1
# Copyright (C) 2007-2010 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20
from bzrlib.lazy_import import lazy_import
21
lazy_import(globals(), """
22
from itertools import izip
38
from bzrlib.index import (
40
GraphIndexPrefixAdapter,
42
from bzrlib.knit import (
48
from bzrlib import tsort
55
revision as _mod_revision,
58
from bzrlib.decorators import needs_write_lock, only_raises
59
from bzrlib.btree_index import (
63
from bzrlib.index import (
67
from bzrlib.repofmt.knitrepo import KnitRepository
68
from bzrlib.repository import (
70
MetaDirRepositoryFormat,
75
from bzrlib.trace import (
82
class PackCommitBuilder(CommitBuilder):
83
"""A subclass of CommitBuilder to add texts with pack semantics.
85
Specifically this uses one knit object rather than one knit object per
86
added text, reducing memory and object pressure.
89
def __init__(self, repository, parents, config, timestamp=None,
90
timezone=None, committer=None, revprops=None,
92
CommitBuilder.__init__(self, repository, parents, config,
93
timestamp=timestamp, timezone=timezone, committer=committer,
94
revprops=revprops, revision_id=revision_id)
95
self._file_graph = graph.Graph(
96
repository._pack_collection.text_index.combined_index)
98
def _heads(self, file_id, revision_ids):
99
keys = [(file_id, revision_id) for revision_id in revision_ids]
100
return set([key[1] for key in self._file_graph.heads(keys)])
103
class PackRootCommitBuilder(RootCommitBuilder):
104
"""A subclass of RootCommitBuilder to add texts with pack semantics.
106
Specifically this uses one knit object rather than one knit object per
107
added text, reducing memory and object pressure.
110
def __init__(self, repository, parents, config, timestamp=None,
111
timezone=None, committer=None, revprops=None,
113
CommitBuilder.__init__(self, repository, parents, config,
114
timestamp=timestamp, timezone=timezone, committer=committer,
115
revprops=revprops, revision_id=revision_id)
116
self._file_graph = graph.Graph(
117
repository._pack_collection.text_index.combined_index)
119
def _heads(self, file_id, revision_ids):
120
keys = [(file_id, revision_id) for revision_id in revision_ids]
121
return set([key[1] for key in self._file_graph.heads(keys)])
125
"""An in memory proxy for a pack and its indices.
127
This is a base class that is not directly used, instead the classes
128
ExistingPack and NewPack are used.
131
# A map of index 'type' to the file extension and position in the
133
index_definitions = {
135
'revision': ('.rix', 0),
136
'inventory': ('.iix', 1),
138
'signature': ('.six', 3),
141
def __init__(self, revision_index, inventory_index, text_index,
142
signature_index, chk_index=None):
143
"""Create a pack instance.
145
:param revision_index: A GraphIndex for determining what revisions are
146
present in the Pack and accessing the locations of their texts.
147
:param inventory_index: A GraphIndex for determining what inventories are
148
present in the Pack and accessing the locations of their
150
:param text_index: A GraphIndex for determining what file texts
151
are present in the pack and accessing the locations of their
152
texts/deltas (via (fileid, revisionid) tuples).
153
:param signature_index: A GraphIndex for determining what signatures are
154
present in the Pack and accessing the locations of their texts.
155
:param chk_index: A GraphIndex for accessing content by CHK, if the
158
self.revision_index = revision_index
159
self.inventory_index = inventory_index
160
self.text_index = text_index
161
self.signature_index = signature_index
162
self.chk_index = chk_index
164
def access_tuple(self):
165
"""Return a tuple (transport, name) for the pack content."""
166
return self.pack_transport, self.file_name()
168
def _check_references(self):
169
"""Make sure our external references are present.
171
Packs are allowed to have deltas whose base is not in the pack, but it
172
must be present somewhere in this collection. It is not allowed to
173
have deltas based on a fallback repository.
174
(See <https://bugs.launchpad.net/bzr/+bug/288751>)
177
for (index_name, external_refs, index) in [
179
self._get_external_refs(self.text_index),
180
self._pack_collection.text_index.combined_index),
182
self._get_external_refs(self.inventory_index),
183
self._pack_collection.inventory_index.combined_index),
185
missing = external_refs.difference(
186
k for (idx, k, v, r) in
187
index.iter_entries(external_refs))
189
missing_items[index_name] = sorted(list(missing))
191
from pprint import pformat
192
raise errors.BzrCheckError(
193
"Newly created pack file %r has delta references to "
194
"items not in its repository:\n%s"
195
% (self, pformat(missing_items)))
198
"""Get the file name for the pack on disk."""
199
return self.name + '.pack'
201
def get_revision_count(self):
202
return self.revision_index.key_count()
204
def index_name(self, index_type, name):
205
"""Get the disk name of an index type for pack name 'name'."""
206
return name + Pack.index_definitions[index_type][0]
208
def index_offset(self, index_type):
209
"""Get the position in a index_size array for a given index type."""
210
return Pack.index_definitions[index_type][1]
212
def inventory_index_name(self, name):
213
"""The inv index is the name + .iix."""
214
return self.index_name('inventory', name)
216
def revision_index_name(self, name):
217
"""The revision index is the name + .rix."""
218
return self.index_name('revision', name)
220
def signature_index_name(self, name):
221
"""The signature index is the name + .six."""
222
return self.index_name('signature', name)
224
def text_index_name(self, name):
225
"""The text index is the name + .tix."""
226
return self.index_name('text', name)
228
def _replace_index_with_readonly(self, index_type):
229
unlimited_cache = False
230
if index_type == 'chk':
231
unlimited_cache = True
232
setattr(self, index_type + '_index',
233
self.index_class(self.index_transport,
234
self.index_name(index_type, self.name),
235
self.index_sizes[self.index_offset(index_type)],
236
unlimited_cache=unlimited_cache))
239
class ExistingPack(Pack):
240
"""An in memory proxy for an existing .pack and its disk indices."""
242
def __init__(self, pack_transport, name, revision_index, inventory_index,
243
text_index, signature_index, chk_index=None):
244
"""Create an ExistingPack object.
246
:param pack_transport: The transport where the pack file resides.
247
:param name: The name of the pack on disk in the pack_transport.
249
Pack.__init__(self, revision_index, inventory_index, text_index,
250
signature_index, chk_index)
252
self.pack_transport = pack_transport
253
if None in (revision_index, inventory_index, text_index,
254
signature_index, name, pack_transport):
255
raise AssertionError()
257
def __eq__(self, other):
258
return self.__dict__ == other.__dict__
260
def __ne__(self, other):
261
return not self.__eq__(other)
264
return "<%s.%s object at 0x%x, %s, %s" % (
265
self.__class__.__module__, self.__class__.__name__, id(self),
266
self.pack_transport, self.name)
269
class ResumedPack(ExistingPack):
271
def __init__(self, name, revision_index, inventory_index, text_index,
272
signature_index, upload_transport, pack_transport, index_transport,
273
pack_collection, chk_index=None):
274
"""Create a ResumedPack object."""
275
ExistingPack.__init__(self, pack_transport, name, revision_index,
276
inventory_index, text_index, signature_index,
278
self.upload_transport = upload_transport
279
self.index_transport = index_transport
280
self.index_sizes = [None, None, None, None]
282
('revision', revision_index),
283
('inventory', inventory_index),
284
('text', text_index),
285
('signature', signature_index),
287
if chk_index is not None:
288
indices.append(('chk', chk_index))
289
self.index_sizes.append(None)
290
for index_type, index in indices:
291
offset = self.index_offset(index_type)
292
self.index_sizes[offset] = index._size
293
self.index_class = pack_collection._index_class
294
self._pack_collection = pack_collection
295
self._state = 'resumed'
296
# XXX: perhaps check that the .pack file exists?
298
def access_tuple(self):
299
if self._state == 'finished':
300
return Pack.access_tuple(self)
301
elif self._state == 'resumed':
302
return self.upload_transport, self.file_name()
304
raise AssertionError(self._state)
307
self.upload_transport.delete(self.file_name())
308
indices = [self.revision_index, self.inventory_index, self.text_index,
309
self.signature_index]
310
if self.chk_index is not None:
311
indices.append(self.chk_index)
312
for index in indices:
313
index._transport.delete(index._name)
316
self._check_references()
317
index_types = ['revision', 'inventory', 'text', 'signature']
318
if self.chk_index is not None:
319
index_types.append('chk')
320
for index_type in index_types:
321
old_name = self.index_name(index_type, self.name)
322
new_name = '../indices/' + old_name
323
self.upload_transport.rename(old_name, new_name)
324
self._replace_index_with_readonly(index_type)
325
new_name = '../packs/' + self.file_name()
326
self.upload_transport.rename(self.file_name(), new_name)
327
self._state = 'finished'
329
def _get_external_refs(self, index):
330
"""Return compression parents for this index that are not present.
332
This returns any compression parents that are referenced by this index,
333
which are not contained *in* this index. They may be present elsewhere.
335
return index.external_references(1)
339
"""An in memory proxy for a pack which is being created."""
341
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
342
"""Create a NewPack instance.
344
:param pack_collection: A PackCollection into which this is being inserted.
345
:param upload_suffix: An optional suffix to be given to any temporary
346
files created during the pack creation. e.g '.autopack'
347
:param file_mode: Unix permissions for newly created file.
349
# The relative locations of the packs are constrained, but all are
350
# passed in because the caller has them, so as to avoid object churn.
351
index_builder_class = pack_collection._index_builder_class
352
if pack_collection.chk_index is not None:
353
chk_index = index_builder_class(reference_lists=0)
357
# Revisions: parents list, no text compression.
358
index_builder_class(reference_lists=1),
359
# Inventory: We want to map compression only, but currently the
360
# knit code hasn't been updated enough to understand that, so we
361
# have a regular 2-list index giving parents and compression
363
index_builder_class(reference_lists=2),
364
# Texts: compression and per file graph, for all fileids - so two
365
# reference lists and two elements in the key tuple.
366
index_builder_class(reference_lists=2, key_elements=2),
367
# Signatures: Just blobs to store, no compression, no parents
369
index_builder_class(reference_lists=0),
370
# CHK based storage - just blobs, no compression or parents.
373
self._pack_collection = pack_collection
374
# When we make readonly indices, we need this.
375
self.index_class = pack_collection._index_class
376
# where should the new pack be opened
377
self.upload_transport = pack_collection._upload_transport
378
# where are indices written out to
379
self.index_transport = pack_collection._index_transport
380
# where is the pack renamed to when it is finished?
381
self.pack_transport = pack_collection._pack_transport
382
# What file mode to upload the pack and indices with.
383
self._file_mode = file_mode
384
# tracks the content written to the .pack file.
385
self._hash = osutils.md5()
386
# a tuple with the length in bytes of the indices, once the pack
387
# is finalised. (rev, inv, text, sigs, chk_if_in_use)
388
self.index_sizes = None
389
# How much data to cache when writing packs. Note that this is not
390
# synchronised with reads, because it's not in the transport layer, so
391
# is not safe unless the client knows it won't be reading from the pack
393
self._cache_limit = 0
394
# the temporary pack file name.
395
self.random_name = osutils.rand_chars(20) + upload_suffix
396
# when was this pack started ?
397
self.start_time = time.time()
398
# open an output stream for the data added to the pack.
399
self.write_stream = self.upload_transport.open_write_stream(
400
self.random_name, mode=self._file_mode)
401
if 'pack' in debug.debug_flags:
402
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
403
time.ctime(), self.upload_transport.base, self.random_name,
404
time.time() - self.start_time)
405
# A list of byte sequences to be written to the new pack, and the
406
# aggregate size of them. Stored as a list rather than separate
407
# variables so that the _write_data closure below can update them.
408
self._buffer = [[], 0]
409
# create a callable for adding data
411
# robertc says- this is a closure rather than a method on the object
412
# so that the variables are locals, and faster than accessing object
414
def _write_data(bytes, flush=False, _buffer=self._buffer,
415
_write=self.write_stream.write, _update=self._hash.update):
416
_buffer[0].append(bytes)
417
_buffer[1] += len(bytes)
419
if _buffer[1] > self._cache_limit or flush:
420
bytes = ''.join(_buffer[0])
424
# expose this on self, for the occasion when clients want to add data.
425
self._write_data = _write_data
426
# a pack writer object to serialise pack records.
427
self._writer = pack.ContainerWriter(self._write_data)
429
# what state is the pack in? (open, finished, aborted)
431
# no name until we finish writing the content
435
"""Cancel creating this pack."""
