1
# Copyright (C) 2007-2010 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
38
from bzrlib.index import (
40
GraphIndexPrefixAdapter,
42
from bzrlib.knit import (
48
from bzrlib import tsort
55
revision as _mod_revision,
58
from bzrlib.decorators import needs_write_lock, only_raises
59
from bzrlib.btree_index import (
63
from bzrlib.index import (
67
from bzrlib.repofmt.knitrepo import KnitRepository
68
from bzrlib.repository import (
70
MetaDirRepositoryFormat,
75
from bzrlib.trace import (
82
class PackCommitBuilder(CommitBuilder):
83
"""A subclass of CommitBuilder to add texts with pack semantics.
85
Specifically this uses one knit object rather than one knit object per
86
added text, reducing memory and object pressure.
89
def __init__(self, repository, parents, config, timestamp=None,
90
timezone=None, committer=None, revprops=None,
92
CommitBuilder.__init__(self, repository, parents, config,
93
timestamp=timestamp, timezone=timezone, committer=committer,
94
revprops=revprops, revision_id=revision_id)
95
self._file_graph = graph.Graph(
96
repository._pack_collection.text_index.combined_index)
98
def _heads(self, file_id, revision_ids):
99
keys = [(file_id, revision_id) for revision_id in revision_ids]
100
return set([key[1] for key in self._file_graph.heads(keys)])
103
class PackRootCommitBuilder(RootCommitBuilder):
104
"""A subclass of RootCommitBuilder to add texts with pack semantics.
106
Specifically this uses one knit object rather than one knit object per
107
added text, reducing memory and object pressure.
110
def __init__(self, repository, parents, config, timestamp=None,
111
timezone=None, committer=None, revprops=None,
113
CommitBuilder.__init__(self, repository, parents, config,
114
timestamp=timestamp, timezone=timezone, committer=committer,
115
revprops=revprops, revision_id=revision_id)
116
self._file_graph = graph.Graph(
117
repository._pack_collection.text_index.combined_index)
119
def _heads(self, file_id, revision_ids):
120
keys = [(file_id, revision_id) for revision_id in revision_ids]
121
return set([key[1] for key in self._file_graph.heads(keys)])
125
"""An in memory proxy for a pack and its indices.
127
This is a base class that is not directly used, instead the classes
128
ExistingPack and NewPack are used.
131
# A map of index 'type' to the file extension and position in the
133
index_definitions = {
135
'revision': ('.rix', 0),
136
'inventory': ('.iix', 1),
138
'signature': ('.six', 3),
141
def __init__(self, revision_index, inventory_index, text_index,
142
signature_index, chk_index=None):
143
"""Create a pack instance.
145
:param revision_index: A GraphIndex for determining what revisions are
146
present in the Pack and accessing the locations of their texts.
147
:param inventory_index: A GraphIndex for determining what inventories are
148
present in the Pack and accessing the locations of their
150
:param text_index: A GraphIndex for determining what file texts
151
are present in the pack and accessing the locations of their
152
texts/deltas (via (fileid, revisionid) tuples).
153
:param signature_index: A GraphIndex for determining what signatures are
154
present in the Pack and accessing the locations of their texts.
155
:param chk_index: A GraphIndex for accessing content by CHK, if the
158
self.revision_index = revision_index
159
self.inventory_index = inventory_index
160
self.text_index = text_index
161
self.signature_index = signature_index
162
self.chk_index = chk_index
164
def access_tuple(self):
165
"""Return a tuple (transport, name) for the pack content."""
166
return self.pack_transport, self.file_name()
168
def _check_references(self):
169
"""Make sure our external references are present.
171
Packs are allowed to have deltas whose base is not in the pack, but it
172
must be present somewhere in this collection. It is not allowed to
173
have deltas based on a fallback repository.
174
(See <https://bugs.launchpad.net/bzr/+bug/288751>)
177
for (index_name, external_refs, index) in [
179
self._get_external_refs(self.text_index),
180
self._pack_collection.text_index.combined_index),
182
self._get_external_refs(self.inventory_index),
183
self._pack_collection.inventory_index.combined_index),
185
missing = external_refs.difference(
186
k for (idx, k, v, r) in
187
index.iter_entries(external_refs))
189
missing_items[index_name] = sorted(list(missing))
191
from pprint import pformat
192
raise errors.BzrCheckError(
193
"Newly created pack file %r has delta references to "
194
"items not in its repository:\n%s"
195
% (self, pformat(missing_items)))
198
"""Get the file name for the pack on disk."""
199
return self.name + '.pack'
201
def get_revision_count(self):
202
return self.revision_index.key_count()
204
def index_name(self, index_type, name):
205
"""Get the disk name of an index type for pack name 'name'."""
206
return name + Pack.index_definitions[index_type][0]
208
def index_offset(self, index_type):
209
"""Get the position in a index_size array for a given index type."""
210
return Pack.index_definitions[index_type][1]
212
def inventory_index_name(self, name):
213
"""The inv index is the name + .iix."""
214
return self.index_name('inventory', name)
216
def revision_index_name(self, name):
217
"""The revision index is the name + .rix."""
218
return self.index_name('revision', name)
220
def signature_index_name(self, name):
221
"""The signature index is the name + .six."""
222
return self.index_name('signature', name)
224
def text_index_name(self, name):
225
"""The text index is the name + .tix."""
226
return self.index_name('text', name)
228
def _replace_index_with_readonly(self, index_type):
229
unlimited_cache = False
230
if index_type == 'chk':
231
unlimited_cache = True
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)],
236
unlimited_cache=unlimited_cache))
239
class ExistingPack(Pack):
240
"""An in memory proxy for an existing .pack and its disk indices."""
242
def __init__(self, pack_transport, name, revision_index, inventory_index,
243
text_index, signature_index, chk_index=None):
244
"""Create an ExistingPack object.
246
:param pack_transport: The transport where the pack file resides.
247
:param name: The name of the pack on disk in the pack_transport.
249
Pack.__init__(self, revision_index, inventory_index, text_index,
250
signature_index, chk_index)
252
self.pack_transport = pack_transport
253
if None in (revision_index, inventory_index, text_index,
254
signature_index, name, pack_transport):
255
raise AssertionError()
257
def __eq__(self, other):
258
return self.__dict__ == other.__dict__
260
def __ne__(self, other):
261
return not self.__eq__(other)
264
return "<%s.%s object at 0x%x, %s, %s" % (
265
self.__class__.__module__, self.__class__.__name__, id(self),
266
self.pack_transport, self.name)
269
class ResumedPack(ExistingPack):
271
def __init__(self, name, revision_index, inventory_index, text_index,
272
signature_index, upload_transport, pack_transport, index_transport,
273
pack_collection, chk_index=None):
274
"""Create a ResumedPack object."""
275
ExistingPack.__init__(self, pack_transport, name, revision_index,
276
inventory_index, text_index, signature_index,
278
self.upload_transport = upload_transport
279
self.index_transport = index_transport
280
self.index_sizes = [None, None, None, None]
282
('revision', revision_index),
283
('inventory', inventory_index),
284
('text', text_index),
285
('signature', signature_index),
287
if chk_index is not None:
288
indices.append(('chk', chk_index))
289
self.index_sizes.append(None)
290
for index_type, index in indices:
291
offset = self.index_offset(index_type)
292
self.index_sizes[offset] = index._size
293
self.index_class = pack_collection._index_class
294
self._pack_collection = pack_collection
295
self._state = 'resumed'
296
# XXX: perhaps check that the .pack file exists?
298
def access_tuple(self):
299
if self._state == 'finished':
300
return Pack.access_tuple(self)
301
elif self._state == 'resumed':
302
return self.upload_transport, self.file_name()
304
raise AssertionError(self._state)
307
self.upload_transport.delete(self.file_name())
308
indices = [self.revision_index, self.inventory_index, self.text_index,
309
self.signature_index]
310
if self.chk_index is not None:
311
indices.append(self.chk_index)
312
for index in indices:
313
index._transport.delete(index._name)
316
self._check_references()
317
index_types = ['revision', 'inventory', 'text', 'signature']
318
if self.chk_index is not None:
319
index_types.append('chk')
320
for index_type in index_types:
321
old_name = self.index_name(index_type, self.name)
322
new_name = '../indices/' + old_name
323
self.upload_transport.rename(old_name, new_name)
324
self._replace_index_with_readonly(index_type)
325
new_name = '../packs/' + self.file_name()
326
self.upload_transport.rename(self.file_name(), new_name)
327
self._state = 'finished'
329
def _get_external_refs(self, index):
330
"""Return compression parents for this index that are not present.
332
This returns any compression parents that are referenced by this index,
333
which are not contained *in* this index. They may be present elsewhere.
335
return index.external_references(1)
339
"""An in memory proxy for a pack which is being created."""
341
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
342
"""Create a NewPack instance.
344
:param pack_collection: A PackCollection into which this is being inserted.
345
:param upload_suffix: An optional suffix to be given to any temporary
346
files created during the pack creation. e.g '.autopack'
347
:param file_mode: Unix permissions for newly created file.
349
# The relative locations of the packs are constrained, but all are
350
# passed in because the caller has them, so as to avoid object churn.
351
index_builder_class = pack_collection._index_builder_class
352
if pack_collection.chk_index is not None:
353
chk_index = index_builder_class(reference_lists=0)
357
# Revisions: parents list, no text compression.
358
index_builder_class(reference_lists=1),
359
# Inventory: We want to map compression only, but currently the
360
# knit code hasn't been updated enough to understand that, so we
361
# have a regular 2-list index giving parents and compression
363
index_builder_class(reference_lists=2),
364
# Texts: compression and per file graph, for all fileids - so two
365
# reference lists and two elements in the key tuple.
366
index_builder_class(reference_lists=2, key_elements=2),
367
# Signatures: Just blobs to store, no compression, no parents
369
index_builder_class(reference_lists=0),
370
# CHK based storage - just blobs, no compression or parents.
373
self._pack_collection = pack_collection
374
# When we make readonly indices, we need this.
375
self.index_class = pack_collection._index_class
376
# where should the new pack be opened
377
self.upload_transport = pack_collection._upload_transport
378
# where are indices written out to
379
self.index_transport = pack_collection._index_transport
380
# where is the pack renamed to when it is finished?
381
self.pack_transport = pack_collection._pack_transport
382
# What file mode to upload the pack and indices with.
383
self._file_mode = file_mode
384
# tracks the content written to the .pack file.
385
self._hash = osutils.md5()
386
# a tuple with the length in bytes of the indices, once the pack
387
# is finalised. (rev, inv, text, sigs, chk_if_in_use)
388
self.index_sizes = None
389
# How much data to cache when writing packs. Note that this is not
390
# synchronised with reads, because it's not in the transport layer, so
391
# is not safe unless the client knows it won't be reading from the pack
393
self._cache_limit = 0
394
# the temporary pack file name.
395
self.random_name = osutils.rand_chars(20) + upload_suffix
396
# when was this pack started ?
397
self.start_time = time.time()
398
# open an output stream for the data added to the pack.
399
self.write_stream = self.upload_transport.open_write_stream(
400
self.random_name, mode=self._file_mode)
401
if 'pack' in debug.debug_flags:
402
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
403
time.ctime(), self.upload_transport.base, self.random_name,
404
time.time() - self.start_time)
405
# A list of byte sequences to be written to the new pack, and the
406
# aggregate size of them. Stored as a list rather than separate
407
# variables so that the _write_data closure below can update them.
408
self._buffer = [[], 0]
409
# create a callable for adding data
411
# robertc says- this is a closure rather than a method on the object
412
# so that the variables are locals, and faster than accessing object
414
def _write_data(bytes, flush=False, _buffer=self._buffer,
415
_write=self.write_stream.write, _update=self._hash.update):
416
_buffer[0].append(bytes)
417
_buffer[1] += len(bytes)
419
if _buffer[1] > self._cache_limit or flush:
420
bytes = ''.join(_buffer[0])
424
# expose this on self, for the occasion when clients want to add data.
425
self._write_data = _write_data
426
# a pack writer object to serialise pack records.
427
self._writer = pack.ContainerWriter(self._write_data)
429
# what state is the pack in? (open, finished, aborted)
431
# no name until we finish writing the content
435
"""Cancel creating this pack."""
