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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
from bzrlib.lazy_import import lazy_import
18
lazy_import(globals(), """
19
from itertools import izip
30
from bzrlib.index import (
35
GraphIndexPrefixAdapter,
37
from bzrlib.knit import KnitGraphIndex, _PackAccess, _KnitData
38
from bzrlib.osutils import rand_chars
39
from bzrlib.pack import ContainerWriter
40
from bzrlib.store import revision
41
from bzrlib import tsort
58
from bzrlib.decorators import needs_read_lock, needs_write_lock
59
from bzrlib.repofmt.knitrepo import KnitRepository
60
from bzrlib.repository import (
63
MetaDirRepositoryFormat,
67
import bzrlib.revision as _mod_revision
68
from bzrlib.store.revision.knit import KnitRevisionStore
69
from bzrlib.store.versioned import VersionedFileStore
70
from bzrlib.trace import (
78
class PackCommitBuilder(CommitBuilder):
79
"""A subclass of CommitBuilder to add texts with pack semantics.
81
Specifically this uses one knit object rather than one knit object per
82
added text, reducing memory and object pressure.
85
def __init__(self, repository, parents, config, timestamp=None,
86
timezone=None, committer=None, revprops=None,
88
CommitBuilder.__init__(self, repository, parents, config,
89
timestamp=timestamp, timezone=timezone, committer=committer,
90
revprops=revprops, revision_id=revision_id)
91
self._file_graph = graph.Graph(
92
repository._pack_collection.text_index.combined_index)
94
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
95
return self.repository._pack_collection._add_text_to_weave(file_id,
96
self._new_revision_id, new_lines, parents, nostore_sha,
99
def _heads(self, file_id, revision_ids):
100
keys = [(file_id, revision_id) for revision_id in revision_ids]
101
return set([key[1] for key in self._file_graph.heads(keys)])
104
class PackRootCommitBuilder(RootCommitBuilder):
105
"""A subclass of RootCommitBuilder to add texts with pack semantics.
107
Specifically this uses one knit object rather than one knit object per
108
added text, reducing memory and object pressure.
111
def __init__(self, repository, parents, config, timestamp=None,
112
timezone=None, committer=None, revprops=None,
114
CommitBuilder.__init__(self, repository, parents, config,
115
timestamp=timestamp, timezone=timezone, committer=committer,
116
revprops=revprops, revision_id=revision_id)
117
self._file_graph = graph.Graph(
118
repository._pack_collection.text_index.combined_index)
120
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
121
return self.repository._pack_collection._add_text_to_weave(file_id,
122
self._new_revision_id, new_lines, parents, nostore_sha,
125
def _heads(self, file_id, revision_ids):
126
keys = [(file_id, revision_id) for revision_id in revision_ids]
127
return set([key[1] for key in self._file_graph.heads(keys)])
131
"""An in memory proxy for a pack and its indices.
133
This is a base class that is not directly used, instead the classes
134
ExistingPack and NewPack are used.
137
def __init__(self, revision_index, inventory_index, text_index,
139
"""Create a pack instance.
141
:param revision_index: A GraphIndex for determining what revisions are
142
present in the Pack and accessing the locations of their texts.
143
:param inventory_index: A GraphIndex for determining what inventories are
144
present in the Pack and accessing the locations of their
146
:param text_index: A GraphIndex for determining what file texts
147
are present in the pack and accessing the locations of their
148
texts/deltas (via (fileid, revisionid) tuples).
149
:param revision_index: A GraphIndex for determining what signatures are
150
present in the Pack and accessing the locations of their texts.
152
self.revision_index = revision_index
153
self.inventory_index = inventory_index
154
self.text_index = text_index
155
self.signature_index = signature_index
157
def access_tuple(self):
158
"""Return a tuple (transport, name) for the pack content."""
159
return self.pack_transport, self.file_name()
162
"""Get the file name for the pack on disk."""
163
return self.name + '.pack'
165
def get_revision_count(self):
166
return self.revision_index.key_count()
168
def inventory_index_name(self, name):
169
"""The inv index is the name + .iix."""
170
return self.index_name('inventory', name)
172
def revision_index_name(self, name):
173
"""The revision index is the name + .rix."""
174
return self.index_name('revision', name)
176
def signature_index_name(self, name):
177
"""The signature index is the name + .six."""
178
return self.index_name('signature', name)
180
def text_index_name(self, name):
181
"""The text index is the name + .tix."""
182
return self.index_name('text', name)
184
def _external_compression_parents_of_texts(self):
187
for node in self.text_index.iter_all_entries():
189
refs.update(node[3][1])
193
class ExistingPack(Pack):
194
"""An in memory proxy for an existing .pack and its disk indices."""
196
def __init__(self, pack_transport, name, revision_index, inventory_index,
197
text_index, signature_index):
198
"""Create an ExistingPack object.
200
:param pack_transport: The transport where the pack file resides.
201
:param name: The name of the pack on disk in the pack_transport.
203
Pack.__init__(self, revision_index, inventory_index, text_index,
206
self.pack_transport = pack_transport
207
if None in (revision_index, inventory_index, text_index,
208
signature_index, name, pack_transport):
209
raise AssertionError()
211
def __eq__(self, other):
212
return self.__dict__ == other.__dict__
214
def __ne__(self, other):
215
return not self.__eq__(other)
218
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
219
id(self), self.transport, self.name)
223
"""An in memory proxy for a pack which is being created."""
225
# A map of index 'type' to the file extension and position in the
227
index_definitions = {
228
'revision': ('.rix', 0),
229
'inventory': ('.iix', 1),
231
'signature': ('.six', 3),
234
def __init__(self, upload_transport, index_transport, pack_transport,
235
upload_suffix='', file_mode=None):
236
"""Create a NewPack instance.
238
:param upload_transport: A writable transport for the pack to be
239
incrementally uploaded to.
240
:param index_transport: A writable transport for the pack's indices to
241
be written to when the pack is finished.
242
:param pack_transport: A writable transport for the pack to be renamed
243
to when the upload is complete. This *must* be the same as
244
upload_transport.clone('../packs').
245
:param upload_suffix: An optional suffix to be given to any temporary
246
files created during the pack creation. e.g '.autopack'
247
:param file_mode: An optional file mode to create the new files with.
249
# The relative locations of the packs are constrained, but all are
250
# passed in because the caller has them, so as to avoid object churn.
252
# Revisions: parents list, no text compression.
253
InMemoryGraphIndex(reference_lists=1),
254
# Inventory: We want to map compression only, but currently the
255
# knit code hasn't been updated enough to understand that, so we
256
# have a regular 2-list index giving parents and compression
258
InMemoryGraphIndex(reference_lists=2),
259
# Texts: compression and per file graph, for all fileids - so two
260
# reference lists and two elements in the key tuple.
261
InMemoryGraphIndex(reference_lists=2, key_elements=2),
262
# Signatures: Just blobs to store, no compression, no parents
264
InMemoryGraphIndex(reference_lists=0),
266
# where should the new pack be opened
267
self.upload_transport = upload_transport
268
# where are indices written out to
269
self.index_transport = index_transport
270
# where is the pack renamed to when it is finished?
271
self.pack_transport = pack_transport
272
# What file mode to upload the pack and indices with.
273
self._file_mode = file_mode
274
# tracks the content written to the .pack file.
275
self._hash = md5.new()
276
# a four-tuple with the length in bytes of the indices, once the pack
277
# is finalised. (rev, inv, text, sigs)
278
self.index_sizes = None
279
# How much data to cache when writing packs. Note that this is not
280
# synchronised with reads, because it's not in the transport layer, so
281
# is not safe unless the client knows it won't be reading from the pack
283
self._cache_limit = 0
284
# the temporary pack file name.
285
self.random_name = rand_chars(20) + upload_suffix
286
# when was this pack started ?
287
self.start_time = time.time()
288
# open an output stream for the data added to the pack.
289
self.write_stream = self.upload_transport.open_write_stream(
290
self.random_name, mode=self._file_mode)
291
if 'pack' in debug.debug_flags:
292
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
293
time.ctime(), self.upload_transport.base, self.random_name,
294
time.time() - self.start_time)
295
# A list of byte sequences to be written to the new pack, and the
296
# aggregate size of them. Stored as a list rather than separate
297
# variables so that the _write_data closure below can update them.
298
self._buffer = [[], 0]
299
# create a callable for adding data
301
# robertc says- this is a closure rather than a method on the object
302
# so that the variables are locals, and faster than accessing object
304
def _write_data(bytes, flush=False, _buffer=self._buffer,
305
_write=self.write_stream.write, _update=self._hash.update):
306
_buffer[0].append(bytes)
307
_buffer[1] += len(bytes)
309
if _buffer[1] > self._cache_limit or flush:
310
bytes = ''.join(_buffer[0])
314
# expose this on self, for the occasion when clients want to add data.
315
self._write_data = _write_data
316
# a pack writer object to serialise pack records.
317
self._writer = pack.ContainerWriter(self._write_data)
319
# what state is the pack in? (open, finished, aborted)
323
"""Cancel creating this pack."""
