.BzrCheckError(
2077
"Repository %s has missing compression parent(s) %r "
2078
% (self.repo, sorted(all_missing)))
2079
self._remove_pack_indices(self._new_pack)
2080
should_autopack = False
2081
if self._new_pack.data_inserted():
2082
# get all the data to disk and read to use
2083
self._new_pack.finish()
2084
self.allocate(self._new_pack)
2085
self._new_pack = None
2086
should_autopack = True
2088
self._new_pack.abort()
2089
self._new_pack = None
2090
for resumed_pack in self._resumed_packs:
2091
# XXX: this is a pretty ugly way to turn the resumed pack into a
2092
# properly committed pack.
2093
self._names[resumed_pack.name] = None
2094
self._remove_pack_from_memory(resumed_pack)
2095
resumed_pack.finish()
2096
self.allocate(resumed_pack)
2097
should_autopack = True
2098
del self._resumed_packs[:]
2100
if not self.autopack():
2101
# when autopack takes no steps, the names list is still
2103
return self._save_pack_names()
2105
def _suspend_write_group(self):
2106
tokens = [pack.name for pack in self._resumed_packs]
2107
self._remove_pack_indices(self._new_pack)
2108
if self._new_pack.data_inserted():
2109
# get all the data to disk and read to use
2110
self._new_pack.finish(suspend=True)
2111
tokens.append(self._new_pack.name)
2112
self._new_pack = None
2114
self._new_pack.abort()
2115
self._new_pack = None
2116
self._remove_resumed_pack_indices()
2119
def _resume_write_group(self, tokens):
2120
for token in tokens:
2121
self._resume_pack(token)
2124
class KnitPackRepository(KnitRepository):
2125
"""Repository with knit objects stored inside pack containers.
2127
The layering for a KnitPackRepository is:
2129
Graph | HPSS | Repository public layer |
2130
===================================================
2131
Tuple based apis below, string based, and key based apis above
2132
---------------------------------------------------
2134
Provides .texts, .revisions etc
2135
This adapts the N-tuple keys to physical knit records which only have a
2136
single string identifier (for historical reasons), which in older formats
2137
was always the revision_id, and in the mapped code for packs is always
2138
the last element of key tuples.
2139
---------------------------------------------------
2141
A separate GraphIndex is used for each of the
2142
texts/inventories/revisions/signatures contained within each individual
2143
pack file. The GraphIndex layer works in N-tuples and is unaware of any
2145
===================================================
2149
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
2151
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2152
_commit_builder_class, _serializer)
2153
index_transport = self._transport.clone('indices')
2154
self._pack_collection = RepositoryPackCollection(self, self._transport,
2156
self._transport.clone('upload'),
2157
self._transport.clone('packs'),
2158
_format.index_builder_class,
2159
_format.index_class,
2160
use_chk_index=self._format.supports_chks,
2162
self.inventories = KnitVersionedFiles(
2163
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2164
add_callback=self._pack_collection.inventory_index.add_callback,
2165
deltas=True, parents=True, is_locked=self.is_locked),
2166
data_access=self._pack_collection.inventory_index.data_access,
2167
max_delta_chain=200)
2168
self.revisions = KnitVersionedFiles(
2169
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2170
add_callback=self._pack_collection.revision_index.add_callback,
2171
deltas=False, parents=True, is_locked=self.is_locked,
2172
track_external_parent_refs=True),
2173
data_access=self._pack_collection.revision_index.data_access,
2175
self.signatures = KnitVersionedFiles(
2176
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2177
add_callback=self._pack_collection.signature_index.add_callback,
2178
deltas=False, parents=False, is_locked=self.is_locked),
2179
data_access=self._pack_collection.signature_index.data_access,
2181
self.texts = KnitVersionedFiles(
2182
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2183
add_callback=self._pack_collection.text_index.add_callback,
2184
deltas=True, parents=True, is_locked=self.is_locked),
2185
data_access=self._pack_collection.text_index.data_access,
2186
max_delta_chain=200)
2187
if _format.supports_chks:
2188
# No graph, no compression:- references from chks are between
2189
# different objects not temporal versions of the same; and without
2190
# some sort of temporal structure knit compression will just fail.
2191
self.chk_bytes = KnitVersionedFiles(
2192
_KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2193
add_callback=self._pack_collection.chk_index.add_callback,
2194
deltas=False, parents=False, is_locked=self.is_locked),
2195
data_access=self._pack_collection.chk_index.data_access,
2198
self.chk_bytes = None
2199
# True when the repository object is 'write locked' (as opposed to the
2200
# physical lock only taken out around changes to the pack-names list.)
2201
# Another way to represent this would be a decorator around the control
2202
# files object that presents logical locks as physical ones - if this
2203
# gets ugly consider that alternative design. RBC 20071011
2204
self._write_lock_count = 0
2205
self._transaction = None
2207
self._reconcile_does_inventory_gc = True
2208
self._reconcile_fixes_text_parents = True
2209
self._reconcile_backsup_inventory = False
2211
def _warn_if_deprecated(self):
2212
# This class isn't deprecated, but one sub-format is
2213
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2214
from bzrlib import repository
2215
if repository._deprecation_warning_done:
2217
repository._deprecation_warning_done = True
2218
warning("Format %s for %s is deprecated - please use"
2219
" 'bzr upgrade --1.6.1-rich-root'"
2220
% (self._format, self.bzrdir.transport.base))
2222
def _abort_write_group(self):
2223
self.revisions._index._key_dependencies.refs.clear()
2224
self._pack_collection._abort_write_group()
2226
def _find_inconsistent_revision_parents(self):
2227
"""Find revisions with incorrectly cached parents.
2229
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
2230
parents-in-revision).
2232
if not self.is_locked():
2233
raise errors.ObjectNotLocked(self)
2234
pb = ui.ui_factory.nested_progress_bar()
2237
revision_nodes = self._pack_collection.revision_index \
2238
.combined_index.iter_all_entries()
2239
index_positions = []
2240
# Get the cached index values for all revisions, and also the
2241
# location in each index of the revision text so we can perform
2243
for index, key, value, refs in revision_nodes:
2244
node = (index, key, value, refs)
2245
index_memo = self.revisions._index._node_to_position(node)
2246
if index_memo[0] != index:
2247
raise AssertionError('%r != %r' % (index_memo[0], index))
2248
index_positions.append((index_memo, key[0],
2249
tuple(parent[0] for parent in refs[0])))
2250
pb.update("Reading revision index", 0, 0)
2251
index_positions.sort()
2253
pb.update("Checking cached revision graph", 0,
2254
len(index_positions))
2255
for offset in xrange(0, len(index_positions), 1000):
2256
pb.update("Checking cached revision graph", offset)
2257
to_query = index_positions[offset:offset + batch_size]
2260
rev_ids = [item[1] for item in to_query]
2261
revs = self.get_revisions(rev_ids)
2262
for revision, item in zip(revs, to_query):
2263
index_parents = item[2]
2264
rev_parents = tuple(revision.parent_ids)
2265
if index_parents != rev_parents:
2266
result.append((revision.revision_id, index_parents,
2272
def _get_source(self, to_format):
2273
if to_format.network_name() == self._format.network_name():
2274
return KnitPackStreamSource(self, to_format)
2275
return super(KnitPackRepository, self)._get_source(to_format)
2277
def _make_parents_provider(self):
2278
return graph.CachingParentsProvider(self)
2280
def _refresh_data(self):
2281
if not self.is_locked():
2283
self._pack_collection.reload_pack_names()
2285
def _start_write_group(self):
2286
self._pack_collection._start_write_group()
2288
def _commit_write_group(self):
2289
self.revisions._index._key_dependencies.refs.clear()
2290
return self._pack_collection._commit_write_group()
2292
def suspend_write_group(self):