436
self._state = 'aborted'
437
self.write_stream.close()
438
# Remove the temporary pack file.
439
self.upload_transport.delete(self.random_name)
440
# The indices have no state on disk.
442
def access_tuple(self):
443
"""Return a tuple (transport, name) for the pack content."""
444
if self._state == 'finished':
445
return Pack.access_tuple(self)
446
elif self._state == 'open':
447
return self.upload_transport, self.random_name
449
raise AssertionError(self._state)
451
def data_inserted(self):
452
"""True if data has been added to this pack."""
453
return bool(self.get_revision_count() or
454
self.inventory_index.key_count() or
455
self.text_index.key_count() or
456
self.signature_index.key_count() or
457
(self.chk_index is not None and self.chk_index.key_count()))
459
def finish_content(self):
460
if self.name is not None:
464
self._write_data('', flush=True)
465
self.name = self._hash.hexdigest()
467
def finish(self, suspend=False):
468
"""Finish the new pack.
471
- finalises the content
472
- assigns a name (the md5 of the content, currently)
473
- writes out the associated indices
474
- renames the pack into place.
475
- stores the index size tuple for the pack in the index_sizes
478
self.finish_content()
480
self._check_references()
482
# XXX: It'd be better to write them all to temporary names, then
483
# rename them all into place, so that the window when only some are
484
# visible is smaller. On the other hand none will be seen until
485
# they're in the names list.
486
self.index_sizes = [None, None, None, None]
487
self._write_index('revision', self.revision_index, 'revision', suspend)
488
self._write_index('inventory', self.inventory_index, 'inventory',
490
self._write_index('text', self.text_index, 'file texts', suspend)
491
self._write_index('signature', self.signature_index,
492
'revision signatures', suspend)
493
if self.chk_index is not None:
494
self.index_sizes.append(None)
495
self._write_index('chk', self.chk_index,
496
'content hash bytes', suspend)
497
self.write_stream.close()
498
# Note that this will clobber an existing pack with the same name,
499
# without checking for hash collisions. While this is undesirable this
500
# is something that can be rectified in a subsequent release. One way
501
# to rectify it may be to leave the pack at the original name, writing
502
# its pack-names entry as something like 'HASH: index-sizes
503
# temporary-name'. Allocate that and check for collisions, if it is
504
# collision free then rename it into place. If clients know this scheme
505
# they can handle missing-file errors by:
506
# - try for HASH.pack
507
# - try for temporary-name
508
# - refresh the pack-list to see if the pack is now absent
509
new_name = self.name + '.pack'
511
new_name = '../packs/' + new_name
512
self.upload_transport.rename(self.random_name, new_name)
513
self._state = 'finished'
514
if 'pack' in debug.debug_flags:
515
# XXX: size might be interesting?
516
mutter('%s: create_pack: pack finished: %s%s->%s t+%6.3fs',
517
time.ctime(), self.upload_transport.base, self.random_name,
518
new_name, time.time() - self.start_time)
521
"""Flush any current data."""
523
bytes = ''.join(self._buffer[0])
524
self.write_stream.write(bytes)
525
self._hash.update(bytes)
526
self._buffer[:] = [[], 0]
528
def _get_external_refs(self, index):
529
return index._external_references()
531
def set_write_cache_size(self, size):
532
self._cache_limit = size
534
def _write_index(self, index_type, index, label, suspend=False):
535
"""Write out an index.
537
:param index_type: The type of index to write - e.g. 'revision'.
538
:param index: The index object to serialise.
539
:param label: What label to give the index e.g. 'revision'.
541
index_name = self.index_name(index_type, self.name)
543
transport = self.upload_transport
545
transport = self.index_transport
546
self.index_sizes[self.index_offset(index_type)] = transport.put_file(
547
index_name, index.finish(), mode=self._file_mode)
548
if 'pack' in debug.debug_flags:
549
# XXX: size might be interesting?
550
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
551
time.ctime(), label, self.upload_transport.base,
552
self.random_name, time.time() - self.start_time)
553
# Replace the writable index on this object with a readonly,
554
# presently unloaded index. We should alter
555
# the index layer to make its finish() error if add_node is
556
# subsequently used. RBC
557
self._replace_index_with_readonly(index_type)
560
class AggregateIndex(object):
561
"""An aggregated index for the RepositoryPackCollection.
563
AggregateIndex is reponsible for managing the PackAccess object,
564
Index-To-Pack mapping, and all indices list for a specific type of index
565
such as 'revision index'.
567
A CombinedIndex provides an index on a single key space built up
568
from several on-disk indices. The AggregateIndex builds on this
569
to provide a knit access layer, and allows having up to one writable
570
index within the collection.
572
# XXX: Probably 'can be written to' could/should be separated from 'acts
573
# like a knit index' -- mbp 20071024
575
def __init__(self, reload_func=None, flush_func=None):
576
"""Create an AggregateIndex.
578
:param reload_func: A function to call if we find we are missing an
579
index. Should have the form reload_func() => True if the list of
580
active pack files has changed.
582
self._reload_func = reload_func
583
self.index_to_pack = {}
584
self.combined_index = CombinedGraphIndex([], reload_func=reload_func)
585
self.data_access = _DirectPackAccess(self.index_to_pack,
586
reload_func=reload_func,
587
flush_func=flush_func)
588
self.add_callback = None
590
def add_index(self, index, pack):
591
"""Add index to the aggregate, which is an index for Pack pack.
593
Future searches on the aggregate index will seach this new index
594
before all previously inserted indices.
596
:param index: An Index for the pack.
597
:param pack: A Pack instance.
599
# expose it to the index map
600
self.index_to_pack[index] = pack.access_tuple()
601
# put it at the front of the linear index list
602
self.combined_index.insert_index(0, index, pack.name)
604
def add_writable_index(self, index, pack):
605
"""Add an index which is able to have data added to it.
607
There can be at most one writable index at any time. Any
608
modifications made to the knit are put into this index.
610
:param index: An index from the pack parameter.
611
:param pack: A Pack instance.
613
if self.add_callback is not None:
614
raise AssertionError(
615
"%s already has a writable index through %s" % \
616
(self, self.add_callback))
617
# allow writing: queue writes to a new index
618
self.add_index(index, pack)
619
# Updates the index to packs mapping as a side effect,
620
self.data_access.set_writer(pack._writer, index, pack.access_tuple())
621
self.add_callback = index.add_nodes
624
"""Reset all the aggregate data to nothing."""
625
self.data_access.set_writer(None, None, (None, None))
626
self.index_to_pack.clear()
627
del self.combined_index._indices[:]
628
del self.combined_index._index_names[:]
629
self.add_callback = None
631
def remove_index(self, index):
632
"""Remove index from the indices used to answer queries.
634
:param index: An index from the pack parameter.
636
del self.index_to_pack[index]
637
pos = self.combined_index._indices.index(index)
638
del self.combined_index._indices[pos]
639
del self.combined_index._index_names[pos]
640
if (self.add_callback is not None and
641
getattr(index, 'add_nodes', None) == self.add_callback):
642
self.add_callback = None
643
self.data_access.set_writer(None, None, (None, None))
646
class Packer(object):
647
"""Create a pack from packs."""
649
def __init__(self, pack_collection, packs, suffix, revision_ids=None,
653
:param pack_collection: A RepositoryPackCollection object where the
654
new pack is being written to.
655
:param packs: The packs to combine.
656
:param suffix: The suffix to use on the temporary files for the pack.
657
:param revision_ids: Revision ids to limit the pack to.
658
:param reload_func: A function to call if a pack file/index goes
659
missing. The side effect of calling this function should be to
660
update self.packs. See also AggregateIndex
664
self.revision_ids = revision_ids
665
# The pack object we are creating.
667
self._pack_collection = pack_collection
668
self._reload_func = reload_func
669
# The index layer keys for the revisions being copied. None for 'all
671
self._revision_keys = None
672
# What text keys to copy. None for 'all texts'. This is set by
673
# _copy_inventory_texts
674
self._text_filter = None
677
def _extra_init(self):
678
"""A template hook to allow extending the constructor trivially."""
680
def _pack_map_and_index_list(self, index_attribute):
681
"""Convert a list of packs to an index pack map and index list.
683
:param index_attribute: The attribute that the desired index is found
685
:return: A tuple (map, list) where map contains the dict from
686
index:pack_tuple, and list contains the indices in the preferred
691
for pack_obj in self.packs:
692
index = getattr(pack_obj, index_attribute)
693
indices.append(index)
694
pack_map[index] = pack_obj
695
return pack_map, indices
697
def _index_contents(self, indices, key_filter=None):
698
"""Get an iterable of the index contents from a pack_map.
700
:param indices: The list of indices to query
701
:param key_filter: An optional filter to limit the keys returned.
703
all_index = CombinedGraphIndex(indices)
704
if key_filter is None:
705
return all_index.iter_all_entries()
707
return all_index.iter_entries(key_filter)
709
def pack(self, pb=None):
710
"""Create a new pack by reading data from other packs.
712
This does little more than a bulk copy of data. One key difference
713
is that data with the same item key across multiple packs is elided
714
from the output. The new pack is written into the current pack store
715
along with its indices, and the name added to the pack names. The
716
source packs are not altered and are not required to be in the current
719
:param pb: An optional progress bar to use. A nested bar is created if
721
:return: A Pack object, or None if nothing was copied.
723
# open a pack - using the same name as the last temporary file
724
# - which has already been flushed, so its safe.
725
# XXX: - duplicate code warning with start_write_group; fix before
726
# considering 'done'.
727
if self._pack_collection._new_pack is not None:
728
raise errors.BzrError('call to %s.pack() while another pack is'
730
% (self.__class__.__name__,))
731
if self.revision_ids is not None:
732
if len(self.revision_ids) == 0:
733
# silly fetch request.
736
self.revision_ids = frozenset(self.revision_ids)
737
self.revision_keys = frozenset((revid,) for revid in
740
self.pb = ui.ui_factory.nested_progress_bar()
744
return self._create_pack_from_packs()
750
"""Open a pack for the pack we are creating."""
751
new_pack = self._pack_collection.pack_factory(self._pack_collection,
752
upload_suffix=self.suffix,
753
file_mode=self._pack_collection.repo.bzrdir._get_file_mode())
754
# We know that we will process all nodes in order, and don't need to
755
# query, so don't combine any indices spilled to disk until we are done
756
new_pack.revision_index.set_optimize(combine_backing_indices=False)
757
new_pack.inventory_index.set_optimize(combine_backing_indices=False)
758
new_pack.text_index.set_optimize(combine_backing_indices=False)
759
new_pack.signature_index.set_optimize(combine_backing_indices=False)
762
def _update_pack_order(self, entries, index_to_pack_map):
763
"""Determine how we want our packs to be ordered.
765
This changes the sort order of the self.packs list so that packs unused
766
by 'entries' will be at the end of the list, so that future requests
767
can avoid probing them. Used packs will be at the front of the
768
self.packs list, in the order of their first use in 'entries'.
770
:param entries: A list of (index, ...) tuples
771
:param index_to_pack_map: A mapping from index objects to pack objects.
775
for entry in entries:
777
if index not in seen_indexes:
778
packs.append(index_to_pack_map[index])
779
seen_indexes.add(index)
780
if len(packs) == len(self.packs):
781
if 'pack' in debug.debug_flags:
782
mutter('Not changing pack list, all packs used.')
