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.lock import LogicalLockResult
68
from bzrlib.repofmt.knitrepo import KnitRepository
69
from bzrlib.repository import (
71
MetaDirRepositoryFormat,
73
RepositoryWriteLockResult,
77
from bzrlib.trace import (
84
class PackCommitBuilder(CommitBuilder):
85
"""A subclass of CommitBuilder to add texts with pack semantics.
87
Specifically this uses one knit object rather than one knit object per
88
added text, reducing memory and object pressure.
91
def __init__(self, repository, parents, config, timestamp=None,
92
timezone=None, committer=None, revprops=None,
94
CommitBuilder.__init__(self, repository, parents, config,
95
timestamp=timestamp, timezone=timezone, committer=committer,
96
revprops=revprops, revision_id=revision_id)
97
self._file_graph = graph.Graph(
98
repository._pack_collection.text_index.combined_index)
100
def _heads(self, file_id, revision_ids):
101
keys = [(file_id, revision_id) for revision_id in revision_ids]
102
return set([key[1] for key in self._file_graph.heads(keys)])
105
class PackRootCommitBuilder(RootCommitBuilder):
106
"""A subclass of RootCommitBuilder to add texts with pack semantics.
108
Specifically this uses one knit object rather than one knit object per
109
added text, reducing memory and object pressure.
112
def __init__(self, repository, parents, config, timestamp=None,
113
timezone=None, committer=None, revprops=None,
115
CommitBuilder.__init__(self, repository, parents, config,
116
timestamp=timestamp, timezone=timezone, committer=committer,
117
revprops=revprops, revision_id=revision_id)
118
self._file_graph = graph.Graph(
119
repository._pack_collection.text_index.combined_index)
121
def _heads(self, file_id, revision_ids):
122
keys = [(file_id, revision_id) for revision_id in revision_ids]
123
return set([key[1] for key in self._file_graph.heads(keys)])
127
"""An in memory proxy for a pack and its indices.
129
This is a base class that is not directly used, instead the classes
130
ExistingPack and NewPack are used.
133
# A map of index 'type' to the file extension and position in the
135
index_definitions = {
137
'revision': ('.rix', 0),
138
'inventory': ('.iix', 1),
140
'signature': ('.six', 3),
143
def __init__(self, revision_index, inventory_index, text_index,
144
signature_index, chk_index=None):
145
"""Create a pack instance.
147
:param revision_index: A GraphIndex for determining what revisions are
148
present in the Pack and accessing the locations of their texts.
149
:param inventory_index: A GraphIndex for determining what inventories are
150
present in the Pack and accessing the locations of their
152
:param text_index: A GraphIndex for determining what file texts
153
are present in the pack and accessing the locations of their
154
texts/deltas (via (fileid, revisionid) tuples).
155
:param signature_index: A GraphIndex for determining what signatures are
156
present in the Pack and accessing the locations of their texts.
157
:param chk_index: A GraphIndex for accessing content by CHK, if the
160
self.revision_index = revision_index
161
self.inventory_index = inventory_index
162
self.text_index = text_index
163
self.signature_index = signature_index
164
self.chk_index = chk_index
166
def access_tuple(self):
167
"""Return a tuple (transport, name) for the pack content."""
168
return self.pack_transport, self.file_name()
170
def _check_references(self):
171
"""Make sure our external references are present.
173
Packs are allowed to have deltas whose base is not in the pack, but it
174
must be present somewhere in this collection. It is not allowed to
175
have deltas based on a fallback repository.
176
(See <https://bugs.launchpad.net/bzr/+bug/288751>)
179
for (index_name, external_refs, index) in [
181
self._get_external_refs(self.text_index),
182
self._pack_collection.text_index.combined_index),
184
self._get_external_refs(self.inventory_index),
185
self._pack_collection.inventory_index.combined_index),
187
missing = external_refs.difference(
188
k for (idx, k, v, r) in
189
index.iter_entries(external_refs))
191
missing_items[index_name] = sorted(list(missing))
193
from pprint import pformat
194
raise errors.BzrCheckError(
195
"Newly created pack file %r has delta references to "
196
"items not in its repository:\n%s"
197
% (self, pformat(missing_items)))
200
"""Get the file name for the pack on disk."""
201
return self.name + '.pack'
203
def get_revision_count(self):
204
return self.revision_index.key_count()
206
def index_name(self, index_type, name):
207
"""Get the disk name of an index type for pack name 'name'."""
208
return name + Pack.index_definitions[index_type][0]
210
def index_offset(self, index_type):
211
"""Get the position in a index_size array for a given index type."""
212
return Pack.index_definitions[index_type][1]
214
def inventory_index_name(self, name):
215
"""The inv index is the name + .iix."""
216
return self.index_name('inventory', name)
218
def revision_index_name(self, name):
219
"""The revision index is the name + .rix."""
220
return self.index_name('revision', name)
222
def signature_index_name(self, name):
223
"""The signature index is the name + .six."""
224
return self.index_name('signature', name)
226
def text_index_name(self, name):
227
"""The text index is the name + .tix."""
228
return self.index_name('text', name)
230
def _replace_index_with_readonly(self, index_type):
231
unlimited_cache = False
232
if index_type == 'chk':
233
unlimited_cache = True
234
setattr(self, index_type + '_index',
235
self.index_class(self.index_transport,
236
self.index_name(index_type, self.name),
237
self.index_sizes[self.index_offset(index_type)],
238
unlimited_cache=unlimited_cache))
241
class ExistingPack(Pack):
242
"""An in memory proxy for an existing .pack and its disk indices."""
244
def __init__(self, pack_transport, name, revision_index, inventory_index,
245
text_index, signature_index, chk_index=None):
246
"""Create an ExistingPack object.
248
:param pack_transport: The transport where the pack file resides.
249
:param name: The name of the pack on disk in the pack_transport.
251
Pack.__init__(self, revision_index, inventory_index, text_index,
252
signature_index, chk_index)
254
self.pack_transport = pack_transport
255
if None in (revision_index, inventory_index, text_index,
256
signature_index, name, pack_transport):
257
raise AssertionError()
259
def __eq__(self, other):
260
return self.__dict__ == other.__dict__
262
def __ne__(self, other):
263
return not self.__eq__(other)
266
return "<%s.%s object at 0x%x, %s, %s" % (
267
self.__class__.__module__, self.__class__.__name__, id(self),
268
self.pack_transport, self.name)
271
class ResumedPack(ExistingPack):
273
def __init__(self, name, revision_index, inventory_index, text_index,
274
signature_index, upload_transport, pack_transport, index_transport,
275
pack_collection, chk_index=None):
276
"""Create a ResumedPack object."""
277
ExistingPack.__init__(self, pack_transport, name, revision_index,
278
inventory_index, text_index, signature_index,
280
self.upload_transport = upload_transport
281
self.index_transport = index_transport
282
self.index_sizes = [None, None, None, None]
284
('revision', revision_index),
285
('inventory', inventory_index),
286
('text', text_index),
287
('signature', signature_index),
289
if chk_index is not None:
290
indices.append(('chk', chk_index))
291
self.index_sizes.append(None)
292
for index_type, index in indices:
293
offset = self.index_offset(index_type)
294
self.index_sizes[offset] = index._size
295
self.index_class = pack_collection._index_class
296
self._pack_collection = pack_collection
297
self._state = 'resumed'
298
# XXX: perhaps check that the .pack file exists?
300
def access_tuple(self):
301
if self._state == 'finished':
302
return Pack.access_tuple(self)
303
elif self._state == 'resumed':
304
return self.upload_transport, self.file_name()
306
raise AssertionError(self._state)
309
self.upload_transport.delete(self.file_name())
310
indices = [self.revision_index, self.inventory_index, self.text_index,
311
self.signature_index]
312
if self.chk_index is not None:
313
indices.append(self.chk_index)
314
for index in indices:
315
index._transport.delete(index._name)
318
self._check_references()
319
index_types = ['revision', 'inventory', 'text', 'signature']
320
if self.chk_index is not None:
321
index_types.append('chk')
322
for index_type in index_types:
323
old_name = self.index_name(index_type, self.name)
324
new_name = '../indices/' + old_name
325
self.upload_transport.rename(old_name, new_name)
326
self._replace_index_with_readonly(index_type)
327
new_name = '../packs/' + self.file_name()
328
self.upload_transport.rename(self.file_name(), new_name)
329
self._state = 'finished'
331
def _get_external_refs(self, index):
332
"""Return compression parents for this index that are not present.
334
This returns any compression parents that are referenced by this index,
335
which are not contained *in* this index. They may be present elsewhere.
337
return index.external_references(1)
341
"""An in memory proxy for a pack which is being created."""
343
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
344
"""Create a NewPack instance.
346
:param pack_collection: A PackCollection into which this is being inserted.
347
:param upload_suffix: An optional suffix to be given to any temporary
348
files created during the pack creation. e.g '.autopack'
349
:param file_mode: Unix permissions for newly created file.
351
# The relative locations of the packs are constrained, but all are
352
# passed in because the caller has them, so as to avoid object churn.
353
index_builder_class = pack_collection._index_builder_class
354
if pack_collection.chk_index is not None:
355
chk_index = index_builder_class(reference_lists=0)
359
# Revisions: parents list, no text compression.
360
index_builder_class(reference_lists=1),
361
# Inventory: We want to map compression only, but currently the
362
# knit code hasn't been updated enough to understand that, so we
363
# have a regular 2-list index giving parents and compression
365
index_builder_class(reference_lists=2),
366
# Texts: compression and per file graph, for all fileids - so two
367
# reference lists and two elements in the key tuple.
368
index_builder_class(reference_lists=2, key_elements=2),
369
# Signatures: Just blobs to store, no compression, no parents
371
index_builder_class(reference_lists=0),
372
# CHK based storage - just blobs, no compression or parents.
375
self._pack_collection = pack_collection
376
# When we make readonly indices, we need this.
377
self.index_class = pack_collection._index_class
378
# where should the new pack be opened
379
self.upload_transport = pack_collection._upload_transport
380
# where are indices written out to
381
self.index_transport = pack_collection._index_transport
382
# where is the pack renamed to when it is finished?
383
self.pack_transport = pack_collection._pack_transport
384
# What file mode to upload the pack and indices with.
385
self._file_mode = file_mode
386
# tracks the content written to the .pack file.
387
self._hash = osutils.md5()
388
# a tuple with the length in bytes of the indices, once the pack
389
# is finalised. (rev, inv, text, sigs, chk_if_in_use)
390
self.index_sizes = None
391
# How much data to cache when writing packs. Note that this is not
392
# synchronised with reads, because it's not in the transport layer, so
393
# is not safe unless the client knows it won't be reading from the pack
395
self._cache_limit = 0
396
# the temporary pack file name.
397
self.random_name = osutils.rand_chars(20) + upload_suffix
398
# when was this pack started ?
399
self.start_time = time.time()
400
# open an output stream for the data added to the pack.
401
self.write_stream = self.upload_transport.open_write_stream(
402
self.random_name, mode=self._file_mode)
403
if 'pack' in debug.debug_flags:
404
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
405
time.ctime(), self.upload_transport.base, self.random_name,
406
time.time() - self.start_time)
407
# A list of byte sequences to be written to the new pack, and the
408
# aggregate size of them. Stored as a list rather than separate
409
# variables so that the _write_data closure below can update them.
410
self._buffer = [[], 0]
411
# create a callable for adding data
413
# robertc says- this is a closure rather than a method on the object
414
# so that the variables are locals, and faster than accessing object
416
def _write_data(bytes, flush=False, _buffer=self._buffer,
417
_write=self.write_stream.write, _update=self._hash.update):
418
_buffer[0].append(bytes)
419
_buffer[1] += len(bytes)
421
if _buffer[1] > self._cache_limit or flush:
422
bytes = ''.join(_buffer[0])
426
# expose this on self, for the occasion when clients want to add data.
427
self._write_data = _write_data
428
# a pack writer object to serialise pack records.
429
self._writer = pack.ContainerWriter(self._write_data)
431
# what state is the pack in? (open, finished, aborted)
433
# no name until we finish writing the content
437
"""Cancel creating this pack."""
