1
# Copyright (C) 2005, 2006, 2007, 2008 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20
from bzrlib.lazy_import import lazy_import
21
lazy_import(globals(), """
22
from itertools import izip
36
from bzrlib.index import (
40
GraphIndexPrefixAdapter,
43
from bzrlib.knit import (
49
from bzrlib import tsort
59
from bzrlib.decorators import needs_write_lock
60
from bzrlib.btree_index import (
64
from bzrlib.index import (
68
from bzrlib.repofmt.knitrepo import KnitRepository
69
from bzrlib.repository import (
71
MetaDirRepositoryFormat,
75
import bzrlib.revision as _mod_revision
76
from bzrlib.trace import (
82
class PackCommitBuilder(CommitBuilder):
83
"""A subclass of CommitBuilder to add texts with pack semantics.
85
Specifically this uses one knit object rather than one knit object per
86
added text, reducing memory and object pressure.
89
def __init__(self, repository, parents, config, timestamp=None,
90
timezone=None, committer=None, revprops=None,
92
CommitBuilder.__init__(self, repository, parents, config,
93
timestamp=timestamp, timezone=timezone, committer=committer,
94
revprops=revprops, revision_id=revision_id)
95
self._file_graph = graph.Graph(
96
repository._pack_collection.text_index.combined_index)
98
def _heads(self, file_id, revision_ids):
99
keys = [(file_id, revision_id) for revision_id in revision_ids]
100
return set([key[1] for key in self._file_graph.heads(keys)])
103
class PackRootCommitBuilder(RootCommitBuilder):
104
"""A subclass of RootCommitBuilder to add texts with pack semantics.
106
Specifically this uses one knit object rather than one knit object per
107
added text, reducing memory and object pressure.
110
def __init__(self, repository, parents, config, timestamp=None,
111
timezone=None, committer=None, revprops=None,
113
CommitBuilder.__init__(self, repository, parents, config,
114
timestamp=timestamp, timezone=timezone, committer=committer,
115
revprops=revprops, revision_id=revision_id)
116
self._file_graph = graph.Graph(
117
repository._pack_collection.text_index.combined_index)
119
def _heads(self, file_id, revision_ids):
120
keys = [(file_id, revision_id) for revision_id in revision_ids]
121
return set([key[1] for key in self._file_graph.heads(keys)])
125
"""An in memory proxy for a pack and its indices.
127
This is a base class that is not directly used, instead the classes
128
ExistingPack and NewPack are used.
131
# A map of index 'type' to the file extension and position in the
133
index_definitions = {
134
'revision': ('.rix', 0),
135
'inventory': ('.iix', 1),
137
'signature': ('.six', 3),
140
def __init__(self, revision_index, inventory_index, text_index,
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.
155
self.revision_index = revision_index
156
self.inventory_index = inventory_index
157
self.text_index = text_index
158
self.signature_index = signature_index
160
def access_tuple(self):
161
"""Return a tuple (transport, name) for the pack content."""
162
return self.pack_transport, self.file_name()
164
def _check_references(self):
165
"""Make sure our external references are present.
167
Packs are allowed to have deltas whose base is not in the pack, but it
168
must be present somewhere in this collection. It is not allowed to
169
have deltas based on a fallback repository.
170
(See <https://bugs.launchpad.net/bzr/+bug/288751>)
173
for (index_name, external_refs, index) in [
175
self._get_external_refs(self.text_index),
176
self._pack_collection.text_index.combined_index),
178
self._get_external_refs(self.inventory_index),
179
self._pack_collection.inventory_index.combined_index),
181
missing = external_refs.difference(
182
k for (idx, k, v, r) in
183
index.iter_entries(external_refs))
185
missing_items[index_name] = sorted(list(missing))
187
from pprint import pformat
188
raise errors.BzrCheckError(
189
"Newly created pack file %r has delta references to "
190
"items not in its repository:\n%s"
191
% (self, pformat(missing_items)))
194
"""Get the file name for the pack on disk."""
195
return self.name + '.pack'
197
def get_revision_count(self):
198
return self.revision_index.key_count()
200
def index_name(self, index_type, name):
201
"""Get the disk name of an index type for pack name 'name'."""
202
return name + Pack.index_definitions[index_type][0]
204
def index_offset(self, index_type):
205
"""Get the position in a index_size array for a given index type."""
206
return Pack.index_definitions[index_type][1]
208
def inventory_index_name(self, name):
209
"""The inv index is the name + .iix."""
210
return self.index_name('inventory', name)
212
def revision_index_name(self, name):
213
"""The revision index is the name + .rix."""
214
return self.index_name('revision', name)
216
def signature_index_name(self, name):
217
"""The signature index is the name + .six."""
218
return self.index_name('signature', name)
220
def text_index_name(self, name):
221
"""The text index is the name + .tix."""
222
return self.index_name('text', name)
224
def _replace_index_with_readonly(self, index_type):
225
setattr(self, index_type + '_index',
226
self.index_class(self.index_transport,
227
self.index_name(index_type, self.name),
228
self.index_sizes[self.index_offset(index_type)]))
231
class ExistingPack(Pack):
232
"""An in memory proxy for an existing .pack and its disk indices."""
234
def __init__(self, pack_transport, name, revision_index, inventory_index,
235
text_index, signature_index):
236
"""Create an ExistingPack object.
238
:param pack_transport: The transport where the pack file resides.
239
:param name: The name of the pack on disk in the pack_transport.
241
Pack.__init__(self, revision_index, inventory_index, text_index,
244
self.pack_transport = pack_transport
245
if None in (revision_index, inventory_index, text_index,
246
signature_index, name, pack_transport):
247
raise AssertionError()
249
def __eq__(self, other):
250
return self.__dict__ == other.__dict__
252
def __ne__(self, other):
253
return not self.__eq__(other)
256
return "<%s.%s object at 0x%x, %s, %s" % (
257
self.__class__.__module__, self.__class__.__name__, id(self),
258
self.pack_transport, self.name)
261
class ResumedPack(ExistingPack):
263
def __init__(self, name, revision_index, inventory_index, text_index,
264
signature_index, upload_transport, pack_transport, index_transport,
266
"""Create a ResumedPack object."""
267
ExistingPack.__init__(self, pack_transport, name, revision_index,
268
inventory_index, text_index, signature_index)
269
self.upload_transport = upload_transport
270
self.index_transport = index_transport
271
self.index_sizes = [None, None, None, None]
273
('revision', revision_index),
274
('inventory', inventory_index),
275
('text', text_index),
276
('signature', signature_index),
278
for index_type, index in indices:
279
offset = self.index_offset(index_type)
280
self.index_sizes[offset] = index._size
281
self.index_class = pack_collection._index_class
282
self._pack_collection = pack_collection
283
self._state = 'resumed'
284
# XXX: perhaps check that the .pack file exists?
286
def access_tuple(self):
287
if self._state == 'finished':
288
return Pack.access_tuple(self)
289
elif self._state == 'resumed':
290
return self.upload_transport, self.file_name()
292
raise AssertionError(self._state)
295
self.upload_transport.delete(self.file_name())
296
indices = [self.revision_index, self.inventory_index, self.text_index,
297
self.signature_index]
298
for index in indices:
299
index._transport.delete(index._name)
302
self._check_references()
303
new_name = '../packs/' + self.file_name()
304
self.upload_transport.rename(self.file_name(), new_name)
305
for index_type in ['revision', 'inventory', 'text', 'signature']:
306
old_name = self.index_name(index_type, self.name)
307
new_name = '../indices/' + old_name
308
self.upload_transport.rename(old_name, new_name)
309
self._replace_index_with_readonly(index_type)
310
self._state = 'finished'
312
def _get_external_refs(self, index):
313
return index.external_references(1)
317
"""An in memory proxy for a pack which is being created."""
319
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
320
"""Create a NewPack instance.
322
:param pack_collection: A PackCollection into which this is being inserted.
323
:param upload_suffix: An optional suffix to be given to any temporary
324
files created during the pack creation. e.g '.autopack'
325
:param file_mode: Unix permissions for newly created file.
327
# The relative locations of the packs are constrained, but all are
328
# passed in because the caller has them, so as to avoid object churn.
329
index_builder_class = pack_collection._index_builder_class
331
# Revisions: parents list, no text compression.
332
index_builder_class(reference_lists=1),
333
# Inventory: We want to map compression only, but currently the
334
# knit code hasn't been updated enough to understand that, so we
335
# have a regular 2-list index giving parents and compression
337
index_builder_class(reference_lists=2),
338
# Texts: compression and per file graph, for all fileids - so two
339
# reference lists and two elements in the key tuple.
340
index_builder_class(reference_lists=2, key_elements=2),
341
# Signatures: Just blobs to store, no compression, no parents
343
index_builder_class(reference_lists=0),
345
self._pack_collection = pack_collection
346
# When we make readonly indices, we need this.
347
self.index_class = pack_collection._index_class
348
# where should the new pack be opened
349
self.upload_transport = pack_collection._upload_transport
350
# where are indices written out to
351
self.index_transport = pack_collection._index_transport
352
# where is the pack renamed to when it is finished?
353
self.pack_transport = pack_collection._pack_transport
354
# What file mode to upload the pack and indices with.
355
self._file_mode = file_mode
356
# tracks the content written to the .pack file.
357
self._hash = osutils.md5()
358
# a four-tuple with the length in bytes of the indices, once the pack
359
# is finalised. (rev, inv, text, sigs)
360
self.index_sizes = None
361
# How much data to cache when writing packs. Note that this is not
362
# synchronised with reads, because it's not in the transport layer, so
363
# is not safe unless the client knows it won't be reading from the pack
365
self._cache_limit = 0
366
# the temporary pack file name.
367
self.random_name = osutils.rand_chars(20) + upload_suffix
368
# when was this pack started ?
369
self.start_time = time.time()
370
# open an output stream for the data added to the pack.
371
self.write_stream = self.upload_transport.open_write_stream(
372
self.random_name, mode=self._file_mode)
373
if 'pack' in debug.debug_flags:
374
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
375
time.ctime(), self.upload_transport.base, self.random_name,
376
time.time() - self.start_time)
377
# A list of byte sequences to be written to the new pack, and the
378
# aggregate size of them. Stored as a list rather than separate
379
# variables so that the _write_data closure below can update them.
380
self._buffer = [[], 0]
381
# create a callable for adding data
383
# robertc says- this is a closure rather than a method on the object
384
# so that the variables are locals, and faster than accessing object
386
def _write_data(bytes, flush=False, _buffer=self._buffer,
387
_write=self.write_stream.write, _update=self._hash.update):
388
_buffer[0].append(bytes)
389
_buffer[1] += len(bytes)
391
if _buffer[1] > self._cache_limit or flush:
392
bytes = ''.join(_buffer[0])
396
# expose this on self, for the occasion when clients want to add data.
397
self._write_data = _write_data
398
# a pack writer object to serialise pack records.
399
self._writer = pack.ContainerWriter(self._write_data)
401
# what state is the pack in? (open, finished, aborted)
405
"""Cancel creating this pack."""