436
self._state = 'aborted'
437
self.write_stream.close()
438
# Remove the temporary pack file.
439
self.upload_transport.delete(self.random_name)
440
# The indices have no state on disk.
442
def access_tuple(self):
443
"""Return a tuple (transport, name) for the pack content."""
444
if self._state == 'finished':
445
return Pack.access_tuple(self)
446
elif self._state == 'open':
447
return self.upload_transport, self.random_name
449
raise AssertionError(self._state)
451
def data_inserted(self):
452
"""True if data has been added to this pack."""
453
return bool(self.get_revision_count() or
454
self.inventory_index.key_count() or
455
self.text_index.key_count() or
456
self.signature_index.key_count() or
457
(self.chk_index is not None and self.chk_index.key_count()))
459
def finish_content(self):
460
if self.name is not None:
464
self._write_data('', flush=True)
465
self.name = self._hash.hexdigest()
467
def finish(self, suspend=False):
468
"""Finish the new pack.
471
- finalises the content
472
- assigns a name (the md5 of the content, currently)
473
- writes out the associated indices
474
- renames the pack into place.
475
- stores the index size tuple for the pack in the index_sizes
478
self.finish_content()
480
self._check_references()
482
# XXX: It'd be better to write them all to temporary names, then
483
# rename them all into place, so that the window when only some are
484
# visible is smaller. On the other hand none will be seen until
485
# they're in the names list.
486
self.index_sizes = [None, None, None, None]
487
self._write_index('revision', self.revision_index, 'revision', suspend)
488
self._write_index('inventory', self.inventory_index, 'inventory',
490
self._write_index('text', self.text_index, 'file texts', suspend)
491
self._write_index('signature', self.signature_index,
492
'revision signatures', suspend)
493
if self.chk_index is not None:
494
self.index_sizes.append(None)
495
self._write_index('chk', self.chk_index,
496
'content hash bytes', suspend)
497
self.write_stream.close()
498
# Note that this will clobber an existing pack with the same name,
499
# without checking for hash collisions. While this is undesirable this
500
# is something that can be rectified in a subsequent release. One way
501
# to rectify it may be to leave the pack at the original name, writing
502
# its pack-names entry as something like 'HASH: index-sizes
503
# temporary-name'. Allocate that and check for collisions, if it is
504
# collision free then rename it into place. If clients know this scheme
505
# they can handle missing-file errors by:
506
# - try for HASH.pack
507
# - try for temporary-name
508
# - refresh the pack-list to see if the pack is now absent
509
new_name = self.name + '.pack'
511
new_name = '../packs/' + new_name
512
self.upload_transport.rename(self.random_name, new_name)
513
self._state = 'finished'
514
if 'pack' in debug.debug_flags:
515
# XXX: size might be interesting?
516
mutter('%s: create_pack: pack finished: %s%s->%s t+%6.3fs',
517
time.ctime(), self.upload_transport.base, self.random_name,
518
new_name, time.time() - self.start_time)
521
"""Flush any current data."""
523
bytes = ''.join(self._buffer[0])
524
self.write_stream.write(bytes)
525
self._hash.update(bytes)
526
self._buffer[:] = [[], 0]
528
def _get_external_refs(self, index):
529
return index._external_references()
531
def set_write_cache_size(self, size):
532
self._cache_limit = size
534
def _write_index(self, index_type, index, label, suspend=False):
535
"""Write out an index.
537
:param index_type: The type of index to write - e.g. 'revision'.
538
:param index: The index object to serialise.
539
:param label: What label to give the index e.g. 'revision'.
541
index_name = self.index_name(index_type, self.name)
543
transport = self.upload_transport
545
transport = self.index_transport
546
self.index_sizes[self.index_offset(index_type)] = transport.put_file(
547
index_name, index.finish(), mode=self._file_mode)
548
if 'pack' in debug.debug_flags:
549
# XXX: size might be interesting?
550
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
551
time.ctime(), label, self.upload_transport.base,
552
self.random_name, time.time() - self.start_time)
553
# Replace the writable index on this object with a readonly,
554
# presently unloaded index. We should alter
555
# the index layer to make its finish() error if add_node is
556
# subsequently used. RBC
557
self._replace_index_with_readonly(index_type)
560
class AggregateIndex(object):
561
"""An aggregated index for the RepositoryPackCollection.
563
AggregateIndex is reponsible for managing the PackAccess object,
564
Index-To-Pack mapping, and all indices list for a specific type of index
565
such as 'revision index'.
567
A CombinedIndex provides an index on a single key space built up
568
from several on-disk indices. The AggregateIndex builds on this
569
to provide a knit access layer, and allows having up to one writable
570
index within the collection.
572
# XXX: Probably 'can be written to' could/should be separated from 'acts
573
# like a knit index' -- mbp 20071024
575
def __init__(self, reload_func=None, flush_func=None):
576
"""Create an AggregateIndex.
578
:param reload_func: A function to call if we find we are missing an
579
index. Should have the form reload_func() => True if the list of
580
active pack files has changed.
582
self._reload_func = reload_func
583
self.index_to_pack = {}
584
self.combined_index = CombinedGraphIndex([], reload_func=reload_func)
585
self.data_access = _DirectPackAccess(self.index_to_pack,
586
reload_func=reload_func,
587
flush_func=flush_func)
588
self.add_callback = None
590
def replace_indices(self, index_to_pack, indices):
591
"""Replace the current mappings with fresh ones.
593
This should probably not be used eventually, rather incremental add and
594
removal of indices. It has been added during refactoring of existing
597
:param index_to_pack: A mapping from index objects to
598
(transport, name) tuples for the pack file data.
599
:param indices: A list of indices.
601
# refresh the revision pack map dict without replacing the instance.
602
self.index_to_pack.clear()
603
self.index_to_pack.update(index_to_pack)
604
# XXX: API break - clearly a 'replace' method would be good?
605
self.combined_index._indices[:] = indices
606
# the current add nodes callback for the current writable index if
608
self.add_callback = None
610
def add_index(self, index, pack):
611
"""Add index to the aggregate, which is an index for Pack pack.
613
Future searches on the aggregate index will seach this new index
614
before all previously inserted indices.
616
:param index: An Index for the pack.
617
:param pack: A Pack instance.
619
# expose it to the index map
620
self.index_to_pack[index] = pack.access_tuple()
621
# put it at the front of the linear index list
622
self.combined_index.insert_index(0, index)
624
def add_writable_index(self, index, pack):
625
"""Add an index which is able to have data added to it.
627
There can be at most one writable index at any time. Any
628
modifications made to the knit are put into this index.
630
:param index: An index from the pack parameter.
631
:param pack: A Pack instance.
633
if self.add_callback is not None:
634
raise AssertionError(
635
"%s already has a writable index through %s" % \
636
(self, self.add_callback))
637
# allow writing: queue writes to a new index
638
self.add_index(index, pack)
639
# Updates the index to packs mapping as a side effect,
640
self.data_access.set_writer(pack._writer, index, pack.access_tuple())
641
self.add_callback = index.add_nodes
644
"""Reset all the aggregate data to nothing."""
645
self.data_access.set_writer(None, None, (None, None))
646
self.index_to_pack.clear()
647
del self.combined_index._indices[:]
648
self.add_callback = None
650
def remove_index(self, index):
651
"""Remove index from the indices used to answer queries.
653
:param index: An index from the pack parameter.
655
del self.index_to_pack[index]
656
self.combined_index._indices.remove(index)
657
if (self.add_callback is not None and
658
getattr(index, 'add_nodes', None) == self.add_callback):
659
self.add_callback = None
660
self.data_access.set_writer(None, None, (None, None))
663
class Packer(object):
664
"""Create a pack from packs."""
666
def __init__(self, pack_collection, packs, suffix, revision_ids=None,
670
:param pack_collection: A RepositoryPackCollection object where the
671
new pack is being written to.
672
:param packs: The packs to combine.
673
:param suffix: The suffix to use on the temporary files for the pack.
674
:param revision_ids: Revision ids to limit the pack to.
675
:param reload_func: A function to call if a pack file/index goes
676
missing. The side effect of calling this function should be to
677
update self.packs. See also AggregateIndex
681
self.revision_ids = revision_ids
682
# The pack object we are creating.
684
self._pack_collection = pack_collection
685
self._reload_func = reload_func
686
# The index layer keys for the revisions being copied. None for 'all
688
self._revision_keys = None
689
# What text keys to copy. None for 'all texts'. This is set by
690
# _copy_inventory_texts
691
self._text_filter = None
694
def _extra_init(self):
695
"""A template hook to allow extending the constructor trivially."""
697
def _pack_map_and_index_list(self, index_attribute):
698
"""Convert a list of packs to an index pack map and index list.
700
:param index_attribute: The attribute that the desired index is found
702
:return: A tuple (map, list) where map contains the dict from
703
index:pack_tuple, and list contains the indices in the preferred
708
for pack_obj in self.packs:
709
index = getattr(pack_obj, index_attribute)
710
indices.append(index)
711
pack_map[index] = pack_obj
712
return pack_map, indices
714
def _index_contents(self, indices, key_filter=None):
715
"""Get an iterable of the index contents from a pack_map.
717
:param indices: The list of indices to query
718
:param key_filter: An optional filter to limit the keys returned.
720
all_index = CombinedGraphIndex(indices)
721
if key_filter is None:
722
return all_index.iter_all_entries()
724
return all_index.iter_entries(key_filter)
726
def pack(self, pb=None):
727
"""Create a new pack by reading data from other packs.
729
This does little more than a bulk copy of data. One key difference
730
is that data with the same item key across multiple packs is elided
731
from the output. The new pack is written into the current pack store
732
along with its indices, and the name added to the pack names. The
733
source packs are not altered and are not required to be in the current
736
:param pb: An optional progress bar to use. A nested bar is created if
738
:return: A Pack object, or None if nothing was copied.
740
# open a pack - using the same name as the last temporary file
741
# - which has already been flushed, so its safe.
742
# XXX: - duplicate code warning with start_write_group; fix before
743
# considering 'done'.
744
if self._pack_collection._new_pack is not None:
745
raise errors.BzrError('call to %s.pack() while another pack is'
747
% (self.__class__.__name__,))
748
if self.revision_ids is not None:
749
if len(self.revision_ids) == 0:
750
# silly fetch request.
753
self.revision_ids = frozenset(self.revision_ids)
754
self.revision_keys = frozenset((revid,) for revid in
757
self.pb = ui.ui_factory.nested_progress_bar()
761
return self._create_pack_from_packs()
767
"""Open a pack for the pack we are creating."""
768
new_pack = self._pack_collection.pack_factory(self._pack_collection,
769
upload_suffix=self.suffix,
770
file_mode=self._pack_collection.repo.bzrdir._get_file_mode())
771
# We know that we will process all nodes in order, and don't need to
772
# query, so don't combine any indices spilled to disk until we are done
773
new_pack.revision_index.set_optimize(combine_backing_indices=False)
774
new_pack.inventory_index.set_optimize(combine_backing_indices=False)
775
new_pack.text_index.set_optimize(combine_backing_indices=False)
776
new_pack.signature_index.set_optimize(combine_backing_indices=False)
779
def _update_pack_order(self, entries, index_to_pack_map):
780
"""Determine how we want our packs to be ordered.