2293
# XXX check self._write_group is self.get_transaction()?
2294
tokens = self._pack_collection._suspend_write_group()
2295
self.revisions._index._key_dependencies.refs.clear()
2296
self._write_group = None
2299
def _resume_write_group(self, tokens):
2300
self._start_write_group()
2302
self._pack_collection._resume_write_group(tokens)
2303
except errors.UnresumableWriteGroup:
2304
self._abort_write_group()
2306
for pack in self._pack_collection._resumed_packs:
2307
self.revisions._index.scan_unvalidated_index(pack.revision_index)
2309
def get_transaction(self):
2310
if self._write_lock_count:
2311
return self._transaction
2313
return self.control_files.get_transaction()
2315
def is_locked(self):
2316
return self._write_lock_count or self.control_files.is_locked()
2318
def is_write_locked(self):
2319
return self._write_lock_count
2321
def lock_write(self, token=None):
2322
locked = self.is_locked()
2323
if not self._write_lock_count and locked:
2324
raise errors.ReadOnlyError(self)
2325
self._write_lock_count += 1
2326
if self._write_lock_count == 1:
2327
self._transaction = transactions.WriteTransaction()
2329
for repo in self._fallback_repositories:
2330
# Writes don't affect fallback repos
2332
self._refresh_data()
2334
def lock_read(self):
2335
locked = self.is_locked()
2336
if self._write_lock_count:
2337
self._write_lock_count += 1
2339
self.control_files.lock_read()
2341
for repo in self._fallback_repositories:
2343
self._refresh_data()
2345
def leave_lock_in_place(self):
2346
# not supported - raise an error
2347
raise NotImplementedError(self.leave_lock_in_place)
2349
def dont_leave_lock_in_place(self):
2350
# not supported - raise an error
2351
raise NotImplementedError(self.dont_leave_lock_in_place)
2354
def pack(self, hint=None):
2355
"""Compress the data within the repository.
2357
This will pack all the data to a single pack. In future it may
2358
recompress deltas or do other such expensive operations.
2360
self._pack_collection.pack(hint=hint)
2363
def reconcile(self, other=None, thorough=False):
2364
"""Reconcile this repository."""
2365
from bzrlib.reconcile import PackReconciler
2366
reconciler = PackReconciler(self, thorough=thorough)
2367
reconciler.reconcile()
2370
def _reconcile_pack(self, collection, packs, extension, revs, pb):
2371
packer = ReconcilePacker(collection, packs, extension, revs)
2372
return packer.pack(pb)
2375
if self._write_lock_count == 1 and self._write_group is not None:
2376
self.abort_write_group()
2377
self._transaction = None
2378
self._write_lock_count = 0
2379
raise errors.BzrError(
2380
'Must end write group before releasing write lock on %s'
2382
if self._write_lock_count:
2383
self._write_lock_count -= 1
2384
if not self._write_lock_count:
2385
transaction = self._transaction
2386
self._transaction = None
2387
transaction.finish()
2389
self.control_files.unlock()
2391
if not self.is_locked():
2392
for repo in self._fallback_repositories:
2396
class KnitPackStreamSource(StreamSource):
2397
"""A StreamSource used to transfer data between same-format KnitPack repos.