406
self._state = 'aborted'
407
self.write_stream.close()
408
# Remove the temporary pack file.
409
self.upload_transport.delete(self.random_name)
410
# The indices have no state on disk.
412
def access_tuple(self):
413
"""Return a tuple (transport, name) for the pack content."""
414
if self._state == 'finished':
415
return Pack.access_tuple(self)
416
elif self._state == 'open':
417
return self.upload_transport, self.random_name
419
raise AssertionError(self._state)
421
def data_inserted(self):
422
"""True if data has been added to this pack."""
423
return bool(self.get_revision_count() or
424
self.inventory_index.key_count() or
425
self.text_index.key_count() or
426
self.signature_index.key_count())
428
def finish(self, suspend=False):
429
"""Finish the new pack.
432
- finalises the content
433
- assigns a name (the md5 of the content, currently)
434
- writes out the associated indices
435
- renames the pack into place.
436
- stores the index size tuple for the pack in the index_sizes
441
self._write_data('', flush=True)
442
self.name = self._hash.hexdigest()
444
self._check_references()
446
# XXX: It'd be better to write them all to temporary names, then
447
# rename them all into place, so that the window when only some are
448
# visible is smaller. On the other hand none will be seen until
449
# they're in the names list.
450
self.index_sizes = [None, None, None, None]
451
self._write_index('revision', self.revision_index, 'revision', suspend)
452
self._write_index('inventory', self.inventory_index, 'inventory',
454
self._write_index('text', self.text_index, 'file texts', suspend)
455
self._write_index('signature', self.signature_index,
456
'revision signatures', suspend)
457
self.write_stream.close()
458
# Note that this will clobber an existing pack with the same name,
459
# without checking for hash collisions. While this is undesirable this
460
# is something that can be rectified in a subsequent release. One way
461
# to rectify it may be to leave the pack at the original name, writing
462
# its pack-names entry as something like 'HASH: index-sizes
463
# temporary-name'. Allocate that and check for collisions, if it is
464
# collision free then rename it into place. If clients know this scheme
465
# they can handle missing-file errors by:
466
# - try for HASH.pack
467
# - try for temporary-name
468
# - refresh the pack-list to see if the pack is now absent
469
new_name = self.name + '.pack'
471
new_name = '../packs/' + new_name
472
self.upload_transport.rename(self.random_name, new_name)
473
self._state = 'finished'
474
if 'pack' in debug.debug_flags:
475
# XXX: size might be interesting?
476
mutter('%s: create_pack: pack finished: %s%s->%s t+%6.3fs',
477
time.ctime(), self.upload_transport.base, self.random_name,
478
new_name, time.time() - self.start_time)
481
"""Flush any current data."""
483
bytes = ''.join(self._buffer[0])
484
self.write_stream.write(bytes)
485
self._hash.update(bytes)
486
self._buffer[:] = [[], 0]
488
def _get_external_refs(self, index):
489
return index._external_references()
491
def set_write_cache_size(self, size):
492
self._cache_limit = size
494
def _write_index(self, index_type, index, label, suspend=False):
495
"""Write out an index.
497
:param index_type: The type of index to write - e.g. 'revision'.
498
:param index: The index object to serialise.
499
:param label: What label to give the index e.g. 'revision'.
501
index_name = self.index_name(index_type, self.name)
503
transport = self.upload_transport
505
transport = self.index_transport
506
self.index_sizes[self.index_offset(index_type)] = transport.put_file(
507
index_name, index.finish(), mode=self._file_mode)
508
if 'pack' in debug.debug_flags:
509
# XXX: size might be interesting?
510
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
511
time.ctime(), label, self.upload_transport.base,
512
self.random_name, time.time() - self.start_time)
513
# Replace the writable index on this object with a readonly,
514
# presently unloaded index. We should alter
515
# the index layer to make its finish() error if add_node is
516
# subsequently used. RBC
517
self._replace_index_with_readonly(index_type)
520
class AggregateIndex(object):
521
"""An aggregated index for the RepositoryPackCollection.
523
AggregateIndex is reponsible for managing the PackAccess object,
524
Index-To-Pack mapping, and all indices list for a specific type of index
525
such as 'revision index'.
527
A CombinedIndex provides an index on a single key space built up
528
from several on-disk indices. The AggregateIndex builds on this
529
to provide a knit access layer, and allows having up to one writable
530
index within the collection.
532
# XXX: Probably 'can be written to' could/should be separated from 'acts
533
# like a knit index' -- mbp 20071024
535
def __init__(self, reload_func=None):
536
"""Create an AggregateIndex.
538
:param reload_func: A function to call if we find we are missing an
539
index. Should have the form reload_func() => True if the list of
540
active pack files has changed.
542
self._reload_func = reload_func
543
self.index_to_pack = {}
544
self.combined_index = CombinedGraphIndex([], reload_func=reload_func)
545
self.data_access = _DirectPackAccess(self.index_to_pack,
546
reload_func=reload_func)
547
self.add_callback = None
549
def replace_indices(self, index_to_pack, indices):
550
"""Replace the current mappings with fresh ones.
552
This should probably not be used eventually, rather incremental add and
553
removal of indices. It has been added during refactoring of existing
556
:param index_to_pack: A mapping from index objects to
557
(transport, name) tuples for the pack file data.
558
:param indices: A list of indices.
560
# refresh the revision pack map dict without replacing the instance.
561
self.index_to_pack.clear()
562
self.index_to_pack.update(index_to_pack)
563
# XXX: API break - clearly a 'replace' method would be good?
564
self.combined_index._indices[:] = indices
565
# the current add nodes callback for the current writable index if
567
self.add_callback = None
569
def add_index(self, index, pack):
570
"""Add index to the aggregate, which is an index for Pack pack.
572
Future searches on the aggregate index will seach this new index
573
before all previously inserted indices.
575
:param index: An Index for the pack.
576
:param pack: A Pack instance.
578
# expose it to the index map
579
self.index_to_pack[index] = pack.access_tuple()
580
# put it at the front of the linear index list
581
self.combined_index.insert_index(0, index)
583
def add_writable_index(self, index, pack):
584
"""Add an index which is able to have data added to it.
586
There can be at most one writable index at any time. Any
587
modifications made to the knit are put into this index.
589
:param index: An index from the pack parameter.
590
:param pack: A Pack instance.
592
if self.add_callback is not None:
593
raise AssertionError(
594
"%s already has a writable index through %s" % \
595
(self, self.add_callback))
596
# allow writing: queue writes to a new index
597
self.add_index(index, pack)
598
# Updates the index to packs mapping as a side effect,
599
self.data_access.set_writer(pack._writer, index, pack.access_tuple())
600
self.add_callback = index.add_nodes
603
"""Reset all the aggregate data to nothing."""
604
self.data_access.set_writer(None, None, (None, None))
605
self.index_to_pack.clear()
606
del self.combined_index._indices[:]
607
self.add_callback = None
609
def remove_index(self, index, pack):
610
"""Remove index from the indices used to answer queries.
612
:param index: An index from the pack parameter.
613
:param pack: A Pack instance.
615
del self.index_to_pack[index]
616
self.combined_index._indices.remove(index)
617
if (self.add_callback is not None and
618
getattr(index, 'add_nodes', None) == self.add_callback):
619
self.add_callback = None
620
self.data_access.set_writer(None, None, (None, None))
623
class Packer(object):
624
"""Create a pack from packs."""
626
def __init__(self, pack_collection, packs, suffix, revision_ids=None,
630
:param pack_collection: A RepositoryPackCollection object where the
631
new pack is being written to.
632
:param packs: The packs to combine.
633
:param suffix: The suffix to use on the temporary files for the pack.
634
:param revision_ids: Revision ids to limit the pack to.
635
:param reload_func: A function to call if a pack file/index goes
636
missing. The side effect of calling this function should be to
637
update self.packs. See also AggregateIndex
641
self.revision_ids = revision_ids
642
# The pack object we are creating.
644
self._pack_collection = pack_collection
645
self._reload_func = reload_func
646
# The index layer keys for the revisions being copied. None for 'all
648
self._revision_keys = None
649
# What text keys to copy. None for 'all texts'. This is set by
650
# _copy_inventory_texts
651
self._text_filter = None
654
def _extra_init(self):
655
"""A template hook to allow extending the constructor trivially."""
657
def _pack_map_and_index_list(self, index_attribute):
658
"""Convert a list of packs to an index pack map and index list.
660
:param index_attribute: The attribute that the desired index is found
662
:return: A tuple (map, list) where map contains the dict from
663
index:pack_tuple, and list contains the indices in the preferred
668
for pack_obj in self.packs:
669
index = getattr(pack_obj, index_attribute)
670
indices.append(index)
671
pack_map[index] = pack_obj
672
return pack_map, indices
674
def _index_contents(self, indices, key_filter=None):
675
"""Get an iterable of the index contents from a pack_map.
677
:param indices: The list of indices to query
678
:param key_filter: An optional filter to limit the keys returned.
680
all_index = CombinedGraphIndex(indices)
681
if key_filter is None:
682
return all_index.iter_all_entries()
684
return all_index.iter_entries(key_filter)
686
def pack(self, pb=None):
687
"""Create a new pack by reading data from other packs.
689
This does little more than a bulk copy of data. One key difference
690
is that data with the same item key across multiple packs is elided
691
from the output. The new pack is written into the current pack store
692
along with its indices, and the name added to the pack names. The
693
source packs are not altered and are not required to be in the current
696
:param pb: An optional progress bar to use. A nested bar is created if
698
:return: A Pack object, or None if nothing was copied.
700
# open a pack - using the same name as the last temporary file
701
# - which has already been flushed, so its safe.
702
# XXX: - duplicate code warning with start_write_group; fix before
703
# considering 'done'.
704
if self._pack_collection._new_pack is not None:
705
raise errors.BzrError('call to %s.pack() while another pack is'
707
% (self.__class__.__name__,))
708
if self.revision_ids is not None:
709
if len(self.revision_ids) == 0:
710
# silly fetch request.
713
self.revision_ids = frozenset(self.revision_ids)
714
self.revision_keys = frozenset((revid,) for revid in
717
self.pb = ui.ui_factory.nested_progress_bar()
721
return self._create_pack_from_packs()
727
"""Open a pack for the pack we are creating."""
728
new_pack = NewPack(self._pack_collection, upload_suffix=self.suffix,
729
file_mode=self._pack_collection.repo.bzrdir._get_file_mode())
730
# We know that we will process all nodes in order, and don't need to
731
# query, so don't combine any indices spilled to disk until we are done
732
new_pack.revision_index.set_optimize(combine_backing_indices=False)
733
new_pack.inventory_index.set_optimize(combine_backing_indices=False)
734
new_pack.text_index.set_optimize(combine_backing_indices=False)
735
new_pack.signature_index.set_optimize(combine_backing_indices=False)
738
def _update_pack_order(self, entries, index_to_pack_map):
739
"""Determine how we want our packs to be ordered.
