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
56
revision as _mod_revision,
59
from bzrlib.decorators import needs_write_lock, only_raises
60
from bzrlib.index import (
64
from bzrlib.lock import LogicalLockResult
65
from bzrlib.repofmt.knitrepo import KnitRepository
66
from bzrlib.repository import (
68
MetaDirRepositoryFormat,
70
RepositoryWriteLockResult,
74
from bzrlib.trace import (
81
class PackCommitBuilder(CommitBuilder):
82
"""A subclass of CommitBuilder to add texts with pack semantics.
84
Specifically this uses one knit object rather than one knit object per
85
added text, reducing memory and object pressure.
88
def __init__(self, repository, parents, config, timestamp=None,
89
timezone=None, committer=None, revprops=None,
91
CommitBuilder.__init__(self, repository, parents, config,
92
timestamp=timestamp, timezone=timezone, committer=committer,
93
revprops=revprops, revision_id=revision_id)
94
self._file_graph = graph.Graph(
95
repository._pack_collection.text_index.combined_index)
97
def _heads(self, file_id, revision_ids):
98
keys = [(file_id, revision_id) for revision_id in revision_ids]
99
return set([key[1] for key in self._file_graph.heads(keys)])
102
class PackRootCommitBuilder(RootCommitBuilder):
103
"""A subclass of RootCommitBuilder to add texts with pack semantics.
105
Specifically this uses one knit object rather than one knit object per
106
added text, reducing memory and object pressure.
109
def __init__(self, repository, parents, config, timestamp=None,
110
timezone=None, committer=None, revprops=None,
112
CommitBuilder.__init__(self, repository, parents, config,
113
timestamp=timestamp, timezone=timezone, committer=committer,
114
revprops=revprops, revision_id=revision_id)
115
self._file_graph = graph.Graph(
116
repository._pack_collection.text_index.combined_index)
118
def _heads(self, file_id, revision_ids):
119
keys = [(file_id, revision_id) for revision_id in revision_ids]
120
return set([key[1] for key in self._file_graph.heads(keys)])
124
"""An in memory proxy for a pack and its indices.
126
This is a base class that is not directly used, instead the classes
127
ExistingPack and NewPack are used.
130
# A map of index 'type' to the file extension and position in the
132
index_definitions = {
134
'revision': ('.rix', 0),
135
'inventory': ('.iix', 1),
137
'signature': ('.six', 3),
140
def __init__(self, revision_index, inventory_index, text_index,
141
signature_index, chk_index=None):
142
"""Create a pack instance.
144
:param revision_index: A GraphIndex for determining what revisions are
145
present in the Pack and accessing the locations of their texts.
146
:param inventory_index: A GraphIndex for determining what inventories are
147
present in the Pack and accessing the locations of their
149
:param text_index: A GraphIndex for determining what file texts
150
are present in the pack and accessing the locations of their
151
texts/deltas (via (fileid, revisionid) tuples).
152
:param signature_index: A GraphIndex for determining what signatures are
153
present in the Pack and accessing the locations of their texts.
154
:param chk_index: A GraphIndex for accessing content by CHK, if the
157
self.revision_index = revision_index
158
self.inventory_index = inventory_index
159
self.text_index = text_index
160
self.signature_index = signature_index
161
self.chk_index = chk_index
163
def access_tuple(self):
164
"""Return a tuple (transport, name) for the pack content."""
165
return self.pack_transport, self.file_name()
167
def _check_references(self):
168
"""Make sure our external references are present.
170
Packs are allowed to have deltas whose base is not in the pack, but it
171
must be present somewhere in this collection. It is not allowed to
172
have deltas based on a fallback repository.
173
(See <https://bugs.launchpad.net/bzr/+bug/288751>)
176
for (index_name, external_refs, index) in [
178
self._get_external_refs(self.text_index),
179
self._pack_collection.text_index.combined_index),
181
self._get_external_refs(self.inventory_index),
182
self._pack_collection.inventory_index.combined_index),
184
missing = external_refs.difference(
185
k for (idx, k, v, r) in
186
index.iter_entries(external_refs))
188
missing_items[index_name] = sorted(list(missing))
190
from pprint import pformat
191
raise errors.BzrCheckError(
192
"Newly created pack file %r has delta references to "
193
"items not in its repository:\n%s"
194
% (self, pformat(missing_items)))
197
"""Get the file name for the pack on disk."""
198
return self.name + '.pack'
200
def get_revision_count(self):
201
return self.revision_index.key_count()
203
def index_name(self, index_type, name):
204
"""Get the disk name of an index type for pack name 'name'."""
205
return name + Pack.index_definitions[index_type][0]
207
def index_offset(self, index_type):
208
"""Get the position in a index_size array for a given index type."""
209
return Pack.index_definitions[index_type][1]
211
def inventory_index_name(self, name):
212
"""The inv index is the name + .iix."""
213
return self.index_name('inventory', name)
215
def revision_index_name(self, name):
216
"""The revision index is the name + .rix."""
217
return self.index_name('revision', name)
219
def signature_index_name(self, name):
220
"""The signature index is the name + .six."""
221
return self.index_name('signature', name)
223
def text_index_name(self, name):
224
"""The text index is the name + .tix."""
225
return self.index_name('text', name)
227
def _replace_index_with_readonly(self, index_type):
228
unlimited_cache = False
229
if index_type == 'chk':
230
unlimited_cache = True
231
index = self.index_class(self.index_transport,
232
self.index_name(index_type, self.name),
233
self.index_sizes[self.index_offset(index_type)],
234
unlimited_cache=unlimited_cache)
235
if index_type == 'chk':
236
index._leaf_factory = btree_index._gcchk_factory
237
setattr(self, index_type + '_index', index)
240
class ExistingPack(Pack):
241
"""An in memory proxy for an existing .pack and its disk indices."""
243
def __init__(self, pack_transport, name, revision_index, inventory_index,
244
text_index, signature_index, chk_index=None):
245
"""Create an ExistingPack object.
247
:param pack_transport: The transport where the pack file resides.
248
:param name: The name of the pack on disk in the pack_transport.
250
Pack.__init__(self, revision_index, inventory_index, text_index,
251
signature_index, chk_index)
253
self.pack_transport = pack_transport
254
if None in (revision_index, inventory_index, text_index,
255
signature_index, name, pack_transport):
256
raise AssertionError()
258
def __eq__(self, other):
259
return self.__dict__ == other.__dict__
261
def __ne__(self, other):
262
return not self.__eq__(other)
265
return "<%s.%s object at 0x%x, %s, %s" % (
266
self.__class__.__module__, self.__class__.__name__, id(self),
267
self.pack_transport, self.name)
270
class ResumedPack(ExistingPack):
272
def __init__(self, name, revision_index, inventory_index, text_index,
273
signature_index, upload_transport, pack_transport, index_transport,
274
pack_collection, chk_index=None):
275
"""Create a ResumedPack object."""
276
ExistingPack.__init__(self, pack_transport, name, revision_index,
277
inventory_index, text_index, signature_index,
279
self.upload_transport = upload_transport
280
self.index_transport = index_transport
281
self.index_sizes = [None, None, None, None]
283
('revision', revision_index),
284
('inventory', inventory_index),
285
('text', text_index),
286
('signature', signature_index),
288
if chk_index is not None:
289
indices.append(('chk', chk_index))
290
self.index_sizes.append(None)
291
for index_type, index in indices:
292
offset = self.index_offset(index_type)
293
self.index_sizes[offset] = index._size
294
self.index_class = pack_collection._index_class
295
self._pack_collection = pack_collection
296
self._state = 'resumed'
297
# XXX: perhaps check that the .pack file exists?
299
def access_tuple(self):
300
if self._state == 'finished':
301
return Pack.access_tuple(self)
302
elif self._state == 'resumed':
303
return self.upload_transport, self.file_name()
305
raise AssertionError(self._state)
308
self.upload_transport.delete(self.file_name())
309
indices = [self.revision_index, self.inventory_index, self.text_index,
310
self.signature_index]
311
if self.chk_index is not None:
312
indices.append(self.chk_index)
313
for index in indices:
314
index._transport.delete(index._name)
317
self._check_references()
318
index_types = ['revision', 'inventory', 'text', 'signature']
319
if self.chk_index is not None:
320
index_types.append('chk')
321
for index_type in index_types:
322
old_name = self.index_name(index_type, self.name)
323
new_name = '../indices/' + old_name
324
self.upload_transport.rename(old_name, new_name)
325
self._replace_index_with_readonly(index_type)
326
new_name = '../packs/' + self.file_name()
327
self.upload_transport.rename(self.file_name(), new_name)
328
self._state = 'finished'
330
def _get_external_refs(self, index):
331
"""Return compression parents for this index that are not present.
333
This returns any compression parents that are referenced by this index,
334
which are not contained *in* this index. They may be present elsewhere.
336
return index.external_references(1)
340
"""An in memory proxy for a pack which is being created."""
342
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
343
"""Create a NewPack instance.
345
:param pack_collection: A PackCollection into which this is being inserted.
346
:param upload_suffix: An optional suffix to be given to any temporary
347
files created during the pack creation. e.g '.autopack'
348
:param file_mode: Unix permissions for newly created file.
350
# The relative locations of the packs are constrained, but all are
351
# passed in because the caller has them, so as to avoid object churn.
352
index_builder_class = pack_collection._index_builder_class
353
if pack_collection.chk_index is not None:
354
chk_index = index_builder_class(reference_lists=0)
358
# Revisions: parents list, no text compression.
359
index_builder_class(reference_lists=1),
360
# Inventory: We want to map compression only, but currently the
361
# knit code hasn't been updated enough to understand that, so we
362
# have a regular 2-list index giving parents and compression
364
index_builder_class(reference_lists=2),
365
# Texts: compression and per file graph, for all fileids - so two
366
# reference lists and two elements in the key tuple.
367
index_builder_class(reference_lists=2, key_elements=2),
368
# Signatures: Just blobs to store, no compression, no parents
370
index_builder_class(reference_lists=0),
371
# CHK based storage - just blobs, no compression or parents.
374
self._pack_collection = pack_collection
375
# When we make readonly indices, we need this.
376
self.index_class = pack_collection._index_class
377
# where should the new pack be opened
378
self.upload_transport = pack_collection._upload_transport
379
# where are indices written out to
380
self.index_transport = pack_collection._index_transport
381
# where is the pack renamed to when it is finished?
382
self.pack_transport = pack_collection._pack_transport
383
# What file mode to upload the pack and indices with.
384
self._file_mode = file_mode
385
# tracks the content written to the .pack file.
386
self._hash = osutils.md5()
387
# a tuple with the length in bytes of the indices, once the pack
388
# is finalised. (rev, inv, text, sigs, chk_if_in_use)
389
self.index_sizes = None
390
# How much data to cache when writing packs. Note that this is not
391
# synchronised with reads, because it's not in the transport layer, so
392
# is not safe unless the client knows it won't be reading from the pack
394
self._cache_limit = 0
395
# the temporary pack file name.
396
self.random_name = osutils.rand_chars(20) + upload_suffix
397
# when was this pack started ?
398
self.start_time = time.time()
399
# open an output stream for the data added to the pack.
400
self.write_stream = self.upload_transport.open_write_stream(
401
self.random_name, mode=self._file_mode)
402
if 'pack' in debug.debug_flags:
403
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
404
time.ctime(), self.upload_transport.base, self.random_name,
405
time.time() - self.start_time)
406
# A list of byte sequences to be written to the new pack, and the
407
# aggregate size of them. Stored as a list rather than separate
408
# variables so that the _write_data closure below can update them.
409
self._buffer = [[], 0]
410
# create a callable for adding data
412
# robertc says- this is a closure rather than a method on the object
413
# so that the variables are locals, and faster than accessing object
415
def _write_data(bytes, flush=False, _buffer=self._buffer,
416
_write=self.write_stream.write, _update=self._hash.update):
417
_buffer[0].append(bytes)
418
_buffer[1] += len(bytes)
420
if _buffer[1] > self._cache_limit or flush:
421
bytes = ''.join(_buffer[0])
425
# expose this on self, for the occasion when clients want to add data.
426
self._write_data = _write_data
427
# a pack writer object to serialise pack records.
428
self._writer = pack.ContainerWriter(self._write_data)
430
# what state is the pack in? (open, finished, aborted)
432
# no name until we finish writing the content
436
"""Cancel creating this pack."""