784
seen_packs = set(packs)
785
for pack in self.packs:
786
if pack not in seen_packs:
789
if 'pack' in debug.debug_flags:
790
old_names = [p.access_tuple()[1] for p in self.packs]
791
new_names = [p.access_tuple()[1] for p in packs]
792
mutter('Reordering packs\nfrom: %s\n to: %s',
793
old_names, new_names)
796
def _copy_revision_texts(self):
797
"""Copy revision data to the new pack."""
799
if self.revision_ids:
800
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
803
# select revision keys
804
revision_index_map, revision_indices = self._pack_map_and_index_list(
806
revision_nodes = self._index_contents(revision_indices, revision_keys)
807
revision_nodes = list(revision_nodes)
808
self._update_pack_order(revision_nodes, revision_index_map)
809
# copy revision keys and adjust values
810
self.pb.update("Copying revision texts", 1)
811
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
812
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
813
self.new_pack.revision_index, readv_group_iter, total_items))
814
if 'pack' in debug.debug_flags:
815
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
816
time.ctime(), self._pack_collection._upload_transport.base,
817
self.new_pack.random_name,
818
self.new_pack.revision_index.key_count(),
819
time.time() - self.new_pack.start_time)
820
self._revision_keys = revision_keys
822
def _copy_inventory_texts(self):
823
"""Copy the inventory texts to the new pack.
825
self._revision_keys is used to determine what inventories to copy.
827
Sets self._text_filter appropriately.
829
# select inventory keys
830
inv_keys = self._revision_keys # currently the same keyspace, and note that
831
# querying for keys here could introduce a bug where an inventory item
832
# is missed, so do not change it to query separately without cross
833
# checking like the text key check below.
834
inventory_index_map, inventory_indices = self._pack_map_and_index_list(
836
inv_nodes = self._index_contents(inventory_indices, inv_keys)
837
# copy inventory keys and adjust values
838
# XXX: Should be a helper function to allow different inv representation
840
self.pb.update("Copying inventory texts", 2)
841
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
842
# Only grab the output lines if we will be processing them
843
output_lines = bool(self.revision_ids)
844
inv_lines = self._copy_nodes_graph(inventory_index_map,
845
self.new_pack._writer, self.new_pack.inventory_index,
846
readv_group_iter, total_items, output_lines=output_lines)
847
if self.revision_ids:
848
self._process_inventory_lines(inv_lines)
850
# eat the iterator to cause it to execute.
852
self._text_filter = None
853
if 'pack' in debug.debug_flags:
854
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
855
time.ctime(), self._pack_collection._upload_transport.base,
856
self.new_pack.random_name,
857
self.new_pack.inventory_index.key_count(),
858
time.time() - self.new_pack.start_time)
860
def _copy_text_texts(self):
862
text_index_map, text_nodes = self._get_text_nodes()
863
if self._text_filter is not None:
864
# We could return the keys copied as part of the return value from
865
# _copy_nodes_graph but this doesn't work all that well with the
866
# need to get line output too, so we check separately, and as we're
867
# going to buffer everything anyway, we check beforehand, which
868
# saves reading knit data over the wire when we know there are
870
text_nodes = set(text_nodes)
871
present_text_keys = set(_node[1] for _node in text_nodes)
872
missing_text_keys = set(self._text_filter) - present_text_keys
873
if missing_text_keys:
874
# TODO: raise a specific error that can handle many missing
876
mutter("missing keys during fetch: %r", missing_text_keys)
877
a_missing_key = missing_text_keys.pop()
878
raise errors.RevisionNotPresent(a_missing_key[1],
880
# copy text keys and adjust values
881
self.pb.update("Copying content texts", 3)
882
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
883
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
884
self.new_pack.text_index, readv_group_iter, total_items))
885
self._log_copied_texts()
887
def _create_pack_from_packs(self):
888
self.pb.update("Opening pack", 0, 5)
889
self.new_pack = self.open_pack()
890
new_pack = self.new_pack
891
# buffer data - we won't be reading-back during the pack creation and
892
# this makes a significant difference on sftp pushes.
893
new_pack.set_write_cache_size(1024*1024)
894
if 'pack' in debug.debug_flags:
895
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
896
for a_pack in self.packs]
897
if self.revision_ids is not None:
898
rev_count = len(self.revision_ids)
901
mutter('%s: create_pack: creating pack from source packs: '
902
'%s%s %s revisions wanted %s t=0',
903
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
904
plain_pack_list, rev_count)
905
self._copy_revision_texts()
906
self._copy_inventory_texts()
907
self._copy_text_texts()
908
# select signature keys
909
signature_filter = self._revision_keys # same keyspace
910
signature_index_map, signature_indices = self._pack_map_and_index_list(
912
signature_nodes = self._index_contents(signature_indices,
914
# copy signature keys and adjust values
915
self.pb.update("Copying signature texts", 4)
916
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
917
new_pack.signature_index)
918
if 'pack' in debug.debug_flags:
919
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
920
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
921
new_pack.signature_index.key_count(),
922
time.time() - new_pack.start_time)
924
# NB XXX: how to check CHK references are present? perhaps by yielding
925
# the items? How should that interact with stacked repos?
926
if new_pack.chk_index is not None:
928
if 'pack' in debug.debug_flags:
929
mutter('%s: create_pack: chk content copied: %s%s %d items t+%6.3fs',
930
time.ctime(), self._pack_collection._upload_transport.base,
931
new_pack.random_name,
932
new_pack.chk_index.key_count(),
933
time.time() - new_pack.start_time)
934
new_pack._check_references()
935
if not self._use_pack(new_pack):
938
self.pb.update("Finishing pack", 5)
940
self._pack_collection.allocate(new_pack)
943
def _copy_chks(self, refs=None):
944
# XXX: Todo, recursive follow-pointers facility when fetching some
946
chk_index_map, chk_indices = self._pack_map_and_index_list(
948
chk_nodes = self._index_contents(chk_indices, refs)
950
# TODO: This isn't strictly tasteful as we are accessing some private
951
# variables (_serializer). Perhaps a better way would be to have
952
# Repository._deserialise_chk_node()
953
search_key_func = chk_map.search_key_registry.get(
954
self._pack_collection.repo._serializer.search_key_name)
955
def accumlate_refs(lines):
956
# XXX: move to a generic location
958
bytes = ''.join(lines)
959
node = chk_map._deserialise(bytes, ("unknown",), search_key_func)
960
new_refs.update(node.refs())
961
self._copy_nodes(chk_nodes, chk_index_map, self.new_pack._writer,
962
self.new_pack.chk_index, output_lines=accumlate_refs)
965
def _copy_nodes(self, nodes, index_map, writer, write_index,
967
"""Copy knit nodes between packs with no graph references.
969
:param output_lines: Output full texts of copied items.
971
pb = ui.ui_factory.nested_progress_bar()
973
return self._do_copy_nodes(nodes, index_map, writer,
974
write_index, pb, output_lines=output_lines)
978
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb,
980
# for record verification
981
knit = KnitVersionedFiles(None, None)
982
# plan a readv on each source pack:
984
nodes = sorted(nodes)
985
# how to map this into knit.py - or knit.py into this?
986
# we don't want the typical knit logic, we want grouping by pack
987
# at this point - perhaps a helper library for the following code
988
# duplication points?
990
for index, key, value in nodes:
991
if index not in request_groups:
992
request_groups[index] = []
993
request_groups[index].append((key, value))
995
pb.update("Copied record", record_index, len(nodes))
996
for index, items in request_groups.iteritems():
997
pack_readv_requests = []
998
for key, value in items:
999
# ---- KnitGraphIndex.get_position
1000
bits = value[1:].split(' ')
1001
offset, length = int(bits[0]), int(bits[1])
1002
pack_readv_requests.append((offset, length, (key, value[0])))
1003
# linear scan up the pack
1004
pack_readv_requests.sort()
1006
pack_obj = index_map[index]
1007
transport, path = pack_obj.access_tuple()
1009
reader = pack.make_readv_reader(transport, path,
1010
[offset[0:2] for offset in pack_readv_requests])
1011
except errors.NoSuchFile:
1012
if self._reload_func is not None:
1015
for (names, read_func), (_1, _2, (key, eol_flag)) in \
1016
izip(reader.iter_records(), pack_readv_requests):
1017
raw_data = read_func(None)
1018
# check the header only
1019
if output_lines is not None:
1020
output_lines(knit._parse_record(key[-1], raw_data)[0])
1022
df, _ = knit._parse_record_header(key, raw_data)
1024
pos, size = writer.add_bytes_record(raw_data, names)
1025
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
1026
pb.update("Copied record", record_index)
1029
def _copy_nodes_graph(self, index_map, writer, write_index,
1030
readv_group_iter, total_items, output_lines=False):
1031
"""Copy knit nodes between packs.
1033
:param output_lines: Return lines present in the copied data as
1034
an iterator of line,version_id.
1036
pb = ui.ui_factory.nested_progress_bar()
1038
for result in self._do_copy_nodes_graph(index_map, writer,
1039
write_index, output_lines, pb, readv_group_iter, total_items):
1042
# Python 2.4 does not permit try:finally: in a generator.
1048
def _do_copy_nodes_graph(self, index_map, writer, write_index,
1049
output_lines, pb, readv_group_iter, total_items):
1050
# for record verification
1051
knit = KnitVersionedFiles(None, None)
1052
# for line extraction when requested (inventories only)
1054
factory = KnitPlainFactory()
1056
pb.update("Copied record", record_index, total_items)
1057
for index, readv_vector, node_vector in readv_group_iter:
1059
pack_obj = index_map[index]
1060
transport, path = pack_obj.access_tuple()
1062
reader = pack.make_readv_reader(transport, path, readv_vector)
1063
except errors.NoSuchFile:
1064
if self._reload_func is not None:
1067
for (names, read_func), (key, eol_flag, references) in \
1068
izip(reader.iter_records(), node_vector):
1069
raw_data = read_func(None)
1071
# read the entire thing
1072
content, _ = knit._parse_record(key[-1], raw_data)
1073
if len(references[-1]) == 0:
1074
line_iterator = factory.get_fulltext_content(content)
1076
line_iterator = factory.get_linedelta_content(content)
1077
for line in line_iterator:
1080
# check the header only
1081
df, _ = knit._parse_record_header(key, raw_data)
1083
pos, size = writer.add_bytes_record(raw_data, names)
1084
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1085
pb.update("Copied record", record_index)
1088
def _get_text_nodes(self):
1089
text_index_map, text_indices = self._pack_map_and_index_list(
1091
return text_index_map, self._index_contents(text_indices,
1094
def _least_readv_node_readv(self, nodes):
1095
"""Generate request groups for nodes using the least readv's.
1097
:param nodes: An iterable of graph index nodes.
1098
:return: Total node count and an iterator of the data needed to perform
1099
readvs to obtain the data for nodes. Each item yielded by the
1100
iterator is a tuple with:
1101
index, readv_vector, node_vector. readv_vector is a list ready to
1102
hand to the transport readv method, and node_vector is a list of
1103
(key, eol_flag, references) for the node retrieved by the
1104
matching readv_vector.
1106
# group by pack so we do one readv per pack
1107
nodes = sorted(nodes)
1110
for index, key, value, references in nodes:
1111
if index not in request_groups:
1112
request_groups[index] = []
1113
request_groups[index].append((key, value, references))
1115
for index, items in request_groups.iteritems():
1116
pack_readv_requests = []
1117
for key, value, references in items:
1118
# ---- KnitGraphIndex.get_position
1119
bits = value[1:].split(' ')
1120
offset, length = int(bits[0]), int(bits[1])
1121
pack_readv_requests.append(
1122
((offset, length), (key, value[0], references)))
1123
# linear scan up the pack to maximum range combining.
1124
pack_readv_requests.sort()
1125
# split out the readv and the node data.
1126
pack_readv = [readv for readv, node in pack_readv_requests]
1127
node_vector = [node for readv, node in pack_readv_requests]
1128
result.append((index, pack_readv, node_vector))
1129
return total, result
1131
def _log_copied_texts(self):
1132
if 'pack' in debug.debug_flags:
1133
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
1134
time.ctime(), self._pack_collection._upload_transport.base,
1135
self.new_pack.random_name,
1136
self.new_pack.text_index.key_count(),
1137
time.time() - self.new_pack.start_time)
1139
def _process_inventory_lines(self, inv_lines):
1140
"""Use up the inv_lines generator and setup a text key filter."""