2399
This source assumes:
2400
1) Same serialization format for all objects
2401
2) Same root information
2402
3) XML format inventories
2403
4) Atomic inserts (so we can stream inventory texts before text
2408
def __init__(self, from_repository, to_format):
2409
super(KnitPackStreamSource, self).__init__(from_repository, to_format)
2410
self._text_keys = None
2411
self._text_fetch_order = 'unordered'
2413
def _get_filtered_inv_stream(self, revision_ids):
2414
from_repo = self.from_repository
2415
parent_ids = from_repo._find_parent_ids_of_revisions(revision_ids)
2416
parent_keys = [(p,) for p in parent_ids]
2417
find_text_keys = from_repo._find_text_key_references_from_xml_inventory_lines
2418
parent_text_keys = set(find_text_keys(
2419
from_repo._inventory_xml_lines_for_keys(parent_keys)))
2420
content_text_keys = set()
2421
knit = KnitVersionedFiles(None, None)
2422
factory = KnitPlainFactory()
2423
def find_text_keys_from_content(record):
2424
if record.storage_kind not in ('knit-delta-gz', 'knit-ft-gz'):
2425
raise ValueError("Unknown content storage kind for"
2426
" inventory text: %s" % (record.storage_kind,))
2427
# It's a knit record, it has a _raw_record field (even if it was
2428
# reconstituted from a network stream).
2429
raw_data = record._raw_record
2430
# read the entire thing
2431
revision_id = record.key[-1]
2432
content, _ = knit._parse_record(revision_id, raw_data)
2433
if record.storage_kind == 'knit-delta-gz':
2434
line_iterator = factory.get_linedelta_content(content)
2435
elif record.storage_kind == 'knit-ft-gz':
2436
line_iterator = factory.get_fulltext_content(content)
2437
content_text_keys.update(find_text_keys(
2438
[(line, revision_id) for line in line_iterator]))
2439
revision_keys = [(r,) for r in revision_ids]
2440
def _filtered_inv_stream():
2441
source_vf = from_repo.inventories
2442
stream = source_vf.get_record_stream(revision_keys,
2444
for record in stream:
2445
if record.storage_kind == 'absent':
2446
raise errors.NoSuchRevision(from_repo, record.key)
2447
find_text_keys_from_content(record)
2449
self._text_keys = content_text_keys - parent_text_keys
2450
return ('inventories', _filtered_inv_stream())
2452
def _get_text_stream(self):
2453
# Note: We know we don't have to handle adding root keys, because both
2454
# the source and target are the identical network name.
2455
text_stream = self.from_repository.texts.get_record_stream(
2456
self._text_keys, self._text_fetch_order, False)
2457
return ('texts', text_stream)
2459
def get_stream(self, search):
2460
revision_ids = search.get_keys()
2461
for stream_info in self._fetch_revision_texts(revision_ids):
2463
self._revision_keys = [(rev_id,) for rev_id in revision_ids]
2464
yield self._get_filtered_inv_stream(revision_ids)
2465
yield self._get_text_stream()
2469
class RepositoryFormatPack(MetaDirRepositoryFormat):
2470
"""Format logic for pack structured repositories.
2472
This repository format has:
2473
- a list of packs in pack-names
2474
- packs in packs/NAME.pack
2475
- indices in indices/NAME.{iix,six,tix,rix}
2476
- knit deltas in the packs, knit indices mapped to the indices.
2477
- thunk objects to support the knits programming API.
2478
- a format marker of its own
2479
- an optional 'shared-storage' flag
2480
- an optional 'no-working-trees' flag
2484
# Set this attribute in derived classes to control the repository class
2485
# created by open and initialize.
2486
repository_class = None
2487
# Set this attribute in derived classes to control the
2488
# _commit_builder_class that the repository objects will have passed to
2489
# their constructor.
2490
_commit_builder_class = None
2491
# Set this attribute in derived clases to control the _serializer that the
2492
# repository objects will have passed to their constructor.
2494
# Packs are not confused by ghosts.
2495
supports_ghosts = True
2496
# External references are not supported in pack repositories yet.
2497
supports_external_lookups = False
2498
# Most pack formats do not use chk lookups.
2499
supports_chks = False
2500
# What index classes to use
2501
index_builder_class = None
2503
_fetch_uses_deltas = True
2506
def initialize(self, a_bzrdir, shared=False):
2507
"""Create a pack based repository.