741
This changes the sort order of the self.packs list so that packs unused
742
by 'entries' will be at the end of the list, so that future requests
743
can avoid probing them. Used packs will be at the front of the
744
self.packs list, in the order of their first use in 'entries'.
746
:param entries: A list of (index, ...) tuples
747
:param index_to_pack_map: A mapping from index objects to pack objects.
751
for entry in entries:
753
if index not in seen_indexes:
754
packs.append(index_to_pack_map[index])
755
seen_indexes.add(index)
756
if len(packs) == len(self.packs):
757
if 'pack' in debug.debug_flags:
758
mutter('Not changing pack list, all packs used.')
760
seen_packs = set(packs)
761
for pack in self.packs:
762
if pack not in seen_packs:
765
if 'pack' in debug.debug_flags:
766
old_names = [p.access_tuple()[1] for p in self.packs]
767
new_names = [p.access_tuple()[1] for p in packs]
768
mutter('Reordering packs\nfrom: %s\n to: %s',
769
old_names, new_names)
772
def _copy_revision_texts(self):
773
"""Copy revision data to the new pack."""
775
if self.revision_ids:
776
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
779
# select revision keys
780
revision_index_map, revision_indices = self._pack_map_and_index_list(
782
revision_nodes = self._index_contents(revision_indices, revision_keys)
783
revision_nodes = list(revision_nodes)
784
self._update_pack_order(revision_nodes, revision_index_map)
785
# copy revision keys and adjust values
786
self.pb.update("Copying revision texts", 1)
787
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
788
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
789
self.new_pack.revision_index, readv_group_iter, total_items))
790
if 'pack' in debug.debug_flags:
791
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
792
time.ctime(), self._pack_collection._upload_transport.base,
793
self.new_pack.random_name,
794
self.new_pack.revision_index.key_count(),
795
time.time() - self.new_pack.start_time)
796
self._revision_keys = revision_keys
798
def _copy_inventory_texts(self):
799
"""Copy the inventory texts to the new pack.
801
self._revision_keys is used to determine what inventories to copy.
803
Sets self._text_filter appropriately.
805
# select inventory keys
806
inv_keys = self._revision_keys # currently the same keyspace, and note that
807
# querying for keys here could introduce a bug where an inventory item
808
# is missed, so do not change it to query separately without cross
809
# checking like the text key check below.
810
inventory_index_map, inventory_indices = self._pack_map_and_index_list(
812
inv_nodes = self._index_contents(inventory_indices, inv_keys)
813
# copy inventory keys and adjust values
814
# XXX: Should be a helper function to allow different inv representation
816
self.pb.update("Copying inventory texts", 2)
817
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
818
# Only grab the output lines if we will be processing them
819
output_lines = bool(self.revision_ids)
820
inv_lines = self._copy_nodes_graph(inventory_index_map,
821
self.new_pack._writer, self.new_pack.inventory_index,
822
readv_group_iter, total_items, output_lines=output_lines)
823
if self.revision_ids:
824
self._process_inventory_lines(inv_lines)
826
# eat the iterator to cause it to execute.
828
self._text_filter = None
829
if 'pack' in debug.debug_flags:
830
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
831
time.ctime(), self._pack_collection._upload_transport.base,
832
self.new_pack.random_name,
833
self.new_pack.inventory_index.key_count(),
834
time.time() - self.new_pack.start_time)
836
def _copy_text_texts(self):
838
text_index_map, text_nodes = self._get_text_nodes()
839
if self._text_filter is not None:
840
# We could return the keys copied as part of the return value from
841
# _copy_nodes_graph but this doesn't work all that well with the
842
# need to get line output too, so we check separately, and as we're
843
# going to buffer everything anyway, we check beforehand, which
844
# saves reading knit data over the wire when we know there are
846
text_nodes = set(text_nodes)
847
present_text_keys = set(_node[1] for _node in text_nodes)
848
missing_text_keys = set(self._text_filter) - present_text_keys
849
if missing_text_keys:
850
# TODO: raise a specific error that can handle many missing
852
mutter("missing keys during fetch: %r", missing_text_keys)
853
a_missing_key = missing_text_keys.pop()
854
raise errors.RevisionNotPresent(a_missing_key[1],
856
# copy text keys and adjust values
857
self.pb.update("Copying content texts", 3)
858
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
859
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
860
self.new_pack.text_index, readv_group_iter, total_items))
861
self._log_copied_texts()
863
def _create_pack_from_packs(self):
864
self.pb.update("Opening pack", 0, 5)
865
self.new_pack = self.open_pack()
866
new_pack = self.new_pack
867
# buffer data - we won't be reading-back during the pack creation and
868
# this makes a significant difference on sftp pushes.
869
new_pack.set_write_cache_size(1024*1024)
870
if 'pack' in debug.debug_flags:
871
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
872
for a_pack in self.packs]
873
if self.revision_ids is not None:
874
rev_count = len(self.revision_ids)
877
mutter('%s: create_pack: creating pack from source packs: '
878
'%s%s %s revisions wanted %s t=0',
879
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
880
plain_pack_list, rev_count)
881
self._copy_revision_texts()
882
self._copy_inventory_texts()
883
self._copy_text_texts()
884
# select signature keys
885
signature_filter = self._revision_keys # same keyspace
886
signature_index_map, signature_indices = self._pack_map_and_index_list(
888
signature_nodes = self._index_contents(signature_indices,
890
# copy signature keys and adjust values
891
self.pb.update("Copying signature texts", 4)
892
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
893
new_pack.signature_index)
894
if 'pack' in debug.debug_flags:
895
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
896
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
897
new_pack.signature_index.key_count(),
898
time.time() - new_pack.start_time)
899
new_pack._check_references()
900
if not self._use_pack(new_pack):
903
self.pb.update("Finishing pack", 5)
905
self._pack_collection.allocate(new_pack)
908
def _copy_nodes(self, nodes, index_map, writer, write_index):
909
"""Copy knit nodes between packs with no graph references."""
910
pb = ui.ui_factory.nested_progress_bar()
912
return self._do_copy_nodes(nodes, index_map, writer,
917
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb):
918
# for record verification
919
knit = KnitVersionedFiles(None, None)
920
# plan a readv on each source pack:
922
nodes = sorted(nodes)
923
# how to map this into knit.py - or knit.py into this?
924
# we don't want the typical knit logic, we want grouping by pack
925
# at this point - perhaps a helper library for the following code
926
# duplication points?
928
for index, key, value in nodes:
929
if index not in request_groups:
930
request_groups[index] = []
931
request_groups[index].append((key, value))
933
pb.update("Copied record", record_index, len(nodes))
934
for index, items in request_groups.iteritems():
935
pack_readv_requests = []
936
for key, value in items:
937
# ---- KnitGraphIndex.get_position
938
bits = value[1:].split(' ')
939
offset, length = int(bits[0]), int(bits[1])
940
pack_readv_requests.append((offset, length, (key, value[0])))
941
# linear scan up the pack
942
pack_readv_requests.sort()
944
pack_obj = index_map[index]
945
transport, path = pack_obj.access_tuple()
947
reader = pack.make_readv_reader(transport, path,
948
[offset[0:2] for offset in pack_readv_requests])
949
except errors.NoSuchFile:
950
if self._reload_func is not None:
953
for (names, read_func), (_1, _2, (key, eol_flag)) in \
954
izip(reader.iter_records(), pack_readv_requests):
955
raw_data = read_func(None)
956
# check the header only
957
df, _ = knit._parse_record_header(key, raw_data)
959
pos, size = writer.add_bytes_record(raw_data, names)
960
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
961
pb.update("Copied record", record_index)
964
def _copy_nodes_graph(self, index_map, writer, write_index,
965
readv_group_iter, total_items, output_lines=False):
966
"""Copy knit nodes between packs.
968
:param output_lines: Return lines present in the copied data as
969
an iterator of line,version_id.
971
pb = ui.ui_factory.nested_progress_bar()
973
for result in self._do_copy_nodes_graph(index_map, writer,
974
write_index, output_lines, pb, readv_group_iter, total_items):
977
# Python 2.4 does not permit try:finally: in a generator.
983
def _do_copy_nodes_graph(self, index_map, writer, write_index,
984
output_lines, pb, readv_group_iter, total_items):
985
# for record verification
986
knit = KnitVersionedFiles(None, None)
987
# for line extraction when requested (inventories only)
989
factory = KnitPlainFactory()
991
pb.update("Copied record", record_index, total_items)
992
for index, readv_vector, node_vector in readv_group_iter:
994
pack_obj = index_map[index]
995
transport, path = pack_obj.access_tuple()
997
reader = pack.make_readv_reader(transport, path, readv_vector)
998
except errors.NoSuchFile:
999
if self._reload_func is not None:
1002
for (names, read_func), (key, eol_flag, references) in \
1003
izip(reader.iter_records(), node_vector):
1004
raw_data = read_func(None)
1006
# read the entire thing
1007
content, _ = knit._parse_record(key[-1], raw_data)
1008
if len(references[-1]) == 0:
1009
line_iterator = factory.get_fulltext_content(content)
1011
line_iterator = factory.get_linedelta_content(content)
1012
for line in line_iterator:
1015
# check the header only
1016
df, _ = knit._parse_record_header(key, raw_data)
1018
pos, size = writer.add_bytes_record(raw_data, names)
1019
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1020
pb.update("Copied record", record_index)
1023
def _get_text_nodes(self):
1024
text_index_map, text_indices = self._pack_map_and_index_list(
1026
return text_index_map, self._index_contents(text_indices,
1029
def _least_readv_node_readv(self, nodes):
1030
"""Generate request groups for nodes using the least readv's.
1032
:param nodes: An iterable of graph index nodes.
1033
:return: Total node count and an iterator of the data needed to perform
1034
readvs to obtain the data for nodes. Each item yielded by the
1035
iterator is a tuple with:
1036
index, readv_vector, node_vector. readv_vector is a list ready to
1037
hand to the transport readv method, and node_vector is a list of
1038
(key, eol_flag, references) for the the node retrieved by the
1039
matching readv_vector.