782
This changes the sort order of the self.packs list so that packs unused
783
by 'entries' will be at the end of the list, so that future requests
784
can avoid probing them. Used packs will be at the front of the
785
self.packs list, in the order of their first use in 'entries'.
787
:param entries: A list of (index, ...) tuples
788
:param index_to_pack_map: A mapping from index objects to pack objects.
792
for entry in entries:
794
if index not in seen_indexes:
795
packs.append(index_to_pack_map[index])
796
seen_indexes.add(index)
797
if len(packs) == len(self.packs):
798
if 'pack' in debug.debug_flags:
799
mutter('Not changing pack list, all packs used.')
801
seen_packs = set(packs)
802
for pack in self.packs:
803
if pack not in seen_packs:
806
if 'pack' in debug.debug_flags:
807
old_names = [p.access_tuple()[1] for p in self.packs]
808
new_names = [p.access_tuple()[1] for p in packs]
809
mutter('Reordering packs\nfrom: %s\n to: %s',
810
old_names, new_names)
813
def _copy_revision_texts(self):
814
"""Copy revision data to the new pack."""
816
if self.revision_ids:
817
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
820
# select revision keys
821
revision_index_map, revision_indices = self._pack_map_and_index_list(
823
revision_nodes = self._index_contents(revision_indices, revision_keys)
824
revision_nodes = list(revision_nodes)
825
self._update_pack_order(revision_nodes, revision_index_map)
826
# copy revision keys and adjust values
827
self.pb.update("Copying revision texts", 1)
828
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
829
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
830
self.new_pack.revision_index, readv_group_iter, total_items))
831
if 'pack' in debug.debug_flags:
832
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
833
time.ctime(), self._pack_collection._upload_transport.base,
834
self.new_pack.random_name,
835
self.new_pack.revision_index.key_count(),
836
time.time() - self.new_pack.start_time)
837
self._revision_keys = revision_keys
839
def _copy_inventory_texts(self):
840
"""Copy the inventory texts to the new pack.
842
self._revision_keys is used to determine what inventories to copy.
844
Sets self._text_filter appropriately.
846
# select inventory keys
847
inv_keys = self._revision_keys # currently the same keyspace, and note that
848
# querying for keys here could introduce a bug where an inventory item
849
# is missed, so do not change it to query separately without cross
850
# checking like the text key check below.
851
inventory_index_map, inventory_indices = self._pack_map_and_index_list(
853
inv_nodes = self._index_contents(inventory_indices, inv_keys)
854
# copy inventory keys and adjust values
855
# XXX: Should be a helper function to allow different inv representation
857
self.pb.update("Copying inventory texts", 2)
858
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
859
# Only grab the output lines if we will be processing them
860
output_lines = bool(self.revision_ids)
861
inv_lines = self._copy_nodes_graph(inventory_index_map,
862
self.new_pack._writer, self.new_pack.inventory_index,
863
readv_group_iter, total_items, output_lines=output_lines)
864
if self.revision_ids:
865
self._process_inventory_lines(inv_lines)
867
# eat the iterator to cause it to execute.
869
self._text_filter = None
870
if 'pack' in debug.debug_flags:
871
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
872
time.ctime(), self._pack_collection._upload_transport.base,
873
self.new_pack.random_name,
874
self.new_pack.inventory_index.key_count(),
875
time.time() - self.new_pack.start_time)
877
def _copy_text_texts(self):
879
text_index_map, text_nodes = self._get_text_nodes()
880
if self._text_filter is not None:
881
# We could return the keys copied as part of the return value from
882
# _copy_nodes_graph but this doesn't work all that well with the
883
# need to get line output too, so we check separately, and as we're
884
# going to buffer everything anyway, we check beforehand, which
885
# saves reading knit data over the wire when we know there are
887
text_nodes = set(text_nodes)
888
present_text_keys = set(_node[1] for _node in text_nodes)
889
missing_text_keys = set(self._text_filter) - present_text_keys
890
if missing_text_keys:
891
# TODO: raise a specific error that can handle many missing
893
mutter("missing keys during fetch: %r", missing_text_keys)
894
a_missing_key = missing_text_keys.pop()
895
raise errors.RevisionNotPresent(a_missing_key[1],
897
# copy text keys and adjust values
898
self.pb.update("Copying content texts", 3)
899
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
900
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
901
self.new_pack.text_index, readv_group_iter, total_items))
902
self._log_copied_texts()
904
def _create_pack_from_packs(self):
905
self.pb.update("Opening pack", 0, 5)
906
self.new_pack = self.open_pack()
907
new_pack = self.new_pack
908
# buffer data - we won't be reading-back during the pack creation and
909
# this makes a significant difference on sftp pushes.
910
new_pack.set_write_cache_size(1024*1024)
911
if 'pack' in debug.debug_flags:
912
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
913
for a_pack in self.packs]
914
if self.revision_ids is not None:
915
rev_count = len(self.revision_ids)
918
mutter('%s: create_pack: creating pack from source packs: '
919
'%s%s %s revisions wanted %s t=0',
920
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
921
plain_pack_list, rev_count)
922
self._copy_revision_texts()
923
self._copy_inventory_texts()
924
self._copy_text_texts()
925
# select signature keys
926
signature_filter = self._revision_keys # same keyspace
927
signature_index_map, signature_indices = self._pack_map_and_index_list(
929
signature_nodes = self._index_contents(signature_indices,
931
# copy signature keys and adjust values
932
self.pb.update("Copying signature texts", 4)
933
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
934
new_pack.signature_index)
935
if 'pack' in debug.debug_flags:
936
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
937
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
938
new_pack.signature_index.key_count(),
939
time.time() - new_pack.start_time)
941
# NB XXX: how to check CHK references are present? perhaps by yielding
942
# the items? How should that interact with stacked repos?
943
if new_pack.chk_index is not None:
945
if 'pack' in debug.debug_flags:
946
mutter('%s: create_pack: chk content copied: %s%s %d items t+%6.3fs',
947
time.ctime(), self._pack_collection._upload_transport.base,
948
new_pack.random_name,
949
new_pack.chk_index.key_count(),
950
time.time() - new_pack.start_time)
951
new_pack._check_references()
952
if not self._use_pack(new_pack):
955
self.pb.update("Finishing pack", 5)
957
self._pack_collection.allocate(new_pack)
960
def _copy_chks(self, refs=None):
961
# XXX: Todo, recursive follow-pointers facility when fetching some
963
chk_index_map, chk_indices = self._pack_map_and_index_list(
965
chk_nodes = self._index_contents(chk_indices, refs)
967
# TODO: This isn't strictly tasteful as we are accessing some private
968
# variables (_serializer). Perhaps a better way would be to have
969
# Repository._deserialise_chk_node()
970
search_key_func = chk_map.search_key_registry.get(
971
self._pack_collection.repo._serializer.search_key_name)
972
def accumlate_refs(lines):
973
# XXX: move to a generic location
975
bytes = ''.join(lines)
976
node = chk_map._deserialise(bytes, ("unknown",), search_key_func)
977
new_refs.update(node.refs())
978
self._copy_nodes(chk_nodes, chk_index_map, self.new_pack._writer,
979
self.new_pack.chk_index, output_lines=accumlate_refs)
982
def _copy_nodes(self, nodes, index_map, writer, write_index,
984
"""Copy knit nodes between packs with no graph references.
986
:param output_lines: Output full texts of copied items.
988
pb = ui.ui_factory.nested_progress_bar()
990
return self._do_copy_nodes(nodes, index_map, writer,
991
write_index, pb, output_lines=output_lines)
995
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb,
997
# for record verification
998
knit = KnitVersionedFiles(None, None)
999
# plan a readv on each source pack:
1001
nodes = sorted(nodes)
1002
# how to map this into knit.py - or knit.py into this?
1003
# we don't want the typical knit logic, we want grouping by pack
1004
# at this point - perhaps a helper library for the following code
1005
# duplication points?
1007
for index, key, value in nodes:
1008
if index not in request_groups:
1009
request_groups[index] = []
1010
request_groups[index].append((key, value))
1012
pb.update("Copied record", record_index, len(nodes))
1013
for index, items in request_groups.iteritems():
1014
pack_readv_requests = []
1015
for key, value in items:
1016
# ---- KnitGraphIndex.get_position
1017
bits = value[1:].split(' ')
1018
offset, length = int(bits[0]), int(bits[1])
1019
pack_readv_requests.append((offset, length, (key, value[0])))
1020
# linear scan up the pack
1021
pack_readv_requests.sort()
1023
pack_obj = index_map[index]
1024
transport, path = pack_obj.access_tuple()
1026
reader = pack.make_readv_reader(transport, path,
1027
[offset[0:2] for offset in pack_readv_requests])
1028
except errors.NoSuchFile:
1029
if self._reload_func is not None:
1032
for (names, read_func), (_1, _2, (key, eol_flag)) in \
1033
izip(reader.iter_records(), pack_readv_requests):
1034
raw_data = read_func(None)
1035
# check the header only
1036
if output_lines is not None:
1037
output_lines(knit._parse_record(key[-1], raw_data)[0])
1039
df, _ = knit._parse_record_header(key, raw_data)
1041
pos, size = writer.add_bytes_record(raw_data, names)
1042
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
1043
pb.update("Copied record", record_index)
1046
def _copy_nodes_graph(self, index_map, writer, write_index,
1047
readv_group_iter, total_items, output_lines=False):
1048
"""Copy knit nodes between packs.
1050
:param output_lines: Return lines present in the copied data as
1051
an iterator of line,version_id.
1053
pb = ui.ui_factory.nested_progress_bar()
1055
for result in self._do_copy_nodes_graph(index_map, writer,
1056
write_index, output_lines, pb, readv_group_iter, total_items):
1059
# Python 2.4 does not permit try:finally: in a generator.
1065
def _do_copy_nodes_graph(self, index_map, writer, write_index,
1066
output_lines, pb, readv_group_iter, total_items):
1067
# for record verification
1068
knit = KnitVersionedFiles(None, None)
1069
# for line extraction when requested (inventories only)
1071
factory = KnitPlainFactory()
1073
pb.update("Copied record", record_index, total_items)
1074
for index, readv_vector, node_vector in readv_group_iter:
1076
pack_obj = index_map[index]
1077
transport, path = pack_obj.access_tuple()
1079
reader = pack.make_readv_reader(transport, path, readv_vector)
1080
except errors.NoSuchFile:
1081
if self._reload_func is not None:
1084
for (names, read_func), (key, eol_flag, references) in \
1085
izip(reader.iter_records(), node_vector):
1086
raw_data = read_func(None)
1088
# read the entire thing
1089
content, _ = knit._parse_record(key[-1], raw_data)
1090
if len(references[-1]) == 0:
1091
line_iterator = factory.get_fulltext_content(content)
1093
line_iterator = factory.get_linedelta_content(content)
1094
for line in line_iterator:
1097
# check the header only
1098
df, _ = knit._parse_record_header(key, raw_data)
1100
pos, size = writer.add_bytes_record(raw_data, names)
1101
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1102
pb.update("Copied record", record_index)
1105
def _get_text_nodes(self):
1106
text_index_map, text_indices = self._pack_map_and_index_list(
1108
return text_index_map, self._index_contents(text_indices,
1111
def _least_readv_node_readv(self, nodes):
1112
"""Generate request groups for nodes using the least readv's.
1114
:param nodes: An iterable of graph index nodes.
1115
:return: Total node count and an iterator of the data needed to perform
1116
readvs to obtain the data for nodes. Each item yielded by the
1117
iterator is a tuple with:
1118
index, readv_vector, node_vector. readv_vector is a list ready to
1119
hand to the transport readv method, and node_vector is a list of
1120
(key, eol_flag, references) for the node retrieved by the
1121
matching readv_vector.