324
self._state = 'aborted'
325
self.write_stream.close()
326
# Remove the temporary pack file.
327
self.upload_transport.delete(self.random_name)
328
# The indices have no state on disk.
330
def access_tuple(self):
331
"""Return a tuple (transport, name) for the pack content."""
332
if self._state == 'finished':
333
return Pack.access_tuple(self)
334
elif self._state == 'open':
335
return self.upload_transport, self.random_name
337
raise AssertionError(self._state)
339
def data_inserted(self):
340
"""True if data has been added to this pack."""
341
return bool(self.get_revision_count() or
342
self.inventory_index.key_count() or
343
self.text_index.key_count() or
344
self.signature_index.key_count())
347
"""Finish the new pack.
350
- finalises the content
351
- assigns a name (the md5 of the content, currently)
352
- writes out the associated indices
353
- renames the pack into place.
354
- stores the index size tuple for the pack in the index_sizes
359
self._write_data('', flush=True)
360
self.name = self._hash.hexdigest()
362
# XXX: It'd be better to write them all to temporary names, then
363
# rename them all into place, so that the window when only some are
364
# visible is smaller. On the other hand none will be seen until
365
# they're in the names list.
366
self.index_sizes = [None, None, None, None]
367
self._write_index('revision', self.revision_index, 'revision')
368
self._write_index('inventory', self.inventory_index, 'inventory')
369
self._write_index('text', self.text_index, 'file texts')
370
self._write_index('signature', self.signature_index,
371
'revision signatures')
372
self.write_stream.close()
373
# Note that this will clobber an existing pack with the same name,
374
# without checking for hash collisions. While this is undesirable this
375
# is something that can be rectified in a subsequent release. One way
376
# to rectify it may be to leave the pack at the original name, writing
377
# its pack-names entry as something like 'HASH: index-sizes
378
# temporary-name'. Allocate that and check for collisions, if it is
379
# collision free then rename it into place. If clients know this scheme
380
# they can handle missing-file errors by:
381
# - try for HASH.pack
382
# - try for temporary-name
383
# - refresh the pack-list to see if the pack is now absent
384
self.upload_transport.rename(self.random_name,
385
'../packs/' + self.name + '.pack')
386
self._state = 'finished'
387
if 'pack' in debug.debug_flags:
388
# XXX: size might be interesting?
389
mutter('%s: create_pack: pack renamed into place: %s%s->%s%s t+%6.3fs',
390
time.ctime(), self.upload_transport.base, self.random_name,
391
self.pack_transport, self.name,
392
time.time() - self.start_time)
395
"""Flush any current data."""
397
bytes = ''.join(self._buffer[0])
398
self.write_stream.write(bytes)
399
self._hash.update(bytes)
400
self._buffer[:] = [[], 0]
402
def index_name(self, index_type, name):
403
"""Get the disk name of an index type for pack name 'name'."""
404
return name + NewPack.index_definitions[index_type][0]
406
def index_offset(self, index_type):
407
"""Get the position in a index_size array for a given index type."""
408
return NewPack.index_definitions[index_type][1]
410
def _replace_index_with_readonly(self, index_type):
411
setattr(self, index_type + '_index',
412
GraphIndex(self.index_transport,
413
self.index_name(index_type, self.name),
414
self.index_sizes[self.index_offset(index_type)]))
416
def set_write_cache_size(self, size):
417
self._cache_limit = size
419
def _write_index(self, index_type, index, label):
420
"""Write out an index.
422
:param index_type: The type of index to write - e.g. 'revision'.
423
:param index: The index object to serialise.
424
:param label: What label to give the index e.g. 'revision'.
426
index_name = self.index_name(index_type, self.name)
427
self.index_sizes[self.index_offset(index_type)] = \
428
self.index_transport.put_file(index_name, index.finish(),
429
mode=self._file_mode)
430
if 'pack' in debug.debug_flags:
431
# XXX: size might be interesting?
432
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
433
time.ctime(), label, self.upload_transport.base,
434
self.random_name, time.time() - self.start_time)
435
# Replace the writable index on this object with a readonly,
436
# presently unloaded index. We should alter
437
# the index layer to make its finish() error if add_node is
438
# subsequently used. RBC
439
self._replace_index_with_readonly(index_type)
442
class AggregateIndex(object):
443
"""An aggregated index for the RepositoryPackCollection.
445
AggregateIndex is reponsible for managing the PackAccess object,
446
Index-To-Pack mapping, and all indices list for a specific type of index
447
such as 'revision index'.
449
A CombinedIndex provides an index on a single key space built up
450
from several on-disk indices. The AggregateIndex builds on this
451
to provide a knit access layer, and allows having up to one writable
452
index within the collection.
454
# XXX: Probably 'can be written to' could/should be separated from 'acts
455
# like a knit index' -- mbp 20071024
458
"""Create an AggregateIndex."""
459
self.index_to_pack = {}
460
self.combined_index = CombinedGraphIndex([])
461
self.knit_access = _PackAccess(self.index_to_pack)
463
def replace_indices(self, index_to_pack, indices):
464
"""Replace the current mappings with fresh ones.
466
This should probably not be used eventually, rather incremental add and
467
removal of indices. It has been added during refactoring of existing
470
:param index_to_pack: A mapping from index objects to
471
(transport, name) tuples for the pack file data.
472
:param indices: A list of indices.
474
# refresh the revision pack map dict without replacing the instance.
475
self.index_to_pack.clear()
476
self.index_to_pack.update(index_to_pack)
477
# XXX: API break - clearly a 'replace' method would be good?
478
self.combined_index._indices[:] = indices
479
# the current add nodes callback for the current writable index if
481
self.add_callback = None
483
def add_index(self, index, pack):
484
"""Add index to the aggregate, which is an index for Pack pack.
486
Future searches on the aggregate index will seach this new index
487
before all previously inserted indices.
489
:param index: An Index for the pack.
490
:param pack: A Pack instance.
492
# expose it to the index map
493
self.index_to_pack[index] = pack.access_tuple()
494
# put it at the front of the linear index list
495
self.combined_index.insert_index(0, index)
497
def add_writable_index(self, index, pack):
498
"""Add an index which is able to have data added to it.
500
There can be at most one writable index at any time. Any
501
modifications made to the knit are put into this index.
503
:param index: An index from the pack parameter.
504
:param pack: A Pack instance.
506
if self.add_callback is not None:
507
raise AssertionError(
508
"%s already has a writable index through %s" % \
509
(self, self.add_callback))
510
# allow writing: queue writes to a new index
511
self.add_index(index, pack)
512
# Updates the index to packs mapping as a side effect,
513
self.knit_access.set_writer(pack._writer, index, pack.access_tuple())
514
self.add_callback = index.add_nodes
517
"""Reset all the aggregate data to nothing."""