1041
# group by pack so we do one readv per pack
1042
nodes = sorted(nodes)
1045
for index, key, value, references in nodes:
1046
if index not in request_groups:
1047
request_groups[index] = []
1048
request_groups[index].append((key, value, references))
1050
for index, items in request_groups.iteritems():
1051
pack_readv_requests = []
1052
for key, value, references in items:
1053
# ---- KnitGraphIndex.get_position
1054
bits = value[1:].split(' ')
1055
offset, length = int(bits[0]), int(bits[1])
1056
pack_readv_requests.append(
1057
((offset, length), (key, value[0], references)))
1058
# linear scan up the pack to maximum range combining.
1059
pack_readv_requests.sort()
1060
# split out the readv and the node data.
1061
pack_readv = [readv for readv, node in pack_readv_requests]
1062
node_vector = [node for readv, node in pack_readv_requests]
1063
result.append((index, pack_readv, node_vector))
1064
return total, result
1066
def _log_copied_texts(self):
1067
if 'pack' in debug.debug_flags:
1068
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
1069
time.ctime(), self._pack_collection._upload_transport.base,
1070
self.new_pack.random_name,
1071
self.new_pack.text_index.key_count(),
1072
time.time() - self.new_pack.start_time)
1074
def _process_inventory_lines(self, inv_lines):
1075
"""Use up the inv_lines generator and setup a text key filter."""
1076
repo = self._pack_collection.repo
1077
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
1078
inv_lines, self.revision_keys)
1080
for fileid, file_revids in fileid_revisions.iteritems():
1081
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
1082
self._text_filter = text_filter
1084
def _revision_node_readv(self, revision_nodes):
1085
"""Return the total revisions and the readv's to issue.
1087
:param revision_nodes: The revision index contents for the packs being
1088
incorporated into the new pack.
1089
:return: As per _least_readv_node_readv.
1091
return self._least_readv_node_readv(revision_nodes)
1093
def _use_pack(self, new_pack):
1094
"""Return True if new_pack should be used.
1096
:param new_pack: The pack that has just been created.
1097
:return: True if the pack should be used.
1099
return new_pack.data_inserted()
1102
class OptimisingPacker(Packer):
1103
"""A packer which spends more time to create better disk layouts."""
1105
def _revision_node_readv(self, revision_nodes):
1106
"""Return the total revisions and the readv's to issue.
1108
This sort places revisions in topological order with the ancestors
1111
:param revision_nodes: The revision index contents for the packs being
1112
incorporated into the new pack.
1113
:return: As per _least_readv_node_readv.
1115
# build an ancestors dict
1118
for index, key, value, references in revision_nodes:
1119
ancestors[key] = references[0]
1120
by_key[key] = (index, value, references)
1121
order = tsort.topo_sort(ancestors)
1123
# Single IO is pathological, but it will work as a starting point.
1125
for key in reversed(order):
1126
index, value, references = by_key[key]
1127
# ---- KnitGraphIndex.get_position
1128
bits = value[1:].split(' ')
1129
offset, length = int(bits[0]), int(bits[1])
1131
(index, [(offset, length)], [(key, value[0], references)]))
1132
# TODO: combine requests in the same index that are in ascending order.
1133
return total, requests
1135
def open_pack(self):
1136
"""Open a pack for the pack we are creating."""
1137
new_pack = super(OptimisingPacker, self).open_pack()
1138
# Turn on the optimization flags for all the index builders.
1139
new_pack.revision_index.set_optimize(for_size=True)
1140
new_pack.inventory_index.set_optimize(for_size=True)
1141
new_pack.text_index.set_optimize(for_size=True)
1142
new_pack.signature_index.set_optimize(for_size=True)
1146
class ReconcilePacker(Packer):
1147
"""A packer which regenerates indices etc as it copies.
1149
This is used by ``bzr reconcile`` to cause parent text pointers to be
1153
def _extra_init(self):
1154
self._data_changed = False
1156
def _process_inventory_lines(self, inv_lines):
1157
"""Generate a text key reference map rather for reconciling with."""
1158
repo = self._pack_collection.repo
1159
refs = repo._find_text_key_references_from_xml_inventory_lines(
1161
self._text_refs = refs
1162
# during reconcile we:
1163
# - convert unreferenced texts to full texts
1164
# - correct texts which reference a text not copied to be full texts
1165
# - copy all others as-is but with corrected parents.
1166
# - so at this point we don't know enough to decide what becomes a full
1168
self._text_filter = None
1170
def _copy_text_texts(self):
1171
"""generate what texts we should have and then copy."""
1172
self.pb.update("Copying content texts", 3)
1173
# we have three major tasks here:
1174
# 1) generate the ideal index
1175
repo = self._pack_collection.repo
1176
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1177
_1, key, _2, refs in
1178
self.new_pack.revision_index.iter_all_entries()])
1179
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1180
# 2) generate a text_nodes list that contains all the deltas that can
1181
# be used as-is, with corrected parents.
1184
discarded_nodes = []
1185
NULL_REVISION = _mod_revision.NULL_REVISION
1186
text_index_map, text_nodes = self._get_text_nodes()
1187
for node in text_nodes:
1193
ideal_parents = tuple(ideal_index[node[1]])
1195
discarded_nodes.append(node)
1196
self._data_changed = True
1198
if ideal_parents == (NULL_REVISION,):
1200
if ideal_parents == node[3][0]:
1202
ok_nodes.append(node)
1203
elif ideal_parents[0:1] == node[3][0][0:1]:
1204
# the left most parent is the same, or there are no parents
1205
# today. Either way, we can preserve the representation as
1206
# long as we change the refs to be inserted.
1207
self._data_changed = True
1208
ok_nodes.append((node[0], node[1], node[2],
1209
(ideal_parents, node[3][1])))
1210
self._data_changed = True
1212
# Reinsert this text completely
1213
bad_texts.append((node[1], ideal_parents))
1214
self._data_changed = True
1215
# we're finished with some data.
1218
# 3) bulk copy the ok data
1219
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1220
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1221
self.new_pack.text_index, readv_group_iter, total_items))
1222
# 4) adhoc copy all the other texts.
1223
# We have to topologically insert all texts otherwise we can fail to
1224
# reconcile when parts of a single delta chain are preserved intact,
1225
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1226
# reinserted, and if d3 has incorrect parents it will also be
1227
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1228
# copied), so we will try to delta, but d2 is not currently able to be
1229
# extracted because it's basis d1 is not present. Topologically sorting
1230
# addresses this. The following generates a sort for all the texts that
1231
# are being inserted without having to reference the entire text key
1232
# space (we only topo sort the revisions, which is smaller).
1233
topo_order = tsort.topo_sort(ancestors)
1234
rev_order = dict(zip(topo_order, range(len(topo_order))))
1235
bad_texts.sort(key=lambda key:rev_order[key[0][1]])
1236
transaction = repo.get_transaction()
1237
file_id_index = GraphIndexPrefixAdapter(
1238
self.new_pack.text_index,
1240
add_nodes_callback=self.new_pack.text_index.add_nodes)
1241
data_access = _DirectPackAccess(
1242
{self.new_pack.text_index:self.new_pack.access_tuple()})
1243
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1244
self.new_pack.access_tuple())
1245
output_texts = KnitVersionedFiles(
1246
_KnitGraphIndex(self.new_pack.text_index,
1247
add_callback=self.new_pack.text_index.add_nodes,
1248
deltas=True, parents=True, is_locked=repo.is_locked),
1249
data_access=data_access, max_delta_chain=200)
1250
for key, parent_keys in bad_texts:
1251
# We refer to the new pack to delta data being output.
1252
# A possible improvement would be to catch errors on short reads
1253
# and only flush then.
1254
self.new_pack.flush()
1256
for parent_key in parent_keys:
1257
if parent_key[0] != key[0]:
1258
# Graph parents must match the fileid
1259
raise errors.BzrError('Mismatched key parent %r:%r' %
1261
parents.append(parent_key[1])
1262
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1263
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1264
output_texts.add_lines(key, parent_keys, text_lines,
1265
random_id=True, check_content=False)
1266
# 5) check that nothing inserted has a reference outside the keyspace.
1267
missing_text_keys = self.new_pack.text_index._external_references()
1268
if missing_text_keys:
1269
raise errors.BzrCheckError('Reference to missing compression parents %r'
1270
% (missing_text_keys,))
1271
self._log_copied_texts()
1273
def _use_pack(self, new_pack):
1274
"""Override _use_pack to check for reconcile having changed content."""
1275
# XXX: we might be better checking this at the copy time.
1276
original_inventory_keys = set()
1277
inv_index = self._pack_collection.inventory_index.combined_index
1278
for entry in inv_index.iter_all_entries():
1279
original_inventory_keys.add(entry[1])
1280
new_inventory_keys = set()
1281
for entry in new_pack.inventory_index.iter_all_entries():
1282
new_inventory_keys.add(entry[1])
1283
if new_inventory_keys != original_inventory_keys:
1284
self._data_changed = True
1285
return new_pack.data_inserted() and self._data_changed
1288
class RepositoryPackCollection(object):
1289
"""Management of packs within a repository.
1291
:ivar _names: map of {pack_name: (index_size,)}
1294
def __init__(self, repo, transport, index_transport, upload_transport,
1295
pack_transport, index_builder_class, index_class):
1296
"""Create a new RepositoryPackCollection.
1298
:param transport: Addresses the repository base directory
1299
(typically .bzr/repository/).
1300
:param index_transport: Addresses the directory containing indices.
1301
:param upload_transport: Addresses the directory into which packs are written
1302
while they're being created.
1303
:param pack_transport: Addresses the directory of existing complete packs.
1304
:param index_builder_class: The index builder class to use.
1305
:param index_class: The index class to use.
1307
# XXX: This should call self.reset()
1309
self.transport = transport
1310
self._index_transport = index_transport
1311
self._upload_transport = upload_transport
1312
self._pack_transport = pack_transport
1313
self._index_builder_class = index_builder_class
1314
self._index_class = index_class
1315
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
1319
self._packs_by_name = {}
1320
# the previous pack-names content
1321
self._packs_at_load = None
1322
# when a pack is being created by this object, the state of that pack.
1323
self._new_pack = None
1324
# aggregated revision index data
1325
self.revision_index = AggregateIndex(self.reload_pack_names)
1326
self.inventory_index = AggregateIndex(self.reload_pack_names)
1327
self.text_index = AggregateIndex(self.reload_pack_names)
1328
self.signature_index = AggregateIndex(self.reload_pack_names)
1330
self._resumed_packs = []
1332
def add_pack_to_memory(self, pack):
1333
"""Make a Pack object available to the repository to satisfy queries.
1335
:param pack: A Pack object.
1337
if pack.name in self._packs_by_name:
1338
raise AssertionError(
1339
'pack %s already in _packs_by_name' % (pack.name,))
1340
self.packs.append(pack)
1341
self._packs_by_name[pack.name] = pack
1342
self.revision_index.add_index(pack.revision_index, pack)
1343
self.inventory_index.add_index(pack.inventory_index, pack)
1344
self.text_index.add_index(pack.text_index, pack)
1345
self.signature_index.add_index(pack.signature_index, pack)
1347
def all_packs(self):
1348
"""Return a list of all the Pack objects this repository has.
1350
Note that an in-progress pack being created is not returned.
1352
:return: A list of Pack objects for all the packs in the repository.
1355
for name in self.names():
1356
result.append(self.get_pack_by_name(name))
1360
"""Pack the pack collection incrementally.