1141
repo = self._pack_collection.repo
1142
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
1143
inv_lines, self.revision_keys)
1145
for fileid, file_revids in fileid_revisions.iteritems():
1146
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
1147
self._text_filter = text_filter
1149
def _revision_node_readv(self, revision_nodes):
1150
"""Return the total revisions and the readv's to issue.
1152
:param revision_nodes: The revision index contents for the packs being
1153
incorporated into the new pack.
1154
:return: As per _least_readv_node_readv.
1156
return self._least_readv_node_readv(revision_nodes)
1158
def _use_pack(self, new_pack):
1159
"""Return True if new_pack should be used.
1161
:param new_pack: The pack that has just been created.
1162
:return: True if the pack should be used.
1164
return new_pack.data_inserted()
1167
class OptimisingPacker(Packer):
1168
"""A packer which spends more time to create better disk layouts."""
1170
def _revision_node_readv(self, revision_nodes):
1171
"""Return the total revisions and the readv's to issue.
1173
This sort places revisions in topological order with the ancestors
1176
:param revision_nodes: The revision index contents for the packs being
1177
incorporated into the new pack.
1178
:return: As per _least_readv_node_readv.
1180
# build an ancestors dict
1183
for index, key, value, references in revision_nodes:
1184
ancestors[key] = references[0]
1185
by_key[key] = (index, value, references)
1186
order = tsort.topo_sort(ancestors)
1188
# Single IO is pathological, but it will work as a starting point.
1190
for key in reversed(order):
1191
index, value, references = by_key[key]
1192
# ---- KnitGraphIndex.get_position
1193
bits = value[1:].split(' ')
1194
offset, length = int(bits[0]), int(bits[1])
1196
(index, [(offset, length)], [(key, value[0], references)]))
1197
# TODO: combine requests in the same index that are in ascending order.
1198
return total, requests
1200
def open_pack(self):
1201
"""Open a pack for the pack we are creating."""
1202
new_pack = super(OptimisingPacker, self).open_pack()
1203
# Turn on the optimization flags for all the index builders.
1204
new_pack.revision_index.set_optimize(for_size=True)
1205
new_pack.inventory_index.set_optimize(for_size=True)
1206
new_pack.text_index.set_optimize(for_size=True)
1207
new_pack.signature_index.set_optimize(for_size=True)
1211
class ReconcilePacker(Packer):
1212
"""A packer which regenerates indices etc as it copies.
1214
This is used by ``bzr reconcile`` to cause parent text pointers to be
1218
def _extra_init(self):
1219
self._data_changed = False
1221
def _process_inventory_lines(self, inv_lines):
1222
"""Generate a text key reference map rather for reconciling with."""
1223
repo = self._pack_collection.repo
1224
refs = repo._find_text_key_references_from_xml_inventory_lines(
1226
self._text_refs = refs
1227
# during reconcile we:
1228
# - convert unreferenced texts to full texts
1229
# - correct texts which reference a text not copied to be full texts
1230
# - copy all others as-is but with corrected parents.
1231
# - so at this point we don't know enough to decide what becomes a full
1233
self._text_filter = None
1235
def _copy_text_texts(self):
1236
"""generate what texts we should have and then copy."""
1237
self.pb.update("Copying content texts", 3)
1238
# we have three major tasks here:
1239
# 1) generate the ideal index
1240
repo = self._pack_collection.repo
1241
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1242
_1, key, _2, refs in
1243
self.new_pack.revision_index.iter_all_entries()])
1244
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1245
# 2) generate a text_nodes list that contains all the deltas that can
1246
# be used as-is, with corrected parents.
1249
discarded_nodes = []
1250
NULL_REVISION = _mod_revision.NULL_REVISION
1251
text_index_map, text_nodes = self._get_text_nodes()
1252
for node in text_nodes:
1258
ideal_parents = tuple(ideal_index[node[1]])
1260
discarded_nodes.append(node)
1261
self._data_changed = True
1263
if ideal_parents == (NULL_REVISION,):
1265
if ideal_parents == node[3][0]:
1267
ok_nodes.append(node)
1268
elif ideal_parents[0:1] == node[3][0][0:1]:
1269
# the left most parent is the same, or there are no parents
1270
# today. Either way, we can preserve the representation as
1271
# long as we change the refs to be inserted.
1272
self._data_changed = True
1273
ok_nodes.append((node[0], node[1], node[2],
1274
(ideal_parents, node[3][1])))
1275
self._data_changed = True
1277
# Reinsert this text completely
1278
bad_texts.append((node[1], ideal_parents))
1279
self._data_changed = True
1280
# we're finished with some data.
1283
# 3) bulk copy the ok data
1284
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1285
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1286
self.new_pack.text_index, readv_group_iter, total_items))
1287
# 4) adhoc copy all the other texts.
1288
# We have to topologically insert all texts otherwise we can fail to
1289
# reconcile when parts of a single delta chain are preserved intact,
1290
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1291
# reinserted, and if d3 has incorrect parents it will also be
1292
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1293
# copied), so we will try to delta, but d2 is not currently able to be
1294
# extracted because it's basis d1 is not present. Topologically sorting
1295
# addresses this. The following generates a sort for all the texts that
1296
# are being inserted without having to reference the entire text key
1297
# space (we only topo sort the revisions, which is smaller).
1298
topo_order = tsort.topo_sort(ancestors)
1299
rev_order = dict(zip(topo_order, range(len(topo_order))))
1300
bad_texts.sort(key=lambda key:rev_order.get(key[0][1], 0))
1301
transaction = repo.get_transaction()
1302
file_id_index = GraphIndexPrefixAdapter(
1303
self.new_pack.text_index,
1305
add_nodes_callback=self.new_pack.text_index.add_nodes)
1306
data_access = _DirectPackAccess(
1307
{self.new_pack.text_index:self.new_pack.access_tuple()})
1308
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1309
self.new_pack.access_tuple())
1310
output_texts = KnitVersionedFiles(
1311
_KnitGraphIndex(self.new_pack.text_index,
1312
add_callback=self.new_pack.text_index.add_nodes,
1313
deltas=True, parents=True, is_locked=repo.is_locked),
1314
data_access=data_access, max_delta_chain=200)
1315
for key, parent_keys in bad_texts:
1316
# We refer to the new pack to delta data being output.
1317
# A possible improvement would be to catch errors on short reads
1318
# and only flush then.
1319
self.new_pack.flush()
1321
for parent_key in parent_keys:
1322
if parent_key[0] != key[0]:
1323
# Graph parents must match the fileid
1324
raise errors.BzrError('Mismatched key parent %r:%r' %
1326
parents.append(parent_key[1])
1327
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1328
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1329
output_texts.add_lines(key, parent_keys, text_lines,
1330
random_id=True, check_content=False)
1331
# 5) check that nothing inserted has a reference outside the keyspace.
1332
missing_text_keys = self.new_pack.text_index._external_references()
1333
if missing_text_keys:
1334
raise errors.BzrCheckError('Reference to missing compression parents %r'
1335
% (missing_text_keys,))
1336
self._log_copied_texts()
1338
def _use_pack(self, new_pack):
1339
"""Override _use_pack to check for reconcile having changed content."""
1340
# XXX: we might be better checking this at the copy time.
1341
original_inventory_keys = set()
1342
inv_index = self._pack_collection.inventory_index.combined_index
1343
for entry in inv_index.iter_all_entries():
1344
original_inventory_keys.add(entry[1])
1345
new_inventory_keys = set()
1346
for entry in new_pack.inventory_index.iter_all_entries():
1347
new_inventory_keys.add(entry[1])
1348
if new_inventory_keys != original_inventory_keys:
1349
self._data_changed = True
1350
return new_pack.data_inserted() and self._data_changed
1353
class RepositoryPackCollection(object):
1354
"""Management of packs within a repository.
1356
:ivar _names: map of {pack_name: (index_size,)}
1359
pack_factory = NewPack
1360
resumed_pack_factory = ResumedPack
1362
def __init__(self, repo, transport, index_transport, upload_transport,
1363
pack_transport, index_builder_class, index_class,
1365
"""Create a new RepositoryPackCollection.
1367
:param transport: Addresses the repository base directory
1368
(typically .bzr/repository/).
1369
:param index_transport: Addresses the directory containing indices.
1370
:param upload_transport: Addresses the directory into which packs are written
1371
while they're being created.
1372
:param pack_transport: Addresses the directory of existing complete packs.
1373
:param index_builder_class: The index builder class to use.
1374
:param index_class: The index class to use.
1375
:param use_chk_index: Whether to setup and manage a CHK index.
1377
# XXX: This should call self.reset()
1379
self.transport = transport
1380
self._index_transport = index_transport
1381
self._upload_transport = upload_transport
1382
self._pack_transport = pack_transport
1383
self._index_builder_class = index_builder_class
1384
self._index_class = index_class
1385
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3,
1390
self._packs_by_name = {}
1391
# the previous pack-names content
1392
self._packs_at_load = None
1393
# when a pack is being created by this object, the state of that pack.
1394
self._new_pack = None
1395
# aggregated revision index data
1396
flush = self._flush_new_pack
1397
self.revision_index = AggregateIndex(self.reload_pack_names, flush)
1398
self.inventory_index = AggregateIndex(self.reload_pack_names, flush)
1399
self.text_index = AggregateIndex(self.reload_pack_names, flush)
1400
self.signature_index = AggregateIndex(self.reload_pack_names, flush)
1401
all_indices = [self.revision_index, self.inventory_index,
1402
self.text_index, self.signature_index]
1404
self.chk_index = AggregateIndex(self.reload_pack_names, flush)
1405
all_indices.append(self.chk_index)
1407
# used to determine if we're using a chk_index elsewhere.
1408
self.chk_index = None
1409
# Tell all the CombinedGraphIndex objects about each other, so they can
1410
# share hints about which pack names to search first.
1411
all_combined = [agg_idx.combined_index for agg_idx in all_indices]
1412
for combined_idx in all_combined:
1413
combined_idx.set_sibling_indices(
1414
set(all_combined).difference([combined_idx]))
1416
self._resumed_packs = []
1419
return '%s(%r)' % (self.__class__.__name__, self.repo)
1421
def add_pack_to_memory(self, pack):
1422
"""Make a Pack object available to the repository to satisfy queries.
1424
:param pack: A Pack object.
1426
if pack.name in self._packs_by_name:
1427
raise AssertionError(
1428
'pack %s already in _packs_by_name' % (pack.name,))
1429
self.packs.append(pack)
1430
self._packs_by_name[pack.name] = pack
1431
self.revision_index.add_index(pack.revision_index, pack)
1432
self.inventory_index.add_index(pack.inventory_index, pack)
1433
self.text_index.add_index(pack.text_index, pack)
1434
self.signature_index.add_index(pack.signature_index, pack)
1435
if self.chk_index is not None:
1436
self.chk_index.add_index(pack.chk_index, pack)
1438
def all_packs(self):
1439
"""Return a list of all the Pack objects this repository has.
1441
Note that an in-progress pack being created is not returned.
1443
:return: A list of Pack objects for all the packs in the repository.
1446
for name in self.names():
1447
result.append(self.get_pack_by_name(name))
1451
"""Pack the pack collection incrementally.