2509
:param a_bzrdir: bzrdir to contain the new repository; must already
2511
:param shared: If true the repository will be initialized as a shared
2514
mutter('creating repository in %s.', a_bzrdir.transport.base)
2515
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
2516
builder = self.index_builder_class()
2517
files = [('pack-names', builder.finish())]
2518
utf8_files = [('format', self.get_format_string())]
2520
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
2521
return self.open(a_bzrdir=a_bzrdir, _found=True)
2523
def open(self, a_bzrdir, _found=False, _override_transport=None):
2524
"""See RepositoryFormat.open().
2526
:param _override_transport: INTERNAL USE ONLY. Allows opening the
2527
repository at a slightly different url
2528
than normal. I.e. during 'upgrade'.
2531
format = RepositoryFormat.find_format(a_bzrdir)
2532
if _override_transport is not None:
2533
repo_transport = _override_transport
2535
repo_transport = a_bzrdir.get_repository_transport(None)
2536
control_files = lockable_files.LockableFiles(repo_transport,
2537
'lock', lockdir.LockDir)
2538
return self.repository_class(_format=self,
2540
control_files=control_files,
2541
_commit_builder_class=self._commit_builder_class,
2542
_serializer=self._serializer)
2545
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2546
"""A no-subtrees parameterized Pack repository.
2548
This format was introduced in 0.92.
2551
repository_class = KnitPackRepository
2552
_commit_builder_class = PackCommitBuilder
2554
def _serializer(self):
2555
return xml5.serializer_v5
2556
# What index classes to use
2557
index_builder_class = InMemoryGraphIndex
2558
index_class = GraphIndex
2560
def _get_matching_bzrdir(self):
2561
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2563
def _ignore_setting_bzrdir(self, format):
2566
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2568
def get_format_string(self):
2569
"""See RepositoryFormat.get_format_string()."""
2570
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2572
def get_format_description(self):
2573
"""See RepositoryFormat.get_format_description()."""
2574
return "Packs containing knits without subtree support"
2576
def check_conversion_target(self, target_format):
2580
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2581
"""A subtrees parameterized Pack repository.
2583
This repository format uses the xml7 serializer to get:
2584
- support for recording full info about the tree root
2585
- support for recording tree-references
2587
This format was introduced in 0.92.
2590
repository_class = KnitPackRepository
2591
_commit_builder_class = PackRootCommitBuilder
2592
rich_root_data = True
2593
supports_tree_reference = True
2595
def _serializer(self):
2596
return xml7.serializer_v7
2597
# What index classes to use
2598
index_builder_class = InMemoryGraphIndex
2599
index_class = GraphIndex
2601
def _get_matching_bzrdir(self):
2602
return bzrdir.format_registry.make_bzrdir(
2603
'pack-0.92-subtree')
2605
def _ignore_setting_bzrdir(self, format):
2608
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2610
def check_conversion_target(self, target_format):
2611
if not target_format.rich_root_data:
2612
raise errors.BadConversionTarget(
2613
'Does not support rich root data.', target_format)
2614
if not getattr(target_format, 'supports_tree_reference', False):
2615
raise errors.BadConversionTarget(
2616
'Does not support nested trees', target_format)
2618
def get_format_string(self):
2619
"""See RepositoryFormat.get_format_string()."""
2620
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2622
def get_format_description(self):
2623
"""See RepositoryFormat.get_format_description()."""
2624
return "Packs containing knits with subtree support\n"
2627
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2628
"""A rich-root, no subtrees parameterized Pack repository.
2630
This repository format uses the xml6 serializer to get:
2631
- support for recording full info about the tree root
2633
This format was introduced in 1.0.