1362
This will not attempt global reorganisation or recompression,
1363
rather it will just ensure that the total number of packs does
1364
not grow without bound. It uses the _max_pack_count method to
1365
determine if autopacking is needed, and the pack_distribution
1366
method to determine the number of revisions in each pack.
1368
If autopacking takes place then the packs name collection will have
1369
been flushed to disk - packing requires updating the name collection
1370
in synchronisation with certain steps. Otherwise the names collection
1373
:return: True if packing took place.
1377
return self._do_autopack()
1378
except errors.RetryAutopack, e:
1379
# If we get a RetryAutopack exception, we should abort the
1380
# current action, and retry.
1383
def _do_autopack(self):
1384
# XXX: Should not be needed when the management of indices is sane.
1385
total_revisions = self.revision_index.combined_index.key_count()
1386
total_packs = len(self._names)
1387
if self._max_pack_count(total_revisions) >= total_packs:
1389
# XXX: the following may want to be a class, to pack with a given
1391
# determine which packs need changing
1392
pack_distribution = self.pack_distribution(total_revisions)
1394
for pack in self.all_packs():
1395
revision_count = pack.get_revision_count()
1396
if revision_count == 0:
1397
# revision less packs are not generated by normal operation,
1398
# only by operations like sign-my-commits, and thus will not
1399
# tend to grow rapdily or without bound like commit containing
1400
# packs do - leave them alone as packing them really should
1401
# group their data with the relevant commit, and that may
1402
# involve rewriting ancient history - which autopack tries to
1403
# avoid. Alternatively we could not group the data but treat
1404
# each of these as having a single revision, and thus add
1405
# one revision for each to the total revision count, to get
1406
# a matching distribution.
1408
existing_packs.append((revision_count, pack))
1409
pack_operations = self.plan_autopack_combinations(
1410
existing_packs, pack_distribution)
1411
num_new_packs = len(pack_operations)
1412
num_old_packs = sum([len(po[1]) for po in pack_operations])
1413
num_revs_affected = sum([po[0] for po in pack_operations])
1414
mutter('Auto-packing repository %s, which has %d pack files, '
1415
'containing %d revisions. Packing %d files into %d affecting %d'
1416
' revisions', self, total_packs, total_revisions, num_old_packs,
1417
num_new_packs, num_revs_affected)
1418
self._execute_pack_operations(pack_operations,
1419
reload_func=self._restart_autopack)
1422
def _execute_pack_operations(self, pack_operations, _packer_class=Packer,
1424
"""Execute a series of pack operations.
1426
:param pack_operations: A list of [revision_count, packs_to_combine].
1427
:param _packer_class: The class of packer to use (default: Packer).
1430
for revision_count, packs in pack_operations:
1431
# we may have no-ops from the setup logic
1434
packer = _packer_class(self, packs, '.autopack',
1435
reload_func=reload_func)
1438
except errors.RetryWithNewPacks:
1439
# An exception is propagating out of this context, make sure
1440
# this packer has cleaned up. Packer() doesn't set its new_pack
1441
# state into the RepositoryPackCollection object, so we only
1442
# have access to it directly here.
1443
if packer.new_pack is not None:
1444
packer.new_pack.abort()
1447
self._remove_pack_from_memory(pack)
1448
# record the newly available packs and stop advertising the old
1450
self._save_pack_names(clear_obsolete_packs=True)
1451
# Move the old packs out of the way now they are no longer referenced.
1452
for revision_count, packs in pack_operations:
1453
self._obsolete_packs(packs)
1455
def lock_names(self):
1456
"""Acquire the mutex around the pack-names index.
1458
This cannot be used in the middle of a read-only transaction on the
1461
self.repo.control_files.lock_write()
1464
"""Pack the pack collection totally."""
1465
self.ensure_loaded()
1466
total_packs = len(self._names)
1468
# This is arguably wrong because we might not be optimal, but for
1469
# now lets leave it in. (e.g. reconcile -> one pack. But not
1472
total_revisions = self.revision_index.combined_index.key_count()
1473
# XXX: the following may want to be a class, to pack with a given
1475
mutter('Packing repository %s, which has %d pack files, '
1476
'containing %d revisions into 1 packs.', self, total_packs,
1478
# determine which packs need changing
1479
pack_distribution = [1]
1480
pack_operations = [[0, []]]
1481
for pack in self.all_packs():
1482
pack_operations[-1][0] += pack.get_revision_count()
1483
pack_operations[-1][1].append(pack)
1484
self._execute_pack_operations(pack_operations, OptimisingPacker)
1486
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1487
"""Plan a pack operation.
1489
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1491
:param pack_distribution: A list with the number of revisions desired
1494
if len(existing_packs) <= len(pack_distribution):
1496
existing_packs.sort(reverse=True)
1497
pack_operations = [[0, []]]
1498
# plan out what packs to keep, and what to reorganise
1499
while len(existing_packs):
1500
# take the largest pack, and if its less than the head of the
1501
# distribution chart we will include its contents in the new pack
1502
# for that position. If its larger, we remove its size from the
1503
# distribution chart
1504
next_pack_rev_count, next_pack = existing_packs.pop(0)
1505
if next_pack_rev_count >= pack_distribution[0]:
1506
# this is already packed 'better' than this, so we can
1507
# not waste time packing it.
1508
while next_pack_rev_count > 0:
1509
next_pack_rev_count -= pack_distribution[0]
1510
if next_pack_rev_count >= 0:
1512
del pack_distribution[0]
1514
# didn't use that entire bucket up
1515
pack_distribution[0] = -next_pack_rev_count
1517
# add the revisions we're going to add to the next output pack
1518
pack_operations[-1][0] += next_pack_rev_count
1519
# allocate this pack to the next pack sub operation
1520
pack_operations[-1][1].append(next_pack)
1521
if pack_operations[-1][0] >= pack_distribution[0]:
1522
# this pack is used up, shift left.
1523
del pack_distribution[0]
1524
pack_operations.append([0, []])
1525
# Now that we know which pack files we want to move, shove them all
1526
# into a single pack file.
1528
final_pack_list = []
1529
for num_revs, pack_files in pack_operations:
1530
final_rev_count += num_revs
1531
final_pack_list.extend(pack_files)
1532
if len(final_pack_list) == 1:
1533
raise AssertionError('We somehow generated an autopack with a'
1534
' single pack file being moved.')
1536
return [[final_rev_count, final_pack_list]]
1538
def ensure_loaded(self):
1539
"""Ensure we have read names from disk.
1541
:return: True if the disk names had not been previously read.
1543
# NB: if you see an assertion error here, its probably access against
1544
# an unlocked repo. Naughty.
1545
if not self.repo.is_locked():
1546
raise errors.ObjectNotLocked(self.repo)
1547
if self._names is None:
1549
self._packs_at_load = set()
1550
for index, key, value in self._iter_disk_pack_index():
1552
self._names[name] = self._parse_index_sizes(value)
1553
self._packs_at_load.add((key, value))
1557
# populate all the metadata.
1561
def _parse_index_sizes(self, value):
1562
"""Parse a string of index sizes."""
1563
return tuple([int(digits) for digits in value.split(' ')])
1565
def get_pack_by_name(self, name):
1566
"""Get a Pack object by name.
1568
:param name: The name of the pack - e.g. '123456'
1569
:return: A Pack object.
1572
return self._packs_by_name[name]
1574
rev_index = self._make_index(name, '.rix')
1575
inv_index = self._make_index(name, '.iix')
1576
txt_index = self._make_index(name, '.tix')
1577
sig_index = self._make_index(name, '.six')
1578
result = ExistingPack(self._pack_transport, name, rev_index,
1579
inv_index, txt_index, sig_index)
1580
self.add_pack_to_memory(result)
1583
def _resume_pack(self, name):
1584
"""Get a suspended Pack object by name.
1586
:param name: The name of the pack - e.g. '123456'
1587
:return: A Pack object.
1589
if not re.match('[a-f0-9]{32}', name):
1590
# Tokens should be md5sums of the suspended pack file, i.e. 32 hex
1592
raise errors.UnresumableWriteGroup(
1593
self.repo, [name], 'Malformed write group token')
1595
rev_index = self._make_index(name, '.rix', resume=True)
1596
inv_index = self._make_index(name, '.iix', resume=True)
1597
txt_index = self._make_index(name, '.tix', resume=True)
1598
sig_index = self._make_index(name, '.six', resume=True)
1599
result = ResumedPack(name, rev_index, inv_index, txt_index,
1600
sig_index, self._upload_transport, self._pack_transport,
1601
self._index_transport, self)
1602
except errors.NoSuchFile, e:
1603
raise errors.UnresumableWriteGroup(self.repo, [name], str(e))
1604
self.add_pack_to_memory(result)
1605
self._resumed_packs.append(result)
1608
def allocate(self, a_new_pack):
1609
"""Allocate name in the list of packs.
1611
:param a_new_pack: A NewPack instance to be added to the collection of
1612
packs for this repository.
1614
self.ensure_loaded()
1615
if a_new_pack.name in self._names:
1616
raise errors.BzrError(
1617
'Pack %r already exists in %s' % (a_new_pack.name, self))
1618
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1619
self.add_pack_to_memory(a_new_pack)
1621
def _iter_disk_pack_index(self):
1622
"""Iterate over the contents of the pack-names index.
1624
This is used when loading the list from disk, and before writing to
1625
detect updates from others during our write operation.
1626
:return: An iterator of the index contents.
1628
return self._index_class(self.transport, 'pack-names', None
1629
).iter_all_entries()
1631
def _make_index(self, name, suffix, resume=False):
1632
size_offset = self._suffix_offsets[suffix]
1633
index_name = name + suffix
1635
transport = self._upload_transport
1636
index_size = transport.stat(index_name).st_size
1638
transport = self._index_transport
1639
index_size = self._names[name][size_offset]
1640
return self._index_class(transport, index_name, index_size)
1642
def _max_pack_count(self, total_revisions):
1643
"""Return the maximum number of packs to use for total revisions.
1645
:param total_revisions: The total number of revisions in the
1648
if not total_revisions:
1650
digits = str(total_revisions)
1652
for digit in digits:
1653
result += int(digit)
1657
"""Provide an order to the underlying names."""
1658
return sorted(self._names.keys())
1660
def _obsolete_packs(self, packs):
1661
"""Move a number of packs which have been obsoleted out of the way.
1663
Each pack and its associated indices are moved out of the way.
1665
Note: for correctness this function should only be called after a new
1666
pack names index has been written without these pack names, and with
1667
the names of packs that contain the data previously available via these
1670
:param packs: The packs to obsolete.
1671
:param return: None.