1453
This will not attempt global reorganisation or recompression,
1454
rather it will just ensure that the total number of packs does
1455
not grow without bound. It uses the _max_pack_count method to
1456
determine if autopacking is needed, and the pack_distribution
1457
method to determine the number of revisions in each pack.
1459
If autopacking takes place then the packs name collection will have
1460
been flushed to disk - packing requires updating the name collection
1461
in synchronisation with certain steps. Otherwise the names collection
1464
:return: Something evaluating true if packing took place.
1468
return self._do_autopack()
1469
except errors.RetryAutopack:
1470
# If we get a RetryAutopack exception, we should abort the
1471
# current action, and retry.
1474
def _do_autopack(self):
1475
# XXX: Should not be needed when the management of indices is sane.
1476
total_revisions = self.revision_index.combined_index.key_count()
1477
total_packs = len(self._names)
1478
if self._max_pack_count(total_revisions) >= total_packs:
1480
# determine which packs need changing
1481
pack_distribution = self.pack_distribution(total_revisions)
1483
for pack in self.all_packs():
1484
revision_count = pack.get_revision_count()
1485
if revision_count == 0:
1486
# revision less packs are not generated by normal operation,
1487
# only by operations like sign-my-commits, and thus will not
1488
# tend to grow rapdily or without bound like commit containing
1489
# packs do - leave them alone as packing them really should
1490
# group their data with the relevant commit, and that may
1491
# involve rewriting ancient history - which autopack tries to
1492
# avoid. Alternatively we could not group the data but treat
1493
# each of these as having a single revision, and thus add
1494
# one revision for each to the total revision count, to get
1495
# a matching distribution.
1497
existing_packs.append((revision_count, pack))
1498
pack_operations = self.plan_autopack_combinations(
1499
existing_packs, pack_distribution)
1500
num_new_packs = len(pack_operations)
1501
num_old_packs = sum([len(po[1]) for po in pack_operations])
1502
num_revs_affected = sum([po[0] for po in pack_operations])
1503
mutter('Auto-packing repository %s, which has %d pack files, '
1504
'containing %d revisions. Packing %d files into %d affecting %d'
1505
' revisions', self, total_packs, total_revisions, num_old_packs,
1506
num_new_packs, num_revs_affected)
1507
result = self._execute_pack_operations(pack_operations,
1508
reload_func=self._restart_autopack)
1509
mutter('Auto-packing repository %s completed', self)
1512
def _execute_pack_operations(self, pack_operations, _packer_class=Packer,
1514
"""Execute a series of pack operations.
1516
:param pack_operations: A list of [revision_count, packs_to_combine].
1517
:param _packer_class: The class of packer to use (default: Packer).
1518
:return: The new pack names.
1520
for revision_count, packs in pack_operations:
1521
# we may have no-ops from the setup logic
1524
packer = _packer_class(self, packs, '.autopack',
1525
reload_func=reload_func)
1528
except errors.RetryWithNewPacks:
1529
# An exception is propagating out of this context, make sure
1530
# this packer has cleaned up. Packer() doesn't set its new_pack
1531
# state into the RepositoryPackCollection object, so we only
1532
# have access to it directly here.
1533
if packer.new_pack is not None:
1534
packer.new_pack.abort()
1537
self._remove_pack_from_memory(pack)
1538
# record the newly available packs and stop advertising the old
1540
to_be_obsoleted = []
1541
for _, packs in pack_operations:
1542
to_be_obsoleted.extend(packs)
1543
result = self._save_pack_names(clear_obsolete_packs=True,
1544
obsolete_packs=to_be_obsoleted)
1547
def _flush_new_pack(self):
1548
if self._new_pack is not None:
1549
self._new_pack.flush()
1551
def lock_names(self):
1552
"""Acquire the mutex around the pack-names index.
1554
This cannot be used in the middle of a read-only transaction on the
1557
self.repo.control_files.lock_write()
1559
def _already_packed(self):
1560
"""Is the collection already packed?"""
1561
return not (self.repo._format.pack_compresses or (len(self._names) > 1))
1563
def pack(self, hint=None):
1564
"""Pack the pack collection totally."""
1565
self.ensure_loaded()
1566
total_packs = len(self._names)
1567
if self._already_packed():
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 with hint %r.', self, total_packs,
1574
total_revisions, hint)
1575
# determine which packs need changing
1576
pack_operations = [[0, []]]
1577
for pack in self.all_packs():
1578
if hint is None or pack.name in hint:
1579
# Either no hint was provided (so we are packing everything),
1580
# or this pack was included in the hint.
1581
pack_operations[-1][0] += pack.get_revision_count()
1582
pack_operations[-1][1].append(pack)
1583
self._execute_pack_operations(pack_operations, OptimisingPacker)
1585
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1586
"""Plan a pack operation.
1588
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1590
:param pack_distribution: A list with the number of revisions desired
1593
if len(existing_packs) <= len(pack_distribution):
1595
existing_packs.sort(reverse=True)
1596
pack_operations = [[0, []]]
1597
# plan out what packs to keep, and what to reorganise
1598
while len(existing_packs):
1599
# take the largest pack, and if its less than the head of the
1600
# distribution chart we will include its contents in the new pack
1601
# for that position. If its larger, we remove its size from the
1602
# distribution chart
1603
next_pack_rev_count, next_pack = existing_packs.pop(0)
1604
if next_pack_rev_count >= pack_distribution[0]:
1605
# this is already packed 'better' than this, so we can
1606
# not waste time packing it.
1607
while next_pack_rev_count > 0:
1608
next_pack_rev_count -= pack_distribution[0]
1609
if next_pack_rev_count >= 0:
1611
del pack_distribution[0]
1613
# didn't use that entire bucket up
1614
pack_distribution[0] = -next_pack_rev_count
1616
# add the revisions we're going to add to the next output pack
1617
pack_operations[-1][0] += next_pack_rev_count
1618
# allocate this pack to the next pack sub operation
1619
pack_operations[-1][1].append(next_pack)
1620
if pack_operations[-1][0] >= pack_distribution[0]:
1621
# this pack is used up, shift left.
1622
del pack_distribution[0]
1623
pack_operations.append([0, []])
1624
# Now that we know which pack files we want to move, shove them all
1625
# into a single pack file.
1627
final_pack_list = []
1628
for num_revs, pack_files in pack_operations:
1629
final_rev_count += num_revs
1630
final_pack_list.extend(pack_files)
1631
if len(final_pack_list) == 1:
1632
raise AssertionError('We somehow generated an autopack with a'
1633
' single pack file being moved.')
1635
return [[final_rev_count, final_pack_list]]
1637
def ensure_loaded(self):
1638
"""Ensure we have read names from disk.
1640
:return: True if the disk names had not been previously read.
1642
# NB: if you see an assertion error here, its probably access against
1643
# an unlocked repo. Naughty.
1644
if not self.repo.is_locked():
1645
raise errors.ObjectNotLocked(self.repo)
1646
if self._names is None:
1648
self._packs_at_load = set()
1649
for index, key, value in self._iter_disk_pack_index():
1651
self._names[name] = self._parse_index_sizes(value)
1652
self._packs_at_load.add((key, value))
1656
# populate all the metadata.
1660
def _parse_index_sizes(self, value):
1661
"""Parse a string of index sizes."""
1662
return tuple([int(digits) for digits in value.split(' ')])
1664
def get_pack_by_name(self, name):
1665
"""Get a Pack object by name.
1667
:param name: The name of the pack - e.g. '123456'
1668
:return: A Pack object.
1671
return self._packs_by_name[name]
1673
rev_index = self._make_index(name, '.rix')
1674
inv_index = self._make_index(name, '.iix')
1675
txt_index = self._make_index(name, '.tix')
1676
sig_index = self._make_index(name, '.six')
1677
if self.chk_index is not None:
1678
chk_index = self._make_index(name, '.cix', unlimited_cache=True)
1681
result = ExistingPack(self._pack_transport, name, rev_index,
1682
inv_index, txt_index, sig_index, chk_index)
1683
self.add_pack_to_memory(result)
1686
def _resume_pack(self, name):
1687
"""Get a suspended Pack object by name.
1689
:param name: The name of the pack - e.g. '123456'
1690
:return: A Pack object.
1692
if not re.match('[a-f0-9]{32}', name):
1693
# Tokens should be md5sums of the suspended pack file, i.e. 32 hex
1695
raise errors.UnresumableWriteGroup(
1696
self.repo, [name], 'Malformed write group token')
1698
rev_index = self._make_index(name, '.rix', resume=True)
1699
inv_index = self._make_index(name, '.iix', resume=True)
1700
txt_index = self._make_index(name, '.tix', resume=True)
1701
sig_index = self._make_index(name, '.six', resume=True)
1702
if self.chk_index is not None:
1703
chk_index = self._make_index(name, '.cix', resume=True,
1704
unlimited_cache=True)
1707
result = self.resumed_pack_factory(name, rev_index, inv_index,
1708
txt_index, sig_index, self._upload_transport,
1709
self._pack_transport, self._index_transport, self,
1710
chk_index=chk_index)
1711
except errors.NoSuchFile, e:
1712
raise errors.UnresumableWriteGroup(self.repo, [name], str(e))
1713
self.add_pack_to_memory(result)
1714
self._resumed_packs.append(result)
1717
def allocate(self, a_new_pack):
1718
"""Allocate name in the list of packs.
1720
:param a_new_pack: A NewPack instance to be added to the collection of
1721
packs for this repository.
1723
self.ensure_loaded()
1724
if a_new_pack.name in self._names:
1725
raise errors.BzrError(
1726
'Pack %r already exists in %s' % (a_new_pack.name, self))
1727
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1728
self.add_pack_to_memory(a_new_pack)
1730
def _iter_disk_pack_index(self):
1731
"""Iterate over the contents of the pack-names index.
1733
This is used when loading the list from disk, and before writing to
1734
detect updates from others during our write operation.
1735
:return: An iterator of the index contents.
1737
return self._index_class(self.transport, 'pack-names', None
1738
).iter_all_entries()
1740
def _make_index(self, name, suffix, resume=False, unlimited_cache=False):
1741
size_offset = self._suffix_offsets[suffix]
1742
index_name = name + suffix
1744
transport = self._upload_transport
1745
index_size = transport.stat(index_name).st_size
1747
transport = self._index_transport
1748
index_size = self._names[name][size_offset]
1749
return self._index_class(transport, index_name, index_size,
1750
unlimited_cache=unlimited_cache)
1752
def _max_pack_count(self, total_revisions):
1753
"""Return the maximum number of packs to use for total revisions.
1755
:param total_revisions: The total number of revisions in the
1758
if not total_revisions:
1760
digits = str(total_revisions)
1762
for digit in digits:
1763
result += int(digit)
1767
"""Provide an order to the underlying names."""
1768
return sorted(self._names.keys())
1770
def _obsolete_packs(self, packs):
1771
"""Move a number of packs which have been obsoleted out of the way.
1773
Each pack and its associated indices are moved out of the way.
1775
Note: for correctness this function should only be called after a new
1776
pack names index has been written without these pack names, and with
1777
the names of packs that contain the data previously available via these
1780
:param packs: The packs to obsolete.
1781
:param return: None.
1785
pack.pack_transport.rename(pack.file_name(),
1786
'../obsolete_packs/' + pack.file_name())
1787
except (errors.PathError, errors.TransportError), e:
1788
# TODO: Should these be warnings or mutters?
1789
mutter("couldn't rename obsolete pack, skipping it:\n%s"
1791
# TODO: Probably needs to know all possible indices for this pack
1792
# - or maybe list the directory and move all indices matching this
1793
# name whether we recognize it or not?
1794
suffixes = ['.iix', '.six', '.tix', '.rix']
1795
if self.chk_index is not None:
1796
suffixes.append('.cix')
1797
for suffix in suffixes:
1799
self._index_transport.rename(pack.name + suffix,
1800
'../obsolete_packs/' + pack.name + suffix)
1801
except (errors.PathError, errors.TransportError), e:
1802
mutter("couldn't rename obsolete index, skipping it:\n%s"
1805
def pack_distribution(self, total_revisions):
1806
"""Generate a list of the number of revisions to put in each pack.