518
self.knit_access.set_writer(None, None, (None, None))
519
self.index_to_pack.clear()
520
del self.combined_index._indices[:]
521
self.add_callback = None
523
def remove_index(self, index, pack):
524
"""Remove index from the indices used to answer queries.
526
:param index: An index from the pack parameter.
527
:param pack: A Pack instance.
529
del self.index_to_pack[index]
530
self.combined_index._indices.remove(index)
531
if (self.add_callback is not None and
532
getattr(index, 'add_nodes', None) == self.add_callback):
533
self.add_callback = None
534
self.knit_access.set_writer(None, None, (None, None))
537
class Packer(object):
538
"""Create a pack from packs."""
540
def __init__(self, pack_collection, packs, suffix, revision_ids=None):
543
:param pack_collection: A RepositoryPackCollection object where the
544
new pack is being written to.
545
:param packs: The packs to combine.
546
:param suffix: The suffix to use on the temporary files for the pack.
547
:param revision_ids: Revision ids to limit the pack to.
551
self.revision_ids = revision_ids
552
# The pack object we are creating.
554
self._pack_collection = pack_collection
555
# The index layer keys for the revisions being copied. None for 'all
557
self._revision_keys = None
558
# What text keys to copy. None for 'all texts'. This is set by
559
# _copy_inventory_texts
560
self._text_filter = None
563
def _extra_init(self):
564
"""A template hook to allow extending the constructor trivially."""
566
def pack(self, pb=None):
567
"""Create a new pack by reading data from other packs.
569
This does little more than a bulk copy of data. One key difference
570
is that data with the same item key across multiple packs is elided
571
from the output. The new pack is written into the current pack store
572
along with its indices, and the name added to the pack names. The
573
source packs are not altered and are not required to be in the current
576
:param pb: An optional progress bar to use. A nested bar is created if
578
:return: A Pack object, or None if nothing was copied.
580
# open a pack - using the same name as the last temporary file
581
# - which has already been flushed, so its safe.
582
# XXX: - duplicate code warning with start_write_group; fix before
583
# considering 'done'.
584
if self._pack_collection._new_pack is not None:
585
raise errors.BzrError('call to create_pack_from_packs while '
586
'another pack is being written.')
587
if self.revision_ids is not None:
588
if len(self.revision_ids) == 0:
589
# silly fetch request.
592
self.revision_ids = frozenset(self.revision_ids)
594
self.pb = ui.ui_factory.nested_progress_bar()
598
return self._create_pack_from_packs()
604
"""Open a pack for the pack we are creating."""
605
return NewPack(self._pack_collection._upload_transport,
606
self._pack_collection._index_transport,
607
self._pack_collection._pack_transport, upload_suffix=self.suffix,
608
file_mode=self._pack_collection.repo.control_files._file_mode)
610
def _copy_revision_texts(self):
611
"""Copy revision data to the new pack."""
613
if self.revision_ids:
614
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
617
# select revision keys
618
revision_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
619
self.packs, 'revision_index')[0]
620
revision_nodes = self._pack_collection._index_contents(revision_index_map, revision_keys)
621
# copy revision keys and adjust values
622
self.pb.update("Copying revision texts", 1)
623
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
624
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
625
self.new_pack.revision_index, readv_group_iter, total_items))
626
if 'pack' in debug.debug_flags:
627
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
628
time.ctime(), self._pack_collection._upload_transport.base,
629
self.new_pack.random_name,
630
self.new_pack.revision_index.key_count(),
631
time.time() - self.new_pack.start_time)
632
self._revision_keys = revision_keys
634
def _copy_inventory_texts(self):
635
"""Copy the inventory texts to the new pack.
637
self._revision_keys is used to determine what inventories to copy.
639
Sets self._text_filter appropriately.
641
# select inventory keys
642
inv_keys = self._revision_keys # currently the same keyspace, and note that
643
# querying for keys here could introduce a bug where an inventory item
644
# is missed, so do not change it to query separately without cross
645
# checking like the text key check below.
646
inventory_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
647
self.packs, 'inventory_index')[0]
648
inv_nodes = self._pack_collection._index_contents(inventory_index_map, inv_keys)
649
# copy inventory keys and adjust values
650
# XXX: Should be a helper function to allow different inv representation
652
self.pb.update("Copying inventory texts", 2)
653
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
654
# Only grab the output lines if we will be processing them
655
output_lines = bool(self.revision_ids)
656
inv_lines = self._copy_nodes_graph(inventory_index_map,
657
self.new_pack._writer, self.new_pack.inventory_index,
658
readv_group_iter, total_items, output_lines=output_lines)
659
if self.revision_ids:
660
self._process_inventory_lines(inv_lines)
662
# eat the iterator to cause it to execute.
664
self._text_filter = None
665
if 'pack' in debug.debug_flags:
666
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
667
time.ctime(), self._pack_collection._upload_transport.base,
668
self.new_pack.random_name,
669
self.new_pack.inventory_index.key_count(),
670
time.time() - self.new_pack.start_time)
672
def _copy_text_texts(self):
674
text_index_map, text_nodes = self._get_text_nodes()
675
if self._text_filter is not None:
676
# We could return the keys copied as part of the return value from
677
# _copy_nodes_graph but this doesn't work all that well with the
678
# need to get line output too, so we check separately, and as we're
679
# going to buffer everything anyway, we check beforehand, which
680
# saves reading knit data over the wire when we know there are
682
text_nodes = set(text_nodes)
683
present_text_keys = set(_node[1] for _node in text_nodes)
684
missing_text_keys = set(self._text_filter) - present_text_keys
685
if missing_text_keys:
686
# TODO: raise a specific error that can handle many missing
688
a_missing_key = missing_text_keys.pop()
689
raise errors.RevisionNotPresent(a_missing_key[1],
691
# copy text keys and adjust values
692
self.pb.update("Copying content texts", 3)
693
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
694
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
695
self.new_pack.text_index, readv_group_iter, total_items))
696
self._log_copied_texts()
698
def _check_references(self):
699
"""Make sure our external refereneces are present."""
700
external_refs = self.new_pack._external_compression_parents_of_texts()
702
index = self._pack_collection.text_index.combined_index
703
found_items = list(index.iter_entries(external_refs))
704
if len(found_items) != len(external_refs):
705
found_keys = set(k for idx, k, refs, value in found_items)
706
missing_items = external_refs - found_keys
707
missing_file_id, missing_revision_id = missing_items.pop()
708
raise errors.RevisionNotPresent(missing_revision_id,
711
def _create_pack_from_packs(self):
712
self.pb.update("Opening pack", 0, 5)
713
self.new_pack = self.open_pack()
714
new_pack = self.new_pack
715
# buffer data - we won't be reading-back during the pack creation and
716
# this makes a significant difference on sftp pushes.
717
new_pack.set_write_cache_size(1024*1024)
718
if 'pack' in debug.debug_flags:
719
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
720
for a_pack in self.packs]
721
if self.revision_ids is not None:
722
rev_count = len(self.revision_ids)
725
mutter('%s: create_pack: creating pack from source packs: '
726
'%s%s %s revisions wanted %s t=0',
727
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
728
plain_pack_list, rev_count)
729
self._copy_revision_texts()
730
self._copy_inventory_texts()
731
self._copy_text_texts()
732
# select signature keys
733
signature_filter = self._revision_keys # same keyspace
734
signature_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
735
self.packs, 'signature_index')[0]
736
signature_nodes = self._pack_collection._index_contents(signature_index_map,
738
# copy signature keys and adjust values
739
self.pb.update("Copying signature texts", 4)
740
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
741
new_pack.signature_index)
742
if 'pack' in debug.debug_flags:
743
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
744
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
745
new_pack.signature_index.key_count(),
746
time.time() - new_pack.start_time)
747
self._check_references()
748
if not self._use_pack(new_pack):
751
self.pb.update("Finishing pack", 5)
753
self._pack_collection.allocate(new_pack)
756
def _copy_nodes(self, nodes, index_map, writer, write_index):
757
"""Copy knit nodes between packs with no graph references."""