1674
pack.pack_transport.rename(pack.file_name(),
1675
'../obsolete_packs/' + pack.file_name())
1676
# TODO: Probably needs to know all possible indices for this pack
1677
# - or maybe list the directory and move all indices matching this
1678
# name whether we recognize it or not?
1679
for suffix in ('.iix', '.six', '.tix', '.rix'):
1680
self._index_transport.rename(pack.name + suffix,
1681
'../obsolete_packs/' + pack.name + suffix)
1683
def pack_distribution(self, total_revisions):
1684
"""Generate a list of the number of revisions to put in each pack.
1686
:param total_revisions: The total number of revisions in the
1689
if total_revisions == 0:
1691
digits = reversed(str(total_revisions))
1693
for exponent, count in enumerate(digits):
1694
size = 10 ** exponent
1695
for pos in range(int(count)):
1697
return list(reversed(result))
1699
def _pack_tuple(self, name):
1700
"""Return a tuple with the transport and file name for a pack name."""
1701
return self._pack_transport, name + '.pack'
1703
def _remove_pack_from_memory(self, pack):
1704
"""Remove pack from the packs accessed by this repository.
1706
Only affects memory state, until self._save_pack_names() is invoked.
1708
self._names.pop(pack.name)
1709
self._packs_by_name.pop(pack.name)
1710
self._remove_pack_indices(pack)
1711
self.packs.remove(pack)
1713
def _remove_pack_indices(self, pack):
1714
"""Remove the indices for pack from the aggregated indices."""
1715
self.revision_index.remove_index(pack.revision_index, pack)
1716
self.inventory_index.remove_index(pack.inventory_index, pack)
1717
self.text_index.remove_index(pack.text_index, pack)
1718
self.signature_index.remove_index(pack.signature_index, pack)
1721
"""Clear all cached data."""
1722
# cached revision data
1723
self.repo._revision_knit = None
1724
self.revision_index.clear()
1725
# cached signature data
1726
self.repo._signature_knit = None
1727
self.signature_index.clear()
1728
# cached file text data
1729
self.text_index.clear()
1730
self.repo._text_knit = None
1731
# cached inventory data
1732
self.inventory_index.clear()
1733
# remove the open pack
1734
self._new_pack = None
1735
# information about packs.
1738
self._packs_by_name = {}
1739
self._packs_at_load = None
1741
def _unlock_names(self):
1742
"""Release the mutex around the pack-names index."""
1743
self.repo.control_files.unlock()
1745
def _diff_pack_names(self):
1746
"""Read the pack names from disk, and compare it to the one in memory.
1748
:return: (disk_nodes, deleted_nodes, new_nodes)
1749
disk_nodes The final set of nodes that should be referenced
1750
deleted_nodes Nodes which have been removed from when we started
1751
new_nodes Nodes that are newly introduced
1753
# load the disk nodes across
1755
for index, key, value in self._iter_disk_pack_index():
1756
disk_nodes.add((key, value))
1758
# do a two-way diff against our original content
1759
current_nodes = set()
1760
for name, sizes in self._names.iteritems():
1762
((name, ), ' '.join(str(size) for size in sizes)))
1764
# Packs no longer present in the repository, which were present when we
1765
# locked the repository
1766
deleted_nodes = self._packs_at_load - current_nodes
1767
# Packs which this process is adding
1768
new_nodes = current_nodes - self._packs_at_load
1770
# Update the disk_nodes set to include the ones we are adding, and
1771
# remove the ones which were removed by someone else
1772
disk_nodes.difference_update(deleted_nodes)
1773
disk_nodes.update(new_nodes)
1775
return disk_nodes, deleted_nodes, new_nodes
1777
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1778
"""Given the correct set of pack files, update our saved info.
1780
:return: (removed, added, modified)
1781
removed pack names removed from self._names
1782
added pack names added to self._names
1783
modified pack names that had changed value
1788
## self._packs_at_load = disk_nodes
1789
new_names = dict(disk_nodes)
1790
# drop no longer present nodes
1791
for pack in self.all_packs():
1792
if (pack.name,) not in new_names:
1793
removed.append(pack.name)
1794
self._remove_pack_from_memory(pack)
1795
# add new nodes/refresh existing ones
1796
for key, value in disk_nodes:
1798
sizes = self._parse_index_sizes(value)
1799
if name in self._names:
1801
if sizes != self._names[name]:
1802
# the pack for name has had its indices replaced - rare but
1803
# important to handle. XXX: probably can never happen today
1804
# because the three-way merge code above does not handle it
1805
# - you may end up adding the same key twice to the new
1806
# disk index because the set values are the same, unless
1807
# the only index shows up as deleted by the set difference
1808
# - which it may. Until there is a specific test for this,
1809
# assume its broken. RBC 20071017.
1810
self._remove_pack_from_memory(self.get_pack_by_name(name))
1811
self._names[name] = sizes
1812
self.get_pack_by_name(name)
1813
modified.append(name)
1816
self._names[name] = sizes
1817
self.get_pack_by_name(name)
1819
return removed, added, modified
1821
def _save_pack_names(self, clear_obsolete_packs=False):
1822
"""Save the list of packs.
1824
This will take out the mutex around the pack names list for the
1825
duration of the method call. If concurrent updates have been made, a
1826
three-way merge between the current list and the current in memory list
1829
:param clear_obsolete_packs: If True, clear out the contents of the
1830
obsolete_packs directory.
1834
builder = self._index_builder_class()
1835
disk_nodes, deleted_nodes, new_nodes = self._diff_pack_names()
1836
# TODO: handle same-name, index-size-changes here -
1837
# e.g. use the value from disk, not ours, *unless* we're the one
1839
for key, value in disk_nodes:
1840
builder.add_node(key, value)
1841
self.transport.put_file('pack-names', builder.finish(),
1842
mode=self.repo.bzrdir._get_file_mode())
1843
# move the baseline forward
1844
self._packs_at_load = disk_nodes
1845
if clear_obsolete_packs:
1846
self._clear_obsolete_packs()
1848
self._unlock_names()
1849
# synchronise the memory packs list with what we just wrote:
1850
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1852
def reload_pack_names(self):
1853
"""Sync our pack listing with what is present in the repository.
1855
This should be called when we find out that something we thought was
1856
present is now missing. This happens when another process re-packs the
1859
:return: True if the in-memory list of packs has been altered at all.
1861
# The ensure_loaded call is to handle the case where the first call
1862
# made involving the collection was to reload_pack_names, where we
1863
# don't have a view of disk contents. Its a bit of a bandaid, and
1864
# causes two reads of pack-names, but its a rare corner case not struck
1865
# with regular push/pull etc.
1866
first_read = self.ensure_loaded()
1869
# out the new value.
1870
disk_nodes, _, _ = self._diff_pack_names()
1871
self._packs_at_load = disk_nodes
1873
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1874
if removed or added or modified:
1878
def _restart_autopack(self):
1879
"""Reload the pack names list, and restart the autopack code."""
1880
if not self.reload_pack_names():
1881
# Re-raise the original exception, because something went missing
1882
# and a restart didn't find it
1884
raise errors.RetryAutopack(self.repo, False, sys.exc_info())
1886
def _clear_obsolete_packs(self):
1887
"""Delete everything from the obsolete-packs directory.
1889
obsolete_pack_transport = self.transport.clone('obsolete_packs')
1890
for filename in obsolete_pack_transport.list_dir('.'):
1892
obsolete_pack_transport.delete(filename)
1893
except (errors.PathError, errors.TransportError), e:
1894
warning("couldn't delete obsolete pack, skipping it:\n%s" % (e,))
1896
def _start_write_group(self):
1897
# Do not permit preparation for writing if we're not in a 'write lock'.
1898
if not self.repo.is_write_locked():
1899
raise errors.NotWriteLocked(self)
1900
self._new_pack = NewPack(self, upload_suffix='.pack',
1901
file_mode=self.repo.bzrdir._get_file_mode())
1902
# allow writing: queue writes to a new index
1903
self.revision_index.add_writable_index(self._new_pack.revision_index,
1905
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
1907
self.text_index.add_writable_index(self._new_pack.text_index,
1909
self.signature_index.add_writable_index(self._new_pack.signature_index,
1912
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
1913
self.repo.revisions._index._add_callback = self.revision_index.add_callback
1914
self.repo.signatures._index._add_callback = self.signature_index.add_callback
1915
self.repo.texts._index._add_callback = self.text_index.add_callback
1917
def _abort_write_group(self):
1918
# FIXME: just drop the transient index.
1919
# forget what names there are
1920
if self._new_pack is not None:
1922
self._new_pack.abort()
1924
# XXX: If we aborted while in the middle of finishing the write
1925
# group, _remove_pack_indices can fail because the indexes are
1926
# already gone. If they're not there we shouldn't fail in this
1927
# case. -- mbp 20081113
1928
self._remove_pack_indices(self._new_pack)
1929
self._new_pack = None
1930
for resumed_pack in self._resumed_packs:
1932
resumed_pack.abort()
1934
# See comment in previous finally block.
1936
self._remove_pack_indices(resumed_pack)
1939
del self._resumed_packs[:]
1940
self.repo._text_knit = None
1942
def _remove_resumed_pack_indices(self):
1943
for resumed_pack in self._resumed_packs:
1944
self._remove_pack_indices(resumed_pack)
1945
del self._resumed_packs[:]
1947
def _commit_write_group(self):
1949
for prefix, versioned_file in (
1950
('revisions', self.repo.revisions),
1951
('inventories', self.repo.inventories),
1952
('texts', self.repo.texts),
1953
('signatures', self.repo.signatures),
1955
missing = versioned_file.get_missing_compression_parent_keys()
1956
all_missing.update([(prefix,) + key for key in missing])
1958
raise errors.BzrCheckError(
1959
"Repository %s has missing compression parent(s) %r "
1960
% (self.repo, sorted(all_missing)))
1961
self._remove_pack_indices(self._new_pack)
1962
should_autopack = False
1963
if self._new_pack.data_inserted():
1964
# get all the data to disk and read to use
1965
self._new_pack.finish()
1966
self.allocate(self._new_pack)
1967
self._new_pack = None
1968
should_autopack = True
1970
self._new_pack.abort()
1971
self._new_pack = None
1972
for resumed_pack in self._resumed_packs:
1973
# XXX: this is a pretty ugly way to turn the resumed pack into a
1974
# properly committed pack.
1975
self._names[resumed_pack.name] = None
1976
self._remove_pack_from_memory(resumed_pack)
1977
resumed_pack.finish()
1978
self.allocate(resumed_pack)
1979
should_autopack = True
1980
del self._resumed_packs[:]
1982
if not self.autopack():
1983
# when autopack takes no steps, the names list is still
1985
self._save_pack_names()
1986
self.repo._text_knit = None
1988
def _suspend_write_group(self):
1989
tokens = [pack.name for pack in self._resumed_packs]
1990
self._remove_pack_indices(self._new_pack)
1991
if self._new_pack.data_inserted():
1992
# get all the data to disk and read to use
1993
self._new_pack.finish(suspend=True)
1994
tokens.append(self._new_pack.name)
1995
self._new_pack = None
1997
self._new_pack.abort()
1998
self._new_pack = None
1999
self._remove_resumed_pack_indices()
2000
self.repo._text_knit = None
2003
def _resume_write_group(self, tokens):
2004
for token in tokens:
2005
self._resume_pack(token)
2008
class KnitPackRepository(KnitRepository):
2009
"""Repository with knit objects stored inside pack containers.