1808
:param total_revisions: The total number of revisions in the
1811
if total_revisions == 0:
1813
digits = reversed(str(total_revisions))
1815
for exponent, count in enumerate(digits):
1816
size = 10 ** exponent
1817
for pos in range(int(count)):
1819
return list(reversed(result))
1821
def _pack_tuple(self, name):
1822
"""Return a tuple with the transport and file name for a pack name."""
1823
return self._pack_transport, name + '.pack'
1825
def _remove_pack_from_memory(self, pack):
1826
"""Remove pack from the packs accessed by this repository.
1828
Only affects memory state, until self._save_pack_names() is invoked.
1830
self._names.pop(pack.name)
1831
self._packs_by_name.pop(pack.name)
1832
self._remove_pack_indices(pack)
1833
self.packs.remove(pack)
1835
def _remove_pack_indices(self, pack, ignore_missing=False):
1836
"""Remove the indices for pack from the aggregated indices.
1838
:param ignore_missing: Suppress KeyErrors from calling remove_index.
1840
for index_type in Pack.index_definitions.keys():
1841
attr_name = index_type + '_index'
1842
aggregate_index = getattr(self, attr_name)
1843
if aggregate_index is not None:
1844
pack_index = getattr(pack, attr_name)
1846
aggregate_index.remove_index(pack_index)
1853
"""Clear all cached data."""
1854
# cached revision data
1855
self.revision_index.clear()
1856
# cached signature data
1857
self.signature_index.clear()
1858
# cached file text data
1859
self.text_index.clear()
1860
# cached inventory data
1861
self.inventory_index.clear()
1863
if self.chk_index is not None:
1864
self.chk_index.clear()
1865
# remove the open pack
1866
self._new_pack = None
1867
# information about packs.
1870
self._packs_by_name = {}
1871
self._packs_at_load = None
1873
def _unlock_names(self):
1874
"""Release the mutex around the pack-names index."""
1875
self.repo.control_files.unlock()
1877
def _diff_pack_names(self):
1878
"""Read the pack names from disk, and compare it to the one in memory.
1880
:return: (disk_nodes, deleted_nodes, new_nodes)
1881
disk_nodes The final set of nodes that should be referenced
1882
deleted_nodes Nodes which have been removed from when we started
1883
new_nodes Nodes that are newly introduced
1885
# load the disk nodes across
1887
for index, key, value in self._iter_disk_pack_index():
1888
disk_nodes.add((key, value))
1889
orig_disk_nodes = set(disk_nodes)
1891
# do a two-way diff against our original content
1892
current_nodes = set()
1893
for name, sizes in self._names.iteritems():
1895
((name, ), ' '.join(str(size) for size in sizes)))
1897
# Packs no longer present in the repository, which were present when we
1898
# locked the repository
1899
deleted_nodes = self._packs_at_load - current_nodes
1900
# Packs which this process is adding
1901
new_nodes = current_nodes - self._packs_at_load
1903
# Update the disk_nodes set to include the ones we are adding, and
1904
# remove the ones which were removed by someone else
1905
disk_nodes.difference_update(deleted_nodes)
1906
disk_nodes.update(new_nodes)
1908
return disk_nodes, deleted_nodes, new_nodes, orig_disk_nodes
1910
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1911
"""Given the correct set of pack files, update our saved info.
1913
:return: (removed, added, modified)
1914
removed pack names removed from self._names
1915
added pack names added to self._names
1916
modified pack names that had changed value
1921
## self._packs_at_load = disk_nodes
1922
new_names = dict(disk_nodes)
1923
# drop no longer present nodes
1924
for pack in self.all_packs():
1925
if (pack.name,) not in new_names:
1926
removed.append(pack.name)
1927
self._remove_pack_from_memory(pack)
1928
# add new nodes/refresh existing ones
1929
for key, value in disk_nodes:
1931
sizes = self._parse_index_sizes(value)
1932
if name in self._names:
1934
if sizes != self._names[name]:
1935
# the pack for name has had its indices replaced - rare but
1936
# important to handle. XXX: probably can never happen today
1937
# because the three-way merge code above does not handle it
1938
# - you may end up adding the same key twice to the new
1939
# disk index because the set values are the same, unless
1940
# the only index shows up as deleted by the set difference
1941
# - which it may. Until there is a specific test for this,
1942
# assume its broken. RBC 20071017.
1943
self._remove_pack_from_memory(self.get_pack_by_name(name))
1944
self._names[name] = sizes
1945
self.get_pack_by_name(name)
1946
modified.append(name)
1949
self._names[name] = sizes
1950
self.get_pack_by_name(name)
1952
return removed, added, modified
1954
def _save_pack_names(self, clear_obsolete_packs=False, obsolete_packs=None):
1955
"""Save the list of packs.
1957
This will take out the mutex around the pack names list for the
1958
duration of the method call. If concurrent updates have been made, a
1959
three-way merge between the current list and the current in memory list
1962
:param clear_obsolete_packs: If True, clear out the contents of the
1963
obsolete_packs directory.
1964
:param obsolete_packs: Packs that are obsolete once the new pack-names
1965
file has been written.
1966
:return: A list of the names saved that were not previously on disk.
1968
already_obsolete = []
1971
builder = self._index_builder_class()
1972
(disk_nodes, deleted_nodes, new_nodes,
1973
orig_disk_nodes) = self._diff_pack_names()
1974
# TODO: handle same-name, index-size-changes here -
1975
# e.g. use the value from disk, not ours, *unless* we're the one
1977
for key, value in disk_nodes:
1978
builder.add_node(key, value)
1979
self.transport.put_file('pack-names', builder.finish(),
1980
mode=self.repo.bzrdir._get_file_mode())
1981
self._packs_at_load = disk_nodes
1982
if clear_obsolete_packs:
1985
to_preserve = set([o.name for o in obsolete_packs])
1986
already_obsolete = self._clear_obsolete_packs(to_preserve)
1988
self._unlock_names()
1989
# synchronise the memory packs list with what we just wrote:
1990
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1992
# TODO: We could add one more condition here. "if o.name not in
1993
# orig_disk_nodes and o != the new_pack we haven't written to
1994
# disk yet. However, the new pack object is not easily
1995
# accessible here (it would have to be passed through the
1996
# autopacking code, etc.)
1997
obsolete_packs = [o for o in obsolete_packs
1998
if o.name not in already_obsolete]
1999
self._obsolete_packs(obsolete_packs)
2000
return [new_node[0][0] for new_node in new_nodes]
2002
def reload_pack_names(self):
2003
"""Sync our pack listing with what is present in the repository.
2005
This should be called when we find out that something we thought was
2006
present is now missing. This happens when another process re-packs the
2009
:return: True if the in-memory list of packs has been altered at all.
2011
# The ensure_loaded call is to handle the case where the first call
2012
# made involving the collection was to reload_pack_names, where we
2013
# don't have a view of disk contents. Its a bit of a bandaid, and
2014
# causes two reads of pack-names, but its a rare corner case not struck
2015
# with regular push/pull etc.
2016
first_read = self.ensure_loaded()
2019
# out the new value.
2020
(disk_nodes, deleted_nodes, new_nodes,
2021
orig_disk_nodes) = self._diff_pack_names()
2022
# _packs_at_load is meant to be the explicit list of names in
2023
# 'pack-names' at then start. As such, it should not contain any
2024
# pending names that haven't been written out yet.
2025
self._packs_at_load = orig_disk_nodes
2027
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
2028
if removed or added or modified:
2032
def _restart_autopack(self):
2033
"""Reload the pack names list, and restart the autopack code."""
2034
if not self.reload_pack_names():
2035
# Re-raise the original exception, because something went missing
2036
# and a restart didn't find it
2038
raise errors.RetryAutopack(self.repo, False, sys.exc_info())
2040
def _clear_obsolete_packs(self, preserve=None):
2041
"""Delete everything from the obsolete-packs directory.
2043
:return: A list of pack identifiers (the filename without '.pack') that
2044
were found in obsolete_packs.
2047
obsolete_pack_transport = self.transport.clone('obsolete_packs')
2048
if preserve is None:
2050
for filename in obsolete_pack_transport.list_dir('.'):
2051
name, ext = osutils.splitext(filename)
2054
if name in preserve:
2057
obsolete_pack_transport.delete(filename)
2058
except (errors.PathError, errors.TransportError), e:
2059
warning("couldn't delete obsolete pack, skipping it:\n%s"
2063
def _start_write_group(self):
2064
# Do not permit preparation for writing if we're not in a 'write lock'.
2065
if not self.repo.is_write_locked():
2066
raise errors.NotWriteLocked(self)
2067
self._new_pack = self.pack_factory(self, upload_suffix='.pack',
2068
file_mode=self.repo.bzrdir._get_file_mode())
2069
# allow writing: queue writes to a new index
2070
self.revision_index.add_writable_index(self._new_pack.revision_index,
2072
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
2074
self.text_index.add_writable_index(self._new_pack.text_index,
2076
self._new_pack.text_index.set_optimize(combine_backing_indices=False)
2077
self.signature_index.add_writable_index(self._new_pack.signature_index,
2079
if self.chk_index is not None:
2080
self.chk_index.add_writable_index(self._new_pack.chk_index,
2082
self.repo.chk_bytes._index._add_callback = self.chk_index.add_callback
2083
self._new_pack.chk_index.set_optimize(combine_backing_indices=False)
2085
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
2086
self.repo.revisions._index._add_callback = self.revision_index.add_callback
2087
self.repo.signatures._index._add_callback = self.signature_index.add_callback
2088
self.repo.texts._index._add_callback = self.text_index.add_callback
2090
def _abort_write_group(self):
2091
# FIXME: just drop the transient index.
2092
# forget what names there are
2093
if self._new_pack is not None:
2094
operation = cleanup.OperationWithCleanups(self._new_pack.abort)
2095
operation.add_cleanup(setattr, self, '_new_pack', None)
2096
# If we aborted while in the middle of finishing the write
2097
# group, _remove_pack_indices could fail because the indexes are
2098
# already gone. But they're not there we shouldn't fail in this
2099
# case, so we pass ignore_missing=True.
2100
operation.add_cleanup(self._remove_pack_indices, self._new_pack,
2101
ignore_missing=True)
2102
operation.run_simple()
2103
for resumed_pack in self._resumed_packs:
2104
operation = cleanup.OperationWithCleanups(resumed_pack.abort)
2105
# See comment in previous finally block.
2106
operation.add_cleanup(self._remove_pack_indices, resumed_pack,
2107
ignore_missing=True)
2108
operation.run_simple()
2109
del self._resumed_packs[:]
2111
def _remove_resumed_pack_indices(self):
2112
for resumed_pack in self._resumed_packs:
2113
self._remove_pack_indices(resumed_pack)
2114
del self._resumed_packs[:]
2116
def _check_new_inventories(self):
2117
"""Detect missing inventories in this write group.
2119
:returns: list of strs, summarising any problems found. If the list is
2120
empty no problems were found.