438
self._state = 'aborted'
439
self.write_stream.close()
440
# Remove the temporary pack file.
441
self.upload_transport.delete(self.random_name)
442
# The indices have no state on disk.
444
def access_tuple(self):
445
"""Return a tuple (transport, name) for the pack content."""
446
if self._state == 'finished':
447
return Pack.access_tuple(self)
448
elif self._state == 'open':
449
return self.upload_transport, self.random_name
451
raise AssertionError(self._state)
453
def data_inserted(self):
454
"""True if data has been added to this pack."""
455
return bool(self.get_revision_count() or
456
self.inventory_index.key_count() or
457
self.text_index.key_count() or
458
self.signature_index.key_count() or
459
(self.chk_index is not None and self.chk_index.key_count()))
461
def finish_content(self):
462
if self.name is not None:
466
self._write_data('', flush=True)
467
self.name = self._hash.hexdigest()
469
def finish(self, suspend=False):
470
"""Finish the new pack.
473
- finalises the content
474
- assigns a name (the md5 of the content, currently)
475
- writes out the associated indices
476
- renames the pack into place.
477
- stores the index size tuple for the pack in the index_sizes
480
self.finish_content()
482
self._check_references()
484
# XXX: It'd be better to write them all to temporary names, then
485
# rename them all into place, so that the window when only some are
486
# visible is smaller. On the other hand none will be seen until
487
# they're in the names list.
488
self.index_sizes = [None, None, None, None]
489
self._write_index('revision', self.revision_index, 'revision', suspend)
490
self._write_index('inventory', self.inventory_index, 'inventory',
492
self._write_index('text', self.text_index, 'file texts', suspend)
493
self._write_index('signature', self.signature_index,
494
'revision signatures', suspend)
495
if self.chk_index is not None:
496
self.index_sizes.append(None)
497
self._write_index('chk', self.chk_index,
498
'content hash bytes', suspend)
499
self.write_stream.close()
500
# Note that this will clobber an existing pack with the same name,
501
# without checking for hash collisions. While this is undesirable this
502
# is something that can be rectified in a subsequent release. One way
503
# to rectify it may be to leave the pack at the original name, writing
504
# its pack-names entry as something like 'HASH: index-sizes
505
# temporary-name'. Allocate that and check for collisions, if it is
506
# collision free then rename it into place. If clients know this scheme
507
# they can handle missing-file errors by:
508
# - try for HASH.pack
509
# - try for temporary-name
510
# - refresh the pack-list to see if the pack is now absent
511
new_name = self.name + '.pack'
513
new_name = '../packs/' + new_name
514
self.upload_transport.rename(self.random_name, new_name)
515
self._state = 'finished'
516
if 'pack' in debug.debug_flags:
517
# XXX: size might be interesting?
518
mutter('%s: create_pack: pack finished: %s%s->%s t+%6.3fs',
519
time.ctime(), self.upload_transport.base, self.random_name,
520
new_name, time.time() - self.start_time)
523
"""Flush any current data."""
525
bytes = ''.join(self._buffer[0])
526
self.write_stream.write(bytes)
527
self._hash.update(bytes)
528
self._buffer[:] = [[], 0]
530
def _get_external_refs(self, index):
531
return index._external_references()
533
def set_write_cache_size(self, size):
534
self._cache_limit = size
536
def _write_index(self, index_type, index, label, suspend=False):
537
"""Write out an index.
539
:param index_type: The type of index to write - e.g. 'revision'.
540
:param index: The index object to serialise.
541
:param label: What label to give the index e.g. 'revision'.
543
index_name = self.index_name(index_type, self.name)
545
transport = self.upload_transport
547
transport = self.index_transport
548
self.index_sizes[self.index_offset(index_type)] = transport.put_file(
549
index_name, index.finish(), mode=self._file_mode)
550
if 'pack' in debug.debug_flags:
551
# XXX: size might be interesting?
552
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
553
time.ctime(), label, self.upload_transport.base,
554
self.random_name, time.time() - self.start_time)
555
# Replace the writable index on this object with a readonly,
556
# presently unloaded index. We should alter
557
# the index layer to make its finish() error if add_node is
558
# subsequently used. RBC
559
self._replace_index_with_readonly(index_type)
562
class AggregateIndex(object):
563
"""An aggregated index for the RepositoryPackCollection.
565
AggregateIndex is reponsible for managing the PackAccess object,
566
Index-To-Pack mapping, and all indices list for a specific type of index
567
such as 'revision index'.
569
A CombinedIndex provides an index on a single key space built up
570
from several on-disk indices. The AggregateIndex builds on this
571
to provide a knit access layer, and allows having up to one writable
572
index within the collection.
574
# XXX: Probably 'can be written to' could/should be separated from 'acts
575
# like a knit index' -- mbp 20071024
577
def __init__(self, reload_func=None, flush_func=None):
578
"""Create an AggregateIndex.
580
:param reload_func: A function to call if we find we are missing an
581
index. Should have the form reload_func() => True if the list of
582
active pack files has changed.
584
self._reload_func = reload_func
585
self.index_to_pack = {}
586
self.combined_index = CombinedGraphIndex([], reload_func=reload_func)
587
self.data_access = _DirectPackAccess(self.index_to_pack,
588
reload_func=reload_func,
589
flush_func=flush_func)
590
self.add_callback = None
592
def add_index(self, index, pack):
593
"""Add index to the aggregate, which is an index for Pack pack.
595
Future searches on the aggregate index will seach this new index
596
before all previously inserted indices.
598
:param index: An Index for the pack.
599
:param pack: A Pack instance.
601
# expose it to the index map
602
self.index_to_pack[index] = pack.access_tuple()
603
# put it at the front of the linear index list
604
self.combined_index.insert_index(0, index, pack.name)
606
def add_writable_index(self, index, pack):
607
"""Add an index which is able to have data added to it.
609
There can be at most one writable index at any time. Any
610
modifications made to the knit are put into this index.
612
:param index: An index from the pack parameter.
613
:param pack: A Pack instance.
615
if self.add_callback is not None:
616
raise AssertionError(
617
"%s already has a writable index through %s" % \
618
(self, self.add_callback))
619
# allow writing: queue writes to a new index
620
self.add_index(index, pack)
621
# Updates the index to packs mapping as a side effect,
622
self.data_access.set_writer(pack._writer, index, pack.access_tuple())
623
self.add_callback = index.add_nodes
626
"""Reset all the aggregate data to nothing."""
627
self.data_access.set_writer(None, None, (None, None))
628
self.index_to_pack.clear()
629
del self.combined_index._indices[:]
630
del self.combined_index._index_names[:]
631
self.add_callback = None
633
def remove_index(self, index):
634
"""Remove index from the indices used to answer queries.
636
:param index: An index from the pack parameter.
638
del self.index_to_pack[index]
639
pos = self.combined_index._indices.index(index)
640
del self.combined_index._indices[pos]
641
del self.combined_index._index_names[pos]
642
if (self.add_callback is not None and
643
getattr(index, 'add_nodes', None) == self.add_callback):
644
self.add_callback = None
645
self.data_access.set_writer(None, None, (None, None))
648
class Packer(object):
649
"""Create a pack from packs."""
651
def __init__(self, pack_collection, packs, suffix, revision_ids=None,
655
:param pack_collection: A RepositoryPackCollection object where the
656
new pack is being written to.
657
:param packs: The packs to combine.
658
:param suffix: The suffix to use on the temporary files for the pack.
659
:param revision_ids: Revision ids to limit the pack to.
660
:param reload_func: A function to call if a pack file/index goes
661
missing. The side effect of calling this function should be to
662
update self.packs. See also AggregateIndex
666
self.revision_ids = revision_ids
667
# The pack object we are creating.
669
self._pack_collection = pack_collection
670
self._reload_func = reload_func
671
# The index layer keys for the revisions being copied. None for 'all
673
self._revision_keys = None
674
# What text keys to copy. None for 'all texts'. This is set by
675
# _copy_inventory_texts
676
self._text_filter = None
679
def _extra_init(self):
680
"""A template hook to allow extending the constructor trivially."""
682
def _pack_map_and_index_list(self, index_attribute):
683
"""Convert a list of packs to an index pack map and index list.
685
:param index_attribute: The attribute that the desired index is found
687
:return: A tuple (map, list) where map contains the dict from
688
index:pack_tuple, and list contains the indices in the preferred
693
for pack_obj in self.packs:
694
index = getattr(pack_obj, index_attribute)
695
indices.append(index)
696
pack_map[index] = pack_obj
697
return pack_map, indices
699
def _index_contents(self, indices, key_filter=None):
700
"""Get an iterable of the index contents from a pack_map.
702
:param indices: The list of indices to query
703
:param key_filter: An optional filter to limit the keys returned.
705
all_index = CombinedGraphIndex(indices)
706
if key_filter is None:
707
return all_index.iter_all_entries()
709
return all_index.iter_entries(key_filter)
711
def pack(self, pb=None):
712
"""Create a new pack by reading data from other packs.
714
This does little more than a bulk copy of data. One key difference
715
is that data with the same item key across multiple packs is elided
716
from the output. The new pack is written into the current pack store
717
along with its indices, and the name added to the pack names. The
718
source packs are not altered and are not required to be in the current
721
:param pb: An optional progress bar to use. A nested bar is created if
723
:return: A Pack object, or None if nothing was copied.
725
# open a pack - using the same name as the last temporary file
726
# - which has already been flushed, so its safe.
727
# XXX: - duplicate code warning with start_write_group; fix before
728
# considering 'done'.
729
if self._pack_collection._new_pack is not None:
730
raise errors.BzrError('call to %s.pack() while another pack is'
732
% (self.__class__.__name__,))
733
if self.revision_ids is not None:
734
if len(self.revision_ids) == 0:
735
# silly fetch request.
738
self.revision_ids = frozenset(self.revision_ids)
739
self.revision_keys = frozenset((revid,) for revid in
742
self.pb = ui.ui_factory.nested_progress_bar()
746
return self._create_pack_from_packs()
752
"""Open a pack for the pack we are creating."""
753
new_pack = self._pack_collection.pack_factory(self._pack_collection,
754
upload_suffix=self.suffix,
755
file_mode=self._pack_collection.repo.bzrdir._get_file_mode())
756
# We know that we will process all nodes in order, and don't need to
757
# query, so don't combine any indices spilled to disk until we are done
758
new_pack.revision_index.set_optimize(combine_backing_indices=False)
759
new_pack.inventory_index.set_optimize(combine_backing_indices=False)
760
new_pack.text_index.set_optimize(combine_backing_indices=False)
761
new_pack.signature_index.set_optimize(combine_backing_indices=False)
764
def _update_pack_order(self, entries, index_to_pack_map):
765
"""Determine how we want our packs to be ordered.
767
This changes the sort order of the self.packs list so that packs unused
768
by 'entries' will be at the end of the list, so that future requests
769
can avoid probing them. Used packs will be at the front of the
770
self.packs list, in the order of their first use in 'entries'.
772
:param entries: A list of (index, ...) tuples
773
:param index_to_pack_map: A mapping from index objects to pack objects.
777
for entry in entries:
779
if index not in seen_indexes:
780
packs.append(index_to_pack_map[index])
781
seen_indexes.add(index)
782
if len(packs) == len(self.packs):
783
if 'pack' in debug.debug_flags:
784
mutter('Not changing pack list, all packs used.')
786
seen_packs = set(packs)
787
for pack in self.packs:
788
if pack not in seen_packs:
791
if 'pack' in debug.debug_flags:
792
old_names = [p.access_tuple()[1] for p in self.packs]
793
new_names = [p.access_tuple()[1] for p in packs]
794
mutter('Reordering packs\nfrom: %s\n to: %s',
795
old_names, new_names)
798
def _copy_revision_texts(self):
799
"""Copy revision data to the new pack."""