1123
# group by pack so we do one readv per pack
1124
nodes = sorted(nodes)
1127
for index, key, value, references in nodes:
1128
if index not in request_groups:
1129
request_groups[index] = []
1130
request_groups[index].append((key, value, references))
1132
for index, items in request_groups.iteritems():
1133
pack_readv_requests = []
1134
for key, value, references in items:
1135
# ---- KnitGraphIndex.get_position
1136
bits = value[1:].split(' ')
1137
offset, length = int(bits[0]), int(bits[1])
1138
pack_readv_requests.append(
1139
((offset, length), (key, value[0], references)))
1140
# linear scan up the pack to maximum range combining.
1141
pack_readv_requests.sort()
1142
# split out the readv and the node data.
1143
pack_readv = [readv for readv, node in pack_readv_requests]
1144
node_vector = [node for readv, node in pack_readv_requests]
1145
result.append((index, pack_readv, node_vector))
1146
return total, result
1148
def _log_copied_texts(self):
1149
if 'pack' in debug.debug_flags:
1150
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
1151
time.ctime(), self._pack_collection._upload_transport.base,
1152
self.new_pack.random_name,
1153
self.new_pack.text_index.key_count(),
1154
time.time() - self.new_pack.start_time)
1156
def _process_inventory_lines(self, inv_lines):
1157
"""Use up the inv_lines generator and setup a text key filter."""
1158
repo = self._pack_collection.repo
1159
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
1160
inv_lines, self.revision_keys)
1162
for fileid, file_revids in fileid_revisions.iteritems():
1163
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
1164
self._text_filter = text_filter
1166
def _revision_node_readv(self, revision_nodes):
1167
"""Return the total revisions and the readv's to issue.
1169
:param revision_nodes: The revision index contents for the packs being
1170
incorporated into the new pack.
1171
:return: As per _least_readv_node_readv.
1173
return self._least_readv_node_readv(revision_nodes)
1175
def _use_pack(self, new_pack):
1176
"""Return True if new_pack should be used.
1178
:param new_pack: The pack that has just been created.
1179
:return: True if the pack should be used.
1181
return new_pack.data_inserted()
1184
class OptimisingPacker(Packer):
1185
"""A packer which spends more time to create better disk layouts."""
1187
def _revision_node_readv(self, revision_nodes):
1188
"""Return the total revisions and the readv's to issue.
1190
This sort places revisions in topological order with the ancestors
1193
:param revision_nodes: The revision index contents for the packs being
1194
incorporated into the new pack.
1195
:return: As per _least_readv_node_readv.
1197
# build an ancestors dict
1200
for index, key, value, references in revision_nodes:
1201
ancestors[key] = references[0]
1202
by_key[key] = (index, value, references)
1203
order = tsort.topo_sort(ancestors)
1205
# Single IO is pathological, but it will work as a starting point.
1207
for key in reversed(order):
1208
index, value, references = by_key[key]
1209
# ---- KnitGraphIndex.get_position
1210
bits = value[1:].split(' ')
1211
offset, length = int(bits[0]), int(bits[1])
1213
(index, [(offset, length)], [(key, value[0], references)]))
1214
# TODO: combine requests in the same index that are in ascending order.
1215
return total, requests
1217
def open_pack(self):
1218
"""Open a pack for the pack we are creating."""
1219
new_pack = super(OptimisingPacker, self).open_pack()
1220
# Turn on the optimization flags for all the index builders.
1221
new_pack.revision_index.set_optimize(for_size=True)
1222
new_pack.inventory_index.set_optimize(for_size=True)
1223
new_pack.text_index.set_optimize(for_size=True)
1224
new_pack.signature_index.set_optimize(for_size=True)
1228
class ReconcilePacker(Packer):
1229
"""A packer which regenerates indices etc as it copies.
1231
This is used by ``bzr reconcile`` to cause parent text pointers to be
1235
def _extra_init(self):
1236
self._data_changed = False
1238
def _process_inventory_lines(self, inv_lines):
1239
"""Generate a text key reference map rather for reconciling with."""
1240
repo = self._pack_collection.repo
1241
refs = repo._find_text_key_references_from_xml_inventory_lines(
1243
self._text_refs = refs
1244
# during reconcile we:
1245
# - convert unreferenced texts to full texts
1246
# - correct texts which reference a text not copied to be full texts
1247
# - copy all others as-is but with corrected parents.
1248
# - so at this point we don't know enough to decide what becomes a full
1250
self._text_filter = None
1252
def _copy_text_texts(self):
1253
"""generate what texts we should have and then copy."""
1254
self.pb.update("Copying content texts", 3)
1255
# we have three major tasks here:
1256
# 1) generate the ideal index
1257
repo = self._pack_collection.repo
1258
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1259
_1, key, _2, refs in
1260
self.new_pack.revision_index.iter_all_entries()])
1261
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1262
# 2) generate a text_nodes list that contains all the deltas that can
1263
# be used as-is, with corrected parents.
1266
discarded_nodes = []
1267
NULL_REVISION = _mod_revision.NULL_REVISION
1268
text_index_map, text_nodes = self._get_text_nodes()
1269
for node in text_nodes:
1275
ideal_parents = tuple(ideal_index[node[1]])
1277
discarded_nodes.append(node)
1278
self._data_changed = True
1280
if ideal_parents == (NULL_REVISION,):
1282
if ideal_parents == node[3][0]:
1284
ok_nodes.append(node)
1285
elif ideal_parents[0:1] == node[3][0][0:1]:
1286
# the left most parent is the same, or there are no parents
1287
# today. Either way, we can preserve the representation as
1288
# long as we change the refs to be inserted.
1289
self._data_changed = True
1290
ok_nodes.append((node[0], node[1], node[2],
1291
(ideal_parents, node[3][1])))
1292
self._data_changed = True
1294
# Reinsert this text completely
1295
bad_texts.append((node[1], ideal_parents))
1296
self._data_changed = True
1297
# we're finished with some data.
1300
# 3) bulk copy the ok data
1301
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1302
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1303
self.new_pack.text_index, readv_group_iter, total_items))
1304
# 4) adhoc copy all the other texts.
1305
# We have to topologically insert all texts otherwise we can fail to
1306
# reconcile when parts of a single delta chain are preserved intact,
1307
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1308
# reinserted, and if d3 has incorrect parents it will also be
1309
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1310
# copied), so we will try to delta, but d2 is not currently able to be
1311
# extracted because it's basis d1 is not present. Topologically sorting
1312
# addresses this. The following generates a sort for all the texts that
1313
# are being inserted without having to reference the entire text key
1314
# space (we only topo sort the revisions, which is smaller).
1315
topo_order = tsort.topo_sort(ancestors)
1316
rev_order = dict(zip(topo_order, range(len(topo_order))))
1317
bad_texts.sort(key=lambda key:rev_order.get(key[0][1], 0))
1318
transaction = repo.get_transaction()
1319
file_id_index = GraphIndexPrefixAdapter(
1320
self.new_pack.text_index,
1322
add_nodes_callback=self.new_pack.text_index.add_nodes)
1323
data_access = _DirectPackAccess(
1324
{self.new_pack.text_index:self.new_pack.access_tuple()})
1325
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1326
self.new_pack.access_tuple())
1327
output_texts = KnitVersionedFiles(
1328
_KnitGraphIndex(self.new_pack.text_index,
1329
add_callback=self.new_pack.text_index.add_nodes,
1330
deltas=True, parents=True, is_locked=repo.is_locked),
1331
data_access=data_access, max_delta_chain=200)
1332
for key, parent_keys in bad_texts:
1333
# We refer to the new pack to delta data being output.
1334
# A possible improvement would be to catch errors on short reads
1335
# and only flush then.
1336
self.new_pack.flush()
1338
for parent_key in parent_keys:
1339
if parent_key[0] != key[0]:
1340
# Graph parents must match the fileid
1341
raise errors.BzrError('Mismatched key parent %r:%r' %
1343
parents.append(parent_key[1])
1344
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1345
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1346
output_texts.add_lines(key, parent_keys, text_lines,
1347
random_id=True, check_content=False)
1348
# 5) check that nothing inserted has a reference outside the keyspace.
1349
missing_text_keys = self.new_pack.text_index._external_references()
1350
if missing_text_keys:
1351
raise errors.BzrCheckError('Reference to missing compression parents %r'
1352
% (missing_text_keys,))
1353
self._log_copied_texts()
1355
def _use_pack(self, new_pack):
1356
"""Override _use_pack to check for reconcile having changed content."""
1357
# XXX: we might be better checking this at the copy time.
1358
original_inventory_keys = set()
1359
inv_index = self._pack_collection.inventory_index.combined_index
1360
for entry in inv_index.iter_all_entries():
1361
original_inventory_keys.add(entry[1])
1362
new_inventory_keys = set()
1363
for entry in new_pack.inventory_index.iter_all_entries():
1364
new_inventory_keys.add(entry[1])
1365
if new_inventory_keys != original_inventory_keys:
1366
self._data_changed = True
1367
return new_pack.data_inserted() and self._data_changed
1370
class RepositoryPackCollection(object):
1371
"""Management of packs within a repository.
1373
:ivar _names: map of {pack_name: (index_size,)}
1376
pack_factory = NewPack
1377
resumed_pack_factory = ResumedPack
1379
def __init__(self, repo, transport, index_transport, upload_transport,
1380
pack_transport, index_builder_class, index_class,
1382
"""Create a new RepositoryPackCollection.
1384
:param transport: Addresses the repository base directory
1385
(typically .bzr/repository/).
1386
:param index_transport: Addresses the directory containing indices.
1387
:param upload_transport: Addresses the directory into which packs are written
1388
while they're being created.
1389
:param pack_transport: Addresses the directory of existing complete packs.
1390
:param index_builder_class: The index builder class to use.
1391
:param index_class: The index class to use.
1392
:param use_chk_index: Whether to setup and manage a CHK index.
1394
# XXX: This should call self.reset()
1396
self.transport = transport
1397
self._index_transport = index_transport
1398
self._upload_transport = upload_transport
1399
self._pack_transport = pack_transport
1400
self._index_builder_class = index_builder_class
1401
self._index_class = index_class
1402
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3,
1407
self._packs_by_name = {}
1408
# the previous pack-names content
1409
self._packs_at_load = None
1410
# when a pack is being created by this object, the state of that pack.
1411
self._new_pack = None
1412
# aggregated revision index data
1413
flush = self._flush_new_pack
1414
self.revision_index = AggregateIndex(self.reload_pack_names, flush)
1415
self.inventory_index = AggregateIndex(self.reload_pack_names, flush)
1416
self.text_index = AggregateIndex(self.reload_pack_names, flush)
1417
self.signature_index = AggregateIndex(self.reload_pack_names, flush)
1419
self.chk_index = AggregateIndex(self.reload_pack_names, flush)
1421
# used to determine if we're using a chk_index elsewhere.
1422
self.chk_index = None
1424
self._resumed_packs = []
1427
return '%s(%r)' % (self.__class__.__name__, self.repo)
1429
def add_pack_to_memory(self, pack):
1430
"""Make a Pack object available to the repository to satisfy queries.
1432
:param pack: A Pack object.
1434
if pack.name in self._packs_by_name:
1435
raise AssertionError(
1436
'pack %s already in _packs_by_name' % (pack.name,))
1437
self.packs.append(pack)
1438
self._packs_by_name[pack.name] = pack
1439
self.revision_index.add_index(pack.revision_index, pack)
1440
self.inventory_index.add_index(pack.inventory_index, pack)
1441
self.text_index.add_index(pack.text_index, pack)
1442
self.signature_index.add_index(pack.signature_index, pack)
1443
if self.chk_index is not None:
1444
self.chk_index.add_index(pack.chk_index, pack)
1446
def all_packs(self):
1447
"""Return a list of all the Pack objects this repository has.
1449
Note that an in-progress pack being created is not returned.
1451
:return: A list of Pack objects for all the packs in the repository.
1454
for name in self.names():
1455
result.append(self.get_pack_by_name(name))
1459
"""Pack the pack collection incrementally.