758
pb = ui.ui_factory.nested_progress_bar()
760
return self._do_copy_nodes(nodes, index_map, writer,
765
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb):
766
# for record verification
767
knit_data = _KnitData(None)
768
# plan a readv on each source pack:
770
nodes = sorted(nodes)
771
# how to map this into knit.py - or knit.py into this?
772
# we don't want the typical knit logic, we want grouping by pack
773
# at this point - perhaps a helper library for the following code
774
# duplication points?
776
for index, key, value in nodes:
777
if index not in request_groups:
778
request_groups[index] = []
779
request_groups[index].append((key, value))
781
pb.update("Copied record", record_index, len(nodes))
782
for index, items in request_groups.iteritems():
783
pack_readv_requests = []
784
for key, value in items:
785
# ---- KnitGraphIndex.get_position
786
bits = value[1:].split(' ')
787
offset, length = int(bits[0]), int(bits[1])
788
pack_readv_requests.append((offset, length, (key, value[0])))
789
# linear scan up the pack
790
pack_readv_requests.sort()
792
transport, path = index_map[index]
793
reader = pack.make_readv_reader(transport, path,
794
[offset[0:2] for offset in pack_readv_requests])
795
for (names, read_func), (_1, _2, (key, eol_flag)) in \
796
izip(reader.iter_records(), pack_readv_requests):
797
raw_data = read_func(None)
798
# check the header only
799
df, _ = knit_data._parse_record_header(key[-1], raw_data)
801
pos, size = writer.add_bytes_record(raw_data, names)
802
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
803
pb.update("Copied record", record_index)
806
def _copy_nodes_graph(self, index_map, writer, write_index,
807
readv_group_iter, total_items, output_lines=False):
808
"""Copy knit nodes between packs.
810
:param output_lines: Return lines present in the copied data as
811
an iterator of line,version_id.
813
pb = ui.ui_factory.nested_progress_bar()
815
for result in self._do_copy_nodes_graph(index_map, writer,
816
write_index, output_lines, pb, readv_group_iter, total_items):
819
# Python 2.4 does not permit try:finally: in a generator.
825
def _do_copy_nodes_graph(self, index_map, writer, write_index,
826
output_lines, pb, readv_group_iter, total_items):
827
# for record verification
828
knit_data = _KnitData(None)
829
# for line extraction when requested (inventories only)
831
factory = knit.KnitPlainFactory()
833
pb.update("Copied record", record_index, total_items)
834
for index, readv_vector, node_vector in readv_group_iter:
836
transport, path = index_map[index]
837
reader = pack.make_readv_reader(transport, path, readv_vector)
838
for (names, read_func), (key, eol_flag, references) in \
839
izip(reader.iter_records(), node_vector):
840
raw_data = read_func(None)
843
# read the entire thing
844
content, _ = knit_data._parse_record(version_id, raw_data)
845
if len(references[-1]) == 0:
846
line_iterator = factory.get_fulltext_content(content)
848
line_iterator = factory.get_linedelta_content(content)
849
for line in line_iterator:
850
yield line, version_id
852
# check the header only
853
df, _ = knit_data._parse_record_header(version_id, raw_data)
855
pos, size = writer.add_bytes_record(raw_data, names)
856
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
857
pb.update("Copied record", record_index)
860
def _get_text_nodes(self):
861
text_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
862
self.packs, 'text_index')[0]
863
return text_index_map, self._pack_collection._index_contents(text_index_map,
866
def _least_readv_node_readv(self, nodes):
867
"""Generate request groups for nodes using the least readv's.
869
:param nodes: An iterable of graph index nodes.
870
:return: Total node count and an iterator of the data needed to perform
871
readvs to obtain the data for nodes. Each item yielded by the
872
iterator is a tuple with:
873
index, readv_vector, node_vector. readv_vector is a list ready to
874
hand to the transport readv method, and node_vector is a list of
875
(key, eol_flag, references) for the the node retrieved by the
876
matching readv_vector.
878
# group by pack so we do one readv per pack
879
nodes = sorted(nodes)
882
for index, key, value, references in nodes:
883
if index not in request_groups:
884
request_groups[index] = []
885
request_groups[index].append((key, value, references))
887
for index, items in request_groups.iteritems():
888
pack_readv_requests = []
889
for key, value, references in items:
890
# ---- KnitGraphIndex.get_position
891
bits = value[1:].split(' ')
892
offset, length = int(bits[0]), int(bits[1])
893
pack_readv_requests.append(
894
((offset, length), (key, value[0], references)))
895
# linear scan up the pack to maximum range combining.
896
pack_readv_requests.sort()
897
# split out the readv and the node data.
898
pack_readv = [readv for readv, node in pack_readv_requests]
899
node_vector = [node for readv, node in pack_readv_requests]
900
result.append((index, pack_readv, node_vector))
903
def _log_copied_texts(self):
904
if 'pack' in debug.debug_flags:
905
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
906
time.ctime(), self._pack_collection._upload_transport.base,
907
self.new_pack.random_name,
908
self.new_pack.text_index.key_count(),
909
time.time() - self.new_pack.start_time)
911
def _process_inventory_lines(self, inv_lines):
912
"""Use up the inv_lines generator and setup a text key filter."""
913
repo = self._pack_collection.repo
914
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
915
inv_lines, self.revision_ids)
917
for fileid, file_revids in fileid_revisions.iteritems():
918
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
919
self._text_filter = text_filter
921
def _revision_node_readv(self, revision_nodes):
922
"""Return the total revisions and the readv's to issue.
924
:param revision_nodes: The revision index contents for the packs being
925
incorporated into the new pack.
926
:return: As per _least_readv_node_readv.
928
return self._least_readv_node_readv(revision_nodes)
930
def _use_pack(self, new_pack):
931
"""Return True if new_pack should be used.
933
:param new_pack: The pack that has just been created.
934
:return: True if the pack should be used.
936
return new_pack.data_inserted()
939
class OptimisingPacker(Packer):
940
"""A packer which spends more time to create better disk layouts."""
942
def _revision_node_readv(self, revision_nodes):
943
"""Return the total revisions and the readv's to issue.
945
This sort places revisions in topological order with the ancestors
948
:param revision_nodes: The revision index contents for the packs being
949
incorporated into the new pack.
950
:return: As per _least_readv_node_readv.
952
# build an ancestors dict
955
for index, key, value, references in revision_nodes:
956
ancestors[key] = references[0]
957
by_key[key] = (index, value, references)
958
order = tsort.topo_sort(ancestors)
960
# Single IO is pathological, but it will work as a starting point.
962
for key in reversed(order):
963
index, value, references = by_key[key]
964
# ---- KnitGraphIndex.get_position
965
bits = value[1:].split(' ')
966
offset, length = int(bits[0]), int(bits[1])
968
(index, [(offset, length)], [(key, value[0], references)]))
969
# TODO: combine requests in the same index that are in ascending order.
970
return total, requests
973
class ReconcilePacker(Packer):
974
"""A packer which regenerates indices etc as it copies.
976
This is used by ``bzr reconcile`` to cause parent text pointers to be
980
def _extra_init(self):
981
self._data_changed = False
983
def _process_inventory_lines(self, inv_lines):
984
"""Generate a text key reference map rather for reconciling with."""
985
repo = self._pack_collection.repo
986
refs = repo._find_text_key_references_from_xml_inventory_lines(
988
self._text_refs = refs
989
# during reconcile we:
990
# - convert unreferenced texts to full texts
991
# - correct texts which reference a text not copied to be full texts
992
# - copy all others as-is but with corrected parents.
993
# - so at this point we don't know enough to decide what becomes a full
995
self._text_filter = None
997
def _copy_text_texts(self):
998
"""generate what texts we should have and then copy."""