801
if self.revision_ids:
802
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
805
# select revision keys
806
revision_index_map, revision_indices = self._pack_map_and_index_list(
808
revision_nodes = self._index_contents(revision_indices, revision_keys)
809
revision_nodes = list(revision_nodes)
810
self._update_pack_order(revision_nodes, revision_index_map)
811
# copy revision keys and adjust values
812
self.pb.update("Copying revision texts", 1)
813
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
814
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
815
self.new_pack.revision_index, readv_group_iter, total_items))
816
if 'pack' in debug.debug_flags:
817
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
818
time.ctime(), self._pack_collection._upload_transport.base,
819
self.new_pack.random_name,
820
self.new_pack.revision_index.key_count(),
821
time.time() - self.new_pack.start_time)
822
self._revision_keys = revision_keys
824
def _copy_inventory_texts(self):
825
"""Copy the inventory texts to the new pack.
827
self._revision_keys is used to determine what inventories to copy.
829
Sets self._text_filter appropriately.
831
# select inventory keys
832
inv_keys = self._revision_keys # currently the same keyspace, and note that
833
# querying for keys here could introduce a bug where an inventory item
834
# is missed, so do not change it to query separately without cross
835
# checking like the text key check below.
836
inventory_index_map, inventory_indices = self._pack_map_and_index_list(
838
inv_nodes = self._index_contents(inventory_indices, inv_keys)
839
# copy inventory keys and adjust values
840
# XXX: Should be a helper function to allow different inv representation
842
self.pb.update("Copying inventory texts", 2)
843
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
844
# Only grab the output lines if we will be processing them
845
output_lines = bool(self.revision_ids)
846
inv_lines = self._copy_nodes_graph(inventory_index_map,
847
self.new_pack._writer, self.new_pack.inventory_index,
848
readv_group_iter, total_items, output_lines=output_lines)
849
if self.revision_ids:
850
self._process_inventory_lines(inv_lines)
852
# eat the iterator to cause it to execute.
854
self._text_filter = None
855
if 'pack' in debug.debug_flags:
856
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
857
time.ctime(), self._pack_collection._upload_transport.base,
858
self.new_pack.random_name,
859
self.new_pack.inventory_index.key_count(),
860
time.time() - self.new_pack.start_time)
862
def _copy_text_texts(self):
864
text_index_map, text_nodes = self._get_text_nodes()
865
if self._text_filter is not None:
866
# We could return the keys copied as part of the return value from
867
# _copy_nodes_graph but this doesn't work all that well with the
868
# need to get line output too, so we check separately, and as we're
869
# going to buffer everything anyway, we check beforehand, which
870
# saves reading knit data over the wire when we know there are
872
text_nodes = set(text_nodes)
873
present_text_keys = set(_node[1] for _node in text_nodes)
874
missing_text_keys = set(self._text_filter) - present_text_keys
875
if missing_text_keys:
876
# TODO: raise a specific error that can handle many missing
878
mutter("missing keys during fetch: %r", missing_text_keys)
879
a_missing_key = missing_text_keys.pop()
880
raise errors.RevisionNotPresent(a_missing_key[1],
882
# copy text keys and adjust values
883
self.pb.update("Copying content texts", 3)
884
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
885
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
886
self.new_pack.text_index, readv_group_iter, total_items))
887
self._log_copied_texts()
889
def _create_pack_from_packs(self):
890
self.pb.update("Opening pack", 0, 5)
891
self.new_pack = self.open_pack()
892
new_pack = self.new_pack
893
# buffer data - we won't be reading-back during the pack creation and
894
# this makes a significant difference on sftp pushes.
895
new_pack.set_write_cache_size(1024*1024)
896
if 'pack' in debug.debug_flags:
897
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
898
for a_pack in self.packs]
899
if self.revision_ids is not None:
900
rev_count = len(self.revision_ids)
903
mutter('%s: create_pack: creating pack from source packs: '
904
'%s%s %s revisions wanted %s t=0',
905
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
906
plain_pack_list, rev_count)
907
self._copy_revision_texts()
908
self._copy_inventory_texts()
909
self._copy_text_texts()
910
# select signature keys
911
signature_filter = self._revision_keys # same keyspace
912
signature_index_map, signature_indices = self._pack_map_and_index_list(
914
signature_nodes = self._index_contents(signature_indices,
916
# copy signature keys and adjust values
917
self.pb.update("Copying signature texts", 4)
918
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
919
new_pack.signature_index)
920
if 'pack' in debug.debug_flags:
921
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
922
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
923
new_pack.signature_index.key_count(),
924
time.time() - new_pack.start_time)
926
# NB XXX: how to check CHK references are present? perhaps by yielding
927
# the items? How should that interact with stacked repos?
928
if new_pack.chk_index is not None:
930
if 'pack' in debug.debug_flags:
931
mutter('%s: create_pack: chk content copied: %s%s %d items t+%6.3fs',
932
time.ctime(), self._pack_collection._upload_transport.base,
933
new_pack.random_name,
934
new_pack.chk_index.key_count(),
935
time.time() - new_pack.start_time)
936
new_pack._check_references()
937
if not self._use_pack(new_pack):
940
self.pb.update("Finishing pack", 5)
942
self._pack_collection.allocate(new_pack)
945
def _copy_chks(self, refs=None):
946
# XXX: Todo, recursive follow-pointers facility when fetching some
948
chk_index_map, chk_indices = self._pack_map_and_index_list(
950
chk_nodes = self._index_contents(chk_indices, refs)
952
# TODO: This isn't strictly tasteful as we are accessing some private
953
# variables (_serializer). Perhaps a better way would be to have
954
# Repository._deserialise_chk_node()
955
search_key_func = chk_map.search_key_registry.get(
956
self._pack_collection.repo._serializer.search_key_name)
957
def accumlate_refs(lines):
958
# XXX: move to a generic location
960
bytes = ''.join(lines)
961
node = chk_map._deserialise(bytes, ("unknown",), search_key_func)
962
new_refs.update(node.refs())
963
self._copy_nodes(chk_nodes, chk_index_map, self.new_pack._writer,
964
self.new_pack.chk_index, output_lines=accumlate_refs)
967
def _copy_nodes(self, nodes, index_map, writer, write_index,
969
"""Copy knit nodes between packs with no graph references.
971
:param output_lines: Output full texts of copied items.
973
pb = ui.ui_factory.nested_progress_bar()
975
return self._do_copy_nodes(nodes, index_map, writer,
976
write_index, pb, output_lines=output_lines)
980
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb,
982
# for record verification
983
knit = KnitVersionedFiles(None, None)
984
# plan a readv on each source pack:
986
nodes = sorted(nodes)
987
# how to map this into knit.py - or knit.py into this?
988
# we don't want the typical knit logic, we want grouping by pack
989
# at this point - perhaps a helper library for the following code
990
# duplication points?
992
for index, key, value in nodes:
993
if index not in request_groups:
994
request_groups[index] = []
995
request_groups[index].append((key, value))
997
pb.update("Copied record", record_index, len(nodes))
998
for index, items in request_groups.iteritems():
999
pack_readv_requests = []
1000
for key, value in items:
1001
# ---- KnitGraphIndex.get_position
1002
bits = value[1:].split(' ')
1003
offset, length = int(bits[0]), int(bits[1])
1004
pack_readv_requests.append((offset, length, (key, value[0])))
1005
# linear scan up the pack
1006
pack_readv_requests.sort()
1008
pack_obj = index_map[index]
1009
transport, path = pack_obj.access_tuple()
1011
reader = pack.make_readv_reader(transport, path,
1012
[offset[0:2] for offset in pack_readv_requests])
1013
except errors.NoSuchFile:
1014
if self._reload_func is not None:
1017
for (names, read_func), (_1, _2, (key, eol_flag)) in \
1018
izip(reader.iter_records(), pack_readv_requests):
1019
raw_data = read_func(None)
1020
# check the header only
1021
if output_lines is not None:
1022
output_lines(knit._parse_record(key[-1], raw_data)[0])
1024
df, _ = knit._parse_record_header(key, raw_data)
1026
pos, size = writer.add_bytes_record(raw_data, names)
1027
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
1028
pb.update("Copied record", record_index)
1031
def _copy_nodes_graph(self, index_map, writer, write_index,
1032
readv_group_iter, total_items, output_lines=False):
1033
"""Copy knit nodes between packs.
1035
:param output_lines: Return lines present in the copied data as
1036
an iterator of line,version_id.
1038
pb = ui.ui_factory.nested_progress_bar()
1040
for result in self._do_copy_nodes_graph(index_map, writer,
1041
write_index, output_lines, pb, readv_group_iter, total_items):
1044
# Python 2.4 does not permit try:finally: in a generator.
1050
def _do_copy_nodes_graph(self, index_map, writer, write_index,
1051
output_lines, pb, readv_group_iter, total_items):
1052
# for record verification
1053
knit = KnitVersionedFiles(None, None)
1054
# for line extraction when requested (inventories only)
1056
factory = KnitPlainFactory()
1058
pb.update("Copied record", record_index, total_items)
1059
for index, readv_vector, node_vector in readv_group_iter:
1061
pack_obj = index_map[index]
1062
transport, path = pack_obj.access_tuple()
1064
reader = pack.make_readv_reader(transport, path, readv_vector)
1065
except errors.NoSuchFile:
1066
if self._reload_func is not None:
1069
for (names, read_func), (key, eol_flag, references) in \
1070
izip(reader.iter_records(), node_vector):
1071
raw_data = read_func(None)
1073
# read the entire thing
1074
content, _ = knit._parse_record(key[-1], raw_data)
1075
if len(references[-1]) == 0:
1076
line_iterator = factory.get_fulltext_content(content)
1078
line_iterator = factory.get_linedelta_content(content)
1079
for line in line_iterator:
1082
# check the header only
1083
df, _ = knit._parse_record_header(key, raw_data)
1085
pos, size = writer.add_bytes_record(raw_data, names)
1086
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1087
pb.update("Copied record", record_index)
1090
def _get_text_nodes(self):
1091
text_index_map, text_indices = self._pack_map_and_index_list(
1093
return text_index_map, self._index_contents(text_indices,
1096
def _least_readv_node_readv(self, nodes):
1097
"""Generate request groups for nodes using the least readv's.
1099
:param nodes: An iterable of graph index nodes.
1100
:return: Total node count and an iterator of the data needed to perform
1101
readvs to obtain the data for nodes. Each item yielded by the
1102
iterator is a tuple with:
1103
index, readv_vector, node_vector. readv_vector is a list ready to
1104
hand to the transport readv method, and node_vector is a list of
1105
(key, eol_flag, references) for the node retrieved by the
1106
matching readv_vector.
1108
# group by pack so we do one readv per pack
1109
nodes = sorted(nodes)
1112
for index, key, value, references in nodes:
1113
if index not in request_groups:
1114
request_groups[index] = []
1115
request_groups[index].append((key, value, references))
1117
for index, items in request_groups.iteritems():
1118
pack_readv_requests = []
1119
for key, value, references in items:
1120
# ---- KnitGraphIndex.get_position
1121
bits = value[1:].split(' ')
1122
offset, length = int(bits[0]), int(bits[1])
1123
pack_readv_requests.append(
1124
((offset, length), (key, value[0], references)))
1125
# linear scan up the pack to maximum range combining.
1126
pack_readv_requests.sort()
1127
# split out the readv and the node data.
1128
pack_readv = [readv for readv, node in pack_readv_requests]
1129
node_vector = [node for readv, node in pack_readv_requests]
1130
result.append((index, pack_readv, node_vector))
1131
return total, result
1133
def _log_copied_texts(self):
1134
if 'pack' in debug.debug_flags:
1135
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
1136
time.ctime(), self._pack_collection._upload_transport.base,
1137
self.new_pack.random_name,
1138
self.new_pack.text_index.key_count(),
1139
time.time() - self.new_pack.start_time)
1141
def _process_inventory_lines(self, inv_lines):
1142
"""Use up the inv_lines generator and setup a text key filter."""