1461
This will not attempt global reorganisation or recompression,
1462
rather it will just ensure that the total number of packs does
1463
not grow without bound. It uses the _max_pack_count method to
1464
determine if autopacking is needed, and the pack_distribution
1465
method to determine the number of revisions in each pack.
1467
If autopacking takes place then the packs name collection will have
1468
been flushed to disk - packing requires updating the name collection
1469
in synchronisation with certain steps. Otherwise the names collection
1472
:return: Something evaluating true if packing took place.
1476
return self._do_autopack()
1477
except errors.RetryAutopack:
1478
# If we get a RetryAutopack exception, we should abort the
1479
# current action, and retry.
1482
def _do_autopack(self):
1483
# XXX: Should not be needed when the management of indices is sane.
1484
total_revisions = self.revision_index.combined_index.key_count()
1485
total_packs = len(self._names)
1486
if self._max_pack_count(total_revisions) >= total_packs:
1488
# determine which packs need changing
1489
pack_distribution = self.pack_distribution(total_revisions)
1491
for pack in self.all_packs():
1492
revision_count = pack.get_revision_count()
1493
if revision_count == 0:
1494
# revision less packs are not generated by normal operation,
1495
# only by operations like sign-my-commits, and thus will not
1496
# tend to grow rapdily or without bound like commit containing
1497
# packs do - leave them alone as packing them really should
1498
# group their data with the relevant commit, and that may
1499
# involve rewriting ancient history - which autopack tries to
1500
# avoid. Alternatively we could not group the data but treat
1501
# each of these as having a single revision, and thus add
1502
# one revision for each to the total revision count, to get
1503
# a matching distribution.
1505
existing_packs.append((revision_count, pack))
1506
pack_operations = self.plan_autopack_combinations(
1507
existing_packs, pack_distribution)
1508
num_new_packs = len(pack_operations)
1509
num_old_packs = sum([len(po[1]) for po in pack_operations])
1510
num_revs_affected = sum([po[0] for po in pack_operations])
1511
mutter('Auto-packing repository %s, which has %d pack files, '
1512
'containing %d revisions. Packing %d files into %d affecting %d'
1513
' revisions', self, total_packs, total_revisions, num_old_packs,
1514
num_new_packs, num_revs_affected)
1515
result = self._execute_pack_operations(pack_operations,
1516
reload_func=self._restart_autopack)
1517
mutter('Auto-packing repository %s completed', self)
1520
def _execute_pack_operations(self, pack_operations, _packer_class=Packer,
1522
"""Execute a series of pack operations.
1524
:param pack_operations: A list of [revision_count, packs_to_combine].
1525
:param _packer_class: The class of packer to use (default: Packer).
1526
:return: The new pack names.
1528
for revision_count, packs in pack_operations:
1529
# we may have no-ops from the setup logic
1532
packer = _packer_class(self, packs, '.autopack',
1533
reload_func=reload_func)
1536
except errors.RetryWithNewPacks:
1537
# An exception is propagating out of this context, make sure
1538
# this packer has cleaned up. Packer() doesn't set its new_pack
1539
# state into the RepositoryPackCollection object, so we only
1540
# have access to it directly here.
1541
if packer.new_pack is not None:
1542
packer.new_pack.abort()
1545
self._remove_pack_from_memory(pack)
1546
# record the newly available packs and stop advertising the old
1548
to_be_obsoleted = []
1549
for _, packs in pack_operations:
1550
to_be_obsoleted.extend(packs)
1551
result = self._save_pack_names(clear_obsolete_packs=True,
1552
obsolete_packs=to_be_obsoleted)
1555
def _flush_new_pack(self):
1556
if self._new_pack is not None:
1557
self._new_pack.flush()
1559
def lock_names(self):
1560
"""Acquire the mutex around the pack-names index.
1562
This cannot be used in the middle of a read-only transaction on the
1565
self.repo.control_files.lock_write()
1567
def _already_packed(self):
1568
"""Is the collection already packed?"""
1569
return not (self.repo._format.pack_compresses or (len(self._names) > 1))
1571
def pack(self, hint=None):
1572
"""Pack the pack collection totally."""
1573
self.ensure_loaded()
1574
total_packs = len(self._names)
1575
if self._already_packed():
1577
total_revisions = self.revision_index.combined_index.key_count()
1578
# XXX: the following may want to be a class, to pack with a given
1580
mutter('Packing repository %s, which has %d pack files, '
1581
'containing %d revisions with hint %r.', self, total_packs,
1582
total_revisions, hint)
1583
# determine which packs need changing
1584
pack_operations = [[0, []]]
1585
for pack in self.all_packs():
1586
if hint is None or pack.name in hint:
1587
# Either no hint was provided (so we are packing everything),
1588
# or this pack was included in the hint.
1589
pack_operations[-1][0] += pack.get_revision_count()
1590
pack_operations[-1][1].append(pack)
1591
self._execute_pack_operations(pack_operations, OptimisingPacker)
1593
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1594
"""Plan a pack operation.
1596
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1598
:param pack_distribution: A list with the number of revisions desired
1601
if len(existing_packs) <= len(pack_distribution):
1603
existing_packs.sort(reverse=True)
1604
pack_operations = [[0, []]]
1605
# plan out what packs to keep, and what to reorganise
1606
while len(existing_packs):
1607
# take the largest pack, and if its less than the head of the
1608
# distribution chart we will include its contents in the new pack
1609
# for that position. If its larger, we remove its size from the
1610
# distribution chart
1611
next_pack_rev_count, next_pack = existing_packs.pop(0)
1612
if next_pack_rev_count >= pack_distribution[0]:
1613
# this is already packed 'better' than this, so we can
1614
# not waste time packing it.
1615
while next_pack_rev_count > 0:
1616
next_pack_rev_count -= pack_distribution[0]
1617
if next_pack_rev_count >= 0:
1619
del pack_distribution[0]
1621
# didn't use that entire bucket up
1622
pack_distribution[0] = -next_pack_rev_count
1624
# add the revisions we're going to add to the next output pack
1625
pack_operations[-1][0] += next_pack_rev_count
1626
# allocate this pack to the next pack sub operation
1627
pack_operations[-1][1].append(next_pack)
1628
if pack_operations[-1][0] >= pack_distribution[0]:
1629
# this pack is used up, shift left.
1630
del pack_distribution[0]
1631
pack_operations.append([0, []])
1632
# Now that we know which pack files we want to move, shove them all
1633
# into a single pack file.
1635
final_pack_list = []
1636
for num_revs, pack_files in pack_operations:
1637
final_rev_count += num_revs
1638
final_pack_list.extend(pack_files)
1639
if len(final_pack_list) == 1:
1640
raise AssertionError('We somehow generated an autopack with a'
1641
' single pack file being moved.')
1643
return [[final_rev_count, final_pack_list]]
1645
def ensure_loaded(self):
1646
"""Ensure we have read names from disk.
1648
:return: True if the disk names had not been previously read.
1650
# NB: if you see an assertion error here, its probably access against
1651
# an unlocked repo. Naughty.
1652
if not self.repo.is_locked():
1653
raise errors.ObjectNotLocked(self.repo)
1654
if self._names is None:
1656
self._packs_at_load = set()
1657
for index, key, value in self._iter_disk_pack_index():
1659
self._names[name] = self._parse_index_sizes(value)
1660
self._packs_at_load.add((key, value))
1664
# populate all the metadata.
1668
def _parse_index_sizes(self, value):
1669
"""Parse a string of index sizes."""
1670
return tuple([int(digits) for digits in value.split(' ')])
1672
def get_pack_by_name(self, name):
1673
"""Get a Pack object by name.
1675
:param name: The name of the pack - e.g. '123456'
1676
:return: A Pack object.
1679
return self._packs_by_name[name]
1681
rev_index = self._make_index(name, '.rix')
1682
inv_index = self._make_index(name, '.iix')
1683
txt_index = self._make_index(name, '.tix')
1684
sig_index = self._make_index(name, '.six')
1685
if self.chk_index is not None:
1686
chk_index = self._make_index(name, '.cix', unlimited_cache=True)
1689
result = ExistingPack(self._pack_transport, name, rev_index,
1690
inv_index, txt_index, sig_index, chk_index)
1691
self.add_pack_to_memory(result)
1694
def _resume_pack(self, name):
1695
"""Get a suspended Pack object by name.
1697
:param name: The name of the pack - e.g. '123456'
1698
:return: A Pack object.
1700
if not re.match('[a-f0-9]{32}', name):
1701
# Tokens should be md5sums of the suspended pack file, i.e. 32 hex
1703
raise errors.UnresumableWriteGroup(
1704
self.repo, [name], 'Malformed write group token')
1706
rev_index = self._make_index(name, '.rix', resume=True)
1707
inv_index = self._make_index(name, '.iix', resume=True)
1708
txt_index = self._make_index(name, '.tix', resume=True)
1709
sig_index = self._make_index(name, '.six', resume=True)
1710
if self.chk_index is not None:
1711
chk_index = self._make_index(name, '.cix', resume=True,
1712
unlimited_cache=True)
1715
result = self.resumed_pack_factory(name, rev_index, inv_index,
1716
txt_index, sig_index, self._upload_transport,
1717
self._pack_transport, self._index_transport, self,
1718
chk_index=chk_index)
1719
except errors.NoSuchFile, e:
1720
raise errors.UnresumableWriteGroup(self.repo, [name], str(e))
1721
self.add_pack_to_memory(result)
1722
self._resumed_packs.append(result)
1725
def allocate(self, a_new_pack):
1726
"""Allocate name in the list of packs.
1728
:param a_new_pack: A NewPack instance to be added to the collection of
1729
packs for this repository.
1731
self.ensure_loaded()
1732
if a_new_pack.name in self._names:
1733
raise errors.BzrError(
1734
'Pack %r already exists in %s' % (a_new_pack.name, self))
1735
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1736
self.add_pack_to_memory(a_new_pack)
1738
def _iter_disk_pack_index(self):
1739
"""Iterate over the contents of the pack-names index.
1741
This is used when loading the list from disk, and before writing to
1742
detect updates from others during our write operation.
1743
:return: An iterator of the index contents.
1745
return self._index_class(self.transport, 'pack-names', None
1746
).iter_all_entries()
1748
def _make_index(self, name, suffix, resume=False, unlimited_cache=False):
1749
size_offset = self._suffix_offsets[suffix]
1750
index_name = name + suffix
1752
transport = self._upload_transport
1753
index_size = transport.stat(index_name).st_size
1755
transport = self._index_transport
1756
index_size = self._names[name][size_offset]
1757
return self._index_class(transport, index_name, index_size,
1758
unlimited_cache=unlimited_cache)
1760
def _max_pack_count(self, total_revisions):
1761
"""Return the maximum number of packs to use for total revisions.
1763
:param total_revisions: The total number of revisions in the
1766
if not total_revisions:
1768
digits = str(total_revisions)
1770
for digit in digits:
1771
result += int(digit)
1775
"""Provide an order to the underlying names."""
1776
return sorted(self._names.keys())
1778
def _obsolete_packs(self, packs):
1779
"""Move a number of packs which have been obsoleted out of the way.
1781
Each pack and its associated indices are moved out of the way.
1783
Note: for correctness this function should only be called after a new
1784
pack names index has been written without these pack names, and with
1785
the names of packs that contain the data previously available via these
1788
:param packs: The packs to obsolete.
1789
:param return: None.