2011
The layering for a KnitPackRepository is:
2013
Graph | HPSS | Repository public layer |
2014
===================================================
2015
Tuple based apis below, string based, and key based apis above
2016
---------------------------------------------------
2018
Provides .texts, .revisions etc
2019
This adapts the N-tuple keys to physical knit records which only have a
2020
single string identifier (for historical reasons), which in older formats
2021
was always the revision_id, and in the mapped code for packs is always
2022
the last element of key tuples.
2023
---------------------------------------------------
2025
A separate GraphIndex is used for each of the
2026
texts/inventories/revisions/signatures contained within each individual
2027
pack file. The GraphIndex layer works in N-tuples and is unaware of any
2029
===================================================
2033
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
2035
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2036
_commit_builder_class, _serializer)
2037
index_transport = self._transport.clone('indices')
2038
self._pack_collection = RepositoryPackCollection(self, self._transport,
2040
self._transport.clone('upload'),
2041
self._transport.clone('packs'),
2042
_format.index_builder_class,
2043
_format.index_class)
2044
self.inventories = KnitVersionedFiles(
2045
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2046
add_callback=self._pack_collection.inventory_index.add_callback,
2047
deltas=True, parents=True, is_locked=self.is_locked),
2048
data_access=self._pack_collection.inventory_index.data_access,
2049
max_delta_chain=200)
2050
self.revisions = KnitVersionedFiles(
2051
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2052
add_callback=self._pack_collection.revision_index.add_callback,
2053
deltas=False, parents=True, is_locked=self.is_locked),
2054
data_access=self._pack_collection.revision_index.data_access,
2056
self.signatures = KnitVersionedFiles(
2057
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2058
add_callback=self._pack_collection.signature_index.add_callback,
2059
deltas=False, parents=False, is_locked=self.is_locked),
2060
data_access=self._pack_collection.signature_index.data_access,
2062
self.texts = KnitVersionedFiles(
2063
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2064
add_callback=self._pack_collection.text_index.add_callback,
2065
deltas=True, parents=True, is_locked=self.is_locked),
2066
data_access=self._pack_collection.text_index.data_access,
2067
max_delta_chain=200)
2068
# True when the repository object is 'write locked' (as opposed to the
2069
# physical lock only taken out around changes to the pack-names list.)
2070
# Another way to represent this would be a decorator around the control
2071
# files object that presents logical locks as physical ones - if this
2072
# gets ugly consider that alternative design. RBC 20071011
2073
self._write_lock_count = 0
2074
self._transaction = None
2076
self._reconcile_does_inventory_gc = True
2077
self._reconcile_fixes_text_parents = True
2078
self._reconcile_backsup_inventory = False
2080
def _warn_if_deprecated(self):
2081
# This class isn't deprecated, but one sub-format is
2082
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2083
from bzrlib import repository
2084
if repository._deprecation_warning_done:
2086
repository._deprecation_warning_done = True
2087
warning("Format %s for %s is deprecated - please use"
2088
" 'bzr upgrade --1.6.1-rich-root'"
2089
% (self._format, self.bzrdir.transport.base))
2091
def _abort_write_group(self):
2092
self._pack_collection._abort_write_group()
2094
def _find_inconsistent_revision_parents(self):
2095
"""Find revisions with incorrectly cached parents.
2097
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
2098
parents-in-revision).
2100
if not self.is_locked():
2101
raise errors.ObjectNotLocked(self)
2102
pb = ui.ui_factory.nested_progress_bar()
2105
revision_nodes = self._pack_collection.revision_index \
2106
.combined_index.iter_all_entries()
2107
index_positions = []
2108
# Get the cached index values for all revisions, and also the location
2109
# in each index of the revision text so we can perform linear IO.
2110
for index, key, value, refs in revision_nodes:
2111
pos, length = value[1:].split(' ')
2112
index_positions.append((index, int(pos), key[0],
2113
tuple(parent[0] for parent in refs[0])))
2114
pb.update("Reading revision index", 0, 0)
2115
index_positions.sort()
2116
batch_count = len(index_positions) / 1000 + 1
2117
pb.update("Checking cached revision graph", 0, batch_count)
2118
for offset in xrange(batch_count):
2119
pb.update("Checking cached revision graph", offset)
2120
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
2123
rev_ids = [item[2] for item in to_query]
2124
revs = self.get_revisions(rev_ids)
2125
for revision, item in zip(revs, to_query):
2126
index_parents = item[3]
2127
rev_parents = tuple(revision.parent_ids)
2128
if index_parents != rev_parents:
2129
result.append((revision.revision_id, index_parents, rev_parents))
2134
def _make_parents_provider(self):
2135
return graph.CachingParentsProvider(self)
2137
def _refresh_data(self):
2138
if not self.is_locked():
2140
self._pack_collection.reload_pack_names()
2142
def _start_write_group(self):
2143
self._pack_collection._start_write_group()
2145
def _commit_write_group(self):
2146
return self._pack_collection._commit_write_group()
2148
def suspend_write_group(self):
2149
# XXX check self._write_group is self.get_transaction()?
2150
tokens = self._pack_collection._suspend_write_group()
2151
self._write_group = None
2154
def _resume_write_group(self, tokens):
2155
self._start_write_group()
2156
self._pack_collection._resume_write_group(tokens)
2158
def get_transaction(self):
2159
if self._write_lock_count:
2160
return self._transaction
2162
return self.control_files.get_transaction()
2164
def is_locked(self):
2165
return self._write_lock_count or self.control_files.is_locked()
2167
def is_write_locked(self):
2168
return self._write_lock_count
2170
def lock_write(self, token=None):
2171
locked = self.is_locked()
2172
if not self._write_lock_count and locked:
2173
raise errors.ReadOnlyError(self)
2174
self._write_lock_count += 1
2175
if self._write_lock_count == 1:
2176
self._transaction = transactions.WriteTransaction()
2177
for repo in self._fallback_repositories:
2178
# Writes don't affect fallback repos
2181
self._refresh_data()
2183
def lock_read(self):
2184
locked = self.is_locked()
2185
if self._write_lock_count:
2186
self._write_lock_count += 1
2188
self.control_files.lock_read()
2189
for repo in self._fallback_repositories:
2190
# Writes don't affect fallback repos
2193
self._refresh_data()
2195
def leave_lock_in_place(self):
2196
# not supported - raise an error
2197
raise NotImplementedError(self.leave_lock_in_place)
2199
def dont_leave_lock_in_place(self):
2200
# not supported - raise an error
2201
raise NotImplementedError(self.dont_leave_lock_in_place)
2205
"""Compress the data within the repository.
2207
This will pack all the data to a single pack. In future it may
2208
recompress deltas or do other such expensive operations.
2210
self._pack_collection.pack()
2213
def reconcile(self, other=None, thorough=False):
2214
"""Reconcile this repository."""
2215
from bzrlib.reconcile import PackReconciler
2216
reconciler = PackReconciler(self, thorough=thorough)
2217
reconciler.reconcile()
2221
if self._write_lock_count == 1 and self._write_group is not None:
2222
self.abort_write_group()
2223
self._transaction = None
2224
self._write_lock_count = 0
2225
raise errors.BzrError(
2226
'Must end write group before releasing write lock on %s'
2228
if self._write_lock_count:
2229
self._write_lock_count -= 1
2230
if not self._write_lock_count:
2231
transaction = self._transaction
2232
self._transaction = None
2233
transaction.finish()
2234
for repo in self._fallback_repositories:
2237
self.control_files.unlock()
2238
for repo in self._fallback_repositories:
2242
class RepositoryFormatPack(MetaDirRepositoryFormat):
2243
"""Format logic for pack structured repositories.
2245
This repository format has:
2246
- a list of packs in pack-names
2247
- packs in packs/NAME.pack
2248
- indices in indices/NAME.{iix,six,tix,rix}
2249
- knit deltas in the packs, knit indices mapped to the indices.
2250
- thunk objects to support the knits programming API.
2251
- a format marker of its own
2252
- an optional 'shared-storage' flag
2253
- an optional 'no-working-trees' flag
2257
# Set this attribute in derived classes to control the repository class
2258
# created by open and initialize.
2259
repository_class = None
2260
# Set this attribute in derived classes to control the
2261
# _commit_builder_class that the repository objects will have passed to
2262
# their constructor.
2263
_commit_builder_class = None
2264
# Set this attribute in derived clases to control the _serializer that the
2265
# repository objects will have passed to their constructor.
2267
# Packs are not confused by ghosts.
2268
supports_ghosts = True
2269
# External references are not supported in pack repositories yet.
2270
supports_external_lookups = False
2271
# What index classes to use
2272
index_builder_class = None
2274
_fetch_uses_deltas = True
2277
def initialize(self, a_bzrdir, shared=False):
2278
"""Create a pack based repository.
2280
:param a_bzrdir: bzrdir to contain the new repository; must already
2282
:param shared: If true the repository will be initialized as a shared
2285
mutter('creating repository in %s.', a_bzrdir.transport.base)
2286
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
2287
builder = self.index_builder_class()
2288
files = [('pack-names', builder.finish())]
2289
utf8_files = [('format', self.get_format_string())]
2291
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
2292
return self.open(a_bzrdir=a_bzrdir, _found=True)
2294
def open(self, a_bzrdir, _found=False, _override_transport=None):
2295
"""See RepositoryFormat.open().
2297
:param _override_transport: INTERNAL USE ONLY. Allows opening the
2298
repository at a slightly different url
2299
than normal. I.e. during 'upgrade'.
2302
format = RepositoryFormat.find_format(a_bzrdir)
2303
if _override_transport is not None:
2304
repo_transport = _override_transport
2306
repo_transport = a_bzrdir.get_repository_transport(None)
2307
control_files = lockable_files.LockableFiles(repo_transport,
2308
'lock', lockdir.LockDir)
2309
return self.repository_class(_format=self,
2311
control_files=control_files,
2312
_commit_builder_class=self._commit_builder_class,
2313
_serializer=self._serializer)
2316
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2317
"""A no-subtrees parameterized Pack repository.
2319
This format was introduced in 0.92.
2322
repository_class = KnitPackRepository
2323
_commit_builder_class = PackCommitBuilder
2325
def _serializer(self):
2326
return xml5.serializer_v5
2327
# What index classes to use
2328
index_builder_class = InMemoryGraphIndex
2329
index_class = GraphIndex
2331
def _get_matching_bzrdir(self):
2332
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2334
def _ignore_setting_bzrdir(self, format):
2337
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2339
def get_format_string(self):
2340
"""See RepositoryFormat.get_format_string()."""