1143
repo = self._pack_collection.repo
1144
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
1145
inv_lines, self.revision_keys)
1147
for fileid, file_revids in fileid_revisions.iteritems():
1148
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
1149
self._text_filter = text_filter
1151
def _revision_node_readv(self, revision_nodes):
1152
"""Return the total revisions and the readv's to issue.
1154
:param revision_nodes: The revision index contents for the packs being
1155
incorporated into the new pack.
1156
:return: As per _least_readv_node_readv.
1158
return self._least_readv_node_readv(revision_nodes)
1160
def _use_pack(self, new_pack):
1161
"""Return True if new_pack should be used.
1163
:param new_pack: The pack that has just been created.
1164
:return: True if the pack should be used.
1166
return new_pack.data_inserted()
1169
class OptimisingPacker(Packer):
1170
"""A packer which spends more time to create better disk layouts."""
1172
def _revision_node_readv(self, revision_nodes):
1173
"""Return the total revisions and the readv's to issue.
1175
This sort places revisions in topological order with the ancestors
1178
:param revision_nodes: The revision index contents for the packs being
1179
incorporated into the new pack.
1180
:return: As per _least_readv_node_readv.
1182
# build an ancestors dict
1185
for index, key, value, references in revision_nodes:
1186
ancestors[key] = references[0]
1187
by_key[key] = (index, value, references)
1188
order = tsort.topo_sort(ancestors)
1190
# Single IO is pathological, but it will work as a starting point.
1192
for key in reversed(order):
1193
index, value, references = by_key[key]
1194
# ---- KnitGraphIndex.get_position
1195
bits = value[1:].split(' ')
1196
offset, length = int(bits[0]), int(bits[1])
1198
(index, [(offset, length)], [(key, value[0], references)]))
1199
# TODO: combine requests in the same index that are in ascending order.
1200
return total, requests
1202
def open_pack(self):
1203
"""Open a pack for the pack we are creating."""
1204
new_pack = super(OptimisingPacker, self).open_pack()
1205
# Turn on the optimization flags for all the index builders.
1206
new_pack.revision_index.set_optimize(for_size=True)
1207
new_pack.inventory_index.set_optimize(for_size=True)
1208
new_pack.text_index.set_optimize(for_size=True)
1209
new_pack.signature_index.set_optimize(for_size=True)
1213
class ReconcilePacker(Packer):
1214
"""A packer which regenerates indices etc as it copies.
1216
This is used by ``bzr reconcile`` to cause parent text pointers to be
1220
def _extra_init(self):
1221
self._data_changed = False
1223
def _process_inventory_lines(self, inv_lines):
1224
"""Generate a text key reference map rather for reconciling with."""
1225
repo = self._pack_collection.repo
1226
refs = repo._find_text_key_references_from_xml_inventory_lines(
1228
self._text_refs = refs
1229
# during reconcile we:
1230
# - convert unreferenced texts to full texts
1231
# - correct texts which reference a text not copied to be full texts
1232
# - copy all others as-is but with corrected parents.
1233
# - so at this point we don't know enough to decide what becomes a full
1235
self._text_filter = None
1237
def _copy_text_texts(self):
1238
"""generate what texts we should have and then copy."""
1239
self.pb.update("Copying content texts", 3)
1240
# we have three major tasks here:
1241
# 1) generate the ideal index
1242
repo = self._pack_collection.repo
1243
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1244
_1, key, _2, refs in
1245
self.new_pack.revision_index.iter_all_entries()])
1246
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1247
# 2) generate a text_nodes list that contains all the deltas that can
1248
# be used as-is, with corrected parents.
1251
discarded_nodes = []
1252
NULL_REVISION = _mod_revision.NULL_REVISION
1253
text_index_map, text_nodes = self._get_text_nodes()
1254
for node in text_nodes:
1260
ideal_parents = tuple(ideal_index[node[1]])
1262
discarded_nodes.append(node)
1263
self._data_changed = True
1265
if ideal_parents == (NULL_REVISION,):
1267
if ideal_parents == node[3][0]:
1269
ok_nodes.append(node)
1270
elif ideal_parents[0:1] == node[3][0][0:1]:
1271
# the left most parent is the same, or there are no parents
1272
# today. Either way, we can preserve the representation as
1273
# long as we change the refs to be inserted.
1274
self._data_changed = True
1275
ok_nodes.append((node[0], node[1], node[2],
1276
(ideal_parents, node[3][1])))
1277
self._data_changed = True
1279
# Reinsert this text completely
1280
bad_texts.append((node[1], ideal_parents))
1281
self._data_changed = True
1282
# we're finished with some data.
1285
# 3) bulk copy the ok data
1286
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1287
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1288
self.new_pack.text_index, readv_group_iter, total_items))
1289
# 4) adhoc copy all the other texts.
1290
# We have to topologically insert all texts otherwise we can fail to
1291
# reconcile when parts of a single delta chain are preserved intact,
1292
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1293
# reinserted, and if d3 has incorrect parents it will also be
1294
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1295
# copied), so we will try to delta, but d2 is not currently able to be
1296
# extracted because it's basis d1 is not present. Topologically sorting
1297
# addresses this. The following generates a sort for all the texts that
1298
# are being inserted without having to reference the entire text key
1299
# space (we only topo sort the revisions, which is smaller).
1300
topo_order = tsort.topo_sort(ancestors)
1301
rev_order = dict(zip(topo_order, range(len(topo_order))))
1302
bad_texts.sort(key=lambda key:rev_order.get(key[0][1], 0))
1303
transaction = repo.get_transaction()
1304
file_id_index = GraphIndexPrefixAdapter(
1305
self.new_pack.text_index,
1307
add_nodes_callback=self.new_pack.text_index.add_nodes)
1308
data_access = _DirectPackAccess(
1309
{self.new_pack.text_index:self.new_pack.access_tuple()})
1310
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1311
self.new_pack.access_tuple())
1312
output_texts = KnitVersionedFiles(
1313
_KnitGraphIndex(self.new_pack.text_index,
1314
add_callback=self.new_pack.text_index.add_nodes,
1315
deltas=True, parents=True, is_locked=repo.is_locked),
1316
data_access=data_access, max_delta_chain=200)
1317
for key, parent_keys in bad_texts:
1318
# We refer to the new pack to delta data being output.
1319
# A possible improvement would be to catch errors on short reads
1320
# and only flush then.
1321
self.new_pack.flush()
1323
for parent_key in parent_keys:
1324
if parent_key[0] != key[0]:
1325
# Graph parents must match the fileid
1326
raise errors.BzrError('Mismatched key parent %r:%r' %
1328
parents.append(parent_key[1])
1329
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1330
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1331
output_texts.add_lines(key, parent_keys, text_lines,
1332
random_id=True, check_content=False)
1333
# 5) check that nothing inserted has a reference outside the keyspace.
1334
missing_text_keys = self.new_pack.text_index._external_references()
1335
if missing_text_keys:
1336
raise errors.BzrCheckError('Reference to missing compression parents %r'
1337
% (missing_text_keys,))
1338
self._log_copied_texts()
1340
def _use_pack(self, new_pack):
1341
"""Override _use_pack to check for reconcile having changed content."""
1342
# XXX: we might be better checking this at the copy time.
1343
original_inventory_keys = set()
1344
inv_index = self._pack_collection.inventory_index.combined_index
1345
for entry in inv_index.iter_all_entries():
1346
original_inventory_keys.add(entry[1])
1347
new_inventory_keys = set()
1348
for entry in new_pack.inventory_index.iter_all_entries():
1349
new_inventory_keys.add(entry[1])
1350
if new_inventory_keys != original_inventory_keys:
1351
self._data_changed = True
1352
return new_pack.data_inserted() and self._data_changed
1355
class RepositoryPackCollection(object):
1356
"""Management of packs within a repository.
1358
:ivar _names: map of {pack_name: (index_size,)}
1361
pack_factory = NewPack
1362
resumed_pack_factory = ResumedPack
1364
def __init__(self, repo, transport, index_transport, upload_transport,
1365
pack_transport, index_builder_class, index_class,
1367
"""Create a new RepositoryPackCollection.
1369
:param transport: Addresses the repository base directory
1370
(typically .bzr/repository/).
1371
:param index_transport: Addresses the directory containing indices.
1372
:param upload_transport: Addresses the directory into which packs are written
1373
while they're being created.
1374
:param pack_transport: Addresses the directory of existing complete packs.
1375
:param index_builder_class: The index builder class to use.
1376
:param index_class: The index class to use.
1377
:param use_chk_index: Whether to setup and manage a CHK index.
1379
# XXX: This should call self.reset()
1381
self.transport = transport
1382
self._index_transport = index_transport
1383
self._upload_transport = upload_transport
1384
self._pack_transport = pack_transport
1385
self._index_builder_class = index_builder_class
1386
self._index_class = index_class
1387
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3,
1392
self._packs_by_name = {}
1393
# the previous pack-names content
1394
self._packs_at_load = None
1395
# when a pack is being created by this object, the state of that pack.
1396
self._new_pack = None
1397
# aggregated revision index data
1398
flush = self._flush_new_pack
1399
self.revision_index = AggregateIndex(self.reload_pack_names, flush)
1400
self.inventory_index = AggregateIndex(self.reload_pack_names, flush)
1401
self.text_index = AggregateIndex(self.reload_pack_names, flush)
1402
self.signature_index = AggregateIndex(self.reload_pack_names, flush)
1403
all_indices = [self.revision_index, self.inventory_index,
1404
self.text_index, self.signature_index]
1406
self.chk_index = AggregateIndex(self.reload_pack_names, flush)
1407
all_indices.append(self.chk_index)
1409
# used to determine if we're using a chk_index elsewhere.
1410
self.chk_index = None
1411
# Tell all the CombinedGraphIndex objects about each other, so they can
1412
# share hints about which pack names to search first.
1413
all_combined = [agg_idx.combined_index for agg_idx in all_indices]
1414
for combined_idx in all_combined:
1415
combined_idx.set_sibling_indices(
1416
set(all_combined).difference([combined_idx]))
1418
self._resumed_packs = []
1421
return '%s(%r)' % (self.__class__.__name__, self.repo)
1423
def add_pack_to_memory(self, pack):
1424
"""Make a Pack object available to the repository to satisfy queries.
1426
:param pack: A Pack object.
1428
if pack.name in self._packs_by_name:
1429
raise AssertionError(
1430
'pack %s already in _packs_by_name' % (pack.name,))
1431
self.packs.append(pack)
1432
self._packs_by_name[pack.name] = pack
1433
self.revision_index.add_index(pack.revision_index, pack)
1434
self.inventory_index.add_index(pack.inventory_index, pack)
1435
self.text_index.add_index(pack.text_index, pack)
1436
self.signature_index.add_index(pack.signature_index, pack)
1437
if self.chk_index is not None:
1438
self.chk_index.add_index(pack.chk_index, pack)
1440
def all_packs(self):
1441
"""Return a list of all the Pack objects this repository has.
1443
Note that an in-progress pack being created is not returned.
1445
:return: A list of Pack objects for all the packs in the repository.
1448
for name in self.names():
1449
result.append(self.get_pack_by_name(name))
1453
"""Pack the pack collection incrementally.
1455
This will not attempt global reorganisation or recompression,
1456
rather it will just ensure that the total number of packs does
1457
not grow without bound. It uses the _max_pack_count method to
1458
determine if autopacking is needed, and the pack_distribution
1459
method to determine the number of revisions in each pack.
1461
If autopacking takes place then the packs name collection will have
1462
been flushed to disk - packing requires updating the name collection
1463
in synchronisation with certain steps. Otherwise the names collection
1466
:return: Something evaluating true if packing took place.