999
self.pb.update("Copying content texts", 3)
1000
# we have three major tasks here:
1001
# 1) generate the ideal index
1002
repo = self._pack_collection.repo
1003
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1004
_1, key, _2, refs in
1005
self.new_pack.revision_index.iter_all_entries()])
1006
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1007
# 2) generate a text_nodes list that contains all the deltas that can
1008
# be used as-is, with corrected parents.
1011
discarded_nodes = []
1012
NULL_REVISION = _mod_revision.NULL_REVISION
1013
text_index_map, text_nodes = self._get_text_nodes()
1014
for node in text_nodes:
1020
ideal_parents = tuple(ideal_index[node[1]])
1022
discarded_nodes.append(node)
1023
self._data_changed = True
1025
if ideal_parents == (NULL_REVISION,):
1027
if ideal_parents == node[3][0]:
1029
ok_nodes.append(node)
1030
elif ideal_parents[0:1] == node[3][0][0:1]:
1031
# the left most parent is the same, or there are no parents
1032
# today. Either way, we can preserve the representation as
1033
# long as we change the refs to be inserted.
1034
self._data_changed = True
1035
ok_nodes.append((node[0], node[1], node[2],
1036
(ideal_parents, node[3][1])))
1037
self._data_changed = True
1039
# Reinsert this text completely
1040
bad_texts.append((node[1], ideal_parents))
1041
self._data_changed = True
1042
# we're finished with some data.
1045
# 3) bulk copy the ok data
1046
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1047
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1048
self.new_pack.text_index, readv_group_iter, total_items))
1049
# 4) adhoc copy all the other texts.
1050
# We have to topologically insert all texts otherwise we can fail to
1051
# reconcile when parts of a single delta chain are preserved intact,
1052
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1053
# reinserted, and if d3 has incorrect parents it will also be
1054
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1055
# copied), so we will try to delta, but d2 is not currently able to be
1056
# extracted because it's basis d1 is not present. Topologically sorting
1057
# addresses this. The following generates a sort for all the texts that
1058
# are being inserted without having to reference the entire text key
1059
# space (we only topo sort the revisions, which is smaller).
1060
topo_order = tsort.topo_sort(ancestors)
1061
rev_order = dict(zip(topo_order, range(len(topo_order))))
1062
bad_texts.sort(key=lambda key:rev_order[key[0][1]])
1063
transaction = repo.get_transaction()
1064
file_id_index = GraphIndexPrefixAdapter(
1065
self.new_pack.text_index,
1067
add_nodes_callback=self.new_pack.text_index.add_nodes)
1068
knit_index = KnitGraphIndex(file_id_index,
1069
add_callback=file_id_index.add_nodes,
1070
deltas=True, parents=True)
1071
output_knit = knit.KnitVersionedFile('reconcile-texts',
1072
self._pack_collection.transport,
1074
access_method=_PackAccess(
1075
{self.new_pack.text_index:self.new_pack.access_tuple()},
1076
(self.new_pack._writer, self.new_pack.text_index)),
1077
factory=knit.KnitPlainFactory())
1078
for key, parent_keys in bad_texts:
1079
# We refer to the new pack to delta data being output.
1080
# A possible improvement would be to catch errors on short reads
1081
# and only flush then.
1082
self.new_pack.flush()
1084
for parent_key in parent_keys:
1085
if parent_key[0] != key[0]:
1086
# Graph parents must match the fileid
1087
raise errors.BzrError('Mismatched key parent %r:%r' %
1089
parents.append(parent_key[1])
1090
source_weave = repo.weave_store.get_weave(key[0], transaction)
1091
text_lines = source_weave.get_lines(key[1])
1092
# adapt the 'knit' to the current file_id.
1093
file_id_index = GraphIndexPrefixAdapter(
1094
self.new_pack.text_index,
1096
add_nodes_callback=self.new_pack.text_index.add_nodes)
1097
knit_index._graph_index = file_id_index
1098
knit_index._add_callback = file_id_index.add_nodes
1099
output_knit.add_lines_with_ghosts(
1100
key[1], parents, text_lines, random_id=True, check_content=False)
1101
# 5) check that nothing inserted has a reference outside the keyspace.
1102
missing_text_keys = self.new_pack._external_compression_parents_of_texts()
1103
if missing_text_keys:
1104
raise errors.BzrError('Reference to missing compression parents %r'
1105
% (missing_text_keys,))
1106
self._log_copied_texts()
1108
def _use_pack(self, new_pack):
1109
"""Override _use_pack to check for reconcile having changed content."""
1110
# XXX: we might be better checking this at the copy time.
1111
original_inventory_keys = set()
1112
inv_index = self._pack_collection.inventory_index.combined_index
1113
for entry in inv_index.iter_all_entries():
1114
original_inventory_keys.add(entry[1])
1115
new_inventory_keys = set()
1116
for entry in new_pack.inventory_index.iter_all_entries():
1117
new_inventory_keys.add(entry[1])
1118
if new_inventory_keys != original_inventory_keys:
1119
self._data_changed = True
1120
return new_pack.data_inserted() and self._data_changed
1123
class RepositoryPackCollection(object):
1124
"""Management of packs within a repository."""
1126
def __init__(self, repo, transport, index_transport, upload_transport,
1128
"""Create a new RepositoryPackCollection.
1130
:param transport: Addresses the repository base directory
1131
(typically .bzr/repository/).
1132
:param index_transport: Addresses the directory containing indices.
1133
:param upload_transport: Addresses the directory into which packs are written
1134
while they're being created.
1135
:param pack_transport: Addresses the directory of existing complete packs.
1138
self.transport = transport
1139
self._index_transport = index_transport
1140
self._upload_transport = upload_transport
1141
self._pack_transport = pack_transport
1142
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
1145
self._packs_by_name = {}
1146
# the previous pack-names content
1147
self._packs_at_load = None
1148
# when a pack is being created by this object, the state of that pack.
1149
self._new_pack = None
1150
# aggregated revision index data
1151
self.revision_index = AggregateIndex()
1152
self.inventory_index = AggregateIndex()
1153
self.text_index = AggregateIndex()
1154
self.signature_index = AggregateIndex()
1156
def add_pack_to_memory(self, pack):
1157
"""Make a Pack object available to the repository to satisfy queries.
1159
:param pack: A Pack object.
1161
if pack.name in self._packs_by_name:
1162
raise AssertionError()
1163
self.packs.append(pack)
1164
self._packs_by_name[pack.name] = pack
1165
self.revision_index.add_index(pack.revision_index, pack)
1166
self.inventory_index.add_index(pack.inventory_index, pack)
1167
self.text_index.add_index(pack.text_index, pack)
1168
self.signature_index.add_index(pack.signature_index, pack)
1170
def _add_text_to_weave(self, file_id, revision_id, new_lines, parents,
1171
nostore_sha, random_revid):
1172
file_id_index = GraphIndexPrefixAdapter(
1173
self.text_index.combined_index,
1175
add_nodes_callback=self.text_index.add_callback)
1176
self.repo._text_knit._index._graph_index = file_id_index
1177
self.repo._text_knit._index._add_callback = file_id_index.add_nodes
1178
return self.repo._text_knit.add_lines_with_ghosts(
1179
revision_id, parents, new_lines, nostore_sha=nostore_sha,
1180
random_id=random_revid, check_content=False)[0:2]
1182
def all_packs(self):
1183
"""Return a list of all the Pack objects this repository has.
1185
Note that an in-progress pack being created is not returned.
1187
:return: A list of Pack objects for all the packs in the repository.
1190
for name in self.names():
1191
result.append(self.get_pack_by_name(name))
1195
"""Pack the pack collection incrementally.
1197
This will not attempt global reorganisation or recompression,
1198
rather it will just ensure that the total number of packs does
1199
not grow without bound. It uses the _max_pack_count method to
1200
determine if autopacking is needed, and the pack_distribution
1201
method to determine the number of revisions in each pack.
1203
If autopacking takes place then the packs name collection will have
1204
been flushed to disk - packing requires updating the name collection
1205
in synchronisation with certain steps. Otherwise the names collection
1208
:return: True if packing took place.