437
self._state = 'aborted'
438
self.write_stream.close()
439
# Remove the temporary pack file.
440
self.upload_transport.delete(self.random_name)
441
# The indices have no state on disk.
443
def access_tuple(self):
444
"""Return a tuple (transport, name) for the pack content."""
445
if self._state == 'finished':
446
return Pack.access_tuple(self)
447
elif self._state == 'open':
448
return self.upload_transport, self.random_name
450
raise AssertionError(self._state)
452
def data_inserted(self):
453
"""True if data has been added to this pack."""
454
return bool(self.get_revision_count() or
455
self.inventory_index.key_count() or
456
self.text_index.key_count() or
457
self.signature_index.key_count() or
458
(self.chk_index is not None and self.chk_index.key_count()))
460
def finish_content(self):
461
if self.name is not None:
465
self._write_data('', flush=True)
466
self.name = self._hash.hexdigest()
468
def finish(self, suspend=False):
469
"""Finish the new pack.
472
- finalises the content
473
- assigns a name (the md5 of the content, currently)
474
- writes out the associated indices
475
- renames the pack into place.
476
- stores the index size tuple for the pack in the index_sizes
479
self.finish_content()
481
self._check_references()
483
# XXX: It'd be better to write them all to temporary names, then
484
# rename them all into place, so that the window when only some are
485
# visible is smaller. On the other hand none will be seen until
486
# they're in the names list.
487
self.index_sizes = [None, None, None, None]
488
self._write_index('revision', self.revision_index, 'revision', suspend)
489
self._write_index('inventory', self.inventory_index, 'inventory',
491
self._write_index('text', self.text_index, 'file texts', suspend)
492
self._write_index('signature', self.signature_index,
493
'revision signatures', suspend)
494
if self.chk_index is not None:
495
self.index_sizes.append(None)
496
self._write_index('chk', self.chk_index,
497
'content hash bytes', suspend)
498
self.write_stream.close()
499
# Note that this will clobber an existing pack with the same name,
500
# without checking for hash collisions. While this is undesirable this
501
# is something that can be rectified in a subsequent release. One way
502
# to rectify it may be to leave the pack at the original name, writing
503
# its pack-names entry as something like 'HASH: index-sizes
504
# temporary-name'. Allocate that and check for collisions, if it is
505
# collision free then rename it into place. If clients know this scheme
506
# they can handle missing-file errors by:
507
# - try for HASH.pack
508
# - try for temporary-name
509
# - refresh the pack-list to see if the pack is now absent
510
new_name = self.name + '.pack'
512
new_name = '../packs/' + new_name
513
self.upload_transport.rename(self.random_name, new_name)
514
self._state = 'finished'
515
if 'pack' in debug.debug_flags:
516
# XXX: size might be interesting?
517
mutter('%s: create_pack: pack finished: %s%s->%s t+%6.3fs',
518
time.ctime(), self.upload_transport.base, self.random_name,
519
new_name, time.time() - self.start_time)
522
"""Flush any current data."""
524
bytes = ''.join(self._buffer[0])
525
self.write_stream.write(bytes)
526
self._hash.update(bytes)
527
self._buffer[:] = [[], 0]
529
def _get_external_refs(self, index):
530
return index._external_references()
532
def set_write_cache_size(self, size):
533
self._cache_limit = size
535
def _write_index(self, index_type, index, label, suspend=False):
536
"""Write out an index.
538
:param index_type: The type of index to write - e.g. 'revision'.
539
:param index: The index object to serialise.
540
:param label: What label to give the index e.g. 'revision'.
542
index_name = self.index_name(index_type, self.name)
544
transport = self.upload_transport
546
transport = self.index_transport
547
self.index_sizes[self.index_offset(index_type)] = transport.put_file(
548
index_name, index.finish(), mode=self._file_mode)
549
if 'pack' in debug.debug_flags:
550
# XXX: size might be interesting?
551
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
552
time.ctime(), label, self.upload_transport.base,
553
self.random_name, time.time() - self.start_time)
554
# Replace the writable index on this object with a readonly,
555
# presently unloaded index. We should alter
556
# the index layer to make its finish() error if add_node is
557
# subsequently used. RBC
558
self._replace_index_with_readonly(index_type)
561
class AggregateIndex(object):
562
"""An aggregated index for the RepositoryPackCollection.
564
AggregateIndex is reponsible for managing the PackAccess object,
565
Index-To-Pack mapping, and all indices list for a specific type of index
566
such as 'revision index'.
568
A CombinedIndex provides an index on a single key space built up
569
from several on-disk indices. The AggregateIndex builds on this
570
to provide a knit access layer, and allows having up to one writable
571
index within the collection.
573
# XXX: Probably 'can be written to' could/should be separated from 'acts
574
# like a knit index' -- mbp 20071024
576
def __init__(self, reload_func=None, flush_func=None):
577
"""Create an AggregateIndex.
579
:param reload_func: A function to call if we find we are missing an
580
index. Should have the form reload_func() => True if the list of
581
active pack files has changed.
583
self._reload_func = reload_func
584
self.index_to_pack = {}
585
self.combined_index = CombinedGraphIndex([], reload_func=reload_func)
586
self.data_access = _DirectPackAccess(self.index_to_pack,
587
reload_func=reload_func,
588
flush_func=flush_func)
589
self.add_callback = None
591
def add_index(self, index, pack):
592
"""Add index to the aggregate, which is an index for Pack pack.
594
Future searches on the aggregate index will seach this new index
595
before all previously inserted indices.
597
:param index: An Index for the pack.
598
:param pack: A Pack instance.
600
# expose it to the index map
601
self.index_to_pack[index] = pack.access_tuple()
602
# put it at the front of the linear index list
603
self.combined_index.insert_index(0, index, pack.name)
605
def add_writable_index(self, index, pack):
606
"""Add an index which is able to have data added to it.
608
There can be at most one writable index at any time. Any
609
modifications made to the knit are put into this index.
611
:param index: An index from the pack parameter.
612
:param pack: A Pack instance.
614
if self.add_callback is not None:
615
raise AssertionError(
616
"%s already has a writable index through %s" % \
617
(self, self.add_callback))
618
# allow writing: queue writes to a new index
619
self.add_index(index, pack)
620
# Updates the index to packs mapping as a side effect,
621
self.data_access.set_writer(pack._writer, index, pack.access_tuple())
622
self.add_callback = index.add_nodes
625
"""Reset all the aggregate data to nothing."""
626
self.data_access.set_writer(None, None, (None, None))
627
self.index_to_pack.clear()
628
del self.combined_index._indices[:]
629
del self.combined_index._index_names[:]
630
self.add_callback = None
632
def remove_index(self, index):
633
"""Remove index from the indices used to answer queries.
635
:param index: An index from the pack parameter.
637
del self.index_to_pack[index]
638
pos = self.combined_index._indices.index(index)
639
del self.combined_index._indices[pos]
640
del self.combined_index._index_names[pos]
641
if (self.add_callback is not None and
642
getattr(index, 'add_nodes', None) == self.add_callback):
643
self.add_callback = None
644
self.data_access.set_writer(None, None, (None, None))
647
class Packer(object):
648
"""Create a pack from packs."""
650
def __init__(self, pack_collection, packs, suffix, revision_ids=None,
654
:param pack_collection: A RepositoryPackCollection object where the
655
new pack is being written to.
656
:param packs: The packs to combine.
657
:param suffix: The suffix to use on the temporary files for the pack.
658
:param revision_ids: Revision ids to limit the pack to.
659
:param reload_func: A function to call if a pack file/index goes
660
missing. The side effect of calling this function should be to
661
update self.packs. See also AggregateIndex
665
self.revision_ids = revision_ids
666
# The pack object we are creating.
668
self._pack_collection = pack_collection
669
self._reload_func = reload_func
670
# The index layer keys for the revisions being copied. None for 'all
672
self._revision_keys = None
673
# What text keys to copy. None for 'all texts'. This is set by
674
# _copy_inventory_texts
675
self._text_filter = None
678
def _extra_init(self):
679
"""A template hook to allow extending the constructor trivially."""
681
def _pack_map_and_index_list(self, index_attribute):
682
"""Convert a list of packs to an index pack map and index list.
684
:param index_attribute: The attribute that the desired index is found
686
:return: A tuple (map, list) where map contains the dict from
687
index:pack_tuple, and list contains the indices in the preferred
692
for pack_obj in self.packs:
693
index = getattr(pack_obj, index_attribute)
694
indices.append(index)
695
pack_map[index] = pack_obj
696
return pack_map, indices
698
def _index_contents(self, indices, key_filter=None):
699
"""Get an iterable of the index contents from a pack_map.
701
:param indices: The list of indices to query
702
:param key_filter: An optional filter to limit the keys returned.
704
all_index = CombinedGraphIndex(indices)
705
if key_filter is None:
706
return all_index.iter_all_entries()
708
return all_index.iter_entries(key_filter)
710
def pack(self, pb=None):
711
"""Create a new pack by reading data from other packs.
713
This does little more than a bulk copy of data. One key difference
714
is that data with the same item key across multiple packs is elided
715
from the output. The new pack is written into the current pack store
716
along with its indices, and the name added to the pack names. The
717
source packs are not altered and are not required to be in the current
720
:param pb: An optional progress bar to use. A nested bar is created if
722
:return: A Pack object, or None if nothing was copied.
724
# open a pack - using the same name as the last temporary file
725
# - which has already been flushed, so it's safe.
726
# XXX: - duplicate code warning with start_write_group; fix before
727
# considering 'done'.
728
if self._pack_collection._new_pack is not None:
729
raise errors.BzrError('call to %s.pack() while another pack is'
731
% (self.__class__.__name__,))
732
if self.revision_ids is not None:
733
if len(self.revision_ids) == 0:
734
# silly fetch request.
737
self.revision_ids = frozenset(self.revision_ids)
738
self.revision_keys = frozenset((revid,) for revid in
741
self.pb = ui.ui_factory.nested_progress_bar()
745
return self._create_pack_from_packs()
751
"""Open a pack for the pack we are creating."""
752
new_pack = self._pack_collection.pack_factory(self._pack_collection,
753
upload_suffix=self.suffix,
754
file_mode=self._pack_collection.repo.bzrdir._get_file_mode())
755
# We know that we will process all nodes in order, and don't need to
756
# query, so don't combine any indices spilled to disk until we are done
757
new_pack.revision_index.set_optimize(combine_backing_indices=False)
758
new_pack.inventory_index.set_optimize(combine_backing_indices=False)
759
new_pack.text_index.set_optimize(combine_backing_indices=False)
760
new_pack.signature_index.set_optimize(combine_backing_indices=False)
763
def _update_pack_order(self, entries, index_to_pack_map):
764
"""Determine how we want our packs to be ordered.
766
This changes the sort order of the self.packs list so that packs unused
767
by 'entries' will be at the end of the list, so that future requests
768
can avoid probing them. Used packs will be at the front of the
769
self.packs list, in the order of their first use in 'entries'.
771
:param entries: A list of (index, ...) tuples
772
:param index_to_pack_map: A mapping from index objects to pack objects.
776
for entry in entries:
778
if index not in seen_indexes:
779
packs.append(index_to_pack_map[index])
780
seen_indexes.add(index)
781
if len(packs) == len(self.packs):
782
if 'pack' in debug.debug_flags:
783
mutter('Not changing pack list, all packs used.')
785
seen_packs = set(packs)
786
for pack in self.packs:
787
if pack not in seen_packs:
790
if 'pack' in debug.debug_flags:
791
old_names = [p.access_tuple()[1] for p in self.packs]
792
new_names = [p.access_tuple()[1] for p in packs]
793
mutter('Reordering packs\nfrom: %s\n to: %s',
794
old_names, new_names)
797
def _copy_revision_texts(self):
798
"""Copy revision data to the new pack."""