1793
pack.pack_transport.rename(pack.file_name(),
1794
'../obsolete_packs/' + pack.file_name())
1795
except (errors.PathError, errors.TransportError), e:
1796
# TODO: Should these be warnings or mutters?
1797
mutter("couldn't rename obsolete pack, skipping it:\n%s"
1799
# TODO: Probably needs to know all possible indices for this pack
1800
# - or maybe list the directory and move all indices matching this
1801
# name whether we recognize it or not?
1802
suffixes = ['.iix', '.six', '.tix', '.rix']
1803
if self.chk_index is not None:
1804
suffixes.append('.cix')
1805
for suffix in suffixes:
1807
self._index_transport.rename(pack.name + suffix,
1808
'../obsolete_packs/' + pack.name + suffix)
1809
except (errors.PathError, errors.TransportError), e:
1810
mutter("couldn't rename obsolete index, skipping it:\n%s"
1813
def pack_distribution(self, total_revisions):
1814
"""Generate a list of the number of revisions to put in each pack.
1816
:param total_revisions: The total number of revisions in the
1819
if total_revisions == 0:
1821
digits = reversed(str(total_revisions))
1823
for exponent, count in enumerate(digits):
1824
size = 10 ** exponent
1825
for pos in range(int(count)):
1827
return list(reversed(result))
1829
def _pack_tuple(self, name):
1830
"""Return a tuple with the transport and file name for a pack name."""
1831
return self._pack_transport, name + '.pack'
1833
def _remove_pack_from_memory(self, pack):
1834
"""Remove pack from the packs accessed by this repository.
1836
Only affects memory state, until self._save_pack_names() is invoked.
1838
self._names.pop(pack.name)
1839
self._packs_by_name.pop(pack.name)
1840
self._remove_pack_indices(pack)
1841
self.packs.remove(pack)
1843
def _remove_pack_indices(self, pack, ignore_missing=False):
1844
"""Remove the indices for pack from the aggregated indices.
1846
:param ignore_missing: Suppress KeyErrors from calling remove_index.
1848
for index_type in Pack.index_definitions.keys():
1849
attr_name = index_type + '_index'
1850
aggregate_index = getattr(self, attr_name)
1851
if aggregate_index is not None:
1852
pack_index = getattr(pack, attr_name)
1854
aggregate_index.remove_index(pack_index)
1861
"""Clear all cached data."""
1862
# cached revision data
1863
self.revision_index.clear()
1864
# cached signature data
1865
self.signature_index.clear()
1866
# cached file text data
1867
self.text_index.clear()
1868
# cached inventory data
1869
self.inventory_index.clear()
1871
if self.chk_index is not None:
1872
self.chk_index.clear()
1873
# remove the open pack
1874
self._new_pack = None
1875
# information about packs.
1878
self._packs_by_name = {}
1879
self._packs_at_load = None
1881
def _unlock_names(self):
1882
"""Release the mutex around the pack-names index."""
1883
self.repo.control_files.unlock()
1885
def _diff_pack_names(self):
1886
"""Read the pack names from disk, and compare it to the one in memory.
1888
:return: (disk_nodes, deleted_nodes, new_nodes)
1889
disk_nodes The final set of nodes that should be referenced
1890
deleted_nodes Nodes which have been removed from when we started
1891
new_nodes Nodes that are newly introduced
1893
# load the disk nodes across
1895
for index, key, value in self._iter_disk_pack_index():
1896
disk_nodes.add((key, value))
1897
orig_disk_nodes = set(disk_nodes)
1899
# do a two-way diff against our original content
1900
current_nodes = set()
1901
for name, sizes in self._names.iteritems():
1903
((name, ), ' '.join(str(size) for size in sizes)))
1905
# Packs no longer present in the repository, which were present when we
1906
# locked the repository
1907
deleted_nodes = self._packs_at_load - current_nodes
1908
# Packs which this process is adding
1909
new_nodes = current_nodes - self._packs_at_load
1911
# Update the disk_nodes set to include the ones we are adding, and
1912
# remove the ones which were removed by someone else
1913
disk_nodes.difference_update(deleted_nodes)
1914
disk_nodes.update(new_nodes)
1916
return disk_nodes, deleted_nodes, new_nodes, orig_disk_nodes
1918
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1919
"""Given the correct set of pack files, update our saved info.
1921
:return: (removed, added, modified)
1922
removed pack names removed from self._names
1923
added pack names added to self._names
1924
modified pack names that had changed value
1929
## self._packs_at_load = disk_nodes
1930
new_names = dict(disk_nodes)
1931
# drop no longer present nodes
1932
for pack in self.all_packs():
1933
if (pack.name,) not in new_names:
1934
removed.append(pack.name)
1935
self._remove_pack_from_memory(pack)
1936
# add new nodes/refresh existing ones
1937
for key, value in disk_nodes:
1939
sizes = self._parse_index_sizes(value)
1940
if name in self._names:
1942
if sizes != self._names[name]:
1943
# the pack for name has had its indices replaced - rare but
1944
# important to handle. XXX: probably can never happen today
1945
# because the three-way merge code above does not handle it
1946
# - you may end up adding the same key twice to the new
1947
# disk index because the set values are the same, unless
1948
# the only index shows up as deleted by the set difference
1949
# - which it may. Until there is a specific test for this,
1950
# assume its broken. RBC 20071017.
1951
self._remove_pack_from_memory(self.get_pack_by_name(name))
1952
self._names[name] = sizes
1953
self.get_pack_by_name(name)
1954
modified.append(name)
1957
self._names[name] = sizes
1958
self.get_pack_by_name(name)
1960
return removed, added, modified
1962
def _save_pack_names(self, clear_obsolete_packs=False, obsolete_packs=None):
1963
"""Save the list of packs.
1965
This will take out the mutex around the pack names list for the
1966
duration of the method call. If concurrent updates have been made, a
1967
three-way merge between the current list and the current in memory list
1970
:param clear_obsolete_packs: If True, clear out the contents of the
1971
obsolete_packs directory.
1972
:param obsolete_packs: Packs that are obsolete once the new pack-names
1973
file has been written.
1974
:return: A list of the names saved that were not previously on disk.
1976
already_obsolete = []
1979
builder = self._index_builder_class()
1980
(disk_nodes, deleted_nodes, new_nodes,
1981
orig_disk_nodes) = self._diff_pack_names()
1982
# TODO: handle same-name, index-size-changes here -
1983
# e.g. use the value from disk, not ours, *unless* we're the one
1985
for key, value in disk_nodes:
1986
builder.add_node(key, value)
1987
self.transport.put_file('pack-names', builder.finish(),
1988
mode=self.repo.bzrdir._get_file_mode())
1989
self._packs_at_load = disk_nodes
1990
if clear_obsolete_packs:
1993
to_preserve = set([o.name for o in obsolete_packs])
1994
already_obsolete = self._clear_obsolete_packs(to_preserve)
1996
self._unlock_names()
1997
# synchronise the memory packs list with what we just wrote:
1998
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
2000
# TODO: We could add one more condition here. "if o.name not in
2001
# orig_disk_nodes and o != the new_pack we haven't written to
2002
# disk yet. However, the new pack object is not easily
2003
# accessible here (it would have to be passed through the
2004
# autopacking code, etc.)
2005
obsolete_packs = [o for o in obsolete_packs
2006
if o.name not in already_obsolete]
2007
self._obsolete_packs(obsolete_packs)
2008
return [new_node[0][0] for new_node in new_nodes]
2010
def reload_pack_names(self):
2011
"""Sync our pack listing with what is present in the repository.
2013
This should be called when we find out that something we thought was
2014
present is now missing. This happens when another process re-packs the
2017
:return: True if the in-memory list of packs has been altered at all.
2019
# The ensure_loaded call is to handle the case where the first call
2020
# made involving the collection was to reload_pack_names, where we
2021
# don't have a view of disk contents. Its a bit of a bandaid, and
2022
# causes two reads of pack-names, but its a rare corner case not struck
2023
# with regular push/pull etc.
2024
first_read = self.ensure_loaded()
2027
# out the new value.
2028
(disk_nodes, deleted_nodes, new_nodes,
2029
orig_disk_nodes) = self._diff_pack_names()
2030
# _packs_at_load is meant to be the explicit list of names in
2031
# 'pack-names' at then start. As such, it should not contain any
2032
# pending names that haven't been written out yet.
2033
self._packs_at_load = orig_disk_nodes
2035
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
2036
if removed or added or modified:
2040
def _restart_autopack(self):
2041
"""Reload the pack names list, and restart the autopack code."""
2042
if not self.reload_pack_names():
2043
# Re-raise the original exception, because something went missing
2044
# and a restart didn't find it
2046
raise errors.RetryAutopack(self.repo, False, sys.exc_info())
2048
def _clear_obsolete_packs(self, preserve=None):
2049
"""Delete everything from the obsolete-packs directory.
2051
:return: A list of pack identifiers (the filename without '.pack') that
2052
were found in obsolete_packs.
2055
obsolete_pack_transport = self.transport.clone('obsolete_packs')
2056
if preserve is None:
2058
for filename in obsolete_pack_transport.list_dir('.'):
2059
name, ext = osutils.splitext(filename)
2062
if name in preserve:
2065
obsolete_pack_transport.delete(filename)
2066
except (errors.PathError, errors.TransportError), e:
2067
warning("couldn't delete obsolete pack, skipping it:\n%s"
2071
def _start_write_group(self):
2072
# Do not permit preparation for writing if we're not in a 'write lock'.
2073
if not self.repo.is_write_locked():
2074
raise errors.NotWriteLocked(self)
2075
self._new_pack = self.pack_factory(self, upload_suffix='.pack',
2076
file_mode=self.repo.bzrdir._get_file_mode())
2077
# allow writing: queue writes to a new index
2078
self.revision_index.add_writable_index(self._new_pack.revision_index,
2080
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
2082
self.text_index.add_writable_index(self._new_pack.text_index,
2084
self._new_pack.text_index.set_optimize(combine_backing_indices=False)
2085
self.signature_index.add_writable_index(self._new_pack.signature_index,
2087
if self.chk_index is not None:
2088
self.chk_index.add_writable_index(self._new_pack.chk_index,
2090
self.repo.chk_bytes._index._add_callback = self.chk_index.add_callback
2091
self._new_pack.chk_index.set_optimize(combine_backing_indices=False)
2093
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
2094
self.repo.revisions._index._add_callback = self.revision_index.add_callback
2095
self.repo.signatures._index._add_callback = self.signature_index.add_callback
2096
self.repo.texts._index._add_callback = self.text_index.add_callback
2098
def _abort_write_group(self):
2099
# FIXME: just drop the transient index.
2100
# forget what names there are
2101
if self._new_pack is not None:
2102
operation = cleanup.OperationWithCleanups(self._new_pack.abort)
2103
operation.add_cleanup(setattr, self, '_new_pack', None)
2104
# If we aborted while in the middle of finishing the write
2105
# group, _remove_pack_indices could fail because the indexes are
2106
# already gone. But they're not there we shouldn't fail in this
2107
# case, so we pass ignore_missing=True.
2108
operation.add_cleanup(self._remove_pack_indices, self._new_pack,
2109
ignore_missing=True)
2110
operation.run_simple()
2111
for resumed_pack in self._resumed_packs:
2112
operation = cleanup.OperationWithCleanups(resumed_pack.abort)
2113
# See comment in previous finally block.
2114
operation.add_cleanup(self._remove_pack_indices, resumed_pack,
2115
ignore_missing=True)
2116
operation.run_simple()
2117
del self._resumed_packs[:]
2119
def _remove_resumed_pack_indices(self):
2120
for resumed_pack in self._resumed_packs:
2121
self._remove_pack_indices(resumed_pack)
2122
del self._resumed_packs[:]
2124
def _check_new_inventories(self):
2125
"""Detect missing inventories in this write group.
2127
:returns: list of strs, summarising any problems found. If the list is
2128
empty no problems were found.