1210
# XXX: Should not be needed when the management of indices is sane.
1211
total_revisions = self.revision_index.combined_index.key_count()
1212
total_packs = len(self._names)
1213
if self._max_pack_count(total_revisions) >= total_packs:
1215
# XXX: the following may want to be a class, to pack with a given
1217
mutter('Auto-packing repository %s, which has %d pack files, '
1218
'containing %d revisions into %d packs.', self, total_packs,
1219
total_revisions, self._max_pack_count(total_revisions))
1220
# determine which packs need changing
1221
pack_distribution = self.pack_distribution(total_revisions)
1223
for pack in self.all_packs():
1224
revision_count = pack.get_revision_count()
1225
if revision_count == 0:
1226
# revision less packs are not generated by normal operation,
1227
# only by operations like sign-my-commits, and thus will not
1228
# tend to grow rapdily or without bound like commit containing
1229
# packs do - leave them alone as packing them really should
1230
# group their data with the relevant commit, and that may
1231
# involve rewriting ancient history - which autopack tries to
1232
# avoid. Alternatively we could not group the data but treat
1233
# each of these as having a single revision, and thus add
1234
# one revision for each to the total revision count, to get
1235
# a matching distribution.
1237
existing_packs.append((revision_count, pack))
1238
pack_operations = self.plan_autopack_combinations(
1239
existing_packs, pack_distribution)
1240
self._execute_pack_operations(pack_operations)
1243
def _execute_pack_operations(self, pack_operations, _packer_class=Packer):
1244
"""Execute a series of pack operations.
1246
:param pack_operations: A list of [revision_count, packs_to_combine].
1247
:param _packer_class: The class of packer to use (default: Packer).
1250
for revision_count, packs in pack_operations:
1251
# we may have no-ops from the setup logic
1254
_packer_class(self, packs, '.autopack').pack()
1256
self._remove_pack_from_memory(pack)
1257
# record the newly available packs and stop advertising the old
1259
self._save_pack_names(clear_obsolete_packs=True)
1260
# Move the old packs out of the way now they are no longer referenced.
1261
for revision_count, packs in pack_operations:
1262
self._obsolete_packs(packs)
1264
def lock_names(self):
1265
"""Acquire the mutex around the pack-names index.
1267
This cannot be used in the middle of a read-only transaction on the
1270
self.repo.control_files.lock_write()
1273
"""Pack the pack collection totally."""
1274
self.ensure_loaded()
1275
total_packs = len(self._names)
1277
# This is arguably wrong because we might not be optimal, but for
1278
# now lets leave it in. (e.g. reconcile -> one pack. But not
1281
total_revisions = self.revision_index.combined_index.key_count()
1282
# XXX: the following may want to be a class, to pack with a given
1284
mutter('Packing repository %s, which has %d pack files, '
1285
'containing %d revisions into 1 packs.', self, total_packs,
1287
# determine which packs need changing
1288
pack_distribution = [1]
1289
pack_operations = [[0, []]]
1290
for pack in self.all_packs():
1291
pack_operations[-1][0] += pack.get_revision_count()
1292
pack_operations[-1][1].append(pack)
1293
self._execute_pack_operations(pack_operations, OptimisingPacker)
1295
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1296
"""Plan a pack operation.
1298
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1300
:param pack_distribution: A list with the number of revisions desired
1303
if len(existing_packs) <= len(pack_distribution):
1305
existing_packs.sort(reverse=True)
1306
pack_operations = [[0, []]]
1307
# plan out what packs to keep, and what to reorganise
1308
while len(existing_packs):
1309
# take the largest pack, and if its less than the head of the
1310
# distribution chart we will include its contents in the new pack for
1311
# that position. If its larger, we remove its size from the
1312
# distribution chart
1313
next_pack_rev_count, next_pack = existing_packs.pop(0)
1314
if next_pack_rev_count >= pack_distribution[0]:
1315
# this is already packed 'better' than this, so we can
1316
# not waste time packing it.
1317
while next_pack_rev_count > 0:
1318
next_pack_rev_count -= pack_distribution[0]
1319
if next_pack_rev_count >= 0:
1321
del pack_distribution[0]
1323
# didn't use that entire bucket up
1324
pack_distribution[0] = -next_pack_rev_count
1326
# add the revisions we're going to add to the next output pack
1327
pack_operations[-1][0] += next_pack_rev_count
1328
# allocate this pack to the next pack sub operation
1329
pack_operations[-1][1].append(next_pack)
1330
if pack_operations[-1][0] >= pack_distribution[0]:
1331
# this pack is used up, shift left.
1332
del pack_distribution[0]
1333
pack_operations.append([0, []])
1335
return pack_operations
1337
def ensure_loaded(self):
1338
# NB: if you see an assertion error here, its probably access against
1339
# an unlocked repo. Naughty.
1340
if not self.repo.is_locked():
1341
raise errors.ObjectNotLocked(self.repo)
1342
if self._names is None:
1344
self._packs_at_load = set()
1345
for index, key, value in self._iter_disk_pack_index():
1347
self._names[name] = self._parse_index_sizes(value)
1348
self._packs_at_load.add((key, value))
1349
# populate all the metadata.
1352
def _parse_index_sizes(self, value):
1353
"""Parse a string of index sizes."""
1354
return tuple([int(digits) for digits in value.split(' ')])
1356
def get_pack_by_name(self, name):
1357
"""Get a Pack object by name.
1359
:param name: The name of the pack - e.g. '123456'
1360
:return: A Pack object.
1363
return self._packs_by_name[name]
1365
rev_index = self._make_index(name, '.rix')
1366
inv_index = self._make_index(name, '.iix')
1367
txt_index = self._make_index(name, '.tix')
1368
sig_index = self._make_index(name, '.six')
1369
result = ExistingPack(self._pack_transport, name, rev_index,
1370
inv_index, txt_index, sig_index)
1371
self.add_pack_to_memory(result)
1374
def allocate(self, a_new_pack):
1375
"""Allocate name in the list of packs.
1377
:param a_new_pack: A NewPack instance to be added to the collection of
1378
packs for this repository.
1380
self.ensure_loaded()
1381
if a_new_pack.name in self._names:
1382
raise errors.BzrError(
1383
'Pack %r already exists in %s' % (a_new_pack.name, self))
1384
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1385
self.add_pack_to_memory(a_new_pack)
1387
def _iter_disk_pack_index(self):
1388
"""Iterate over the contents of the pack-names index.
1390
This is used when loading the list from disk, and before writing to
1391
detect updates from others during our write operation.
1392
:return: An iterator of the index contents.
1394
return GraphIndex(self.transport, 'pack-names', None
1395
).iter_all_entries()
1397
def _make_index(self, name, suffix):
1398
size_offset = self._suffix_offsets[suffix]
1399
index_name = name + suffix
1400
index_size = self._names[name][size_offset]
1402
self._index_transport, index_name, index_size)
1404
def _max_pack_count(self, total_revisions):
1405
"""Return the maximum number of packs to use for total revisions.
1407
:param total_revisions: The total number of revisions in the
1410
if not total_revisions:
1412
digits = str(total_revisions)
1414
for digit in digits:
1415
result += int(digit)
1419
"""Provide an order to the underlying names."""
1420
return sorted(self._names.keys())
1422
def _obsolete_packs(self, packs):
1423
"""Move a number of packs which have been obsoleted out of the way.
1425
Each pack and its associated indices are moved out of the way.
1427
Note: for correctness this function should only be called after a new
1428
pack names index has been written without these pack names, and with
1429
the names of packs that contain the data previously available via these
1432
:param packs: The packs to obsolete.
1433
:param return: None.
1436
pack.pack_transport.rename(pack.file_name(),
1437
'../obsolete_packs/' + pack.file_name())
1438
# TODO: Probably needs to know all possible indices for this pack
1439
# - or maybe list the directory and move all indices matching this
1440
# name whether we recognize it or not?
1441
for suffix in ('.iix', '.six', '.tix', '.rix'):
1442
self._index_transport.rename(pack.name + suffix,
1443
'../obsolete_packs/' + pack.name + suffix)
1445
def pack_distribution(self, total_revisions):
1446
"""Generate a list of the number of revisions to put in each pack.