800
if self.revision_ids:
801
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
804
# select revision keys
805
revision_index_map, revision_indices = self._pack_map_and_index_list(
807
revision_nodes = self._index_contents(revision_indices, revision_keys)
808
revision_nodes = list(revision_nodes)
809
self._update_pack_order(revision_nodes, revision_index_map)
810
# copy revision keys and adjust values
811
self.pb.update("Copying revision texts", 1)
812
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
813
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
814
self.new_pack.revision_index, readv_group_iter, total_items))
815
if 'pack' in debug.debug_flags:
816
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
817
time.ctime(), self._pack_collection._upload_transport.base,
818
self.new_pack.random_name,
819
self.new_pack.revision_index.key_count(),
820
time.time() - self.new_pack.start_time)
821
self._revision_keys = revision_keys
823
def _copy_inventory_texts(self):
824
"""Copy the inventory texts to the new pack.
826
self._revision_keys is used to determine what inventories to copy.
828
Sets self._text_filter appropriately.
830
# select inventory keys
831
inv_keys = self._revision_keys # currently the same keyspace, and note that
832
# querying for keys here could introduce a bug where an inventory item
833
# is missed, so do not change it to query separately without cross
834
# checking like the text key check below.
835
inventory_index_map, inventory_indices = self._pack_map_and_index_list(
837
inv_nodes = self._index_contents(inventory_indices, inv_keys)
838
# copy inventory keys and adjust values
839
# XXX: Should be a helper function to allow different inv representation
841
self.pb.update("Copying inventory texts", 2)
842
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
843
# Only grab the output lines if we will be processing them
844
output_lines = bool(self.revision_ids)
845
inv_lines = self._copy_nodes_graph(inventory_index_map,
846
self.new_pack._writer, self.new_pack.inventory_index,
847
readv_group_iter, total_items, output_lines=output_lines)
848
if self.revision_ids:
849
self._process_inventory_lines(inv_lines)
851
# eat the iterator to cause it to execute.
853
self._text_filter = None
854
if 'pack' in debug.debug_flags:
855
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
856
time.ctime(), self._pack_collection._upload_transport.base,
857
self.new_pack.random_name,
858
self.new_pack.inventory_index.key_count(),
859
time.time() - self.new_pack.start_time)
861
def _copy_text_texts(self):
863
text_index_map, text_nodes = self._get_text_nodes()
864
if self._text_filter is not None:
865
# We could return the keys copied as part of the return value from
866
# _copy_nodes_graph but this doesn't work all that well with the
867
# need to get line output too, so we check separately, and as we're
868
# going to buffer everything anyway, we check beforehand, which
869
# saves reading knit data over the wire when we know there are
871
text_nodes = set(text_nodes)
872
present_text_keys = set(_node[1] for _node in text_nodes)
873
missing_text_keys = set(self._text_filter) - present_text_keys
874
if missing_text_keys:
875
# TODO: raise a specific error that can handle many missing
877
mutter("missing keys during fetch: %r", missing_text_keys)
878
a_missing_key = missing_text_keys.pop()
879
raise errors.RevisionNotPresent(a_missing_key[1],
881
# copy text keys and adjust values
882
self.pb.update("Copying content texts", 3)
883
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
884
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
885
self.new_pack.text_index, readv_group_iter, total_items))
886
self._log_copied_texts()
888
def _create_pack_from_packs(self):
889
self.pb.update("Opening pack", 0, 5)
890
self.new_pack = self.open_pack()
891
new_pack = self.new_pack
892
# buffer data - we won't be reading-back during the pack creation and
893
# this makes a significant difference on sftp pushes.
894
new_pack.set_write_cache_size(1024*1024)
895
if 'pack' in debug.debug_flags:
896
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
897
for a_pack in self.packs]
898
if self.revision_ids is not None:
899
rev_count = len(self.revision_ids)
902
mutter('%s: create_pack: creating pack from source packs: '
903
'%s%s %s revisions wanted %s t=0',
904
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
905
plain_pack_list, rev_count)
906
self._copy_revision_texts()
907
self._copy_inventory_texts()
908
self._copy_text_texts()
909
# select signature keys
910
signature_filter = self._revision_keys # same keyspace
911
signature_index_map, signature_indices = self._pack_map_and_index_list(
913
signature_nodes = self._index_contents(signature_indices,
915
# copy signature keys and adjust values
916
self.pb.update("Copying signature texts", 4)
917
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
918
new_pack.signature_index)
919
if 'pack' in debug.debug_flags:
920
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
921
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
922
new_pack.signature_index.key_count(),
923
time.time() - new_pack.start_time)
925
# NB XXX: how to check CHK references are present? perhaps by yielding
926
# the items? How should that interact with stacked repos?
927
if new_pack.chk_index is not None:
929
if 'pack' in debug.debug_flags:
930
mutter('%s: create_pack: chk content copied: %s%s %d items t+%6.3fs',
931
time.ctime(), self._pack_collection._upload_transport.base,
932
new_pack.random_name,
933
new_pack.chk_index.key_count(),
934
time.time() - new_pack.start_time)
935
new_pack._check_references()
936
if not self._use_pack(new_pack):
939
self.pb.update("Finishing pack", 5)
941
self._pack_collection.allocate(new_pack)
944
def _copy_chks(self, refs=None):
945
# XXX: Todo, recursive follow-pointers facility when fetching some
947
chk_index_map, chk_indices = self._pack_map_and_index_list(
949
chk_nodes = self._index_contents(chk_indices, refs)
951
# TODO: This isn't strictly tasteful as we are accessing some private
952
# variables (_serializer). Perhaps a better way would be to have
953
# Repository._deserialise_chk_node()
954
search_key_func = chk_map.search_key_registry.get(
955
self._pack_collection.repo._serializer.search_key_name)
956
def accumlate_refs(lines):
957
# XXX: move to a generic location
959
bytes = ''.join(lines)
960
node = chk_map._deserialise(bytes, ("unknown",), search_key_func)
961
new_refs.update(node.refs())
962
self._copy_nodes(chk_nodes, chk_index_map, self.new_pack._writer,
963
self.new_pack.chk_index, output_lines=accumlate_refs)
966
def _copy_nodes(self, nodes, index_map, writer, write_index,
968
"""Copy knit nodes between packs with no graph references.
970
:param output_lines: Output full texts of copied items.
972
pb = ui.ui_factory.nested_progress_bar()
974
return self._do_copy_nodes(nodes, index_map, writer,
975
write_index, pb, output_lines=output_lines)
979
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb,
981
# for record verification
982
knit = KnitVersionedFiles(None, None)
983
# plan a readv on each source pack:
985
nodes = sorted(nodes)
986
# how to map this into knit.py - or knit.py into this?
987
# we don't want the typical knit logic, we want grouping by pack
988
# at this point - perhaps a helper library for the following code
989
# duplication points?
991
for index, key, value in nodes:
992
if index not in request_groups:
993
request_groups[index] = []
994
request_groups[index].append((key, value))
996
pb.update("Copied record", record_index, len(nodes))
997
for index, items in request_groups.iteritems():
998
pack_readv_requests = []
999
for key, value in items:
1000
# ---- KnitGraphIndex.get_position
1001
bits = value[1:].split(' ')
1002
offset, length = int(bits[0]), int(bits[1])
1003
pack_readv_requests.append((offset, length, (key, value[0])))
1004
# linear scan up the pack
1005
pack_readv_requests.sort()
1007
pack_obj = index_map[index]
1008
transport, path = pack_obj.access_tuple()
1010
reader = pack.make_readv_reader(transport, path,
1011
[offset[0:2] for offset in pack_readv_requests])
1012
except errors.NoSuchFile:
1013
if self._reload_func is not None:
1016
for (names, read_func), (_1, _2, (key, eol_flag)) in \
1017
izip(reader.iter_records(), pack_readv_requests):
1018
raw_data = read_func(None)
1019
# check the header only
1020
if output_lines is not None:
1021
output_lines(knit._parse_record(key[-1], raw_data)[0])
1023
df, _ = knit._parse_record_header(key, raw_data)
1025
pos, size = writer.add_bytes_record(raw_data, names)
1026
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
1027
pb.update("Copied record", record_index)
1030
def _copy_nodes_graph(self, index_map, writer, write_index,
1031
readv_group_iter, total_items, output_lines=False):
1032
"""Copy knit nodes between packs.
1034
:param output_lines: Return lines present in the copied data as
1035
an iterator of line,version_id.
1037
pb = ui.ui_factory.nested_progress_bar()
1039
for result in self._do_copy_nodes_graph(index_map, writer,
1040
write_index, output_lines, pb, readv_group_iter, total_items):
1043
# Python 2.4 does not permit try:finally: in a generator.
1049
def _do_copy_nodes_graph(self, index_map, writer, write_index,
1050
output_lines, pb, readv_group_iter, total_items):
1051
# for record verification
1052
knit = KnitVersionedFiles(None, None)
1053
# for line extraction when requested (inventories only)
1055
factory = KnitPlainFactory()
1057
pb.update("Copied record", record_index, total_items)
1058
for index, readv_vector, node_vector in readv_group_iter:
1060
pack_obj = index_map[index]
1061
transport, path = pack_obj.access_tuple()
1063
reader = pack.make_readv_reader(transport, path, readv_vector)
1064
except errors.NoSuchFile:
1065
if self._reload_func is not None:
1068
for (names, read_func), (key, eol_flag, references) in \
1069
izip(reader.iter_records(), node_vector):
1070
raw_data = read_func(None)
1072
# read the entire thing
1073
content, _ = knit._parse_record(key[-1], raw_data)
1074
if len(references[-1]) == 0:
1075
line_iterator = factory.get_fulltext_content(content)
1077
line_iterator = factory.get_linedelta_content(content)
1078
for line in line_iterator:
1081
# check the header only
1082
df, _ = knit._parse_record_header(key, raw_data)
1084
pos, size = writer.add_bytes_record(raw_data, names)
1085
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1086
pb.update("Copied record", record_index)
1089
def _get_text_nodes(self):
1090
text_index_map, text_indices = self._pack_map_and_index_list(
1092
return text_index_map, self._index_contents(text_indices,
1095
def _least_readv_node_readv(self, nodes):
1096
"""Generate request groups for nodes using the least readv's.
1098
:param nodes: An iterable of graph index nodes.
1099
:return: Total node count and an iterator of the data needed to perform
1100
readvs to obtain the data for nodes. Each item yielded by the
1101
iterator is a tuple with:
1102
index, readv_vector, node_vector. readv_vector is a list ready to
1103
hand to the transport readv method, and node_vector is a list of
1104
(key, eol_flag, references) for the node retrieved by the
1105
matching readv_vector.
1107
# group by pack so we do one readv per pack
1108
nodes = sorted(nodes)
1111
for index, key, value, references in nodes:
1112
if index not in request_groups:
1113
request_groups[index] = []
1114
request_groups[index].append((key, value, references))
1116
for index, items in request_groups.iteritems():
1117
pack_readv_requests = []
1118
for key, value, references in items:
1119
# ---- KnitGraphIndex.get_position
1120
bits = value[1:].split(' ')
1121
offset, length = int(bits[0]), int(bits[1])
1122
pack_readv_requests.append(
1123
((offset, length), (key, value[0], references)))
1124
# linear scan up the pack to maximum range combining.
1125
pack_readv_requests.sort()
1126
# split out the readv and the node data.
1127
pack_readv = [readv for readv, node in pack_readv_requests]
1128
node_vector = [node for readv, node in pack_readv_requests]
1129
result.append((index, pack_readv, node_vector))
1130
return total, result
1132
def _log_copied_texts(self):
1133
if 'pack' in debug.debug_flags:
1134
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
1135
time.ctime(), self._pack_collection._upload_transport.base,
1136
self.new_pack.random_name,
1137
self.new_pack.text_index.key_count(),
1138
time.time() - self.new_pack.start_time)
1140
def _process_inventory_lines(self, inv_lines):
1141
"""Use up the inv_lines generator and setup a text key filter."""