1470
return self._do_autopack()
1471
except errors.RetryAutopack:
1472
# If we get a RetryAutopack exception, we should abort the
1473
# current action, and retry.
1476
def _do_autopack(self):
1477
# XXX: Should not be needed when the management of indices is sane.
1478
total_revisions = self.revision_index.combined_index.key_count()
1479
total_packs = len(self._names)
1480
if self._max_pack_count(total_revisions) >= total_packs:
1482
# determine which packs need changing
1483
pack_distribution = self.pack_distribution(total_revisions)
1485
for pack in self.all_packs():
1486
revision_count = pack.get_revision_count()
1487
if revision_count == 0:
1488
# revision less packs are not generated by normal operation,
1489
# only by operations like sign-my-commits, and thus will not
1490
# tend to grow rapdily or without bound like commit containing
1491
# packs do - leave them alone as packing them really should
1492
# group their data with the relevant commit, and that may
1493
# involve rewriting ancient history - which autopack tries to
1494
# avoid. Alternatively we could not group the data but treat
1495
# each of these as having a single revision, and thus add
1496
# one revision for each to the total revision count, to get
1497
# a matching distribution.
1499
existing_packs.append((revision_count, pack))
1500
pack_operations = self.plan_autopack_combinations(
1501
existing_packs, pack_distribution)
1502
num_new_packs = len(pack_operations)
1503
num_old_packs = sum([len(po[1]) for po in pack_operations])
1504
num_revs_affected = sum([po[0] for po in pack_operations])
1505
mutter('Auto-packing repository %s, which has %d pack files, '
1506
'containing %d revisions. Packing %d files into %d affecting %d'
1507
' revisions', self, total_packs, total_revisions, num_old_packs,
1508
num_new_packs, num_revs_affected)
1509
result = self._execute_pack_operations(pack_operations,
1510
reload_func=self._restart_autopack)
1511
mutter('Auto-packing repository %s completed', self)
1514
def _execute_pack_operations(self, pack_operations, _packer_class=Packer,
1516
"""Execute a series of pack operations.
1518
:param pack_operations: A list of [revision_count, packs_to_combine].
1519
:param _packer_class: The class of packer to use (default: Packer).
1520
:return: The new pack names.
1522
for revision_count, packs in pack_operations:
1523
# we may have no-ops from the setup logic
1526
packer = _packer_class(self, packs, '.autopack',
1527
reload_func=reload_func)
1530
except errors.RetryWithNewPacks:
1531
# An exception is propagating out of this context, make sure
1532
# this packer has cleaned up. Packer() doesn't set its new_pack
1533
# state into the RepositoryPackCollection object, so we only
1534
# have access to it directly here.
1535
if packer.new_pack is not None:
1536
packer.new_pack.abort()
1539
self._remove_pack_from_memory(pack)
1540
# record the newly available packs and stop advertising the old
1542
to_be_obsoleted = []
1543
for _, packs in pack_operations:
1544
to_be_obsoleted.extend(packs)
1545
result = self._save_pack_names(clear_obsolete_packs=True,
1546
obsolete_packs=to_be_obsoleted)
1549
def _flush_new_pack(self):
1550
if self._new_pack is not None:
1551
self._new_pack.flush()
1553
def lock_names(self):
1554
"""Acquire the mutex around the pack-names index.
1556
This cannot be used in the middle of a read-only transaction on the
1559
self.repo.control_files.lock_write()
1561
def _already_packed(self):
1562
"""Is the collection already packed?"""
1563
return not (self.repo._format.pack_compresses or (len(self._names) > 1))
1565
def pack(self, hint=None, clean_obsolete_packs=False):
1566
"""Pack the pack collection totally."""
1567
self.ensure_loaded()
1568
total_packs = len(self._names)
1569
if self._already_packed():
1571
total_revisions = self.revision_index.combined_index.key_count()
1572
# XXX: the following may want to be a class, to pack with a given
1574
mutter('Packing repository %s, which has %d pack files, '
1575
'containing %d revisions with hint %r.', self, total_packs,
1576
total_revisions, hint)
1577
# determine which packs need changing
1578
pack_operations = [[0, []]]
1579
for pack in self.all_packs():
1580
if hint is None or pack.name in hint:
1581
# Either no hint was provided (so we are packing everything),
1582
# or this pack was included in the hint.
1583
pack_operations[-1][0] += pack.get_revision_count()
1584
pack_operations[-1][1].append(pack)
1585
self._execute_pack_operations(pack_operations, OptimisingPacker)
1587
if clean_obsolete_packs:
1588
self._clear_obsolete_packs()
1590
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1591
"""Plan a pack operation.
1593
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1595
:param pack_distribution: A list with the number of revisions desired
1598
if len(existing_packs) <= len(pack_distribution):
1600
existing_packs.sort(reverse=True)
1601
pack_operations = [[0, []]]
1602
# plan out what packs to keep, and what to reorganise
1603
while len(existing_packs):
1604
# take the largest pack, and if its less than the head of the
1605
# distribution chart we will include its contents in the new pack
1606
# for that position. If its larger, we remove its size from the
1607
# distribution chart
1608
next_pack_rev_count, next_pack = existing_packs.pop(0)
1609
if next_pack_rev_count >= pack_distribution[0]:
1610
# this is already packed 'better' than this, so we can
1611
# not waste time packing it.
1612
while next_pack_rev_count > 0:
1613
next_pack_rev_count -= pack_distribution[0]
1614
if next_pack_rev_count >= 0:
1616
del pack_distribution[0]
1618
# didn't use that entire bucket up
1619
pack_distribution[0] = -next_pack_rev_count
1621
# add the revisions we're going to add to the next output pack
1622
pack_operations[-1][0] += next_pack_rev_count
1623
# allocate this pack to the next pack sub operation
1624
pack_operations[-1][1].append(next_pack)
1625
if pack_operations[-1][0] >= pack_distribution[0]:
1626
# this pack is used up, shift left.
1627
del pack_distribution[0]
1628
pack_operations.append([0, []])
1629
# Now that we know which pack files we want to move, shove them all
1630
# into a single pack file.
1632
final_pack_list = []
1633
for num_revs, pack_files in pack_operations:
1634
final_rev_count += num_revs
1635
final_pack_list.extend(pack_files)
1636
if len(final_pack_list) == 1:
1637
raise AssertionError('We somehow generated an autopack with a'
1638
' single pack file being moved.')
1640
return [[final_rev_count, final_pack_list]]
1642
def ensure_loaded(self):
1643
"""Ensure we have read names from disk.
1645
:return: True if the disk names had not been previously read.
1647
# NB: if you see an assertion error here, its probably access against
1648
# an unlocked repo. Naughty.
1649
if not self.repo.is_locked():
1650
raise errors.ObjectNotLocked(self.repo)
1651
if self._names is None:
1653
self._packs_at_load = set()
1654
for index, key, value in self._iter_disk_pack_index():
1656
self._names[name] = self._parse_index_sizes(value)
1657
self._packs_at_load.add((key, value))
1661
# populate all the metadata.
1665
def _parse_index_sizes(self, value):
1666
"""Parse a string of index sizes."""
1667
return tuple([int(digits) for digits in value.split(' ')])
1669
def get_pack_by_name(self, name):
1670
"""Get a Pack object by name.
1672
:param name: The name of the pack - e.g. '123456'
1673
:return: A Pack object.
1676
return self._packs_by_name[name]
1678
rev_index = self._make_index(name, '.rix')
1679
inv_index = self._make_index(name, '.iix')
1680
txt_index = self._make_index(name, '.tix')
1681
sig_index = self._make_index(name, '.six')
1682
if self.chk_index is not None:
1683
chk_index = self._make_index(name, '.cix', unlimited_cache=True)
1686
result = ExistingPack(self._pack_transport, name, rev_index,
1687
inv_index, txt_index, sig_index, chk_index)
1688
self.add_pack_to_memory(result)
1691
def _resume_pack(self, name):
1692
"""Get a suspended Pack object by name.
1694
:param name: The name of the pack - e.g. '123456'
1695
:return: A Pack object.
1697
if not re.match('[a-f0-9]{32}', name):
1698
# Tokens should be md5sums of the suspended pack file, i.e. 32 hex
1700
raise errors.UnresumableWriteGroup(
1701
self.repo, [name], 'Malformed write group token')
1703
rev_index = self._make_index(name, '.rix', resume=True)
1704
inv_index = self._make_index(name, '.iix', resume=True)
1705
txt_index = self._make_index(name, '.tix', resume=True)
1706
sig_index = self._make_index(name, '.six', resume=True)
1707
if self.chk_index is not None:
1708
chk_index = self._make_index(name, '.cix', resume=True,
1709
unlimited_cache=True)
1712
result = self.resumed_pack_factory(name, rev_index, inv_index,
1713
txt_index, sig_index, self._upload_transport,
1714
self._pack_transport, self._index_transport, self,
1715
chk_index=chk_index)
1716
except errors.NoSuchFile, e:
1717
raise errors.UnresumableWriteGroup(self.repo, [name], str(e))
1718
self.add_pack_to_memory(result)
1719
self._resumed_packs.append(result)
1722
def allocate(self, a_new_pack):
1723
"""Allocate name in the list of packs.
1725
:param a_new_pack: A NewPack instance to be added to the collection of
1726
packs for this repository.
1728
self.ensure_loaded()
1729
if a_new_pack.name in self._names:
1730
raise errors.BzrError(
1731
'Pack %r already exists in %s' % (a_new_pack.name, self))
1732
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1733
self.add_pack_to_memory(a_new_pack)
1735
def _iter_disk_pack_index(self):
1736
"""Iterate over the contents of the pack-names index.
1738
This is used when loading the list from disk, and before writing to
1739
detect updates from others during our write operation.
1740
:return: An iterator of the index contents.
1742
return self._index_class(self.transport, 'pack-names', None
1743
).iter_all_entries()
1745
def _make_index(self, name, suffix, resume=False, unlimited_cache=False):
1746
size_offset = self._suffix_offsets[suffix]
1747
index_name = name + suffix
1749
transport = self._upload_transport
1750
index_size = transport.stat(index_name).st_size
1752
transport = self._index_transport
1753
index_size = self._names[name][size_offset]
1754
return self._index_class(transport, index_name, index_size,
1755
unlimited_cache=unlimited_cache)
1757
def _max_pack_count(self, total_revisions):
1758
"""Return the maximum number of packs to use for total revisions.
1760
:param total_revisions: The total number of revisions in the
1763
if not total_revisions:
1765
digits = str(total_revisions)
1767
for digit in digits:
1768
result += int(digit)
1772
"""Provide an order to the underlying names."""
1773
return sorted(self._names.keys())
1775
def _obsolete_packs(self, packs):
1776
"""Move a number of packs which have been obsoleted out of the way.
1778
Each pack and its associated indices are moved out of the way.
1780
Note: for correctness this function should only be called after a new
1781
pack names index has been written without these pack names, and with
1782
the names of packs that contain the data previously available via these
1785
:param packs: The packs to obsolete.
1786
:param return: None.
1790
pack.pack_transport.rename(pack.file_name(),
1791
'../obsolete_packs/' + pack.file_name())
1792
except (errors.PathError, errors.TransportError), e:
1793
# TODO: Should these be warnings or mutters?
1794
mutter("couldn't rename obsolete pack, skipping it:\n%s"
1796
# TODO: Probably needs to know all possible indices for this pack
1797
# - or maybe list the directory and move all indices matching this
1798
# name whether we recognize it or not?
1799
suffixes = ['.iix', '.six', '.tix', '.rix']
1800
if self.chk_index is not None:
1801
suffixes.append('.cix')
1802
for suffix in suffixes:
1804
self._index_transport.rename(pack.name + suffix,
1805
'../obsolete_packs/' + pack.name + suffix)
1806
except (errors.PathError, errors.TransportError), e:
1807
mutter("couldn't rename obsolete index, skipping it:\n%s"
1810
def pack_distribution(self, total_revisions):
1811
"""Generate a list of the number of revisions to put in each pack.