2636
repository_class = KnitPackRepository
2637
_commit_builder_class = PackRootCommitBuilder
2638
rich_root_data = True
2639
supports_tree_reference = False
2641
def _serializer(self):
2642
return xml6.serializer_v6
2643
# What index classes to use
2644
index_builder_class = InMemoryGraphIndex
2645
index_class = GraphIndex
2647
def _get_matching_bzrdir(self):
2648
return bzrdir.format_registry.make_bzrdir(
2651
def _ignore_setting_bzrdir(self, format):
2654
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2656
def check_conversion_target(self, target_format):
2657
if not target_format.rich_root_data:
2658
raise errors.BadConversionTarget(
2659
'Does not support rich root data.', target_format)
2661
def get_format_string(self):
2662
"""See RepositoryFormat.get_format_string()."""
2663
return ("Bazaar pack repository format 1 with rich root"
2664
" (needs bzr 1.0)\n")
2666
def get_format_description(self):
2667
"""See RepositoryFormat.get_format_description()."""
2668
return "Packs containing knits with rich root support\n"
2671
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2672
"""Repository that supports external references to allow stacking.
2676
Supports external lookups, which results in non-truncated ghosts after
2677
reconcile compared to pack-0.92 formats.
2680
repository_class = KnitPackRepository
2681
_commit_builder_class = PackCommitBuilder
2682
supports_external_lookups = True
2683
# What index classes to use
2684
index_builder_class = InMemoryGraphIndex
2685
index_class = GraphIndex
2688
def _serializer(self):
2689
return xml5.serializer_v5
2691
def _get_matching_bzrdir(self):
2692
return bzrdir.format_registry.make_bzrdir('1.6')
2694
def _ignore_setting_bzrdir(self, format):
2697
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2699
def get_format_string(self):
2700
"""See RepositoryFormat.get_format_string()."""
2701
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2703
def get_format_description(self):
2704
"""See RepositoryFormat.get_format_description()."""
2705
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2707
def check_conversion_target(self, target_format):
2711
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2712
"""A repository with rich roots and stacking.
2714
New in release 1.6.1.
2716
Supports stacking on other repositories, allowing data to be accessed
2717
without being stored locally.
2720
repository_class = KnitPackRepository
2721
_commit_builder_class = PackRootCommitBuilder
2722
rich_root_data = True
2723
supports_tree_reference = False # no subtrees
2724
supports_external_lookups = True
2725
# What index classes to use
2726
index_builder_class = InMemoryGraphIndex
2727
index_class = GraphIndex
2730
def _serializer(self):
2731
return xml6.serializer_v6
2733
def _get_matching_bzrdir(self):
2734
return bzrdir.format_registry.make_bzrdir(
2737
def _ignore_setting_bzrdir(self, format):
2740
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2742
def check_conversion_target(self, target_format):
2743
if not target_format.rich_root_data:
2744
raise errors.BadConversionTarget(
2745
'Does not support rich root data.', target_format)
2747
def get_format_string(self):
2748
"""See RepositoryFormat.get_format_string()."""
2749
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2751
def get_format_description(self):
2752
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2755
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2756
"""A repository with rich roots and external references.
2760
Supports external lookups, which results in non-truncated ghosts after
2761
reconcile compared to pack-0.92 formats.
2763
This format was deprecated because the serializer it uses accidentally
2764
supported subtrees, when the format was not intended to. This meant that
2765
someone could accidentally fetch from an incorrect repository.
2768
repository_class = KnitPackRepository
2769
_commit_builder_class = PackRootCommitBuilder
2770
rich_root_data = True
2771
supports_tree_reference = False # no subtrees
2773
supports_external_lookups = True
2774
# What index classes to use
2775
index_builder_class = InMemoryGraphIndex
2776
index_class = GraphIndex
2779
def _serializer(self):
2780
return xml7.serializer_v7
2782
def _get_matching_bzrdir(self):
2783
matching = bzrdir.format_registry.make_bzrdir(
2785
matching.repository_format = self
2788
def _ignore_setting_bzrdir(self, format):
2791
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2793
def check_conversion_target(self, target_format):
2794
if not target_format.rich_root_data:
2795
raise errors.BadConversionTarget(
2796
'Does not support rich root data.', target_format)
2798
def get_format_string(self):
2799
"""See RepositoryFormat.get_format_string()."""