1142
repo = self._pack_collection.repo
1143
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
1144
inv_lines, self.revision_keys)
1146
for fileid, file_revids in fileid_revisions.iteritems():
1147
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
1148
self._text_filter = text_filter
1150
def _revision_node_readv(self, revision_nodes):
1151
"""Return the total revisions and the readv's to issue.
1153
:param revision_nodes: The revision index contents for the packs being
1154
incorporated into the new pack.
1155
:return: As per _least_readv_node_readv.
1157
return self._least_readv_node_readv(revision_nodes)
1159
def _use_pack(self, new_pack):
1160
"""Return True if new_pack should be used.
1162
:param new_pack: The pack that has just been created.
1163
:return: True if the pack should be used.
1165
return new_pack.data_inserted()
1168
class OptimisingPacker(Packer):
1169
"""A packer which spends more time to create better disk layouts."""
1171
def _revision_node_readv(self, revision_nodes):
1172
"""Return the total revisions and the readv's to issue.
1174
This sort places revisions in topological order with the ancestors
1177
:param revision_nodes: The revision index contents for the packs being
1178
incorporated into the new pack.
1179
:return: As per _least_readv_node_readv.
1181
# build an ancestors dict
1184
for index, key, value, references in revision_nodes:
1185
ancestors[key] = references[0]
1186
by_key[key] = (index, value, references)
1187
order = tsort.topo_sort(ancestors)
1189
# Single IO is pathological, but it will work as a starting point.
1191
for key in reversed(order):
1192
index, value, references = by_key[key]
1193
# ---- KnitGraphIndex.get_position
1194
bits = value[1:].split(' ')
1195
offset, length = int(bits[0]), int(bits[1])
1197
(index, [(offset, length)], [(key, value[0], references)]))
1198
# TODO: combine requests in the same index that are in ascending order.
1199
return total, requests
1201
def open_pack(self):
1202
"""Open a pack for the pack we are creating."""
1203
new_pack = super(OptimisingPacker, self).open_pack()
1204
# Turn on the optimization flags for all the index builders.
1205
new_pack.revision_index.set_optimize(for_size=True)
1206
new_pack.inventory_index.set_optimize(for_size=True)
1207
new_pack.text_index.set_optimize(for_size=True)
1208
new_pack.signature_index.set_optimize(for_size=True)
1212
class ReconcilePacker(Packer):
1213
"""A packer which regenerates indices etc as it copies.
1215
This is used by ``bzr reconcile`` to cause parent text pointers to be
1219
def _extra_init(self):
1220
self._data_changed = False
1222
def _process_inventory_lines(self, inv_lines):
1223
"""Generate a text key reference map rather for reconciling with."""
1224
repo = self._pack_collection.repo
1225
refs = repo._find_text_key_references_from_xml_inventory_lines(
1227
self._text_refs = refs
1228
# during reconcile we:
1229
# - convert unreferenced texts to full texts
1230
# - correct texts which reference a text not copied to be full texts
1231
# - copy all others as-is but with corrected parents.
1232
# - so at this point we don't know enough to decide what becomes a full
1234
self._text_filter = None
1236
def _copy_text_texts(self):
1237
"""generate what texts we should have and then copy."""
1238
self.pb.update("Copying content texts", 3)
1239
# we have three major tasks here:
1240
# 1) generate the ideal index
1241
repo = self._pack_collection.repo
1242
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1243
_1, key, _2, refs in
1244
self.new_pack.revision_index.iter_all_entries()])
1245
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1246
# 2) generate a text_nodes list that contains all the deltas that can
1247
# be used as-is, with corrected parents.
1250
discarded_nodes = []
1251
NULL_REVISION = _mod_revision.NULL_REVISION
1252
text_index_map, text_nodes = self._get_text_nodes()
1253
for node in text_nodes:
1259
ideal_parents = tuple(ideal_index[node[1]])
1261
discarded_nodes.append(node)
1262
self._data_changed = True
1264
if ideal_parents == (NULL_REVISION,):
1266
if ideal_parents == node[3][0]:
1268
ok_nodes.append(node)
1269
elif ideal_parents[0:1] == node[3][0][0:1]:
1270
# the left most parent is the same, or there are no parents
1271
# today. Either way, we can preserve the representation as
1272
# long as we change the refs to be inserted.
1273
self._data_changed = True
1274
ok_nodes.append((node[0], node[1], node[2],
1275
(ideal_parents, node[3][1])))
1276
self._data_changed = True
1278
# Reinsert this text completely
1279
bad_texts.append((node[1], ideal_parents))
1280
self._data_changed = True
1281
# we're finished with some data.
1284
# 3) bulk copy the ok data
1285
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1286
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1287
self.new_pack.text_index, readv_group_iter, total_items))
1288
# 4) adhoc copy all the other texts.
1289
# We have to topologically insert all texts otherwise we can fail to
1290
# reconcile when parts of a single delta chain are preserved intact,
1291
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1292
# reinserted, and if d3 has incorrect parents it will also be
1293
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1294
# copied), so we will try to delta, but d2 is not currently able to be
1295
# extracted because its basis d1 is not present. Topologically sorting
1296
# addresses this. The following generates a sort for all the texts that
1297
# are being inserted without having to reference the entire text key
1298
# space (we only topo sort the revisions, which is smaller).
1299
topo_order = tsort.topo_sort(ancestors)
1300
rev_order = dict(zip(topo_order, range(len(topo_order))))
1301
bad_texts.sort(key=lambda key:rev_order.get(key[0][1], 0))
1302
transaction = repo.get_transaction()
1303
file_id_index = GraphIndexPrefixAdapter(
1304
self.new_pack.text_index,
1306
add_nodes_callback=self.new_pack.text_index.add_nodes)
1307
data_access = _DirectPackAccess(
1308
{self.new_pack.text_index:self.new_pack.access_tuple()})
1309
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1310
self.new_pack.access_tuple())
1311
output_texts = KnitVersionedFiles(
1312
_KnitGraphIndex(self.new_pack.text_index,
1313
add_callback=self.new_pack.text_index.add_nodes,
1314
deltas=True, parents=True, is_locked=repo.is_locked),
1315
data_access=data_access, max_delta_chain=200)
1316
for key, parent_keys in bad_texts:
1317
# We refer to the new pack to delta data being output.
1318
# A possible improvement would be to catch errors on short reads
1319
# and only flush then.
1320
self.new_pack.flush()
1322
for parent_key in parent_keys:
1323
if parent_key[0] != key[0]:
1324
# Graph parents must match the fileid
1325
raise errors.BzrError('Mismatched key parent %r:%r' %
1327
parents.append(parent_key[1])
1328
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1329
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1330
output_texts.add_lines(key, parent_keys, text_lines,
1331
random_id=True, check_content=False)
1332
# 5) check that nothing inserted has a reference outside the keyspace.
1333
missing_text_keys = self.new_pack.text_index._external_references()
1334
if missing_text_keys:
1335
raise errors.BzrCheckError('Reference to missing compression parents %r'
1336
% (missing_text_keys,))
1337
self._log_copied_texts()
1339
def _use_pack(self, new_pack):
1340
"""Override _use_pack to check for reconcile having changed content."""
1341
# XXX: we might be better checking this at the copy time.
1342
original_inventory_keys = set()
1343
inv_index = self._pack_collection.inventory_index.combined_index
1344
for entry in inv_index.iter_all_entries():
1345
original_inventory_keys.add(entry[1])
1346
new_inventory_keys = set()
1347
for entry in new_pack.inventory_index.iter_all_entries():
1348
new_inventory_keys.add(entry[1])
1349
if new_inventory_keys != original_inventory_keys:
1350
self._data_changed = True
1351
return new_pack.data_inserted() and self._data_changed
1354
class RepositoryPackCollection(object):
1355
"""Management of packs within a repository.
1357
:ivar _names: map of {pack_name: (index_size,)}
1360
pack_factory = NewPack
1361
resumed_pack_factory = ResumedPack
1363
def __init__(self, repo, transport, index_transport, upload_transport,
1364
pack_transport, index_builder_class, index_class,
1366
"""Create a new RepositoryPackCollection.
1368
:param transport: Addresses the repository base directory
1369
(typically .bzr/repository/).
1370
:param index_transport: Addresses the directory containing indices.
1371
:param upload_transport: Addresses the directory into which packs are written
1372
while they're being created.
1373
:param pack_transport: Addresses the directory of existing complete packs.
1374
:param index_builder_class: The index builder class to use.
1375
:param index_class: The index class to use.
1376
:param use_chk_index: Whether to setup and manage a CHK index.
1378
# XXX: This should call self.reset()
1380
self.transport = transport
1381
self._index_transport = index_transport
1382
self._upload_transport = upload_transport
1383
self._pack_transport = pack_transport
1384
self._index_builder_class = index_builder_class
1385
self._index_class = index_class
1386
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3,
1391
self._packs_by_name = {}
1392
# the previous pack-names content
1393
self._packs_at_load = None
1394
# when a pack is being created by this object, the state of that pack.
1395
self._new_pack = None
1396
# aggregated revision index data
1397
flush = self._flush_new_pack
1398
self.revision_index = AggregateIndex(self.reload_pack_names, flush)
1399
self.inventory_index = AggregateIndex(self.reload_pack_names, flush)
1400
self.text_index = AggregateIndex(self.reload_pack_names, flush)
1401
self.signature_index = AggregateIndex(self.reload_pack_names, flush)
1402
all_indices = [self.revision_index, self.inventory_index,
1403
self.text_index, self.signature_index]
1405
self.chk_index = AggregateIndex(self.reload_pack_names, flush)
1406
all_indices.append(self.chk_index)
1408
# used to determine if we're using a chk_index elsewhere.
1409
self.chk_index = None
1410
# Tell all the CombinedGraphIndex objects about each other, so they can
1411
# share hints about which pack names to search first.
1412
all_combined = [agg_idx.combined_index for agg_idx in all_indices]
1413
for combined_idx in all_combined:
1414
combined_idx.set_sibling_indices(
1415
set(all_combined).difference([combined_idx]))
1417
self._resumed_packs = []
1420
return '%s(%r)' % (self.__class__.__name__, self.repo)
1422
def add_pack_to_memory(self, pack):
1423
"""Make a Pack object available to the repository to satisfy queries.
1425
:param pack: A Pack object.
1427
if pack.name in self._packs_by_name:
1428
raise AssertionError(
1429
'pack %s already in _packs_by_name' % (pack.name,))
1430
self.packs.append(pack)
1431
self._packs_by_name[pack.name] = pack
1432
self.revision_index.add_index(pack.revision_index, pack)
1433
self.inventory_index.add_index(pack.inventory_index, pack)
1434
self.text_index.add_index(pack.text_index, pack)
1435
self.signature_index.add_index(pack.signature_index, pack)
1436
if self.chk_index is not None:
1437
self.chk_index.add_index(pack.chk_index, pack)
1439
def all_packs(self):
1440
"""Return a list of all the Pack objects this repository has.
1442
Note that an in-progress pack being created is not returned.
1444
:return: A list of Pack objects for all the packs in the repository.
1447
for name in self.names():
1448
result.append(self.get_pack_by_name(name))
1452
"""Pack the pack collection incrementally.
1454
This will not attempt global reorganisation or recompression,
1455
rather it will just ensure that the total number of packs does
1456
not grow without bound. It uses the _max_pack_count method to
1457
determine if autopacking is needed, and the pack_distribution
1458
method to determine the number of revisions in each pack.
1460
If autopacking takes place then the packs name collection will have
1461
been flushed to disk - packing requires updating the name collection
1462
in synchronisation with certain steps. Otherwise the names collection
1465
:return: Something evaluating true if packing took place.