1448
:param total_revisions: The total number of revisions in the
1451
if total_revisions == 0:
1453
digits = reversed(str(total_revisions))
1455
for exponent, count in enumerate(digits):
1456
size = 10 ** exponent
1457
for pos in range(int(count)):
1459
return list(reversed(result))
1461
def _pack_tuple(self, name):
1462
"""Return a tuple with the transport and file name for a pack name."""
1463
return self._pack_transport, name + '.pack'
1465
def _remove_pack_from_memory(self, pack):
1466
"""Remove pack from the packs accessed by this repository.
1468
Only affects memory state, until self._save_pack_names() is invoked.
1470
self._names.pop(pack.name)
1471
self._packs_by_name.pop(pack.name)
1472
self._remove_pack_indices(pack)
1474
def _remove_pack_indices(self, pack):
1475
"""Remove the indices for pack from the aggregated indices."""
1476
self.revision_index.remove_index(pack.revision_index, pack)
1477
self.inventory_index.remove_index(pack.inventory_index, pack)
1478
self.text_index.remove_index(pack.text_index, pack)
1479
self.signature_index.remove_index(pack.signature_index, pack)
1482
"""Clear all cached data."""
1483
# cached revision data
1484
self.repo._revision_knit = None
1485
self.revision_index.clear()
1486
# cached signature data
1487
self.repo._signature_knit = None
1488
self.signature_index.clear()
1489
# cached file text data
1490
self.text_index.clear()
1491
self.repo._text_knit = None
1492
# cached inventory data
1493
self.inventory_index.clear()
1494
# remove the open pack
1495
self._new_pack = None
1496
# information about packs.
1499
self._packs_by_name = {}
1500
self._packs_at_load = None
1502
def _make_index_map(self, index_suffix):
1503
"""Return information on existing indices.
1505
:param suffix: Index suffix added to pack name.
1507
:returns: (pack_map, indices) where indices is a list of GraphIndex
1508
objects, and pack_map is a mapping from those objects to the
1509
pack tuple they describe.
1511
# TODO: stop using this; it creates new indices unnecessarily.
1512
self.ensure_loaded()
1513
suffix_map = {'.rix': 'revision_index',
1514
'.six': 'signature_index',
1515
'.iix': 'inventory_index',
1516
'.tix': 'text_index',
1518
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1519
suffix_map[index_suffix])
1521
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1522
"""Convert a list of packs to an index pack map and index list.
1524
:param packs: The packs list to process.
1525
:param index_attribute: The attribute that the desired index is found
1527
:return: A tuple (map, list) where map contains the dict from
1528
index:pack_tuple, and lsit contains the indices in the same order
1534
index = getattr(pack, index_attribute)
1535
indices.append(index)
1536
pack_map[index] = (pack.pack_transport, pack.file_name())
1537
return pack_map, indices
1539
def _index_contents(self, pack_map, key_filter=None):
1540
"""Get an iterable of the index contents from a pack_map.
1542
:param pack_map: A map from indices to pack details.
1543
:param key_filter: An optional filter to limit the
1546
indices = [index for index in pack_map.iterkeys()]
1547
all_index = CombinedGraphIndex(indices)
1548
if key_filter is None:
1549
return all_index.iter_all_entries()
1551
return all_index.iter_entries(key_filter)
1553
def _unlock_names(self):
1554
"""Release the mutex around the pack-names index."""
1555
self.repo.control_files.unlock()
1557
def _save_pack_names(self, clear_obsolete_packs=False):
1558
"""Save the list of packs.
1560
This will take out the mutex around the pack names list for the
1561
duration of the method call. If concurrent updates have been made, a
1562
three-way merge between the current list and the current in memory list
1565
:param clear_obsolete_packs: If True, clear out the contents of the
1566
obsolete_packs directory.
1570
builder = GraphIndexBuilder()
1571
# load the disk nodes across
1573
for index, key, value in self._iter_disk_pack_index():
1574
disk_nodes.add((key, value))
1575
# do a two-way diff against our original content
1576
current_nodes = set()
1577
for name, sizes in self._names.iteritems():
1579
((name, ), ' '.join(str(size) for size in sizes)))
1580
deleted_nodes = self._packs_at_load - current_nodes
1581
new_nodes = current_nodes - self._packs_at_load
1582
disk_nodes.difference_update(deleted_nodes)
1583
disk_nodes.update(new_nodes)
1584
# TODO: handle same-name, index-size-changes here -
1585
# e.g. use the value from disk, not ours, *unless* we're the one
1587
for key, value in disk_nodes:
1588
builder.add_node(key, value)
1589
self.transport.put_file('pack-names', builder.finish(),
1590
mode=self.repo.control_files._file_mode)
1591
# move the baseline forward
1592
self._packs_at_load = disk_nodes
1593
# now clear out the obsolete packs directory
1594
if clear_obsolete_packs:
1595
self.transport.clone('obsolete_packs').delete_multi(
1596
self.transport.list_dir('obsolete_packs'))
1598
self._unlock_names()
1599
# synchronise the memory packs list with what we just wrote:
1600
new_names = dict(disk_nodes)
1601
# drop no longer present nodes
1602
for pack in self.all_packs():
1603
if (pack.name,) not in new_names:
1604
self._remove_pack_from_memory(pack)
1605
# add new nodes/refresh existing ones
1606
for key, value in disk_nodes:
1608
sizes = self._parse_index_sizes(value)
1609
if name in self._names:
1611
if sizes != self._names[name]:
1612
# the pack for name has had its indices replaced - rare but
1613
# important to handle. XXX: probably can never happen today
1614
# because the three-way merge code above does not handle it
1615
# - you may end up adding the same key twice to the new
1616
# disk index because the set values are the same, unless
1617
# the only index shows up as deleted by the set difference
1618
# - which it may. Until there is a specific test for this,
1619
# assume its broken. RBC 20071017.
1620
self._remove_pack_from_memory(self.get_pack_by_name(name))
1621
self._names[name] = sizes
1622
self.get_pack_by_name(name)
1625
self._names[name] = sizes
1626
self.get_pack_by_name(name)
1628
def _start_write_group(self):
1629
# Do not permit preparation for writing if we're not in a 'write lock'.
1630
if not self.repo.is_write_locked():
1631
raise errors.NotWriteLocked(self)
1632
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1633
self._pack_transport, upload_suffix='.pack',
1634
file_mode=self.repo.control_files._file_mode)
1635
# allow writing: queue writes to a new index
1636
self.revision_index.add_writable_index(self._new_pack.revision_index,
1638
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
1640
self.text_index.add_writable_index(self._new_pack.text_index,
1642
self.signature_index.add_writable_index(self._new_pack.signature_index,
1645
# reused revision and signature knits may need updating
1647
# "Hysterical raisins. client code in bzrlib grabs those knits outside
1648
# of write groups and then mutates it inside the write group."
1649
if self.repo._revision_knit is not None:
1650
self.repo._revision_knit._index._add_callback = \
1651
self.revision_index.add_callback
1652
if self.repo._signature_knit is not None:
1653
self.repo._signature_knit._index._add_callback = \
1654
self.signature_index.add_callback
1655
# create a reused knit object for text addition in commit.
1656
self.repo._text_knit = self.repo.weave_store.get_weave_or_empty(
1659
def _abort_write_group(self):
1660
# FIXME: just drop the transient index.
1661
# forget what names there are
1662
if self._new_pack is not None:
1663
self._new_pack.abort()
1664
self._remove_pack_indices(self._new_pack)
1665
self._new_pack = None
1666
self.repo._text_knit = None
1668
def _commit_write_group(self):
1669
self._remove_pack_indices(self._new_pack)
1670
if self._new_pack.data_inserted():
1671
# get all the data to disk and read to use
1672
self._new_pack.finish()
1673
self.allocate(self._new_pack)
1674
self._new_pack = None
1675
if not self.autopack():
1676
# when autopack takes no steps, the names list is still
1678
self._save_pack_names()
1680
self._new_pack.abort()
1681
self._new_pack = None
1682
self.repo._text_knit = None
1685
class KnitPackRevisionStore(KnitRevisionStore):
1686
"""An object to adapt access from RevisionStore's to use KnitPacks.