1813
:param total_revisions: The total number of revisions in the
1816
if total_revisions == 0:
1818
digits = reversed(str(total_revisions))
1820
for exponent, count in enumerate(digits):
1821
size = 10 ** exponent
1822
for pos in range(int(count)):
1824
return list(reversed(result))
1826
def _pack_tuple(self, name):
1827
"""Return a tuple with the transport and file name for a pack name."""
1828
return self._pack_transport, name + '.pack'
1830
def _remove_pack_from_memory(self, pack):
1831
"""Remove pack from the packs accessed by this repository.
1833
Only affects memory state, until self._save_pack_names() is invoked.
1835
self._names.pop(pack.name)
1836
self._packs_by_name.pop(pack.name)
1837
self._remove_pack_indices(pack)
1838
self.packs.remove(pack)
1840
def _remove_pack_indices(self, pack, ignore_missing=False):
1841
"""Remove the indices for pack from the aggregated indices.
1843
:param ignore_missing: Suppress KeyErrors from calling remove_index.
1845
for index_type in Pack.index_definitions.keys():
1846
attr_name = index_type + '_index'
1847
aggregate_index = getattr(self, attr_name)
1848
if aggregate_index is not None:
1849
pack_index = getattr(pack, attr_name)
1851
aggregate_index.remove_index(pack_index)
1858
"""Clear all cached data."""
1859
# cached revision data
1860
self.revision_index.clear()
1861
# cached signature data
1862
self.signature_index.clear()
1863
# cached file text data
1864
self.text_index.clear()
1865
# cached inventory data
1866
self.inventory_index.clear()
1868
if self.chk_index is not None:
1869
self.chk_index.clear()
1870
# remove the open pack
1871
self._new_pack = None
1872
# information about packs.
1875
self._packs_by_name = {}
1876
self._packs_at_load = None
1878
def _unlock_names(self):
1879
"""Release the mutex around the pack-names index."""
1880
self.repo.control_files.unlock()
1882
def _diff_pack_names(self):
1883
"""Read the pack names from disk, and compare it to the one in memory.
1885
:return: (disk_nodes, deleted_nodes, new_nodes)
1886
disk_nodes The final set of nodes that should be referenced
1887
deleted_nodes Nodes which have been removed from when we started
1888
new_nodes Nodes that are newly introduced
1890
# load the disk nodes across
1892
for index, key, value in self._iter_disk_pack_index():
1893
disk_nodes.add((key, value))
1894
orig_disk_nodes = set(disk_nodes)
1896
# do a two-way diff against our original content
1897
current_nodes = set()
1898
for name, sizes in self._names.iteritems():
1900
((name, ), ' '.join(str(size) for size in sizes)))
1902
# Packs no longer present in the repository, which were present when we
1903
# locked the repository
1904
deleted_nodes = self._packs_at_load - current_nodes
1905
# Packs which this process is adding
1906
new_nodes = current_nodes - self._packs_at_load
1908
# Update the disk_nodes set to include the ones we are adding, and
1909
# remove the ones which were removed by someone else
1910
disk_nodes.difference_update(deleted_nodes)
1911
disk_nodes.update(new_nodes)
1913
return disk_nodes, deleted_nodes, new_nodes, orig_disk_nodes
1915
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1916
"""Given the correct set of pack files, update our saved info.
1918
:return: (removed, added, modified)
1919
removed pack names removed from self._names
1920
added pack names added to self._names
1921
modified pack names that had changed value
1926
## self._packs_at_load = disk_nodes
1927
new_names = dict(disk_nodes)
1928
# drop no longer present nodes
1929
for pack in self.all_packs():
1930
if (pack.name,) not in new_names:
1931
removed.append(pack.name)
1932
self._remove_pack_from_memory(pack)
1933
# add new nodes/refresh existing ones
1934
for key, value in disk_nodes:
1936
sizes = self._parse_index_sizes(value)
1937
if name in self._names:
1939
if sizes != self._names[name]:
1940
# the pack for name has had its indices replaced - rare but
1941
# important to handle. XXX: probably can never happen today
1942
# because the three-way merge code above does not handle it
1943
# - you may end up adding the same key twice to the new
1944
# disk index because the set values are the same, unless
1945
# the only index shows up as deleted by the set difference
1946
# - which it may. Until there is a specific test for this,
1947
# assume its broken. RBC 20071017.
1948
self._remove_pack_from_memory(self.get_pack_by_name(name))
1949
self._names[name] = sizes
1950
self.get_pack_by_name(name)
1951
modified.append(name)
1954
self._names[name] = sizes
1955
self.get_pack_by_name(name)
1957
return removed, added, modified
1959
def _save_pack_names(self, clear_obsolete_packs=False, obsolete_packs=None):
1960
"""Save the list of packs.
1962
This will take out the mutex around the pack names list for the
1963
duration of the method call. If concurrent updates have been made, a
1964
three-way merge between the current list and the current in memory list
1967
:param clear_obsolete_packs: If True, clear out the contents of the
1968
obsolete_packs directory.
1969
:param obsolete_packs: Packs that are obsolete once the new pack-names
1970
file has been written.
1971
:return: A list of the names saved that were not previously on disk.
1973
already_obsolete = []
1976
builder = self._index_builder_class()
1977
(disk_nodes, deleted_nodes, new_nodes,
1978
orig_disk_nodes) = self._diff_pack_names()
1979
# TODO: handle same-name, index-size-changes here -
1980
# e.g. use the value from disk, not ours, *unless* we're the one
1982
for key, value in disk_nodes:
1983
builder.add_node(key, value)
1984
self.transport.put_file('pack-names', builder.finish(),
1985
mode=self.repo.bzrdir._get_file_mode())
1986
self._packs_at_load = disk_nodes
1987
if clear_obsolete_packs:
1990
to_preserve = set([o.name for o in obsolete_packs])
1991
already_obsolete = self._clear_obsolete_packs(to_preserve)
1993
self._unlock_names()
1994
# synchronise the memory packs list with what we just wrote:
1995
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1997
# TODO: We could add one more condition here. "if o.name not in
1998
# orig_disk_nodes and o != the new_pack we haven't written to
1999
# disk yet. However, the new pack object is not easily
2000
# accessible here (it would have to be passed through the
2001
# autopacking code, etc.)
2002
obsolete_packs = [o for o in obsolete_packs
2003
if o.name not in already_obsolete]
2004
self._obsolete_packs(obsolete_packs)
2005
return [new_node[0][0] for new_node in new_nodes]
2007
def reload_pack_names(self):
2008
"""Sync our pack listing with what is present in the repository.
2010
This should be called when we find out that something we thought was
2011
present is now missing. This happens when another process re-packs the
2014
:return: True if the in-memory list of packs has been altered at all.
2016
# The ensure_loaded call is to handle the case where the first call
2017
# made involving the collection was to reload_pack_names, where we
2018
# don't have a view of disk contents. Its a bit of a bandaid, and
2019
# causes two reads of pack-names, but its a rare corner case not struck
2020
# with regular push/pull etc.
2021
first_read = self.ensure_loaded()
2024
# out the new value.
2025
(disk_nodes, deleted_nodes, new_nodes,
2026
orig_disk_nodes) = self._diff_pack_names()
2027
# _packs_at_load is meant to be the explicit list of names in
2028
# 'pack-names' at then start. As such, it should not contain any
2029
# pending names that haven't been written out yet.
2030
self._packs_at_load = orig_disk_nodes
2032
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
2033
if removed or added or modified:
2037
def _restart_autopack(self):
2038
"""Reload the pack names list, and restart the autopack code."""
2039
if not self.reload_pack_names():
2040
# Re-raise the original exception, because something went missing
2041
# and a restart didn't find it
2043
raise errors.RetryAutopack(self.repo, False, sys.exc_info())
2045
def _clear_obsolete_packs(self, preserve=None):
2046
"""Delete everything from the obsolete-packs directory.
2048
:return: A list of pack identifiers (the filename without '.pack') that
2049
were found in obsolete_packs.
2052
obsolete_pack_transport = self.transport.clone('obsolete_packs')
2053
if preserve is None:
2055
for filename in obsolete_pack_transport.list_dir('.'):
2056
name, ext = osutils.splitext(filename)
2059
if name in preserve:
2062
obsolete_pack_transport.delete(filename)
2063
except (errors.PathError, errors.TransportError), e:
2064
warning("couldn't delete obsolete pack, skipping it:\n%s"
2068
def _start_write_group(self):
2069
# Do not permit preparation for writing if we're not in a 'write lock'.
2070
if not self.repo.is_write_locked():
2071
raise errors.NotWriteLocked(self)
2072
self._new_pack = self.pack_factory(self, upload_suffix='.pack',
2073
file_mode=self.repo.bzrdir._get_file_mode())
2074
# allow writing: queue writes to a new index
2075
self.revision_index.add_writable_index(self._new_pack.revision_index,
2077
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
2079
self.text_index.add_writable_index(self._new_pack.text_index,
2081
self._new_pack.text_index.set_optimize(combine_backing_indices=False)
2082
self.signature_index.add_writable_index(self._new_pack.signature_index,
2084
if self.chk_index is not None:
2085
self.chk_index.add_writable_index(self._new_pack.chk_index,
2087
self.repo.chk_bytes._index._add_callback = self.chk_index.add_callback
2088
self._new_pack.chk_index.set_optimize(combine_backing_indices=False)
2090
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
2091
self.repo.revisions._index._add_callback = self.revision_index.add_callback
2092
self.repo.signatures._index._add_callback = self.signature_index.add_callback
2093
self.repo.texts._index._add_callback = self.text_index.add_callback
2095
def _abort_write_group(self):
2096
# FIXME: just drop the transient index.
2097
# forget what names there are
2098
if self._new_pack is not None:
2099
operation = cleanup.OperationWithCleanups(self._new_pack.abort)
2100
operation.add_cleanup(setattr, self, '_new_pack', None)
2101
# If we aborted while in the middle of finishing the write
2102
# group, _remove_pack_indices could fail because the indexes are
2103
# already gone. But they're not there we shouldn't fail in this
2104
# case, so we pass ignore_missing=True.
2105
operation.add_cleanup(self._remove_pack_indices, self._new_pack,
2106
ignore_missing=True)
2107
operation.run_simple()
2108
for resumed_pack in self._resumed_packs:
2109
operation = cleanup.OperationWithCleanups(resumed_pack.abort)
2110
# See comment in previous finally block.
2111
operation.add_cleanup(self._remove_pack_indices, resumed_pack,
2112
ignore_missing=True)
2113
operation.run_simple()
2114
del self._resumed_packs[:]
2116
def _remove_resumed_pack_indices(self):
2117
for resumed_pack in self._resumed_packs:
2118
self._remove_pack_indices(resumed_pack)
2119
del self._resumed_packs[:]
2121
def _check_new_inventories(self):
2122
"""Detect missing inventories in this write group.
2124
:returns: list of strs, summarising any problems found. If the list is
2125
empty no problems were found.