1469
return self._do_autopack()
1470
except errors.RetryAutopack:
1471
# If we get a RetryAutopack exception, we should abort the
1472
# current action, and retry.
1475
def _do_autopack(self):
1476
# XXX: Should not be needed when the management of indices is sane.
1477
total_revisions = self.revision_index.combined_index.key_count()
1478
total_packs = len(self._names)
1479
if self._max_pack_count(total_revisions) >= total_packs:
1481
# determine which packs need changing
1482
pack_distribution = self.pack_distribution(total_revisions)
1484
for pack in self.all_packs():
1485
revision_count = pack.get_revision_count()
1486
if revision_count == 0:
1487
# revision less packs are not generated by normal operation,
1488
# only by operations like sign-my-commits, and thus will not
1489
# tend to grow rapdily or without bound like commit containing
1490
# packs do - leave them alone as packing them really should
1491
# group their data with the relevant commit, and that may
1492
# involve rewriting ancient history - which autopack tries to
1493
# avoid. Alternatively we could not group the data but treat
1494
# each of these as having a single revision, and thus add
1495
# one revision for each to the total revision count, to get
1496
# a matching distribution.
1498
existing_packs.append((revision_count, pack))
1499
pack_operations = self.plan_autopack_combinations(
1500
existing_packs, pack_distribution)
1501
num_new_packs = len(pack_operations)
1502
num_old_packs = sum([len(po[1]) for po in pack_operations])
1503
num_revs_affected = sum([po[0] for po in pack_operations])
1504
mutter('Auto-packing repository %s, which has %d pack files, '
1505
'containing %d revisions. Packing %d files into %d affecting %d'
1506
' revisions', self, total_packs, total_revisions, num_old_packs,
1507
num_new_packs, num_revs_affected)
1508
result = self._execute_pack_operations(pack_operations,
1509
reload_func=self._restart_autopack)
1510
mutter('Auto-packing repository %s completed', self)
1513
def _execute_pack_operations(self, pack_operations, _packer_class=Packer,
1515
"""Execute a series of pack operations.
1517
:param pack_operations: A list of [revision_count, packs_to_combine].
1518
:param _packer_class: The class of packer to use (default: Packer).
1519
:return: The new pack names.
1521
for revision_count, packs in pack_operations:
1522
# we may have no-ops from the setup logic
1525
packer = _packer_class(self, packs, '.autopack',
1526
reload_func=reload_func)
1529
except errors.RetryWithNewPacks:
1530
# An exception is propagating out of this context, make sure
1531
# this packer has cleaned up. Packer() doesn't set its new_pack
1532
# state into the RepositoryPackCollection object, so we only
1533
# have access to it directly here.
1534
if packer.new_pack is not None:
1535
packer.new_pack.abort()
1538
self._remove_pack_from_memory(pack)
1539
# record the newly available packs and stop advertising the old
1541
to_be_obsoleted = []
1542
for _, packs in pack_operations:
1543
to_be_obsoleted.extend(packs)
1544
result = self._save_pack_names(clear_obsolete_packs=True,
1545
obsolete_packs=to_be_obsoleted)
1548
def _flush_new_pack(self):
1549
if self._new_pack is not None:
1550
self._new_pack.flush()
1552
def lock_names(self):
1553
"""Acquire the mutex around the pack-names index.
1555
This cannot be used in the middle of a read-only transaction on the
1558
self.repo.control_files.lock_write()
1560
def _already_packed(self):
1561
"""Is the collection already packed?"""
1562
return not (self.repo._format.pack_compresses or (len(self._names) > 1))
1564
def pack(self, hint=None, clean_obsolete_packs=False):
1565
"""Pack the pack collection totally."""
1566
self.ensure_loaded()
1567
total_packs = len(self._names)
1568
if self._already_packed():
1570
total_revisions = self.revision_index.combined_index.key_count()
1571
# XXX: the following may want to be a class, to pack with a given
1573
mutter('Packing repository %s, which has %d pack files, '
1574
'containing %d revisions with hint %r.', self, total_packs,
1575
total_revisions, hint)
1578
self._try_pack_operations(hint)
1579
except RetryPackOperations:
1583
if clean_obsolete_packs:
1584
self._clear_obsolete_packs()
1586
def _try_pack_operations(self, hint):
1587
"""Calculate the pack operations based on the hint (if any), and
1590
# determine which packs need changing
1591
pack_operations = [[0, []]]
1592
for pack in self.all_packs():
1593
if hint is None or pack.name in hint:
1594
# Either no hint was provided (so we are packing everything),
1595
# or this pack was included in the hint.
1596
pack_operations[-1][0] += pack.get_revision_count()
1597
pack_operations[-1][1].append(pack)
1598
self._execute_pack_operations(pack_operations, OptimisingPacker,
1599
reload_func=self._restart_pack_operations)
1601
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1602
"""Plan a pack operation.
1604
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1606
:param pack_distribution: A list with the number of revisions desired
1609
if len(existing_packs) <= len(pack_distribution):
1611
existing_packs.sort(reverse=True)
1612
pack_operations = [[0, []]]
1613
# plan out what packs to keep, and what to reorganise
1614
while len(existing_packs):
1615
# take the largest pack, and if it's less than the head of the
1616
# distribution chart we will include its contents in the new pack
1617
# for that position. If it's larger, we remove its size from the
1618
# distribution chart
1619
next_pack_rev_count, next_pack = existing_packs.pop(0)
1620
if next_pack_rev_count >= pack_distribution[0]:
1621
# this is already packed 'better' than this, so we can
1622
# not waste time packing it.
1623
while next_pack_rev_count > 0:
1624
next_pack_rev_count -= pack_distribution[0]
1625
if next_pack_rev_count >= 0:
1627
del pack_distribution[0]
1629
# didn't use that entire bucket up
1630
pack_distribution[0] = -next_pack_rev_count
1632
# add the revisions we're going to add to the next output pack
1633
pack_operations[-1][0] += next_pack_rev_count
1634
# allocate this pack to the next pack sub operation
1635
pack_operations[-1][1].append(next_pack)
1636
if pack_operations[-1][0] >= pack_distribution[0]:
1637
# this pack is used up, shift left.
1638
del pack_distribution[0]
1639
pack_operations.append([0, []])
1640
# Now that we know which pack files we want to move, shove them all
1641
# into a single pack file.
1643
final_pack_list = []
1644
for num_revs, pack_files in pack_operations:
1645
final_rev_count += num_revs
1646
final_pack_list.extend(pack_files)
1647
if len(final_pack_list) == 1:
1648
raise AssertionError('We somehow generated an autopack with a'
1649
' single pack file being moved.')
1651
return [[final_rev_count, final_pack_list]]
1653
def ensure_loaded(self):
1654
"""Ensure we have read names from disk.
1656
:return: True if the disk names had not been previously read.
1658
# NB: if you see an assertion error here, it's probably access against
1659
# an unlocked repo. Naughty.
1660
if not self.repo.is_locked():
1661
raise errors.ObjectNotLocked(self.repo)
1662
if self._names is None:
1664
self._packs_at_load = set()
1665
for index, key, value in self._iter_disk_pack_index():
1667
self._names[name] = self._parse_index_sizes(value)
1668
self._packs_at_load.add((key, value))
1672
# populate all the metadata.
1676
def _parse_index_sizes(self, value):
1677
"""Parse a string of index sizes."""
1678
return tuple([int(digits) for digits in value.split(' ')])
1680
def get_pack_by_name(self, name):
1681
"""Get a Pack object by name.
1683
:param name: The name of the pack - e.g. '123456'
1684
:return: A Pack object.
1687
return self._packs_by_name[name]
1689
rev_index = self._make_index(name, '.rix')
1690
inv_index = self._make_index(name, '.iix')
1691
txt_index = self._make_index(name, '.tix')
1692
sig_index = self._make_index(name, '.six')
1693
if self.chk_index is not None:
1694
chk_index = self._make_index(name, '.cix', is_chk=True)
1697
result = ExistingPack(self._pack_transport, name, rev_index,
1698
inv_index, txt_index, sig_index, chk_index)
1699
self.add_pack_to_memory(result)
1702
def _resume_pack(self, name):
1703
"""Get a suspended Pack object by name.
1705
:param name: The name of the pack - e.g. '123456'
1706
:return: A Pack object.
1708
if not re.match('[a-f0-9]{32}', name):
1709
# Tokens should be md5sums of the suspended pack file, i.e. 32 hex
1711
raise errors.UnresumableWriteGroup(
1712
self.repo, [name], 'Malformed write group token')
1714
rev_index = self._make_index(name, '.rix', resume=True)
1715
inv_index = self._make_index(name, '.iix', resume=True)
1716
txt_index = self._make_index(name, '.tix', resume=True)
1717
sig_index = self._make_index(name, '.six', resume=True)
1718
if self.chk_index is not None:
1719
chk_index = self._make_index(name, '.cix', resume=True,
1723
result = self.resumed_pack_factory(name, rev_index, inv_index,
1724
txt_index, sig_index, self._upload_transport,
1725
self._pack_transport, self._index_transport, self,
1726
chk_index=chk_index)
1727
except errors.NoSuchFile, e:
1728
raise errors.UnresumableWriteGroup(self.repo, [name], str(e))
1729
self.add_pack_to_memory(result)
1730
self._resumed_packs.append(result)
1733
def allocate(self, a_new_pack):
1734
"""Allocate name in the list of packs.
1736
:param a_new_pack: A NewPack instance to be added to the collection of
1737
packs for this repository.
1739
self.ensure_loaded()
1740
if a_new_pack.name in self._names:
1741
raise errors.BzrError(
1742
'Pack %r already exists in %s' % (a_new_pack.name, self))
1743
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1744
self.add_pack_to_memory(a_new_pack)
1746
def _iter_disk_pack_index(self):
1747
"""Iterate over the contents of the pack-names index.
1749
This is used when loading the list from disk, and before writing to
1750
detect updates from others during our write operation.
1751
:return: An iterator of the index contents.
1753
return self._index_class(self.transport, 'pack-names', None
1754
).iter_all_entries()
1756
def _make_index(self, name, suffix, resume=False, is_chk=False):
1757
size_offset = self._suffix_offsets[suffix]
1758
index_name = name + suffix
1760
transport = self._upload_transport
1761
index_size = transport.stat(index_name).st_size
1763
transport = self._index_transport
1764
index_size = self._names[name][size_offset]
1765
index = self._index_class(transport, index_name, index_size,
1766
unlimited_cache=is_chk)
1767
if is_chk and self._index_class is btree_index.BTreeGraphIndex:
1768
index._leaf_factory = btree_index._gcchk_factory
1771
def _max_pack_count(self, total_revisions):
1772
"""Return the maximum number of packs to use for total revisions.
1774
:param total_revisions: The total number of revisions in the
1777
if not total_revisions:
1779
digits = str(total_revisions)
1781
for digit in digits:
1782
result += int(digit)
1786
"""Provide an order to the underlying names."""
1787
return sorted(self._names.keys())
1789
def _obsolete_packs(self, packs):
1790
"""Move a number of packs which have been obsoleted out of the way.
1792
Each pack and its associated indices are moved out of the way.
1794
Note: for correctness this function should only be called after a new
1795
pack names index has been written without these pack names, and with
1796
the names of packs that contain the data previously available via these
1799
:param packs: The packs to obsolete.
1800
:param return: None.
1804
pack.pack_transport.rename(pack.file_name(),
1805
'../obsolete_packs/' + pack.file_name())
1806
except (errors.PathError, errors.TransportError), e:
1807
# TODO: Should these be warnings or mutters?
1808
mutter("couldn't rename obsolete pack, skipping it:\n%s"
1810
# TODO: Probably needs to know all possible indices for this pack
1811
# - or maybe list the directory and move all indices matching this
1812
# name whether we recognize it or not?
1813
suffixes = ['.iix', '.six', '.tix', '.rix']
1814
if self.chk_index is not None:
1815
suffixes.append('.cix')
1816
for suffix in suffixes:
1818
self._index_transport.rename(pack.name + suffix,
1819
'../obsolete_packs/' + pack.name + suffix)
1820
except (errors.PathError, errors.TransportError), e:
1821
mutter("couldn't rename obsolete index, skipping it:\n%s"
1824
def pack_distribution(self, total_revisions):
1825
"""Generate a list of the number of revisions to put in each pack.