2130
# The base implementation does no checks. GCRepositoryPackCollection
2134
def _commit_write_group(self):
2136
for prefix, versioned_file in (
2137
('revisions', self.repo.revisions),
2138
('inventories', self.repo.inventories),
2139
('texts', self.repo.texts),
2140
('signatures', self.repo.signatures),
2142
missing = versioned_file.get_missing_compression_parent_keys()
2143
all_missing.update([(prefix,) + key for key in missing])
2145
raise errors.BzrCheckError(
2146
"Repository %s has missing compression parent(s) %r "
2147
% (self.repo, sorted(all_missing)))
2148
problems = self._check_new_inventories()
2150
problems_summary = '\n'.join(problems)
2151
raise errors.BzrCheckError(
2152
"Cannot add revision(s) to repository: " + problems_summary)
2153
self._remove_pack_indices(self._new_pack)
2154
any_new_content = False
2155
if self._new_pack.data_inserted():
2156
# get all the data to disk and read to use
2157
self._new_pack.finish()
2158
self.allocate(self._new_pack)
2159
self._new_pack = None
2160
any_new_content = True
2162
self._new_pack.abort()
2163
self._new_pack = None
2164
for resumed_pack in self._resumed_packs:
2165
# XXX: this is a pretty ugly way to turn the resumed pack into a
2166
# properly committed pack.
2167
self._names[resumed_pack.name] = None
2168
self._remove_pack_from_memory(resumed_pack)
2169
resumed_pack.finish()
2170
self.allocate(resumed_pack)
2171
any_new_content = True
2172
del self._resumed_packs[:]
2174
result = self.autopack()
2176
# when autopack takes no steps, the names list is still
2178
return self._save_pack_names()
2182
def _suspend_write_group(self):
2183
tokens = [pack.name for pack in self._resumed_packs]
2184
self._remove_pack_indices(self._new_pack)
2185
if self._new_pack.data_inserted():
2186
# get all the data to disk and read to use
2187
self._new_pack.finish(suspend=True)
2188
tokens.append(self._new_pack.name)
2189
self._new_pack = None
2191
self._new_pack.abort()
2192
self._new_pack = None
2193
self._remove_resumed_pack_indices()
2196
def _resume_write_group(self, tokens):
2197
for token in tokens:
2198
self._resume_pack(token)
2201
class KnitPackRepository(KnitRepository):
2202
"""Repository with knit objects stored inside pack containers.
2204
The layering for a KnitPackRepository is:
2206
Graph | HPSS | Repository public layer |
2207
===================================================
2208
Tuple based apis below, string based, and key based apis above
2209
---------------------------------------------------
2211
Provides .texts, .revisions etc
2212
This adapts the N-tuple keys to physical knit records which only have a
2213
single string identifier (for historical reasons), which in older formats
2214
was always the revision_id, and in the mapped code for packs is always
2215
the last element of key tuples.
2216
---------------------------------------------------
2218
A separate GraphIndex is used for each of the
2219
texts/inventories/revisions/signatures contained within each individual
2220
pack file. The GraphIndex layer works in N-tuples and is unaware of any
2222
===================================================
2226
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
2228
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2229
_commit_builder_class, _serializer)
2230
index_transport = self._transport.clone('indices')
2231
self._pack_collection = RepositoryPackCollection(self, self._transport,
2233
self._transport.clone('upload'),
2234
self._transport.clone('packs'),
2235
_format.index_builder_class,
2236
_format.index_class,
2237
use_chk_index=self._format.supports_chks,
2239
self.inventories = KnitVersionedFiles(
2240
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2241
add_callback=self._pack_collection.inventory_index.add_callback,
2242
deltas=True, parents=True, is_locked=self.is_locked),
2243
data_access=self._pack_collection.inventory_index.data_access,
2244
max_delta_chain=200)
2245
self.revisions = KnitVersionedFiles(
2246
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2247
add_callback=self._pack_collection.revision_index.add_callback,
2248
deltas=False, parents=True, is_locked=self.is_locked,
2249
track_external_parent_refs=True),
2250
data_access=self._pack_collection.revision_index.data_access,
2252
self.signatures = KnitVersionedFiles(
2253
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2254
add_callback=self._pack_collection.signature_index.add_callback,
2255
deltas=False, parents=False, is_locked=self.is_locked),
2256
data_access=self._pack_collection.signature_index.data_access,
2258
self.texts = KnitVersionedFiles(
2259
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2260
add_callback=self._pack_collection.text_index.add_callback,
2261
deltas=True, parents=True, is_locked=self.is_locked),
2262
data_access=self._pack_collection.text_index.data_access,
2263
max_delta_chain=200)
2264
if _format.supports_chks:
2265
# No graph, no compression:- references from chks are between
2266
# different objects not temporal versions of the same; and without
2267
# some sort of temporal structure knit compression will just fail.
2268
self.chk_bytes = KnitVersionedFiles(
2269
_KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2270
add_callback=self._pack_collection.chk_index.add_callback,
2271
deltas=False, parents=False, is_locked=self.is_locked),
2272
data_access=self._pack_collection.chk_index.data_access,
2275
self.chk_bytes = None
2276
# True when the repository object is 'write locked' (as opposed to the
2277
# physical lock only taken out around changes to the pack-names list.)
2278
# Another way to represent this would be a decorator around the control
2279
# files object that presents logical locks as physical ones - if this
2280
# gets ugly consider that alternative design. RBC 20071011
2281
self._write_lock_count = 0
2282
self._transaction = None
2284
self._reconcile_does_inventory_gc = True
2285
self._reconcile_fixes_text_parents = True
2286
self._reconcile_backsup_inventory = False
2288
def _warn_if_deprecated(self, branch=None):
2289
# This class isn't deprecated, but one sub-format is
2290
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2291
super(KnitPackRepository, self)._warn_if_deprecated(branch)
2293
def _abort_write_group(self):
2294
self.revisions._index._key_dependencies.clear()
2295
self._pack_collection._abort_write_group()
2297
def _get_source(self, to_format):
2298
if to_format.network_name() == self._format.network_name():
2299
return KnitPackStreamSource(self, to_format)
2300
return super(KnitPackRepository, self)._get_source(to_format)
2302
def _make_parents_provider(self):
2303
return graph.CachingParentsProvider(self)
2305
def _refresh_data(self):
2306
if not self.is_locked():
2308
self._pack_collection.reload_pack_names()
2310
def _start_write_group(self):
2311
self._pack_collection._start_write_group()
2313
def _commit_write_group(self):
2314
hint = self._pack_collection._commit_write_group()
2315
self.revisions._index._key_dependencies.clear()
2318
def suspend_write_group(self):
2319
# XXX check self._write_group is self.get_transaction()?
2320
tokens = self._pack_collection._suspend_write_group()
2321
self.revisions._index._key_dependencies.clear()
2322
self._write_group = None
2325
def _resume_write_group(self, tokens):
2326
self._start_write_group()
2328
self._pack_collection._resume_write_group(tokens)
2329
except errors.UnresumableWriteGroup:
2330
self._abort_write_group()
2332
for pack in self._pack_collection._resumed_packs:
2333
self.revisions._index.scan_unvalidated_index(pack.revision_index)
2335
def get_transaction(self):
2336
if self._write_lock_count:
2337
return self._transaction
2339
return self.control_files.get_transaction()
2341
def is_locked(self):
2342
return self._write_lock_count or self.control_files.is_locked()
2344
def is_write_locked(self):
2345
return self._write_lock_count
2347
def lock_write(self, token=None):
2348
locked = self.is_locked()
2349
if not self._write_lock_count and locked:
2350
raise errors.ReadOnlyError(self)
2351
self._write_lock_count += 1
2352
if self._write_lock_count == 1:
2353
self._transaction = transactions.WriteTransaction()
2355
if 'relock' in debug.debug_flags and self._prev_lock == 'w':
2356
note('%r was write locked again', self)
2357
self._prev_lock = 'w'
2358
for repo in self._fallback_repositories:
2359
# Writes don't affect fallback repos
2361
self._refresh_data()
2363
def lock_read(self):
2364
locked = self.is_locked()
2365
if self._write_lock_count:
2366
self._write_lock_count += 1
2368
self.control_files.lock_read()
2370
if 'relock' in debug.debug_flags and self._prev_lock == 'r':
2371
note('%r was read locked again', self)
2372
self._prev_lock = 'r'
2373
for repo in self._fallback_repositories:
2375
self._refresh_data()
2377
def leave_lock_in_place(self):
2378
# not supported - raise an error
2379
raise NotImplementedError(self.leave_lock_in_place)
2381
def dont_leave_lock_in_place(self):
2382
# not supported - raise an error
2383
raise NotImplementedError(self.dont_leave_lock_in_place)
2386
def pack(self, hint=None):
2387
"""Compress the data within the repository.
2389
This will pack all the data to a single pack. In future it may
2390
recompress deltas or do other such expensive operations.
2392
self._pack_collection.pack(hint=hint)
2395
def reconcile(self, other=None, thorough=False):
2396
"""Reconcile this repository."""
2397
from bzrlib.reconcile import PackReconciler
2398
reconciler = PackReconciler(self, thorough=thorough)
2399
reconciler.reconcile()
2402
def _reconcile_pack(self, collection, packs, extension, revs, pb):
2403
packer = ReconcilePacker(collection, packs, extension, revs)
2404
return packer.pack(pb)
2406
@only_raises(errors.LockNotHeld, errors.LockBroken)
2408
if self._write_lock_count == 1 and self._write_group is not None:
2409
self.abort_write_group()
2410
self._transaction = None
2411
self._write_lock_count = 0
2412
raise errors.BzrError(
2413
'Must end write group before releasing write lock on %s'
2415
if self._write_lock_count:
2416
self._write_lock_count -= 1
2417
if not self._write_lock_count:
2418
transaction = self._transaction
2419
self._transaction = None
2420
transaction.finish()
2422
self.control_files.unlock()
2424
if not self.is_locked():
2425
for repo in self._fallback_repositories:
2429
class KnitPackStreamSource(StreamSource):
2430
"""A StreamSource used to transfer data between same-format KnitPack repos.
2432
This source assumes:
2433
1) Same serialization format for all objects
2434
2) Same root information
2435
3) XML format inventories
2436
4) Atomic inserts (so we can stream inventory texts before text
2441
def __init__(self, from_repository, to_format):
2442
super(KnitPackStreamSource, self).__init__(from_repository, to_format)
2443
self._text_keys = None
2444
self._text_fetch_order = 'unordered'
2446
def _get_filtered_inv_stream(self, revision_ids):
2447
from_repo = self.from_repository
2448
parent_ids = from_repo._find_parent_ids_of_revisions(revision_ids)
2449
parent_keys = [(p,) for p in parent_ids]
2450
find_text_keys = from_repo._find_text_key_references_from_xml_inventory_lines
2451
parent_text_keys = set(find_text_keys(
2452
from_repo._inventory_xml_lines_for_keys(parent_keys)))
2453
content_text_keys = set()
2454
knit = KnitVersionedFiles(None, None)
2455
factory = KnitPlainFactory()
2456
def find_text_keys_from_content(record):
2457
if record.storage_kind not in ('knit-delta-gz', 'knit-ft-gz'):
2458
raise ValueError("Unknown content storage kind for"
2459
" inventory text: %s" % (record.storage_kind,))
2460
# It's a knit record, it has a _raw_record field (even if it was
2461
# reconstituted from a network stream).
2462
raw_data = record._raw_record
2463
# read the entire thing
2464
revision_id = record.key[-1]
2465
content, _ = knit._parse_record(revision_id, raw_data)
2466
if record.storage_kind == 'knit-delta-gz':
2467
line_iterator = factory.get_linedelta_content(content)
2468
elif record.storage_kind == 'knit-ft-gz':
2469
line_iterator = factory.get_fulltext_content(content)
2470
content_text_keys.update(find_text_keys(
2471
[(line, revision_id) for line in line_iterator]))
2472
revision_keys = [(r,) for r in revision_ids]
2473
def _filtered_inv_stream():
2474
source_vf = from_repo.inventories
2475
stream = source_vf.get_record_stream(revision_keys,
2477
for record in stream:
2478
if record.storage_kind == 'absent':
2479
raise errors.NoSuchRevision(from_repo, record.key)
2480
find_text_keys_from_content(record)
2482
self._text_keys = content_text_keys - parent_text_keys
2483
return ('inventories', _filtered_inv_stream())
2485
def _get_text_stream(self):
2486
# Note: We know we don't have to handle adding root keys, because both
2487
# the source and target are the identical network name.
2488
text_stream = self.from_repository.texts.get_record_stream(
2489
self._text_keys, self._text_fetch_order, False)
2490
return ('texts', text_stream)
2492
def get_stream(self, search):
2493
revision_ids = search.get_keys()
2494
for stream_info in self._fetch_revision_texts(revision_ids):
2496
self._revision_keys = [(rev_id,) for rev_id in revision_ids]
2497
yield self._get_filtered_inv_stream(revision_ids)
2498
yield self._get_text_stream()
2502
class RepositoryFormatPack(MetaDirRepositoryFormat):
2503
"""Format logic for pack structured repositories.