2127
# The base implementation does no checks. GCRepositoryPackCollection
2131
def _commit_write_group(self):
2133
for prefix, versioned_file in (
2134
('revisions', self.repo.revisions),
2135
('inventories', self.repo.inventories),
2136
('texts', self.repo.texts),
2137
('signatures', self.repo.signatures),
2139
missing = versioned_file.get_missing_compression_parent_keys()
2140
all_missing.update([(prefix,) + key for key in missing])
2142
raise errors.BzrCheckError(
2143
"Repository %s has missing compression parent(s) %r "
2144
% (self.repo, sorted(all_missing)))
2145
problems = self._check_new_inventories()
2147
problems_summary = '\n'.join(problems)
2148
raise errors.BzrCheckError(
2149
"Cannot add revision(s) to repository: " + problems_summary)
2150
self._remove_pack_indices(self._new_pack)
2151
any_new_content = False
2152
if self._new_pack.data_inserted():
2153
# get all the data to disk and read to use
2154
self._new_pack.finish()
2155
self.allocate(self._new_pack)
2156
self._new_pack = None
2157
any_new_content = True
2159
self._new_pack.abort()
2160
self._new_pack = None
2161
for resumed_pack in self._resumed_packs:
2162
# XXX: this is a pretty ugly way to turn the resumed pack into a
2163
# properly committed pack.
2164
self._names[resumed_pack.name] = None
2165
self._remove_pack_from_memory(resumed_pack)
2166
resumed_pack.finish()
2167
self.allocate(resumed_pack)
2168
any_new_content = True
2169
del self._resumed_packs[:]
2171
result = self.autopack()
2173
# when autopack takes no steps, the names list is still
2175
return self._save_pack_names()
2179
def _suspend_write_group(self):
2180
tokens = [pack.name for pack in self._resumed_packs]
2181
self._remove_pack_indices(self._new_pack)
2182
if self._new_pack.data_inserted():
2183
# get all the data to disk and read to use
2184
self._new_pack.finish(suspend=True)
2185
tokens.append(self._new_pack.name)
2186
self._new_pack = None
2188
self._new_pack.abort()
2189
self._new_pack = None
2190
self._remove_resumed_pack_indices()
2193
def _resume_write_group(self, tokens):
2194
for token in tokens:
2195
self._resume_pack(token)
2198
class KnitPackRepository(KnitRepository):
2199
"""Repository with knit objects stored inside pack containers.
2201
The layering for a KnitPackRepository is:
2203
Graph | HPSS | Repository public layer |
2204
===================================================
2205
Tuple based apis below, string based, and key based apis above
2206
---------------------------------------------------
2208
Provides .texts, .revisions etc
2209
This adapts the N-tuple keys to physical knit records which only have a
2210
single string identifier (for historical reasons), which in older formats
2211
was always the revision_id, and in the mapped code for packs is always
2212
the last element of key tuples.
2213
---------------------------------------------------
2215
A separate GraphIndex is used for each of the
2216
texts/inventories/revisions/signatures contained within each individual
2217
pack file. The GraphIndex layer works in N-tuples and is unaware of any
2219
===================================================
2223
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
2225
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2226
_commit_builder_class, _serializer)
2227
index_transport = self._transport.clone('indices')
2228
self._pack_collection = RepositoryPackCollection(self, self._transport,
2230
self._transport.clone('upload'),
2231
self._transport.clone('packs'),
2232
_format.index_builder_class,
2233
_format.index_class,
2234
use_chk_index=self._format.supports_chks,
2236
self.inventories = KnitVersionedFiles(
2237
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2238
add_callback=self._pack_collection.inventory_index.add_callback,
2239
deltas=True, parents=True, is_locked=self.is_locked),
2240
data_access=self._pack_collection.inventory_index.data_access,
2241
max_delta_chain=200)
2242
self.revisions = KnitVersionedFiles(
2243
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2244
add_callback=self._pack_collection.revision_index.add_callback,
2245
deltas=False, parents=True, is_locked=self.is_locked,
2246
track_external_parent_refs=True),
2247
data_access=self._pack_collection.revision_index.data_access,
2249
self.signatures = KnitVersionedFiles(
2250
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2251
add_callback=self._pack_collection.signature_index.add_callback,
2252
deltas=False, parents=False, is_locked=self.is_locked),
2253
data_access=self._pack_collection.signature_index.data_access,
2255
self.texts = KnitVersionedFiles(
2256
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2257
add_callback=self._pack_collection.text_index.add_callback,
2258
deltas=True, parents=True, is_locked=self.is_locked),
2259
data_access=self._pack_collection.text_index.data_access,
2260
max_delta_chain=200)
2261
if _format.supports_chks:
2262
# No graph, no compression:- references from chks are between
2263
# different objects not temporal versions of the same; and without
2264
# some sort of temporal structure knit compression will just fail.
2265
self.chk_bytes = KnitVersionedFiles(
2266
_KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2267
add_callback=self._pack_collection.chk_index.add_callback,
2268
deltas=False, parents=False, is_locked=self.is_locked),
2269
data_access=self._pack_collection.chk_index.data_access,
2272
self.chk_bytes = None
2273
# True when the repository object is 'write locked' (as opposed to the
2274
# physical lock only taken out around changes to the pack-names list.)
2275
# Another way to represent this would be a decorator around the control
2276
# files object that presents logical locks as physical ones - if this
2277
# gets ugly consider that alternative design. RBC 20071011
2278
self._write_lock_count = 0
2279
self._transaction = None
2281
self._reconcile_does_inventory_gc = True
2282
self._reconcile_fixes_text_parents = True
2283
self._reconcile_backsup_inventory = False
2285
def _warn_if_deprecated(self, branch=None):
2286
# This class isn't deprecated, but one sub-format is
2287
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2288
super(KnitPackRepository, self)._warn_if_deprecated(branch)
2290
def _abort_write_group(self):
2291
self.revisions._index._key_dependencies.clear()
2292
self._pack_collection._abort_write_group()
2294
def _get_source(self, to_format):
2295
if to_format.network_name() == self._format.network_name():
2296
return KnitPackStreamSource(self, to_format)
2297
return super(KnitPackRepository, self)._get_source(to_format)
2299
def _make_parents_provider(self):
2300
return graph.CachingParentsProvider(self)
2302
def _refresh_data(self):
2303
if not self.is_locked():
2305
self._pack_collection.reload_pack_names()
2307
def _start_write_group(self):
2308
self._pack_collection._start_write_group()
2310
def _commit_write_group(self):
2311
hint = self._pack_collection._commit_write_group()
2312
self.revisions._index._key_dependencies.clear()
2315
def suspend_write_group(self):
2316
# XXX check self._write_group is self.get_transaction()?
2317
tokens = self._pack_collection._suspend_write_group()
2318
self.revisions._index._key_dependencies.clear()
2319
self._write_group = None
2322
def _resume_write_group(self, tokens):
2323
self._start_write_group()
2325
self._pack_collection._resume_write_group(tokens)
2326
except errors.UnresumableWriteGroup:
2327
self._abort_write_group()
2329
for pack in self._pack_collection._resumed_packs:
2330
self.revisions._index.scan_unvalidated_index(pack.revision_index)
2332
def get_transaction(self):
2333
if self._write_lock_count:
2334
return self._transaction
2336
return self.control_files.get_transaction()
2338
def is_locked(self):
2339
return self._write_lock_count or self.control_files.is_locked()
2341
def is_write_locked(self):
2342
return self._write_lock_count
2344
def lock_write(self, token=None):
2345
"""Lock the repository for writes.
2347
:return: A bzrlib.repository.RepositoryWriteLockResult.
2349
locked = self.is_locked()
2350
if not self._write_lock_count and locked:
2351
raise errors.ReadOnlyError(self)
2352
self._write_lock_count += 1
2353
if self._write_lock_count == 1:
2354
self._transaction = transactions.WriteTransaction()
2356
if 'relock' in debug.debug_flags and self._prev_lock == 'w':
2357
note('%r was write locked again', self)
2358
self._prev_lock = 'w'
2359
for repo in self._fallback_repositories:
2360
# Writes don't affect fallback repos
2362
self._refresh_data()
2363
return RepositoryWriteLockResult(self.unlock, None)
2365
def lock_read(self):
2366
"""Lock the repository for reads.
2368
:return: A bzrlib.lock.LogicalLockResult.
2370
locked = self.is_locked()
2371
if self._write_lock_count:
2372
self._write_lock_count += 1
2374
self.control_files.lock_read()
2376
if 'relock' in debug.debug_flags and self._prev_lock == 'r':
2377
note('%r was read locked again', self)
2378
self._prev_lock = 'r'
2379
for repo in self._fallback_repositories:
2381
self._refresh_data()
2382
return LogicalLockResult(self.unlock)
2384
def leave_lock_in_place(self):
2385
# not supported - raise an error
2386
raise NotImplementedError(self.leave_lock_in_place)
2388
def dont_leave_lock_in_place(self):
2389
# not supported - raise an error
2390
raise NotImplementedError(self.dont_leave_lock_in_place)
2393
def pack(self, hint=None, clean_obsolete_packs=False):
2394
"""Compress the data within the repository.
2396
This will pack all the data to a single pack. In future it may
2397
recompress deltas or do other such expensive operations.
2399
self._pack_collection.pack(hint=hint, clean_obsolete_packs=clean_obsolete_packs)
2402
def reconcile(self, other=None, thorough=False):
2403
"""Reconcile this repository."""
2404
from bzrlib.reconcile import PackReconciler
2405
reconciler = PackReconciler(self, thorough=thorough)
2406
reconciler.reconcile()
2409
def _reconcile_pack(self, collection, packs, extension, revs, pb):
2410
packer = ReconcilePacker(collection, packs, extension, revs)
2411
return packer.pack(pb)
2413
@only_raises(errors.LockNotHeld, errors.LockBroken)
2415
if self._write_lock_count == 1 and self._write_group is not None:
2416
self.abort_write_group()
2417
self._transaction = None
2418
self._write_lock_count = 0
2419
raise errors.BzrError(
2420
'Must end write group before releasing write lock on %s'
2422
if self._write_lock_count:
2423
self._write_lock_count -= 1
2424
if not self._write_lock_count:
2425
transaction = self._transaction
2426
self._transaction = None
2427
transaction.finish()
2429
self.control_files.unlock()
2431
if not self.is_locked():
2432
for repo in self._fallback_repositories:
2436
class KnitPackStreamSource(StreamSource):
2437
"""A StreamSource used to transfer data between same-format KnitPack repos.
2439
This source assumes:
2440
1) Same serialization format for all objects
2441
2) Same root information
2442
3) XML format inventories
2443
4) Atomic inserts (so we can stream inventory texts before text
2448
def __init__(self, from_repository, to_format):
2449
super(KnitPackStreamSource, self).__init__(from_repository, to_format)
2450
self._text_keys = None
2451
self._text_fetch_order = 'unordered'
2453
def _get_filtered_inv_stream(self, revision_ids):
2454
from_repo = self.from_repository
2455
parent_ids = from_repo._find_parent_ids_of_revisions(revision_ids)
2456
parent_keys = [(p,) for p in parent_ids]
2457
find_text_keys = from_repo._find_text_key_references_from_xml_inventory_lines
2458
parent_text_keys = set(find_text_keys(
2459
from_repo._inventory_xml_lines_for_keys(parent_keys)))
2460
content_text_keys = set()
2461
knit = KnitVersionedFiles(None, None)
2462
factory = KnitPlainFactory()
2463
def find_text_keys_from_content(record):
2464
if record.storage_kind not in ('knit-delta-gz', 'knit-ft-gz'):
2465
raise ValueError("Unknown content storage kind for"
2466
" inventory text: %s" % (record.storage_kind,))
2467
# It's a knit record, it has a _raw_record field (even if it was
2468
# reconstituted from a network stream).
2469
raw_data = record._raw_record
2470
# read the entire thing
2471
revision_id = record.key[-1]
2472
content, _ = knit._parse_record(revision_id, raw_data)
2473
if record.storage_kind == 'knit-delta-gz':
2474
line_iterator = factory.get_linedelta_content(content)
2475
elif record.storage_kind == 'knit-ft-gz':
2476
line_iterator = factory.get_fulltext_content(content)
2477
content_text_keys.update(find_text_keys(
2478
[(line, revision_id) for line in line_iterator]))
2479
revision_keys = [(r,) for r in revision_ids]
2480
def _filtered_inv_stream():
2481
source_vf = from_repo.inventories
2482
stream = source_vf.get_record_stream(revision_keys,
2484
for record in stream:
2485
if record.storage_kind == 'absent':
2486
raise errors.NoSuchRevision(from_repo, record.key)
2487
find_text_keys_from_content(record)
2489
self._text_keys = content_text_keys - parent_text_keys
2490
return ('inventories', _filtered_inv_stream())
2492
def _get_text_stream(self):
2493
# Note: We know we don't have to handle adding root keys, because both
2494
# the source and target are the identical network name.
2495
text_stream = self.from_repository.texts.get_record_stream(
2496
self._text_keys, self._text_fetch_order, False)
2497
return ('texts', text_stream)
2499
def get_stream(self, search):
2500
revision_ids = search.get_keys()
2501
for stream_info in self._fetch_revision_texts(revision_ids):
2503
self._revision_keys = [(rev_id,) for rev_id in revision_ids]
2504
yield self._get_filtered_inv_stream(revision_ids)
2505
yield self._get_text_stream()
2509
class RepositoryFormatPack(MetaDirRepositoryFormat):
2510
"""Format logic for pack structured repositories.