1827
:param total_revisions: The total number of revisions in the
1830
if total_revisions == 0:
1832
digits = reversed(str(total_revisions))
1834
for exponent, count in enumerate(digits):
1835
size = 10 ** exponent
1836
for pos in range(int(count)):
1838
return list(reversed(result))
1840
def _pack_tuple(self, name):
1841
"""Return a tuple with the transport and file name for a pack name."""
1842
return self._pack_transport, name + '.pack'
1844
def _remove_pack_from_memory(self, pack):
1845
"""Remove pack from the packs accessed by this repository.
1847
Only affects memory state, until self._save_pack_names() is invoked.
1849
self._names.pop(pack.name)
1850
self._packs_by_name.pop(pack.name)
1851
self._remove_pack_indices(pack)
1852
self.packs.remove(pack)
1854
def _remove_pack_indices(self, pack, ignore_missing=False):
1855
"""Remove the indices for pack from the aggregated indices.
1857
:param ignore_missing: Suppress KeyErrors from calling remove_index.
1859
for index_type in Pack.index_definitions.keys():
1860
attr_name = index_type + '_index'
1861
aggregate_index = getattr(self, attr_name)
1862
if aggregate_index is not None:
1863
pack_index = getattr(pack, attr_name)
1865
aggregate_index.remove_index(pack_index)
1872
"""Clear all cached data."""
1873
# cached revision data
1874
self.revision_index.clear()
1875
# cached signature data
1876
self.signature_index.clear()
1877
# cached file text data
1878
self.text_index.clear()
1879
# cached inventory data
1880
self.inventory_index.clear()
1882
if self.chk_index is not None:
1883
self.chk_index.clear()
1884
# remove the open pack
1885
self._new_pack = None
1886
# information about packs.
1889
self._packs_by_name = {}
1890
self._packs_at_load = None
1892
def _unlock_names(self):
1893
"""Release the mutex around the pack-names index."""
1894
self.repo.control_files.unlock()
1896
def _diff_pack_names(self):
1897
"""Read the pack names from disk, and compare it to the one in memory.
1899
:return: (disk_nodes, deleted_nodes, new_nodes)
1900
disk_nodes The final set of nodes that should be referenced
1901
deleted_nodes Nodes which have been removed from when we started
1902
new_nodes Nodes that are newly introduced
1904
# load the disk nodes across
1906
for index, key, value in self._iter_disk_pack_index():
1907
disk_nodes.add((key, value))
1908
orig_disk_nodes = set(disk_nodes)
1910
# do a two-way diff against our original content
1911
current_nodes = set()
1912
for name, sizes in self._names.iteritems():
1914
((name, ), ' '.join(str(size) for size in sizes)))
1916
# Packs no longer present in the repository, which were present when we
1917
# locked the repository
1918
deleted_nodes = self._packs_at_load - current_nodes
1919
# Packs which this process is adding
1920
new_nodes = current_nodes - self._packs_at_load
1922
# Update the disk_nodes set to include the ones we are adding, and
1923
# remove the ones which were removed by someone else
1924
disk_nodes.difference_update(deleted_nodes)
1925
disk_nodes.update(new_nodes)
1927
return disk_nodes, deleted_nodes, new_nodes, orig_disk_nodes
1929
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1930
"""Given the correct set of pack files, update our saved info.
1932
:return: (removed, added, modified)
1933
removed pack names removed from self._names
1934
added pack names added to self._names
1935
modified pack names that had changed value
1940
## self._packs_at_load = disk_nodes
1941
new_names = dict(disk_nodes)
1942
# drop no longer present nodes
1943
for pack in self.all_packs():
1944
if (pack.name,) not in new_names:
1945
removed.append(pack.name)
1946
self._remove_pack_from_memory(pack)
1947
# add new nodes/refresh existing ones
1948
for key, value in disk_nodes:
1950
sizes = self._parse_index_sizes(value)
1951
if name in self._names:
1953
if sizes != self._names[name]:
1954
# the pack for name has had its indices replaced - rare but
1955
# important to handle. XXX: probably can never happen today
1956
# because the three-way merge code above does not handle it
1957
# - you may end up adding the same key twice to the new
1958
# disk index because the set values are the same, unless
1959
# the only index shows up as deleted by the set difference
1960
# - which it may. Until there is a specific test for this,
1961
# assume it's broken. RBC 20071017.
1962
self._remove_pack_from_memory(self.get_pack_by_name(name))
1963
self._names[name] = sizes
1964
self.get_pack_by_name(name)
1965
modified.append(name)
1968
self._names[name] = sizes
1969
self.get_pack_by_name(name)
1971
return removed, added, modified
1973
def _save_pack_names(self, clear_obsolete_packs=False, obsolete_packs=None):
1974
"""Save the list of packs.
1976
This will take out the mutex around the pack names list for the
1977
duration of the method call. If concurrent updates have been made, a
1978
three-way merge between the current list and the current in memory list
1981
:param clear_obsolete_packs: If True, clear out the contents of the
1982
obsolete_packs directory.
1983
:param obsolete_packs: Packs that are obsolete once the new pack-names
1984
file has been written.
1985
:return: A list of the names saved that were not previously on disk.
1987
already_obsolete = []
1990
builder = self._index_builder_class()
1991
(disk_nodes, deleted_nodes, new_nodes,
1992
orig_disk_nodes) = self._diff_pack_names()
1993
# TODO: handle same-name, index-size-changes here -
1994
# e.g. use the value from disk, not ours, *unless* we're the one
1996
for key, value in disk_nodes:
1997
builder.add_node(key, value)
1998
self.transport.put_file('pack-names', builder.finish(),
1999
mode=self.repo.bzrdir._get_file_mode())
2000
self._packs_at_load = disk_nodes
2001
if clear_obsolete_packs:
2004
to_preserve = set([o.name for o in obsolete_packs])
2005
already_obsolete = self._clear_obsolete_packs(to_preserve)
2007
self._unlock_names()
2008
# synchronise the memory packs list with what we just wrote:
2009
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
2011
# TODO: We could add one more condition here. "if o.name not in
2012
# orig_disk_nodes and o != the new_pack we haven't written to
2013
# disk yet. However, the new pack object is not easily
2014
# accessible here (it would have to be passed through the
2015
# autopacking code, etc.)
2016
obsolete_packs = [o for o in obsolete_packs
2017
if o.name not in already_obsolete]
2018
self._obsolete_packs(obsolete_packs)
2019
return [new_node[0][0] for new_node in new_nodes]
2021
def reload_pack_names(self):
2022
"""Sync our pack listing with what is present in the repository.
2024
This should be called when we find out that something we thought was
2025
present is now missing. This happens when another process re-packs the
2028
:return: True if the in-memory list of packs has been altered at all.
2030
# The ensure_loaded call is to handle the case where the first call
2031
# made involving the collection was to reload_pack_names, where we
2032
# don't have a view of disk contents. It's a bit of a bandaid, and
2033
# causes two reads of pack-names, but it's a rare corner case not
2034
# struck with regular push/pull etc.
2035
first_read = self.ensure_loaded()
2038
# out the new value.
2039
(disk_nodes, deleted_nodes, new_nodes,
2040
orig_disk_nodes) = self._diff_pack_names()
2041
# _packs_at_load is meant to be the explicit list of names in
2042
# 'pack-names' at then start. As such, it should not contain any
2043
# pending names that haven't been written out yet.
2044
self._packs_at_load = orig_disk_nodes
2046
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
2047
if removed or added or modified:
2051
def _restart_autopack(self):
2052
"""Reload the pack names list, and restart the autopack code."""
2053
if not self.reload_pack_names():
2054
# Re-raise the original exception, because something went missing
2055
# and a restart didn't find it
2057
raise errors.RetryAutopack(self.repo, False, sys.exc_info())
2059
def _restart_pack_operations(self):
2060
"""Reload the pack names list, and restart the autopack code."""
2061
if not self.reload_pack_names():
2062
# Re-raise the original exception, because something went missing
2063
# and a restart didn't find it
2065
raise RetryPackOperations(self.repo, False, sys.exc_info())
2067
def _clear_obsolete_packs(self, preserve=None):
2068
"""Delete everything from the obsolete-packs directory.
2070
:return: A list of pack identifiers (the filename without '.pack') that
2071
were found in obsolete_packs.
2074
obsolete_pack_transport = self.transport.clone('obsolete_packs')
2075
if preserve is None:
2077
for filename in obsolete_pack_transport.list_dir('.'):
2078
name, ext = osutils.splitext(filename)
2081
if name in preserve:
2084
obsolete_pack_transport.delete(filename)
2085
except (errors.PathError, errors.TransportError), e:
2086
warning("couldn't delete obsolete pack, skipping it:\n%s"
2090
def _start_write_group(self):
2091
# Do not permit preparation for writing if we're not in a 'write lock'.
2092
if not self.repo.is_write_locked():
2093
raise errors.NotWriteLocked(self)
2094
self._new_pack = self.pack_factory(self, upload_suffix='.pack',
2095
file_mode=self.repo.bzrdir._get_file_mode())
2096
# allow writing: queue writes to a new index
2097
self.revision_index.add_writable_index(self._new_pack.revision_index,
2099
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
2101
self.text_index.add_writable_index(self._new_pack.text_index,
2103
self._new_pack.text_index.set_optimize(combine_backing_indices=False)
2104
self.signature_index.add_writable_index(self._new_pack.signature_index,
2106
if self.chk_index is not None:
2107
self.chk_index.add_writable_index(self._new_pack.chk_index,
2109
self.repo.chk_bytes._index._add_callback = self.chk_index.add_callback
2110
self._new_pack.chk_index.set_optimize(combine_backing_indices=False)
2112
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
2113
self.repo.revisions._index._add_callback = self.revision_index.add_callback
2114
self.repo.signatures._index._add_callback = self.signature_index.add_callback
2115
self.repo.texts._index._add_callback = self.text_index.add_callback
2117
def _abort_write_group(self):
2118
# FIXME: just drop the transient index.
2119
# forget what names there are
2120
if self._new_pack is not None:
2121
operation = cleanup.OperationWithCleanups(self._new_pack.abort)
2122
operation.add_cleanup(setattr, self, '_new_pack', None)
2123
# If we aborted while in the middle of finishing the write
2124
# group, _remove_pack_indices could fail because the indexes are
2125
# already gone. But they're not there we shouldn't fail in this
2126
# case, so we pass ignore_missing=True.
2127
operation.add_cleanup(self._remove_pack_indices, self._new_pack,
2128
ignore_missing=True)
2129
operation.run_simple()
2130
for resumed_pack in self._resumed_packs:
2131
operation = cleanup.OperationWithCleanups(resumed_pack.abort)
2132
# See comment in previous finally block.
2133
operation.add_cleanup(self._remove_pack_indices, resumed_pack,
2134
ignore_missing=True)
2135
operation.run_simple()
2136
del self._resumed_packs[:]
2138
def _remove_resumed_pack_indices(self):
2139
for resumed_pack in self._resumed_packs:
2140
self._remove_pack_indices(resumed_pack)
2141
del self._resumed_packs[:]
2143
def _check_new_inventories(self):
2144
"""Detect missing inventories in this write group.
2146
:returns: list of strs, summarising any problems found. If the list is
2147
empty no problems were found.