2122
# The base implementation does no checks. GCRepositoryPackCollection
2126
def _commit_write_group(self):
2128
for prefix, versioned_file in (
2129
('revisions', self.repo.revisions),
2130
('inventories', self.repo.inventories),
2131
('texts', self.repo.texts),
2132
('signatures', self.repo.signatures),
2134
missing = versioned_file.get_missing_compression_parent_keys()
2135
all_missing.update([(prefix,) + key for key in missing])
2137
raise errors.BzrCheckError(
2138
"Repository %s has missing compression parent(s) %r "
2139
% (self.repo, sorted(all_missing)))
2140
problems = self._check_new_inventories()
2142
problems_summary = '\n'.join(problems)
2143
raise errors.BzrCheckError(
2144
"Cannot add revision(s) to repository: " + problems_summary)
2145
self._remove_pack_indices(self._new_pack)
2146
any_new_content = False
2147
if self._new_pack.data_inserted():
2148
# get all the data to disk and read to use
2149
self._new_pack.finish()
2150
self.allocate(self._new_pack)
2151
self._new_pack = None
2152
any_new_content = True
2154
self._new_pack.abort()
2155
self._new_pack = None
2156
for resumed_pack in self._resumed_packs:
2157
# XXX: this is a pretty ugly way to turn the resumed pack into a
2158
# properly committed pack.
2159
self._names[resumed_pack.name] = None
2160
self._remove_pack_from_memory(resumed_pack)
2161
resumed_pack.finish()
2162
self.allocate(resumed_pack)
2163
any_new_content = True
2164
del self._resumed_packs[:]
2166
result = self.autopack()
2168
# when autopack takes no steps, the names list is still
2170
return self._save_pack_names()
2174
def _suspend_write_group(self):
2175
tokens = [pack.name for pack in self._resumed_packs]
2176
self._remove_pack_indices(self._new_pack)
2177
if self._new_pack.data_inserted():
2178
# get all the data to disk and read to use
2179
self._new_pack.finish(suspend=True)
2180
tokens.append(self._new_pack.name)
2181
self._new_pack = None
2183
self._new_pack.abort()
2184
self._new_pack = None
2185
self._remove_resumed_pack_indices()
2188
def _resume_write_group(self, tokens):
2189
for token in tokens:
2190
self._resume_pack(token)
2193
class KnitPackRepository(KnitRepository):
2194
"""Repository with knit objects stored inside pack containers.
2196
The layering for a KnitPackRepository is:
2198
Graph | HPSS | Repository public layer |
2199
===================================================
2200
Tuple based apis below, string based, and key based apis above
2201
---------------------------------------------------
2203
Provides .texts, .revisions etc
2204
This adapts the N-tuple keys to physical knit records which only have a
2205
single string identifier (for historical reasons), which in older formats
2206
was always the revision_id, and in the mapped code for packs is always
2207
the last element of key tuples.
2208
---------------------------------------------------
2210
A separate GraphIndex is used for each of the
2211
texts/inventories/revisions/signatures contained within each individual
2212
pack file. The GraphIndex layer works in N-tuples and is unaware of any
2214
===================================================
2218
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
2220
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2221
_commit_builder_class, _serializer)
2222
index_transport = self._transport.clone('indices')
2223
self._pack_collection = RepositoryPackCollection(self, self._transport,
2225
self._transport.clone('upload'),
2226
self._transport.clone('packs'),
2227
_format.index_builder_class,
2228
_format.index_class,
2229
use_chk_index=self._format.supports_chks,
2231
self.inventories = KnitVersionedFiles(
2232
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2233
add_callback=self._pack_collection.inventory_index.add_callback,
2234
deltas=True, parents=True, is_locked=self.is_locked),
2235
data_access=self._pack_collection.inventory_index.data_access,
2236
max_delta_chain=200)
2237
self.revisions = KnitVersionedFiles(
2238
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2239
add_callback=self._pack_collection.revision_index.add_callback,
2240
deltas=False, parents=True, is_locked=self.is_locked,
2241
track_external_parent_refs=True),
2242
data_access=self._pack_collection.revision_index.data_access,
2244
self.signatures = KnitVersionedFiles(
2245
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2246
add_callback=self._pack_collection.signature_index.add_callback,
2247
deltas=False, parents=False, is_locked=self.is_locked),
2248
data_access=self._pack_collection.signature_index.data_access,
2250
self.texts = KnitVersionedFiles(
2251
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2252
add_callback=self._pack_collection.text_index.add_callback,
2253
deltas=True, parents=True, is_locked=self.is_locked),
2254
data_access=self._pack_collection.text_index.data_access,
2255
max_delta_chain=200)
2256
if _format.supports_chks:
2257
# No graph, no compression:- references from chks are between
2258
# different objects not temporal versions of the same; and without
2259
# some sort of temporal structure knit compression will just fail.
2260
self.chk_bytes = KnitVersionedFiles(
2261
_KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2262
add_callback=self._pack_collection.chk_index.add_callback,
2263
deltas=False, parents=False, is_locked=self.is_locked),
2264
data_access=self._pack_collection.chk_index.data_access,
2267
self.chk_bytes = None
2268
# True when the repository object is 'write locked' (as opposed to the
2269
# physical lock only taken out around changes to the pack-names list.)
2270
# Another way to represent this would be a decorator around the control
2271
# files object that presents logical locks as physical ones - if this
2272
# gets ugly consider that alternative design. RBC 20071011
2273
self._write_lock_count = 0
2274
self._transaction = None
2276
self._reconcile_does_inventory_gc = True
2277
self._reconcile_fixes_text_parents = True
2278
self._reconcile_backsup_inventory = False
2280
def _warn_if_deprecated(self, branch=None):
2281
# This class isn't deprecated, but one sub-format is
2282
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2283
super(KnitPackRepository, self)._warn_if_deprecated(branch)
2285
def _abort_write_group(self):
2286
self.revisions._index._key_dependencies.clear()
2287
self._pack_collection._abort_write_group()
2289
def _get_source(self, to_format):
2290
if to_format.network_name() == self._format.network_name():
2291
return KnitPackStreamSource(self, to_format)
2292
return super(KnitPackRepository, self)._get_source(to_format)
2294
def _make_parents_provider(self):
2295
return graph.CachingParentsProvider(self)
2297
def _refresh_data(self):
2298
if not self.is_locked():
2300
self._pack_collection.reload_pack_names()
2302
def _start_write_group(self):
2303
self._pack_collection._start_write_group()
2305
def _commit_write_group(self):
2306
hint = self._pack_collection._commit_write_group()
2307
self.revisions._index._key_dependencies.clear()
2310
def suspend_write_group(self):
2311
# XXX check self._write_group is self.get_transaction()?
2312
tokens = self._pack_collection._suspend_write_group()
2313
self.revisions._index._key_dependencies.clear()
2314
self._write_group = None
2317
def _resume_write_group(self, tokens):
2318
self._start_write_group()
2320
self._pack_collection._resume_write_group(tokens)
2321
except errors.UnresumableWriteGroup:
2322
self._abort_write_group()
2324
for pack in self._pack_collection._resumed_packs:
2325
self.revisions._index.scan_unvalidated_index(pack.revision_index)
2327
def get_transaction(self):
2328
if self._write_lock_count:
2329
return self._transaction
2331
return self.control_files.get_transaction()
2333
def is_locked(self):
2334
return self._write_lock_count or self.control_files.is_locked()
2336
def is_write_locked(self):
2337
return self._write_lock_count
2339
def lock_write(self, token=None):
2340
locked = self.is_locked()
2341
if not self._write_lock_count and locked:
2342
raise errors.ReadOnlyError(self)
2343
self._write_lock_count += 1
2344
if self._write_lock_count == 1:
2345
self._transaction = transactions.WriteTransaction()
2347
if 'relock' in debug.debug_flags and self._prev_lock == 'w':
2348
note('%r was write locked again', self)
2349
self._prev_lock = 'w'
2350
for repo in self._fallback_repositories:
2351
# Writes don't affect fallback repos
2353
self._refresh_data()
2355
def lock_read(self):
2356
locked = self.is_locked()
2357
if self._write_lock_count:
2358
self._write_lock_count += 1
2360
self.control_files.lock_read()
2362
if 'relock' in debug.debug_flags and self._prev_lock == 'r':
2363
note('%r was read locked again', self)
2364
self._prev_lock = 'r'
2365
for repo in self._fallback_repositories:
2367
self._refresh_data()
2369
def leave_lock_in_place(self):
2370
# not supported - raise an error
2371
raise NotImplementedError(self.leave_lock_in_place)
2373
def dont_leave_lock_in_place(self):
2374
# not supported - raise an error
2375
raise NotImplementedError(self.dont_leave_lock_in_place)
2378
def pack(self, hint=None):
2379
"""Compress the data within the repository.
2381
This will pack all the data to a single pack. In future it may
2382
recompress deltas or do other such expensive operations.
2384
self._pack_collection.pack(hint=hint)
2387
def reconcile(self, other=None, thorough=False):
2388
"""Reconcile this repository."""
2389
from bzrlib.reconcile import PackReconciler
2390
reconciler = PackReconciler(self, thorough=thorough)
2391
reconciler.reconcile()
2394
def _reconcile_pack(self, collection, packs, extension, revs, pb):
2395
packer = ReconcilePacker(collection, packs, extension, revs)
2396
return packer.pack(pb)
2398
@only_raises(errors.LockNotHeld, errors.LockBroken)
2400
if self._write_lock_count == 1 and self._write_group is not None:
2401
self.abort_write_group()
2402
self._transaction = None
2403
self._write_lock_count = 0
2404
raise errors.BzrError(
2405
'Must end write group before releasing write lock on %s'
2407
if self._write_lock_count:
2408
self._write_lock_count -= 1
2409
if not self._write_lock_count:
2410
transaction = self._transaction
2411
self._transaction = None
2412
transaction.finish()
2414
self.control_files.unlock()
2416
if not self.is_locked():
2417
for repo in self._fallback_repositories:
2421
class KnitPackStreamSource(StreamSource):
2422
"""A StreamSource used to transfer data between same-format KnitPack repos.
2424
This source assumes:
2425
1) Same serialization format for all objects
2426
2) Same root information
2427
3) XML format inventories
2428
4) Atomic inserts (so we can stream inventory texts before text
2433
def __init__(self, from_repository, to_format):
2434
super(KnitPackStreamSource, self).__init__(from_repository, to_format)
2435
self._text_keys = None
2436
self._text_fetch_order = 'unordered'
2438
def _get_filtered_inv_stream(self, revision_ids):
2439
from_repo = self.from_repository
2440
parent_ids = from_repo._find_parent_ids_of_revisions(revision_ids)
2441
parent_keys = [(p,) for p in parent_ids]
2442
find_text_keys = from_repo._find_text_key_references_from_xml_inventory_lines
2443
parent_text_keys = set(find_text_keys(
2444
from_repo._inventory_xml_lines_for_keys(parent_keys)))
2445
content_text_keys = set()
2446
knit = KnitVersionedFiles(None, None)
2447
factory = KnitPlainFactory()
2448
def find_text_keys_from_content(record):
2449
if record.storage_kind not in ('knit-delta-gz', 'knit-ft-gz'):
2450
raise ValueError("Unknown content storage kind for"
2451
" inventory text: %s" % (record.storage_kind,))
2452
# It's a knit record, it has a _raw_record field (even if it was
2453
# reconstituted from a network stream).
2454
raw_data = record._raw_record
2455
# read the entire thing
2456
revision_id = record.key[-1]
2457
content, _ = knit._parse_record(revision_id, raw_data)
2458
if record.storage_kind == 'knit-delta-gz':
2459
line_iterator = factory.get_linedelta_content(content)
2460
elif record.storage_kind == 'knit-ft-gz':
2461
line_iterator = factory.get_fulltext_content(content)
2462
content_text_keys.update(find_text_keys(
2463
[(line, revision_id) for line in line_iterator]))
2464
revision_keys = [(r,) for r in revision_ids]
2465
def _filtered_inv_stream():
2466
source_vf = from_repo.inventories
2467
stream = source_vf.get_record_stream(revision_keys,
2469
for record in stream:
2470
if record.storage_kind == 'absent':
2471
raise errors.NoSuchRevision(from_repo, record.key)
2472
find_text_keys_from_content(record)
2474
self._text_keys = content_text_keys - parent_text_keys
2475
return ('inventories', _filtered_inv_stream())
2477
def _get_text_stream(self):
2478
# Note: We know we don't have to handle adding root keys, because both
2479
# the source and target are the identical network name.
2480
text_stream = self.from_repository.texts.get_record_stream(
2481
self._text_keys, self._text_fetch_order, False)
2482
return ('texts', text_stream)
2484
def get_stream(self, search):
2485
revision_ids = search.get_keys()
2486
for stream_info in self._fetch_revision_texts(revision_ids):
2488
self._revision_keys = [(rev_id,) for rev_id in revision_ids]
2489
yield self._get_filtered_inv_stream(revision_ids)
2490
yield self._get_text_stream()
2494
class RepositoryFormatPack(MetaDirRepositoryFormat):
2495
"""Format logic for pack structured repositories.