2505
This repository format has:
2506
- a list of packs in pack-names
2507
- packs in packs/NAME.pack
2508
- indices in indices/NAME.{iix,six,tix,rix}
2509
- knit deltas in the packs, knit indices mapped to the indices.
2510
- thunk objects to support the knits programming API.
2511
- a format marker of its own
2512
- an optional 'shared-storage' flag
2513
- an optional 'no-working-trees' flag
2517
# Set this attribute in derived classes to control the repository class
2518
# created by open and initialize.
2519
repository_class = None
2520
# Set this attribute in derived classes to control the
2521
# _commit_builder_class that the repository objects will have passed to
2522
# their constructor.
2523
_commit_builder_class = None
2524
# Set this attribute in derived clases to control the _serializer that the
2525
# repository objects will have passed to their constructor.
2527
# Packs are not confused by ghosts.
2528
supports_ghosts = True
2529
# External references are not supported in pack repositories yet.
2530
supports_external_lookups = False
2531
# Most pack formats do not use chk lookups.
2532
supports_chks = False
2533
# What index classes to use
2534
index_builder_class = None
2536
_fetch_uses_deltas = True
2539
def initialize(self, a_bzrdir, shared=False):
2540
"""Create a pack based repository.
2542
:param a_bzrdir: bzrdir to contain the new repository; must already
2544
:param shared: If true the repository will be initialized as a shared
2547
mutter('creating repository in %s.', a_bzrdir.transport.base)
2548
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
2549
builder = self.index_builder_class()
2550
files = [('pack-names', builder.finish())]
2551
utf8_files = [('format', self.get_format_string())]
2553
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
2554
return self.open(a_bzrdir=a_bzrdir, _found=True)
2556
def open(self, a_bzrdir, _found=False, _override_transport=None):
2557
"""See RepositoryFormat.open().
2559
:param _override_transport: INTERNAL USE ONLY. Allows opening the
2560
repository at a slightly different url
2561
than normal. I.e. during 'upgrade'.
2564
format = RepositoryFormat.find_format(a_bzrdir)
2565
if _override_transport is not None:
2566
repo_transport = _override_transport
2568
repo_transport = a_bzrdir.get_repository_transport(None)
2569
control_files = lockable_files.LockableFiles(repo_transport,
2570
'lock', lockdir.LockDir)
2571
return self.repository_class(_format=self,
2573
control_files=control_files,
2574
_commit_builder_class=self._commit_builder_class,
2575
_serializer=self._serializer)
2578
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2579
"""A no-subtrees parameterized Pack repository.
2581
This format was introduced in 0.92.
2584
repository_class = KnitPackRepository
2585
_commit_builder_class = PackCommitBuilder
2587
def _serializer(self):
2588
return xml5.serializer_v5
2589
# What index classes to use
2590
index_builder_class = InMemoryGraphIndex
2591
index_class = GraphIndex
2593
def _get_matching_bzrdir(self):
2594
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2596
def _ignore_setting_bzrdir(self, format):
2599
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2601
def get_format_string(self):
2602
"""See RepositoryFormat.get_format_string()."""
2603
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2605
def get_format_description(self):
2606
"""See RepositoryFormat.get_format_description()."""
2607
return "Packs containing knits without subtree support"
2610
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2611
"""A subtrees parameterized Pack repository.
2613
This repository format uses the xml7 serializer to get:
2614
- support for recording full info about the tree root
2615
- support for recording tree-references
2617
This format was introduced in 0.92.
2620
repository_class = KnitPackRepository
2621
_commit_builder_class = PackRootCommitBuilder
2622
rich_root_data = True
2623
supports_tree_reference = True
2625
def _serializer(self):
2626
return xml7.serializer_v7
2627
# What index classes to use
2628
index_builder_class = InMemoryGraphIndex
2629
index_class = GraphIndex
2631
def _get_matching_bzrdir(self):
2632
return bzrdir.format_registry.make_bzrdir(
2633
'pack-0.92-subtree')
2635
def _ignore_setting_bzrdir(self, format):
2638
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2640
def get_format_string(self):
2641
"""See RepositoryFormat.get_format_string()."""
2642
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2644
def get_format_description(self):
2645
"""See RepositoryFormat.get_format_description()."""
2646
return "Packs containing knits with subtree support\n"
2649
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2650
"""A rich-root, no subtrees parameterized Pack repository.
2652
This repository format uses the xml6 serializer to get:
2653
- support for recording full info about the tree root
2655
This format was introduced in 1.0.
2658
repository_class = KnitPackRepository
2659
_commit_builder_class = PackRootCommitBuilder
2660
rich_root_data = True
2661
supports_tree_reference = False
2663
def _serializer(self):
2664
return xml6.serializer_v6
2665
# What index classes to use
2666
index_builder_class = InMemoryGraphIndex
2667
index_class = GraphIndex
2669
def _get_matching_bzrdir(self):
2670
return bzrdir.format_registry.make_bzrdir(
2673
def _ignore_setting_bzrdir(self, format):
2676
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2678
def get_format_string(self):
2679
"""See RepositoryFormat.get_format_string()."""
2680
return ("Bazaar pack repository format 1 with rich root"
2681
" (needs bzr 1.0)\n")
2683
def get_format_description(self):
2684
"""See RepositoryFormat.get_format_description()."""
2685
return "Packs containing knits with rich root support\n"
2688
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2689
"""Repository that supports external references to allow stacking.
2693
Supports external lookups, which results in non-truncated ghosts after
2694
reconcile compared to pack-0.92 formats.
2697
repository_class = KnitPackRepository
2698
_commit_builder_class = PackCommitBuilder
2699
supports_external_lookups = True
2700
# What index classes to use
2701
index_builder_class = InMemoryGraphIndex
2702
index_class = GraphIndex
2705
def _serializer(self):
2706
return xml5.serializer_v5
2708
def _get_matching_bzrdir(self):
2709
return bzrdir.format_registry.make_bzrdir('1.6')
2711
def _ignore_setting_bzrdir(self, format):
2714
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2716
def get_format_string(self):
2717
"""See RepositoryFormat.get_format_string()."""
2718
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2720
def get_format_description(self):
2721
"""See RepositoryFormat.get_format_description()."""
2722
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2725
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2726
"""A repository with rich roots and stacking.
2728
New in release 1.6.1.
2730
Supports stacking on other repositories, allowing data to be accessed
2731
without being stored locally.
2734
repository_class = KnitPackRepository
2735
_commit_builder_class = PackRootCommitBuilder
2736
rich_root_data = True
2737
supports_tree_reference = False # no subtrees
2738
supports_external_lookups = True
2739
# What index classes to use
2740
index_builder_class = InMemoryGraphIndex
2741
index_class = GraphIndex
2744
def _serializer(self):
2745
return xml6.serializer_v6
2747
def _get_matching_bzrdir(self):
2748
return bzrdir.format_registry.make_bzrdir(
2751
def _ignore_setting_bzrdir(self, format):
2754
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2756
def get_format_string(self):
2757
"""See RepositoryFormat.get_format_string()."""
2758
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2760
def get_format_description(self):
2761
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2764
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2765
"""A repository with rich roots and external references.
2769
Supports external lookups, which results in non-truncated ghosts after
2770
reconcile compared to pack-0.92 formats.
2772
This format was deprecated because the serializer it uses accidentally
2773
supported subtrees, when the format was not intended to. This meant that
2774
someone could accidentally fetch from an incorrect repository.
2777
repository_class = KnitPackRepository
2778
_commit_builder_class = PackRootCommitBuilder
2779
rich_root_data = True
2780
supports_tree_reference = False # no subtrees
2782
supports_external_lookups = True
2783
# What index classes to use
2784
index_builder_class = InMemoryGraphIndex
2785
index_class = GraphIndex
2788
def _serializer(self):
2789
return xml7.serializer_v7
2791
def _get_matching_bzrdir(self):
2792
matching = bzrdir.format_registry.make_bzrdir(
2794
matching.repository_format = self
2797
def _ignore_setting_bzrdir(self, format):
2800
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2802
def get_format_string(self):
2803
"""See RepositoryFormat.get_format_string()."""
2804
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2806
def get_format_description(self):
2807
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2811
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2812
"""A repository with stacking and btree indexes,
2813
without rich roots or subtrees.
2815
This is equivalent to pack-1.6 with B+Tree indices.
2818
repository_class = KnitPackRepository
2819
_commit_builder_class = PackCommitBuilder
2820
supports_external_lookups = True
2821
# What index classes to use
2822
index_builder_class = BTreeBuilder
2823
index_class = BTreeGraphIndex
2826
def _serializer(self):
2827
return xml5.serializer_v5
2829
def _get_matching_bzrdir(self):
2830
return bzrdir.format_registry.make_bzrdir('1.9')
2832
def _ignore_setting_bzrdir(self, format):
2835
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2837
def get_format_string(self):
2838
"""See RepositoryFormat.get_format_string()."""
2839
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2841
def get_format_description(self):
2842
"""See RepositoryFormat.get_format_description()."""
2843
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2846
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2847
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2849
1.6-rich-root with B+Tree indices.
2852
repository_class = KnitPackRepository
2853
_commit_builder_class = PackRootCommitBuilder
2854
rich_root_data = True
2855
supports_tree_reference = False # no subtrees
2856
supports_external_lookups = True
2857
# What index classes to use
2858
index_builder_class = BTreeBuilder
2859
index_class = BTreeGraphIndex
2862
def _serializer(self):
2863
return xml6.serializer_v6
2865
def _get_matching_bzrdir(self):
2866
return bzrdir.format_registry.make_bzrdir(
2869
def _ignore_setting_bzrdir(self, format):
2872
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2874
def get_format_string(self):
2875
"""See RepositoryFormat.get_format_string()."""
2876
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2878
def get_format_description(self):
2879
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2882
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2883
"""A subtrees development repository.
2885
This format should be retained until the second release after bzr 1.7.
2887
1.6.1-subtree[as it might have been] with B+Tree indices.
2889
This is [now] retained until we have a CHK based subtree format in
2893
repository_class = KnitPackRepository
2894
_commit_builder_class = PackRootCommitBuilder
2895
rich_root_data = True
2896
supports_tree_reference = True
2897
supports_external_lookups = True
2898
# What index classes to use
2899
index_builder_class = BTreeBuilder
2900
index_class = BTreeGraphIndex
2903
def _serializer(self):
2904
return xml7.serializer_v7
2906
def _get_matching_bzrdir(self):
2907
return bzrdir.format_registry.make_bzrdir(
2908
'development-subtree')
2910
def _ignore_setting_bzrdir(self, format):
2913
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2915
def get_format_string(self):
2916
"""See RepositoryFormat.get_format_string()."""
2917
return ("Bazaar development format 2 with subtree support "
2918
"(needs bzr.dev from before 1.8)\n")
2920
def get_format_description(self):
2921
"""See RepositoryFormat.get_format_description()."""
2922
return ("Development repository format, currently the same as "
2923
"1.6.1-subtree with B+Tree indices.\n")
added
removed
bzrlib/repofmt/pack_repo.py