2149
# The base implementation does no checks. GCRepositoryPackCollection
2153
def _commit_write_group(self):
2155
for prefix, versioned_file in (
2156
('revisions', self.repo.revisions),
2157
('inventories', self.repo.inventories),
2158
('texts', self.repo.texts),
2159
('signatures', self.repo.signatures),
2161
missing = versioned_file.get_missing_compression_parent_keys()
2162
all_missing.update([(prefix,) + key for key in missing])
2164
raise errors.BzrCheckError(
2165
"Repository %s has missing compression parent(s) %r "
2166
% (self.repo, sorted(all_missing)))
2167
problems = self._check_new_inventories()
2169
problems_summary = '\n'.join(problems)
2170
raise errors.BzrCheckError(
2171
"Cannot add revision(s) to repository: " + problems_summary)
2172
self._remove_pack_indices(self._new_pack)
2173
any_new_content = False
2174
if self._new_pack.data_inserted():
2175
# get all the data to disk and read to use
2176
self._new_pack.finish()
2177
self.allocate(self._new_pack)
2178
self._new_pack = None
2179
any_new_content = True
2181
self._new_pack.abort()
2182
self._new_pack = None
2183
for resumed_pack in self._resumed_packs:
2184
# XXX: this is a pretty ugly way to turn the resumed pack into a
2185
# properly committed pack.
2186
self._names[resumed_pack.name] = None
2187
self._remove_pack_from_memory(resumed_pack)
2188
resumed_pack.finish()
2189
self.allocate(resumed_pack)
2190
any_new_content = True
2191
del self._resumed_packs[:]
2193
result = self.autopack()
2195
# when autopack takes no steps, the names list is still
2197
return self._save_pack_names()
2201
def _suspend_write_group(self):
2202
tokens = [pack.name for pack in self._resumed_packs]
2203
self._remove_pack_indices(self._new_pack)
2204
if self._new_pack.data_inserted():
2205
# get all the data to disk and read to use
2206
self._new_pack.finish(suspend=True)
2207
tokens.append(self._new_pack.name)
2208
self._new_pack = None
2210
self._new_pack.abort()
2211
self._new_pack = None
2212
self._remove_resumed_pack_indices()
2215
def _resume_write_group(self, tokens):
2216
for token in tokens:
2217
self._resume_pack(token)
2220
class KnitPackRepository(KnitRepository):
2221
"""Repository with knit objects stored inside pack containers.
2223
The layering for a KnitPackRepository is:
2225
Graph | HPSS | Repository public layer |
2226
===================================================
2227
Tuple based apis below, string based, and key based apis above
2228
---------------------------------------------------
2230
Provides .texts, .revisions etc
2231
This adapts the N-tuple keys to physical knit records which only have a
2232
single string identifier (for historical reasons), which in older formats
2233
was always the revision_id, and in the mapped code for packs is always
2234
the last element of key tuples.
2235
---------------------------------------------------
2237
A separate GraphIndex is used for each of the
2238
texts/inventories/revisions/signatures contained within each individual
2239
pack file. The GraphIndex layer works in N-tuples and is unaware of any
2241
===================================================
2245
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
2247
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2248
_commit_builder_class, _serializer)
2249
index_transport = self._transport.clone('indices')
2250
self._pack_collection = RepositoryPackCollection(self, self._transport,
2252
self._transport.clone('upload'),
2253
self._transport.clone('packs'),
2254
_format.index_builder_class,
2255
_format.index_class,
2256
use_chk_index=self._format.supports_chks,
2258
self.inventories = KnitVersionedFiles(
2259
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2260
add_callback=self._pack_collection.inventory_index.add_callback,
2261
deltas=True, parents=True, is_locked=self.is_locked),
2262
data_access=self._pack_collection.inventory_index.data_access,
2263
max_delta_chain=200)
2264
self.revisions = KnitVersionedFiles(
2265
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2266
add_callback=self._pack_collection.revision_index.add_callback,
2267
deltas=False, parents=True, is_locked=self.is_locked,
2268
track_external_parent_refs=True),
2269
data_access=self._pack_collection.revision_index.data_access,
2271
self.signatures = KnitVersionedFiles(
2272
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2273
add_callback=self._pack_collection.signature_index.add_callback,
2274
deltas=False, parents=False, is_locked=self.is_locked),
2275
data_access=self._pack_collection.signature_index.data_access,
2277
self.texts = KnitVersionedFiles(
2278
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2279
add_callback=self._pack_collection.text_index.add_callback,
2280
deltas=True, parents=True, is_locked=self.is_locked),
2281
data_access=self._pack_collection.text_index.data_access,
2282
max_delta_chain=200)
2283
if _format.supports_chks:
2284
# No graph, no compression:- references from chks are between
2285
# different objects not temporal versions of the same; and without
2286
# some sort of temporal structure knit compression will just fail.
2287
self.chk_bytes = KnitVersionedFiles(
2288
_KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2289
add_callback=self._pack_collection.chk_index.add_callback,
2290
deltas=False, parents=False, is_locked=self.is_locked),
2291
data_access=self._pack_collection.chk_index.data_access,
2294
self.chk_bytes = None
2295
# True when the repository object is 'write locked' (as opposed to the
2296
# physical lock only taken out around changes to the pack-names list.)
2297
# Another way to represent this would be a decorator around the control
2298
# files object that presents logical locks as physical ones - if this
2299
# gets ugly consider that alternative design. RBC 20071011
2300
self._write_lock_count = 0
2301
self._transaction = None
2303
self._reconcile_does_inventory_gc = True
2304
self._reconcile_fixes_text_parents = True
2305
self._reconcile_backsup_inventory = False
2307
def _warn_if_deprecated(self, branch=None):
2308
# This class isn't deprecated, but one sub-format is
2309
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2310
super(KnitPackRepository, self)._warn_if_deprecated(branch)
2312
def _abort_write_group(self):
2313
self.revisions._index._key_dependencies.clear()
2314
self._pack_collection._abort_write_group()
2316
def _get_source(self, to_format):
2317
if to_format.network_name() == self._format.network_name():
2318
return KnitPackStreamSource(self, to_format)
2319
return super(KnitPackRepository, self)._get_source(to_format)
2321
def _make_parents_provider(self):
2322
return graph.CachingParentsProvider(self)
2324
def _refresh_data(self):
2325
if not self.is_locked():
2327
self._pack_collection.reload_pack_names()
2329
def _start_write_group(self):
2330
self._pack_collection._start_write_group()
2332
def _commit_write_group(self):
2333
hint = self._pack_collection._commit_write_group()
2334
self.revisions._index._key_dependencies.clear()
2337
def suspend_write_group(self):
2338
# XXX check self._write_group is self.get_transaction()?
2339
tokens = self._pack_collection._suspend_write_group()
2340
self.revisions._index._key_dependencies.clear()
2341
self._write_group = None
2344
def _resume_write_group(self, tokens):
2345
self._start_write_group()
2347
self._pack_collection._resume_write_group(tokens)
2348
except errors.UnresumableWriteGroup:
2349
self._abort_write_group()
2351
for pack in self._pack_collection._resumed_packs:
2352
self.revisions._index.scan_unvalidated_index(pack.revision_index)
2354
def get_transaction(self):
2355
if self._write_lock_count:
2356
return self._transaction
2358
return self.control_files.get_transaction()
2360
def is_locked(self):
2361
return self._write_lock_count or self.control_files.is_locked()
2363
def is_write_locked(self):
2364
return self._write_lock_count
2366
def lock_write(self, token=None):
2367
"""Lock the repository for writes.
2369
:return: A bzrlib.repository.RepositoryWriteLockResult.
2371
locked = self.is_locked()
2372
if not self._write_lock_count and locked:
2373
raise errors.ReadOnlyError(self)
2374
self._write_lock_count += 1
2375
if self._write_lock_count == 1:
2376
self._transaction = transactions.WriteTransaction()
2378
if 'relock' in debug.debug_flags and self._prev_lock == 'w':
2379
note('%r was write locked again', self)
2380
self._prev_lock = 'w'
2381
for repo in self._fallback_repositories:
2382
# Writes don't affect fallback repos
2384
self._refresh_data()
2385
return RepositoryWriteLockResult(self.unlock, None)
2387
def lock_read(self):
2388
"""Lock the repository for reads.
2390
:return: A bzrlib.lock.LogicalLockResult.
2392
locked = self.is_locked()
2393
if self._write_lock_count:
2394
self._write_lock_count += 1
2396
self.control_files.lock_read()
2398
if 'relock' in debug.debug_flags and self._prev_lock == 'r':
2399
note('%r was read locked again', self)
2400
self._prev_lock = 'r'
2401
for repo in self._fallback_repositories:
2403
self._refresh_data()
2404
return LogicalLockResult(self.unlock)
2406
def leave_lock_in_place(self):
2407
# not supported - raise an error
2408
raise NotImplementedError(self.leave_lock_in_place)
2410
def dont_leave_lock_in_place(self):
2411
# not supported - raise an error
2412
raise NotImplementedError(self.dont_leave_lock_in_place)
2415
def pack(self, hint=None, clean_obsolete_packs=False):
2416
"""Compress the data within the repository.
2418
This will pack all the data to a single pack. In future it may
2419
recompress deltas or do other such expensive operations.
2421
self._pack_collection.pack(hint=hint, clean_obsolete_packs=clean_obsolete_packs)
2424
def reconcile(self, other=None, thorough=False):
2425
"""Reconcile this repository."""
2426
from bzrlib.reconcile import PackReconciler
2427
reconciler = PackReconciler(self, thorough=thorough)
2428
reconciler.reconcile()
2431
def _reconcile_pack(self, collection, packs, extension, revs, pb):
2432
packer = ReconcilePacker(collection, packs, extension, revs)
2433
return packer.pack(pb)
2435
@only_raises(errors.LockNotHeld, errors.LockBroken)
2437
if self._write_lock_count == 1 and self._write_group is not None:
2438
self.abort_write_group()
2439
self._transaction = None
2440
self._write_lock_count = 0
2441
raise errors.BzrError(
2442
'Must end write group before releasing write lock on %s'
2444
if self._write_lock_count:
2445
self._write_lock_count -= 1
2446
if not self._write_lock_count:
2447
transaction = self._transaction
2448
self._transaction = None
2449
transaction.finish()
2451
self.control_files.unlock()
2453
if not self.is_locked():
2454
for repo in self._fallback_repositories:
2458
class KnitPackStreamSource(StreamSource):
2459
"""A StreamSource used to transfer data between same-format KnitPack repos.
2461
This source assumes:
2462
1) Same serialization format for all objects
2463
2) Same root information
2464
3) XML format inventories
2465
4) Atomic inserts (so we can stream inventory texts before text
2470
def __init__(self, from_repository, to_format):
2471
super(KnitPackStreamSource, self).__init__(from_repository, to_format)
2472
self._text_keys = None
2473
self._text_fetch_order = 'unordered'
2475
def _get_filtered_inv_stream(self, revision_ids):
2476
from_repo = self.from_repository
2477
parent_ids = from_repo._find_parent_ids_of_revisions(revision_ids)
2478
parent_keys = [(p,) for p in parent_ids]
2479
find_text_keys = from_repo._find_text_key_references_from_xml_inventory_lines
2480
parent_text_keys = set(find_text_keys(
2481
from_repo._inventory_xml_lines_for_keys(parent_keys)))
2482
content_text_keys = set()
2483
knit = KnitVersionedFiles(None, None)
2484
factory = KnitPlainFactory()
2485
def find_text_keys_from_content(record):
2486
if record.storage_kind not in ('knit-delta-gz', 'knit-ft-gz'):
2487
raise ValueError("Unknown content storage kind for"
2488
" inventory text: %s" % (record.storage_kind,))
2489
# It's a knit record, it has a _raw_record field (even if it was
2490
# reconstituted from a network stream).
2491
raw_data = record._raw_record
2492
# read the entire thing
2493
revision_id = record.key[-1]
2494
content, _ = knit._parse_record(revision_id, raw_data)
2495
if record.storage_kind == 'knit-delta-gz':
2496
line_iterator = factory.get_linedelta_content(content)
2497
elif record.storage_kind == 'knit-ft-gz':
2498
line_iterator = factory.get_fulltext_content(content)
2499
content_text_keys.update(find_text_keys(
2500
[(line, revision_id) for line in line_iterator]))
2501
revision_keys = [(r,) for r in revision_ids]
2502
def _filtered_inv_stream():
2503
source_vf = from_repo.inventories
2504
stream = source_vf.get_record_stream(revision_keys,
2506
for record in stream:
2507
if record.storage_kind == 'absent':
2508
raise errors.NoSuchRevision(from_repo, record.key)
2509
find_text_keys_from_content(record)
2511
self._text_keys = content_text_keys - parent_text_keys
2512
return ('inventories', _filtered_inv_stream())
2514
def _get_text_stream(self):
2515
# Note: We know we don't have to handle adding root keys, because both
2516
# the source and target are the identical network name.
2517
text_stream = self.from_repository.texts.get_record_stream(
2518
self._text_keys, self._text_fetch_order, False)
2519
return ('texts', text_stream)
2521
def get_stream(self, search):
2522
revision_ids = search.get_keys()
2523
for stream_info in self._fetch_revision_texts(revision_ids):
2525
self._revision_keys = [(rev_id,) for rev_id in revision_ids]
2526
yield self._get_filtered_inv_stream(revision_ids)
2527
yield self._get_text_stream()
2531
class RepositoryFormatPack(MetaDirRepositoryFormat):
2532
"""Format logic for pack structured repositories.