2512
This repository format has:
2513
- a list of packs in pack-names
2514
- packs in packs/NAME.pack
2515
- indices in indices/NAME.{iix,six,tix,rix}
2516
- knit deltas in the packs, knit indices mapped to the indices.
2517
- thunk objects to support the knits programming API.
2518
- a format marker of its own
2519
- an optional 'shared-storage' flag
2520
- an optional 'no-working-trees' flag
2524
# Set this attribute in derived classes to control the repository class
2525
# created by open and initialize.
2526
repository_class = None
2527
# Set this attribute in derived classes to control the
2528
# _commit_builder_class that the repository objects will have passed to
2529
# their constructor.
2530
_commit_builder_class = None
2531
# Set this attribute in derived clases to control the _serializer that the
2532
# repository objects will have passed to their constructor.
2534
# Packs are not confused by ghosts.
2535
supports_ghosts = True
2536
# External references are not supported in pack repositories yet.
2537
supports_external_lookups = False
2538
# Most pack formats do not use chk lookups.
2539
supports_chks = False
2540
# What index classes to use
2541
index_builder_class = None
2543
_fetch_uses_deltas = True
2546
def initialize(self, a_bzrdir, shared=False):
2547
"""Create a pack based repository.
2549
:param a_bzrdir: bzrdir to contain the new repository; must already
2551
:param shared: If true the repository will be initialized as a shared
2554
mutter('creating repository in %s.', a_bzrdir.transport.base)
2555
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
2556
builder = self.index_builder_class()
2557
files = [('pack-names', builder.finish())]
2558
utf8_files = [('format', self.get_format_string())]
2560
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
2561
repository = self.open(a_bzrdir=a_bzrdir, _found=True)
2562
self._run_post_repo_init_hooks(repository, a_bzrdir, shared)
2565
def open(self, a_bzrdir, _found=False, _override_transport=None):
2566
"""See RepositoryFormat.open().
2568
:param _override_transport: INTERNAL USE ONLY. Allows opening the
2569
repository at a slightly different url
2570
than normal. I.e. during 'upgrade'.
2573
format = RepositoryFormat.find_format(a_bzrdir)
2574
if _override_transport is not None:
2575
repo_transport = _override_transport
2577
repo_transport = a_bzrdir.get_repository_transport(None)
2578
control_files = lockable_files.LockableFiles(repo_transport,
2579
'lock', lockdir.LockDir)
2580
return self.repository_class(_format=self,
2582
control_files=control_files,
2583
_commit_builder_class=self._commit_builder_class,
2584
_serializer=self._serializer)
2587
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2588
"""A no-subtrees parameterized Pack repository.
2590
This format was introduced in 0.92.
2593
repository_class = KnitPackRepository
2594
_commit_builder_class = PackCommitBuilder
2596
def _serializer(self):
2597
return xml5.serializer_v5
2598
# What index classes to use
2599
index_builder_class = InMemoryGraphIndex
2600
index_class = GraphIndex
2602
def _get_matching_bzrdir(self):
2603
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2605
def _ignore_setting_bzrdir(self, format):
2608
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2610
def get_format_string(self):
2611
"""See RepositoryFormat.get_format_string()."""
2612
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2614
def get_format_description(self):
2615
"""See RepositoryFormat.get_format_description()."""
2616
return "Packs containing knits without subtree support"
2619
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2620
"""A subtrees parameterized Pack repository.
2622
This repository format uses the xml7 serializer to get:
2623
- support for recording full info about the tree root
2624
- support for recording tree-references
2626
This format was introduced in 0.92.
2629
repository_class = KnitPackRepository
2630
_commit_builder_class = PackRootCommitBuilder
2631
rich_root_data = True
2633
supports_tree_reference = True
2635
def _serializer(self):
2636
return xml7.serializer_v7
2637
# What index classes to use
2638
index_builder_class = InMemoryGraphIndex
2639
index_class = GraphIndex
2641
def _get_matching_bzrdir(self):
2642
return bzrdir.format_registry.make_bzrdir(
2643
'pack-0.92-subtree')
2645
def _ignore_setting_bzrdir(self, format):
2648
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2650
def get_format_string(self):
2651
"""See RepositoryFormat.get_format_string()."""
2652
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2654
def get_format_description(self):
2655
"""See RepositoryFormat.get_format_description()."""
2656
return "Packs containing knits with subtree support\n"
2659
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2660
"""A rich-root, no subtrees parameterized Pack repository.
2662
This repository format uses the xml6 serializer to get:
2663
- support for recording full info about the tree root
2665
This format was introduced in 1.0.
2668
repository_class = KnitPackRepository
2669
_commit_builder_class = PackRootCommitBuilder
2670
rich_root_data = True
2671
supports_tree_reference = False
2673
def _serializer(self):
2674
return xml6.serializer_v6
2675
# What index classes to use
2676
index_builder_class = InMemoryGraphIndex
2677
index_class = GraphIndex
2679
def _get_matching_bzrdir(self):
2680
return bzrdir.format_registry.make_bzrdir(
2683
def _ignore_setting_bzrdir(self, format):
2686
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2688
def get_format_string(self):
2689
"""See RepositoryFormat.get_format_string()."""
2690
return ("Bazaar pack repository format 1 with rich root"
2691
" (needs bzr 1.0)\n")
2693
def get_format_description(self):
2694
"""See RepositoryFormat.get_format_description()."""
2695
return "Packs containing knits with rich root support\n"
2698
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2699
"""Repository that supports external references to allow stacking.
2703
Supports external lookups, which results in non-truncated ghosts after
2704
reconcile compared to pack-0.92 formats.
2707
repository_class = KnitPackRepository
2708
_commit_builder_class = PackCommitBuilder
2709
supports_external_lookups = True
2710
# What index classes to use
2711
index_builder_class = InMemoryGraphIndex
2712
index_class = GraphIndex
2715
def _serializer(self):
2716
return xml5.serializer_v5
2718
def _get_matching_bzrdir(self):
2719
return bzrdir.format_registry.make_bzrdir('1.6')
2721
def _ignore_setting_bzrdir(self, format):
2724
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2726
def get_format_string(self):
2727
"""See RepositoryFormat.get_format_string()."""
2728
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2730
def get_format_description(self):
2731
"""See RepositoryFormat.get_format_description()."""
2732
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2735
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2736
"""A repository with rich roots and stacking.
2738
New in release 1.6.1.
2740
Supports stacking on other repositories, allowing data to be accessed
2741
without being stored locally.
2744
repository_class = KnitPackRepository
2745
_commit_builder_class = PackRootCommitBuilder
2746
rich_root_data = True
2747
supports_tree_reference = False # no subtrees
2748
supports_external_lookups = True
2749
# What index classes to use
2750
index_builder_class = InMemoryGraphIndex
2751
index_class = GraphIndex
2754
def _serializer(self):
2755
return xml6.serializer_v6
2757
def _get_matching_bzrdir(self):
2758
return bzrdir.format_registry.make_bzrdir(
2761
def _ignore_setting_bzrdir(self, format):
2764
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2766
def get_format_string(self):
2767
"""See RepositoryFormat.get_format_string()."""
2768
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2770
def get_format_description(self):
2771
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2774
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2775
"""A repository with rich roots and external references.
2779
Supports external lookups, which results in non-truncated ghosts after
2780
reconcile compared to pack-0.92 formats.
2782
This format was deprecated because the serializer it uses accidentally
2783
supported subtrees, when the format was not intended to. This meant that
2784
someone could accidentally fetch from an incorrect repository.
2787
repository_class = KnitPackRepository
2788
_commit_builder_class = PackRootCommitBuilder
2789
rich_root_data = True
2790
supports_tree_reference = False # no subtrees
2792
supports_external_lookups = True
2793
# What index classes to use
2794
index_builder_class = InMemoryGraphIndex
2795
index_class = GraphIndex
2798
def _serializer(self):
2799
return xml7.serializer_v7
2801
def _get_matching_bzrdir(self):
2802
matching = bzrdir.format_registry.make_bzrdir(
2804
matching.repository_format = self
2807
def _ignore_setting_bzrdir(self, format):
2810
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2812
def get_format_string(self):
2813
"""See RepositoryFormat.get_format_string()."""
2814
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2816
def get_format_description(self):
2817
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2821
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2822
"""A repository with stacking and btree indexes,
2823
without rich roots or subtrees.
2825
This is equivalent to pack-1.6 with B+Tree indices.
2828
repository_class = KnitPackRepository
2829
_commit_builder_class = PackCommitBuilder
2830
supports_external_lookups = True
2831
# What index classes to use
2832
index_builder_class = BTreeBuilder
2833
index_class = BTreeGraphIndex
2836
def _serializer(self):
2837
return xml5.serializer_v5
2839
def _get_matching_bzrdir(self):
2840
return bzrdir.format_registry.make_bzrdir('1.9')
2842
def _ignore_setting_bzrdir(self, format):
2845
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2847
def get_format_string(self):
2848
"""See RepositoryFormat.get_format_string()."""
2849
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2851
def get_format_description(self):
2852
"""See RepositoryFormat.get_format_description()."""
2853
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2856
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2857
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2859
1.6-rich-root with B+Tree indices.
2862
repository_class = KnitPackRepository
2863
_commit_builder_class = PackRootCommitBuilder
2864
rich_root_data = True
2865
supports_tree_reference = False # no subtrees
2866
supports_external_lookups = True
2867
# What index classes to use
2868
index_builder_class = BTreeBuilder
2869
index_class = BTreeGraphIndex
2872
def _serializer(self):
2873
return xml6.serializer_v6
2875
def _get_matching_bzrdir(self):
2876
return bzrdir.format_registry.make_bzrdir(
2879
def _ignore_setting_bzrdir(self, format):
2882
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2884
def get_format_string(self):
2885
"""See RepositoryFormat.get_format_string()."""
2886
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2888
def get_format_description(self):
2889
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2892
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2893
"""A subtrees development repository.
2895
This format should be retained until the second release after bzr 1.7.
2897
1.6.1-subtree[as it might have been] with B+Tree indices.
2899
This is [now] retained until we have a CHK based subtree format in
2903
repository_class = KnitPackRepository
2904
_commit_builder_class = PackRootCommitBuilder
2905
rich_root_data = True
2907
supports_tree_reference = True
2908
supports_external_lookups = True
2909
# What index classes to use
2910
index_builder_class = BTreeBuilder
2911
index_class = BTreeGraphIndex
2914
def _serializer(self):
2915
return xml7.serializer_v7
2917
def _get_matching_bzrdir(self):
2918
return bzrdir.format_registry.make_bzrdir(
2919
'development-subtree')
2921
def _ignore_setting_bzrdir(self, format):
2924
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2926
def get_format_string(self):
2927
"""See RepositoryFormat.get_format_string()."""
2928
return ("Bazaar development format 2 with subtree support "
2929
"(needs bzr.dev from before 1.8)\n")
2931
def get_format_description(self):
2932
"""See RepositoryFormat.get_format_description()."""
2933
return ("Development repository format, currently the same as "
2934
"1.6.1-subtree with B+Tree indices.\n")