2800
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2802
def get_format_description(self):
2803
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2807
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2808
"""A repository with stacking and btree indexes,
2809
without rich roots or subtrees.
2811
This is equivalent to pack-1.6 with B+Tree indices.
2814
repository_class = KnitPackRepository
2815
_commit_builder_class = PackCommitBuilder
2816
supports_external_lookups = True
2817
# What index classes to use
2818
index_builder_class = BTreeBuilder
2819
index_class = BTreeGraphIndex
2822
def _serializer(self):
2823
return xml5.serializer_v5
2825
def _get_matching_bzrdir(self):
2826
return bzrdir.format_registry.make_bzrdir('1.9')
2828
def _ignore_setting_bzrdir(self, format):
2831
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2833
def get_format_string(self):
2834
"""See RepositoryFormat.get_format_string()."""
2835
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2837
def get_format_description(self):
2838
"""See RepositoryFormat.get_format_description()."""
2839
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2841
def check_conversion_target(self, target_format):
2845
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2846
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2848
1.6-rich-root with B+Tree indices.
2851
repository_class = KnitPackRepository
2852
_commit_builder_class = PackRootCommitBuilder
2853
rich_root_data = True
2854
supports_tree_reference = False # no subtrees
2855
supports_external_lookups = True
2856
# What index classes to use
2857
index_builder_class = BTreeBuilder
2858
index_class = BTreeGraphIndex
2861
def _serializer(self):
2862
return xml6.serializer_v6
2864
def _get_matching_bzrdir(self):
2865
return bzrdir.format_registry.make_bzrdir(
2868
def _ignore_setting_bzrdir(self, format):
2871
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2873
def check_conversion_target(self, target_format):
2874
if not target_format.rich_root_data:
2875
raise errors.BadConversionTarget(
2876
'Does not support rich root data.', target_format)
2878
def get_format_string(self):
2879
"""See RepositoryFormat.get_format_string()."""
2880
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2882
def get_format_description(self):
2883
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2886
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2887
"""A subtrees development repository.
2889
This format should be retained until the second release after bzr 1.7.
2891
1.6.1-subtree[as it might have been] with B+Tree indices.
2893
This is [now] retained until we have a CHK based subtree format in
2897
repository_class = KnitPackRepository
2898
_commit_builder_class = PackRootCommitBuilder
2899
rich_root_data = True
2900
supports_tree_reference = True
2901
supports_external_lookups = True
2902
# What index classes to use
2903
index_builder_class = BTreeBuilder
2904
index_class = BTreeGraphIndex
2907
def _serializer(self):
2908
return xml7.serializer_v7
2910
def _get_matching_bzrdir(self):
2911
return bzrdir.format_registry.make_bzrdir(
2912
'development-subtree')
2914
def _ignore_setting_bzrdir(self, format):
2917
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2919
def check_conversion_target(self, target_format):
2920
if not target_format.rich_root_data:
2921
raise errors.BadConversionTarget(
2922
'Does not support rich root data.', target_format)
2923
if not getattr(target_format, 'supports_tree_reference', False):
2924
raise errors.BadConversionTarget(
2925
'Does not support nested trees', target_format)
2927
def get_format_string(self):
2928
"""See RepositoryFormat.get_format_string()."""
2929
return ("Bazaar development format 2 with subtree support "
2930
"(needs bzr.dev from before 1.8)\n")
2932
def get_format_description(self):
2933
"""See RepositoryFormat.get_format_description()."""
2934
return ("Development repository format, currently the same as "
2935
"1.6.1-subtree with B+Tree indices.\n")
added
removed
bzrlib/repofmt/pack_repo.py