2534
This repository format has:
2535
- a list of packs in pack-names
2536
- packs in packs/NAME.pack
2537
- indices in indices/NAME.{iix,six,tix,rix}
2538
- knit deltas in the packs, knit indices mapped to the indices.
2539
- thunk objects to support the knits programming API.
2540
- a format marker of its own
2541
- an optional 'shared-storage' flag
2542
- an optional 'no-working-trees' flag
2546
# Set this attribute in derived classes to control the repository class
2547
# created by open and initialize.
2548
repository_class = None
2549
# Set this attribute in derived classes to control the
2550
# _commit_builder_class that the repository objects will have passed to
2551
# their constructor.
2552
_commit_builder_class = None
2553
# Set this attribute in derived clases to control the _serializer that the
2554
# repository objects will have passed to their constructor.
2556
# Packs are not confused by ghosts.
2557
supports_ghosts = True
2558
# External references are not supported in pack repositories yet.
2559
supports_external_lookups = False
2560
# Most pack formats do not use chk lookups.
2561
supports_chks = False
2562
# What index classes to use
2563
index_builder_class = None
2565
_fetch_uses_deltas = True
2568
def initialize(self, a_bzrdir, shared=False):
2569
"""Create a pack based repository.
2571
:param a_bzrdir: bzrdir to contain the new repository; must already
2573
:param shared: If true the repository will be initialized as a shared
2576
mutter('creating repository in %s.', a_bzrdir.transport.base)
2577
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
2578
builder = self.index_builder_class()
2579
files = [('pack-names', builder.finish())]
2580
utf8_files = [('format', self.get_format_string())]
2582
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
2583
repository = self.open(a_bzrdir=a_bzrdir, _found=True)
2584
self._run_post_repo_init_hooks(repository, a_bzrdir, shared)
2587
def open(self, a_bzrdir, _found=False, _override_transport=None):
2588
"""See RepositoryFormat.open().
2590
:param _override_transport: INTERNAL USE ONLY. Allows opening the
2591
repository at a slightly different url
2592
than normal. I.e. during 'upgrade'.
2595
format = RepositoryFormat.find_format(a_bzrdir)
2596
if _override_transport is not None:
2597
repo_transport = _override_transport
2599
repo_transport = a_bzrdir.get_repository_transport(None)
2600
control_files = lockable_files.LockableFiles(repo_transport,
2601
'lock', lockdir.LockDir)
2602
return self.repository_class(_format=self,
2604
control_files=control_files,
2605
_commit_builder_class=self._commit_builder_class,
2606
_serializer=self._serializer)
2609
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2610
"""A no-subtrees parameterized Pack repository.
2612
This format was introduced in 0.92.
2615
repository_class = KnitPackRepository
2616
_commit_builder_class = PackCommitBuilder
2618
def _serializer(self):
2619
return xml5.serializer_v5
2620
# What index classes to use
2621
index_builder_class = InMemoryGraphIndex
2622
index_class = GraphIndex
2624
def _get_matching_bzrdir(self):
2625
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2627
def _ignore_setting_bzrdir(self, format):
2630
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2632
def get_format_string(self):
2633
"""See RepositoryFormat.get_format_string()."""
2634
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2636
def get_format_description(self):
2637
"""See RepositoryFormat.get_format_description()."""
2638
return "Packs containing knits without subtree support"
2641
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2642
"""A subtrees parameterized Pack repository.
2644
This repository format uses the xml7 serializer to get:
2645
- support for recording full info about the tree root
2646
- support for recording tree-references
2648
This format was introduced in 0.92.
2651
repository_class = KnitPackRepository
2652
_commit_builder_class = PackRootCommitBuilder
2653
rich_root_data = True
2655
supports_tree_reference = True
2657
def _serializer(self):
2658
return xml7.serializer_v7
2659
# What index classes to use
2660
index_builder_class = InMemoryGraphIndex
2661
index_class = GraphIndex
2663
def _get_matching_bzrdir(self):
2664
return bzrdir.format_registry.make_bzrdir(
2665
'pack-0.92-subtree')
2667
def _ignore_setting_bzrdir(self, format):
2670
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2672
def get_format_string(self):
2673
"""See RepositoryFormat.get_format_string()."""
2674
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2676
def get_format_description(self):
2677
"""See RepositoryFormat.get_format_description()."""
2678
return "Packs containing knits with subtree support\n"
2681
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2682
"""A rich-root, no subtrees parameterized Pack repository.
2684
This repository format uses the xml6 serializer to get:
2685
- support for recording full info about the tree root
2687
This format was introduced in 1.0.
2690
repository_class = KnitPackRepository
2691
_commit_builder_class = PackRootCommitBuilder
2692
rich_root_data = True
2693
supports_tree_reference = False
2695
def _serializer(self):
2696
return xml6.serializer_v6
2697
# What index classes to use
2698
index_builder_class = InMemoryGraphIndex
2699
index_class = GraphIndex
2701
def _get_matching_bzrdir(self):
2702
return bzrdir.format_registry.make_bzrdir(
2705
def _ignore_setting_bzrdir(self, format):
2708
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2710
def get_format_string(self):
2711
"""See RepositoryFormat.get_format_string()."""
2712
return ("Bazaar pack repository format 1 with rich root"
2713
" (needs bzr 1.0)\n")
2715
def get_format_description(self):
2716
"""See RepositoryFormat.get_format_description()."""
2717
return "Packs containing knits with rich root support\n"
2720
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2721
"""Repository that supports external references to allow stacking.
2725
Supports external lookups, which results in non-truncated ghosts after
2726
reconcile compared to pack-0.92 formats.
2729
repository_class = KnitPackRepository
2730
_commit_builder_class = PackCommitBuilder
2731
supports_external_lookups = True
2732
# What index classes to use
2733
index_builder_class = InMemoryGraphIndex
2734
index_class = GraphIndex
2737
def _serializer(self):
2738
return xml5.serializer_v5
2740
def _get_matching_bzrdir(self):
2741
return bzrdir.format_registry.make_bzrdir('1.6')
2743
def _ignore_setting_bzrdir(self, format):
2746
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2748
def get_format_string(self):
2749
"""See RepositoryFormat.get_format_string()."""
2750
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2752
def get_format_description(self):
2753
"""See RepositoryFormat.get_format_description()."""
2754
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2757
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2758
"""A repository with rich roots and stacking.
2760
New in release 1.6.1.
2762
Supports stacking on other repositories, allowing data to be accessed
2763
without being stored locally.
2766
repository_class = KnitPackRepository
2767
_commit_builder_class = PackRootCommitBuilder
2768
rich_root_data = True
2769
supports_tree_reference = False # no subtrees
2770
supports_external_lookups = True
2771
# What index classes to use
2772
index_builder_class = InMemoryGraphIndex
2773
index_class = GraphIndex
2776
def _serializer(self):
2777
return xml6.serializer_v6
2779
def _get_matching_bzrdir(self):
2780
return bzrdir.format_registry.make_bzrdir(
2783
def _ignore_setting_bzrdir(self, format):
2786
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2788
def get_format_string(self):
2789
"""See RepositoryFormat.get_format_string()."""
2790
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2792
def get_format_description(self):
2793
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2796
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2797
"""A repository with rich roots and external references.
2801
Supports external lookups, which results in non-truncated ghosts after
2802
reconcile compared to pack-0.92 formats.
2804
This format was deprecated because the serializer it uses accidentally
2805
supported subtrees, when the format was not intended to. This meant that
2806
someone could accidentally fetch from an incorrect repository.
2809
repository_class = KnitPackRepository
2810
_commit_builder_class = PackRootCommitBuilder
2811
rich_root_data = True
2812
supports_tree_reference = False # no subtrees
2814
supports_external_lookups = True
2815
# What index classes to use
2816
index_builder_class = InMemoryGraphIndex
2817
index_class = GraphIndex
2820
def _serializer(self):
2821
return xml7.serializer_v7
2823
def _get_matching_bzrdir(self):
2824
matching = bzrdir.format_registry.make_bzrdir(
2826
matching.repository_format = self
2829
def _ignore_setting_bzrdir(self, format):
2832
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2834
def get_format_string(self):
2835
"""See RepositoryFormat.get_format_string()."""
2836
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2838
def get_format_description(self):
2839
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2843
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2844
"""A repository with stacking and btree indexes,
2845
without rich roots or subtrees.
2847
This is equivalent to pack-1.6 with B+Tree indices.
2850
repository_class = KnitPackRepository
2851
_commit_builder_class = PackCommitBuilder
2852
supports_external_lookups = True
2853
# What index classes to use
2854
index_builder_class = btree_index.BTreeBuilder
2855
index_class = btree_index.BTreeGraphIndex
2858
def _serializer(self):
2859
return xml5.serializer_v5
2861
def _get_matching_bzrdir(self):
2862
return bzrdir.format_registry.make_bzrdir('1.9')
2864
def _ignore_setting_bzrdir(self, format):
2867
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2869
def get_format_string(self):
2870
"""See RepositoryFormat.get_format_string()."""
2871
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2873
def get_format_description(self):
2874
"""See RepositoryFormat.get_format_description()."""
2875
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2878
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2879
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2881
1.6-rich-root with B+Tree indices.
2884
repository_class = KnitPackRepository
2885
_commit_builder_class = PackRootCommitBuilder
2886
rich_root_data = True
2887
supports_tree_reference = False # no subtrees
2888
supports_external_lookups = True
2889
# What index classes to use
2890
index_builder_class = btree_index.BTreeBuilder
2891
index_class = btree_index.BTreeGraphIndex
2894
def _serializer(self):
2895
return xml6.serializer_v6
2897
def _get_matching_bzrdir(self):
2898
return bzrdir.format_registry.make_bzrdir(
2901
def _ignore_setting_bzrdir(self, format):
2904
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2906
def get_format_string(self):
2907
"""See RepositoryFormat.get_format_string()."""
2908
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2910
def get_format_description(self):
2911
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2914
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2915
"""A subtrees development repository.
2917
This format should be retained in 2.3, to provide an upgrade path from this
2918
to RepositoryFormat2aSubtree. It can be removed in later releases.
2920
1.6.1-subtree[as it might have been] with B+Tree indices.
2923
repository_class = KnitPackRepository
2924
_commit_builder_class = PackRootCommitBuilder
2925
rich_root_data = True
2927
supports_tree_reference = True
2928
supports_external_lookups = True
2929
# What index classes to use
2930
index_builder_class = btree_index.BTreeBuilder
2931
index_class = btree_index.BTreeGraphIndex
2934
def _serializer(self):
2935
return xml7.serializer_v7
2937
def _get_matching_bzrdir(self):
2938
return bzrdir.format_registry.make_bzrdir(
2939
'development5-subtree')
2941
def _ignore_setting_bzrdir(self, format):
2944
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2946
def get_format_string(self):
2947
"""See RepositoryFormat.get_format_string()."""
2948
return ("Bazaar development format 2 with subtree support "
2949
"(needs bzr.dev from before 1.8)\n")
2951
def get_format_description(self):
2952
"""See RepositoryFormat.get_format_description()."""
2953
return ("Development repository format, currently the same as "
2954
"1.6.1-subtree with B+Tree indices.\n")
2957
class RetryPackOperations(errors.RetryWithNewPacks):
2958
"""Raised when we are packing and we find a missing file.
2960
Meant as a signaling exception, to tell the RepositoryPackCollection.pack
2961
code it should try again.
2964
internal_error = True
2966
_fmt = ("Pack files have changed, reload and try pack again."
2967
" context: %(context)s %(orig_error)s")
added
removed
bzrlib/repofmt/pack_repo.py