2497
This repository format has:
2498
- a list of packs in pack-names
2499
- packs in packs/NAME.pack
2500
- indices in indices/NAME.{iix,six,tix,rix}
2501
- knit deltas in the packs, knit indices mapped to the indices.
2502
- thunk objects to support the knits programming API.
2503
- a format marker of its own
2504
- an optional 'shared-storage' flag
2505
- an optional 'no-working-trees' flag
2509
# Set this attribute in derived classes to control the repository class
2510
# created by open and initialize.
2511
repository_class = None
2512
# Set this attribute in derived classes to control the
2513
# _commit_builder_class that the repository objects will have passed to
2514
# their constructor.
2515
_commit_builder_class = None
2516
# Set this attribute in derived clases to control the _serializer that the
2517
# repository objects will have passed to their constructor.
2519
# Packs are not confused by ghosts.
2520
supports_ghosts = True
2521
# External references are not supported in pack repositories yet.
2522
supports_external_lookups = False
2523
# Most pack formats do not use chk lookups.
2524
supports_chks = False
2525
# What index classes to use
2526
index_builder_class = None
2528
_fetch_uses_deltas = True
2531
def initialize(self, a_bzrdir, shared=False):
2532
"""Create a pack based repository.
2534
:param a_bzrdir: bzrdir to contain the new repository; must already
2536
:param shared: If true the repository will be initialized as a shared
2539
mutter('creating repository in %s.', a_bzrdir.transport.base)
2540
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
2541
builder = self.index_builder_class()
2542
files = [('pack-names', builder.finish())]
2543
utf8_files = [('format', self.get_format_string())]
2545
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
2546
return self.open(a_bzrdir=a_bzrdir, _found=True)
2548
def open(self, a_bzrdir, _found=False, _override_transport=None):
2549
"""See RepositoryFormat.open().
2551
:param _override_transport: INTERNAL USE ONLY. Allows opening the
2552
repository at a slightly different url
2553
than normal. I.e. during 'upgrade'.
2556
format = RepositoryFormat.find_format(a_bzrdir)
2557
if _override_transport is not None:
2558
repo_transport = _override_transport
2560
repo_transport = a_bzrdir.get_repository_transport(None)
2561
control_files = lockable_files.LockableFiles(repo_transport,
2562
'lock', lockdir.LockDir)
2563
return self.repository_class(_format=self,
2565
control_files=control_files,
2566
_commit_builder_class=self._commit_builder_class,
2567
_serializer=self._serializer)
2570
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2571
"""A no-subtrees parameterized Pack repository.
2573
This format was introduced in 0.92.
2576
repository_class = KnitPackRepository
2577
_commit_builder_class = PackCommitBuilder
2579
def _serializer(self):
2580
return xml5.serializer_v5
2581
# What index classes to use
2582
index_builder_class = InMemoryGraphIndex
2583
index_class = GraphIndex
2585
def _get_matching_bzrdir(self):
2586
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2588
def _ignore_setting_bzrdir(self, format):
2591
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2593
def get_format_string(self):
2594
"""See RepositoryFormat.get_format_string()."""
2595
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2597
def get_format_description(self):
2598
"""See RepositoryFormat.get_format_description()."""
2599
return "Packs containing knits without subtree support"
2602
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2603
"""A subtrees parameterized Pack repository.
2605
This repository format uses the xml7 serializer to get:
2606
- support for recording full info about the tree root
2607
- support for recording tree-references
2609
This format was introduced in 0.92.
2612
repository_class = KnitPackRepository
2613
_commit_builder_class = PackRootCommitBuilder
2614
rich_root_data = True
2616
supports_tree_reference = True
2618
def _serializer(self):
2619
return xml7.serializer_v7
2620
# What index classes to use
2621
index_builder_class = InMemoryGraphIndex
2622
index_class = GraphIndex
2624
def _get_matching_bzrdir(self):
2625
return bzrdir.format_registry.make_bzrdir(
2626
'pack-0.92-subtree')
2628
def _ignore_setting_bzrdir(self, format):
2631
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2633
def get_format_string(self):
2634
"""See RepositoryFormat.get_format_string()."""
2635
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2637
def get_format_description(self):
2638
"""See RepositoryFormat.get_format_description()."""
2639
return "Packs containing knits with subtree support\n"
2642
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2643
"""A rich-root, no subtrees parameterized Pack repository.
2645
This repository format uses the xml6 serializer to get:
2646
- support for recording full info about the tree root
2648
This format was introduced in 1.0.
2651
repository_class = KnitPackRepository
2652
_commit_builder_class = PackRootCommitBuilder
2653
rich_root_data = True
2654
supports_tree_reference = False
2656
def _serializer(self):
2657
return xml6.serializer_v6
2658
# What index classes to use
2659
index_builder_class = InMemoryGraphIndex
2660
index_class = GraphIndex
2662
def _get_matching_bzrdir(self):
2663
return bzrdir.format_registry.make_bzrdir(
2666
def _ignore_setting_bzrdir(self, format):
2669
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2671
def get_format_string(self):
2672
"""See RepositoryFormat.get_format_string()."""
2673
return ("Bazaar pack repository format 1 with rich root"
2674
" (needs bzr 1.0)\n")
2676
def get_format_description(self):
2677
"""See RepositoryFormat.get_format_description()."""
2678
return "Packs containing knits with rich root support\n"
2681
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2682
"""Repository that supports external references to allow stacking.
2686
Supports external lookups, which results in non-truncated ghosts after
2687
reconcile compared to pack-0.92 formats.
2690
repository_class = KnitPackRepository
2691
_commit_builder_class = PackCommitBuilder
2692
supports_external_lookups = True
2693
# What index classes to use
2694
index_builder_class = InMemoryGraphIndex
2695
index_class = GraphIndex
2698
def _serializer(self):
2699
return xml5.serializer_v5
2701
def _get_matching_bzrdir(self):
2702
return bzrdir.format_registry.make_bzrdir('1.6')
2704
def _ignore_setting_bzrdir(self, format):
2707
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2709
def get_format_string(self):
2710
"""See RepositoryFormat.get_format_string()."""
2711
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2713
def get_format_description(self):
2714
"""See RepositoryFormat.get_format_description()."""
2715
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2718
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2719
"""A repository with rich roots and stacking.
2721
New in release 1.6.1.
2723
Supports stacking on other repositories, allowing data to be accessed
2724
without being stored locally.
2727
repository_class = KnitPackRepository
2728
_commit_builder_class = PackRootCommitBuilder
2729
rich_root_data = True
2730
supports_tree_reference = False # no subtrees
2731
supports_external_lookups = True
2732
# What index classes to use
2733
index_builder_class = InMemoryGraphIndex
2734
index_class = GraphIndex
2737
def _serializer(self):
2738
return xml6.serializer_v6
2740
def _get_matching_bzrdir(self):
2741
return bzrdir.format_registry.make_bzrdir(
2744
def _ignore_setting_bzrdir(self, format):
2747
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2749
def get_format_string(self):
2750
"""See RepositoryFormat.get_format_string()."""
2751
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2753
def get_format_description(self):
2754
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2757
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2758
"""A repository with rich roots and external references.
2762
Supports external lookups, which results in non-truncated ghosts after
2763
reconcile compared to pack-0.92 formats.
2765
This format was deprecated because the serializer it uses accidentally
2766
supported subtrees, when the format was not intended to. This meant that
2767
someone could accidentally fetch from an incorrect repository.
2770
repository_class = KnitPackRepository
2771
_commit_builder_class = PackRootCommitBuilder
2772
rich_root_data = True
2773
supports_tree_reference = False # no subtrees
2775
supports_external_lookups = True
2776
# What index classes to use
2777
index_builder_class = InMemoryGraphIndex
2778
index_class = GraphIndex
2781
def _serializer(self):
2782
return xml7.serializer_v7
2784
def _get_matching_bzrdir(self):
2785
matching = bzrdir.format_registry.make_bzrdir(
2787
matching.repository_format = self
2790
def _ignore_setting_bzrdir(self, format):
2793
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2795
def get_format_string(self):
2796
"""See RepositoryFormat.get_format_string()."""
2797
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2799
def get_format_description(self):
2800
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2804
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2805
"""A repository with stacking and btree indexes,
2806
without rich roots or subtrees.
2808
This is equivalent to pack-1.6 with B+Tree indices.
2811
repository_class = KnitPackRepository
2812
_commit_builder_class = PackCommitBuilder
2813
supports_external_lookups = True
2814
# What index classes to use
2815
index_builder_class = BTreeBuilder
2816
index_class = BTreeGraphIndex
2819
def _serializer(self):
2820
return xml5.serializer_v5
2822
def _get_matching_bzrdir(self):
2823
return bzrdir.format_registry.make_bzrdir('1.9')
2825
def _ignore_setting_bzrdir(self, format):
2828
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2830
def get_format_string(self):
2831
"""See RepositoryFormat.get_format_string()."""
2832
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2834
def get_format_description(self):
2835
"""See RepositoryFormat.get_format_description()."""
2836
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2839
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2840
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2842
1.6-rich-root with B+Tree indices.
2845
repository_class = KnitPackRepository
2846
_commit_builder_class = PackRootCommitBuilder
2847
rich_root_data = True
2848
supports_tree_reference = False # no subtrees
2849
supports_external_lookups = True
2850
# What index classes to use
2851
index_builder_class = BTreeBuilder
2852
index_class = BTreeGraphIndex
2855
def _serializer(self):
2856
return xml6.serializer_v6
2858
def _get_matching_bzrdir(self):
2859
return bzrdir.format_registry.make_bzrdir(
2862
def _ignore_setting_bzrdir(self, format):
2865
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2867
def get_format_string(self):
2868
"""See RepositoryFormat.get_format_string()."""
2869
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2871
def get_format_description(self):
2872
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2875
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2876
"""A subtrees development repository.
2878
This format should be retained until the second release after bzr 1.7.
2880
1.6.1-subtree[as it might have been] with B+Tree indices.
2882
This is [now] retained until we have a CHK based subtree format in
2886
repository_class = KnitPackRepository
2887
_commit_builder_class = PackRootCommitBuilder
2888
rich_root_data = True
2890
supports_tree_reference = True
2891
supports_external_lookups = True
2892
# What index classes to use
2893
index_builder_class = BTreeBuilder
2894
index_class = BTreeGraphIndex
2897
def _serializer(self):
2898
return xml7.serializer_v7
2900
def _get_matching_bzrdir(self):
2901
return bzrdir.format_registry.make_bzrdir(
2902
'development-subtree')
2904
def _ignore_setting_bzrdir(self, format):
2907
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2909
def get_format_string(self):
2910
"""See RepositoryFormat.get_format_string()."""
2911
return ("Bazaar development format 2 with subtree support "
2912
"(needs bzr.dev from before 1.8)\n")
2914
def get_format_description(self):
2915
"""See RepositoryFormat.get_format_description()."""
2916
return ("Development repository format, currently the same as "
2917
"1.6.1-subtree with B+Tree indices.\n")
added
removed
bzrlib/repofmt/pack_repo.py