2341
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2343
def get_format_description(self):
2344
"""See RepositoryFormat.get_format_description()."""
2345
return "Packs containing knits without subtree support"
2347
def check_conversion_target(self, target_format):
2351
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2352
"""A subtrees parameterized Pack repository.
2354
This repository format uses the xml7 serializer to get:
2355
- support for recording full info about the tree root
2356
- support for recording tree-references
2358
This format was introduced in 0.92.
2361
repository_class = KnitPackRepository
2362
_commit_builder_class = PackRootCommitBuilder
2363
rich_root_data = True
2364
supports_tree_reference = True
2366
def _serializer(self):
2367
return xml7.serializer_v7
2368
# What index classes to use
2369
index_builder_class = InMemoryGraphIndex
2370
index_class = GraphIndex
2372
def _get_matching_bzrdir(self):
2373
return bzrdir.format_registry.make_bzrdir(
2374
'pack-0.92-subtree')
2376
def _ignore_setting_bzrdir(self, format):
2379
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2381
def check_conversion_target(self, target_format):
2382
if not target_format.rich_root_data:
2383
raise errors.BadConversionTarget(
2384
'Does not support rich root data.', target_format)
2385
if not getattr(target_format, 'supports_tree_reference', False):
2386
raise errors.BadConversionTarget(
2387
'Does not support nested trees', target_format)
2389
def get_format_string(self):
2390
"""See RepositoryFormat.get_format_string()."""
2391
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2393
def get_format_description(self):
2394
"""See RepositoryFormat.get_format_description()."""
2395
return "Packs containing knits with subtree support\n"
2398
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2399
"""A rich-root, no subtrees parameterized Pack repository.
2401
This repository format uses the xml6 serializer to get:
2402
- support for recording full info about the tree root
2404
This format was introduced in 1.0.
2407
repository_class = KnitPackRepository
2408
_commit_builder_class = PackRootCommitBuilder
2409
rich_root_data = True
2410
supports_tree_reference = False
2412
def _serializer(self):
2413
return xml6.serializer_v6
2414
# What index classes to use
2415
index_builder_class = InMemoryGraphIndex
2416
index_class = GraphIndex
2418
def _get_matching_bzrdir(self):
2419
return bzrdir.format_registry.make_bzrdir(
2422
def _ignore_setting_bzrdir(self, format):
2425
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2427
def check_conversion_target(self, target_format):
2428
if not target_format.rich_root_data:
2429
raise errors.BadConversionTarget(
2430
'Does not support rich root data.', target_format)
2432
def get_format_string(self):
2433
"""See RepositoryFormat.get_format_string()."""
2434
return ("Bazaar pack repository format 1 with rich root"
2435
" (needs bzr 1.0)\n")
2437
def get_format_description(self):
2438
"""See RepositoryFormat.get_format_description()."""
2439
return "Packs containing knits with rich root support\n"
2442
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2443
"""Repository that supports external references to allow stacking.
2447
Supports external lookups, which results in non-truncated ghosts after
2448
reconcile compared to pack-0.92 formats.
2451
repository_class = KnitPackRepository
2452
_commit_builder_class = PackCommitBuilder
2453
supports_external_lookups = True
2454
# What index classes to use
2455
index_builder_class = InMemoryGraphIndex
2456
index_class = GraphIndex
2459
def _serializer(self):
2460
return xml5.serializer_v5
2462
def _get_matching_bzrdir(self):
2463
return bzrdir.format_registry.make_bzrdir('1.6')
2465
def _ignore_setting_bzrdir(self, format):
2468
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2470
def get_format_string(self):
2471
"""See RepositoryFormat.get_format_string()."""
2472
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2474
def get_format_description(self):
2475
"""See RepositoryFormat.get_format_description()."""
2476
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2478
def check_conversion_target(self, target_format):
2482
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2483
"""A repository with rich roots and stacking.
2485
New in release 1.6.1.
2487
Supports stacking on other repositories, allowing data to be accessed
2488
without being stored locally.
2491
repository_class = KnitPackRepository
2492
_commit_builder_class = PackRootCommitBuilder
2493
rich_root_data = True
2494
supports_tree_reference = False # no subtrees
2495
supports_external_lookups = True
2496
# What index classes to use
2497
index_builder_class = InMemoryGraphIndex
2498
index_class = GraphIndex
2501
def _serializer(self):
2502
return xml6.serializer_v6
2504
def _get_matching_bzrdir(self):
2505
return bzrdir.format_registry.make_bzrdir(
2508
def _ignore_setting_bzrdir(self, format):
2511
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2513
def check_conversion_target(self, target_format):
2514
if not target_format.rich_root_data:
2515
raise errors.BadConversionTarget(
2516
'Does not support rich root data.', target_format)
2518
def get_format_string(self):
2519
"""See RepositoryFormat.get_format_string()."""
2520
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2522
def get_format_description(self):
2523
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2526
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2527
"""A repository with rich roots and external references.
2531
Supports external lookups, which results in non-truncated ghosts after
2532
reconcile compared to pack-0.92 formats.
2534
This format was deprecated because the serializer it uses accidentally
2535
supported subtrees, when the format was not intended to. This meant that
2536
someone could accidentally fetch from an incorrect repository.
2539
repository_class = KnitPackRepository
2540
_commit_builder_class = PackRootCommitBuilder
2541
rich_root_data = True
2542
supports_tree_reference = False # no subtrees
2544
supports_external_lookups = True
2545
# What index classes to use
2546
index_builder_class = InMemoryGraphIndex
2547
index_class = GraphIndex
2550
def _serializer(self):
2551
return xml7.serializer_v7
2553
def _get_matching_bzrdir(self):
2554
matching = bzrdir.format_registry.make_bzrdir(
2556
matching.repository_format = self
2559
def _ignore_setting_bzrdir(self, format):
2562
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2564
def check_conversion_target(self, target_format):
2565
if not target_format.rich_root_data:
2566
raise errors.BadConversionTarget(
2567
'Does not support rich root data.', target_format)
2569
def get_format_string(self):
2570
"""See RepositoryFormat.get_format_string()."""
2571
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2573
def get_format_description(self):
2574
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2578
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2579
"""A repository with stacking and btree indexes,
2580
without rich roots or subtrees.
2582
This is equivalent to pack-1.6 with B+Tree indices.
2585
repository_class = KnitPackRepository
2586
_commit_builder_class = PackCommitBuilder
2587
supports_external_lookups = True
2588
# What index classes to use
2589
index_builder_class = BTreeBuilder
2590
index_class = BTreeGraphIndex
2593
def _serializer(self):
2594
return xml5.serializer_v5
2596
def _get_matching_bzrdir(self):
2597
return bzrdir.format_registry.make_bzrdir('1.9')
2599
def _ignore_setting_bzrdir(self, format):
2602
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2604
def get_format_string(self):
2605
"""See RepositoryFormat.get_format_string()."""
2606
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2608
def get_format_description(self):
2609
"""See RepositoryFormat.get_format_description()."""
2610
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2612
def check_conversion_target(self, target_format):
2616
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2617
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2619
1.6-rich-root with B+Tree indices.
2622
repository_class = KnitPackRepository
2623
_commit_builder_class = PackRootCommitBuilder
2624
rich_root_data = True
2625
supports_tree_reference = False # no subtrees
2626
supports_external_lookups = True
2627
# What index classes to use
2628
index_builder_class = BTreeBuilder
2629
index_class = BTreeGraphIndex
2632
def _serializer(self):
2633
return xml6.serializer_v6
2635
def _get_matching_bzrdir(self):
2636
return bzrdir.format_registry.make_bzrdir(
2639
def _ignore_setting_bzrdir(self, format):
2642
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2644
def check_conversion_target(self, target_format):
2645
if not target_format.rich_root_data:
2646
raise errors.BadConversionTarget(
2647
'Does not support rich root data.', target_format)
2649
def get_format_string(self):
2650
"""See RepositoryFormat.get_format_string()."""
2651
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2653
def get_format_description(self):
2654
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2657
class RepositoryFormatPackDevelopment2(RepositoryFormatPack):
2658
"""A no-subtrees development repository.
2660
This format should be retained until the second release after bzr 1.7.
2662
This is pack-1.6.1 with B+Tree indices.
2665
repository_class = KnitPackRepository
2666
_commit_builder_class = PackCommitBuilder
2667
supports_external_lookups = True
2668
# What index classes to use
2669
index_builder_class = BTreeBuilder
2670
index_class = BTreeGraphIndex
2671
# Set to true to get the fast-commit code path tested until a really fast
2672
# format lands in trunk. Not actually fast in this format.
2676
def _serializer(self):
2677
return xml5.serializer_v5
2679
def _get_matching_bzrdir(self):
2680
return bzrdir.format_registry.make_bzrdir('development2')
2682
def _ignore_setting_bzrdir(self, format):
2685
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2687
def get_format_string(self):
2688
"""See RepositoryFormat.get_format_string()."""
2689
return "Bazaar development format 2 (needs bzr.dev from before 1.8)\n"
2691
def get_format_description(self):
2692
"""See RepositoryFormat.get_format_description()."""
2693
return ("Development repository format, currently the same as "
2694
"1.6.1 with B+Trees.\n")
2696
def check_conversion_target(self, target_format):
2700
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2701
"""A subtrees development repository.
2703
This format should be retained until the second release after bzr 1.7.
2705
1.6.1-subtree[as it might have been] with B+Tree indices.
2708
repository_class = KnitPackRepository
2709
_commit_builder_class = PackRootCommitBuilder
2710
rich_root_data = True
2711
supports_tree_reference = True
2712
supports_external_lookups = True
2713
# What index classes to use
2714
index_builder_class = BTreeBuilder
2715
index_class = BTreeGraphIndex
2718
def _serializer(self):
2719
return xml7.serializer_v7
2721
def _get_matching_bzrdir(self):
2722
return bzrdir.format_registry.make_bzrdir(
2723
'development2-subtree')
2725
def _ignore_setting_bzrdir(self, format):
2728
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2730
def check_conversion_target(self, target_format):
2731
if not target_format.rich_root_data:
2732
raise errors.BadConversionTarget(
2733
'Does not support rich root data.', target_format)
2734
if not getattr(target_format, 'supports_tree_reference', False):
2735
raise errors.BadConversionTarget(
2736
'Does not support nested trees', target_format)
2738
def get_format_string(self):
2739
"""See RepositoryFormat.get_format_string()."""
2740
return ("Bazaar development format 2 with subtree support "
2741
"(needs bzr.dev from before 1.8)\n")
2743
def get_format_description(self):
2744
"""See RepositoryFormat.get_format_description()."""
2745
return ("Development repository format, currently the same as "
2746
"1.6.1-subtree with B+Tree indices.\n")
added
removed
bzrlib/repofmt/pack_repo.py