1688
This class works by replacing the original RevisionStore.
1689
We need to do this because the KnitPackRevisionStore is less
1690
isolated in its layering - it uses services from the repo.
1693
def __init__(self, repo, transport, revisionstore):
1694
"""Create a KnitPackRevisionStore on repo with revisionstore.
1696
This will store its state in the Repository, use the
1697
indices to provide a KnitGraphIndex,
1698
and at the end of transactions write new indices.
1700
KnitRevisionStore.__init__(self, revisionstore.versioned_file_store)
1702
self._serializer = revisionstore._serializer
1703
self.transport = transport
1705
def get_revision_file(self, transaction):
1706
"""Get the revision versioned file object."""
1707
if getattr(self.repo, '_revision_knit', None) is not None:
1708
return self.repo._revision_knit
1709
self.repo._pack_collection.ensure_loaded()
1710
add_callback = self.repo._pack_collection.revision_index.add_callback
1711
# setup knit specific objects
1712
knit_index = KnitGraphIndex(
1713
self.repo._pack_collection.revision_index.combined_index,
1714
add_callback=add_callback)
1715
self.repo._revision_knit = knit.KnitVersionedFile(
1716
'revisions', self.transport.clone('..'),
1717
self.repo.control_files._file_mode,
1719
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1720
access_method=self.repo._pack_collection.revision_index.knit_access)
1721
return self.repo._revision_knit
1723
def get_signature_file(self, transaction):
1724
"""Get the signature versioned file object."""
1725
if getattr(self.repo, '_signature_knit', None) is not None:
1726
return self.repo._signature_knit
1727
self.repo._pack_collection.ensure_loaded()
1728
add_callback = self.repo._pack_collection.signature_index.add_callback
1729
# setup knit specific objects
1730
knit_index = KnitGraphIndex(
1731
self.repo._pack_collection.signature_index.combined_index,
1732
add_callback=add_callback, parents=False)
1733
self.repo._signature_knit = knit.KnitVersionedFile(
1734
'signatures', self.transport.clone('..'),
1735
self.repo.control_files._file_mode,
1737
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1738
access_method=self.repo._pack_collection.signature_index.knit_access)
1739
return self.repo._signature_knit
1742
class KnitPackTextStore(VersionedFileStore):
1743
"""Presents a TextStore abstraction on top of packs.
1745
This class works by replacing the original VersionedFileStore.
1746
We need to do this because the KnitPackRevisionStore is less
1747
isolated in its layering - it uses services from the repo and shares them
1748
with all the data written in a single write group.
1751
def __init__(self, repo, transport, weavestore):
1752
"""Create a KnitPackTextStore on repo with weavestore.
1754
This will store its state in the Repository, use the
1755
indices FileNames to provide a KnitGraphIndex,
1756
and at the end of transactions write new indices.
1758
# don't call base class constructor - it's not suitable.
1759
# no transient data stored in the transaction
1761
self._precious = False
1763
self.transport = transport
1764
self.weavestore = weavestore
1765
# XXX for check() which isn't updated yet
1766
self._transport = weavestore._transport
1768
def get_weave_or_empty(self, file_id, transaction):
1769
"""Get a 'Knit' backed by the .tix indices.
1771
The transaction parameter is ignored.
1773
self.repo._pack_collection.ensure_loaded()
1774
add_callback = self.repo._pack_collection.text_index.add_callback
1775
# setup knit specific objects
1776
file_id_index = GraphIndexPrefixAdapter(
1777
self.repo._pack_collection.text_index.combined_index,
1778
(file_id, ), 1, add_nodes_callback=add_callback)
1779
knit_index = KnitGraphIndex(file_id_index,
1780
add_callback=file_id_index.add_nodes,
1781
deltas=True, parents=True)
1782
return knit.KnitVersionedFile('text:' + file_id,
1783
self.transport.clone('..'),
1786
access_method=self.repo._pack_collection.text_index.knit_access,
1787
factory=knit.KnitPlainFactory())
1789
get_weave = get_weave_or_empty
1792
"""Generate a list of the fileids inserted, for use by check."""
1793
self.repo._pack_collection.ensure_loaded()
1795
for index, key, value, refs in \
1796
self.repo._pack_collection.text_index.combined_index.iter_all_entries():
1801
class InventoryKnitThunk(object):
1802
"""An object to manage thunking get_inventory_weave to pack based knits."""
1804
def __init__(self, repo, transport):
1805
"""Create an InventoryKnitThunk for repo at transport.
1807
This will store its state in the Repository, use the
1808
indices FileNames to provide a KnitGraphIndex,
1809
and at the end of transactions write a new index..
1812
self.transport = transport
1814
def get_weave(self):
1815
"""Get a 'Knit' that contains inventory data."""
1816
self.repo._pack_collection.ensure_loaded()
1817
add_callback = self.repo._pack_collection.inventory_index.add_callback
1818
# setup knit specific objects
1819
knit_index = KnitGraphIndex(
1820
self.repo._pack_collection.inventory_index.combined_index,
1821
add_callback=add_callback, deltas=True, parents=True)
1822
return knit.KnitVersionedFile(
1823
'inventory', self.transport.clone('..'),
1824
self.repo.control_files._file_mode,
1826
index=knit_index, delta=True, factory=knit.KnitPlainFactory(),
1827
access_method=self.repo._pack_collection.inventory_index.knit_access)
1830
class KnitPackRepository(KnitRepository):
1831
"""Experimental graph-knit using repository."""
1833
def __init__(self, _format, a_bzrdir, control_files, _revision_store,
1834
control_store, text_store, _commit_builder_class, _serializer):
1835
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
1836
_revision_store, control_store, text_store, _commit_builder_class,
1838
index_transport = control_files._transport.clone('indices')
1839
self._pack_collection = RepositoryPackCollection(self, control_files._transport,
1841
control_files._transport.clone('upload'),
1842
control_files._transport.clone('packs'))
1843
self._revision_store = KnitPackRevisionStore(self, index_transport, self._revision_store)
1844
self.weave_store = KnitPackTextStore(self, index_transport, self.weave_store)
1845
self._inv_thunk = InventoryKnitThunk(self, index_transport)
1846
# True when the repository object is 'write locked' (as opposed to the
1847
# physical lock only taken out around changes to the pack-names list.)
1848
# Another way to represent this would be a decorator around the control
1849
# files object that presents logical locks as physical ones - if this
1850
# gets ugly consider that alternative design. RBC 20071011
1851
self._write_lock_count = 0
1852
self._transaction = None
1854
self._reconcile_does_inventory_gc = True
1855
self._reconcile_fixes_text_parents = True
1856
self._reconcile_backsup_inventory = False
1858
def _abort_write_group(self):
1859
self._pack_collection._abort_write_group()
1861
def _find_inconsistent_revision_parents(self):
1862
"""Find revisions with incorrectly cached parents.
1864
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1865
parents-in-revision).
1867
if not self.is_locked():
1868
raise errors.ObjectNotLocked(self)
1869
pb = ui.ui_factory.nested_progress_bar()
1872
revision_nodes = self._pack_collection.revision_index \
1873
.combined_index.iter_all_entries()
1874
index_positions = []
1875
# Get the cached index values for all revisions, and also the location
1876
# in each index of the revision text so we can perform linear IO.
1877
for index, key, value, refs in revision_nodes:
1878
pos, length = value[1:].split(' ')
1879
index_positions.append((index, int(pos), key[0],
1880
tuple(parent[0] for parent in refs[0])))
1881
pb.update("Reading revision index.", 0, 0)
1882
index_positions.sort()
1883
batch_count = len(index_positions) / 1000 + 1
1884
pb.update("Checking cached revision graph.", 0, batch_count)
1885
for offset in xrange(batch_count):
1886
pb.update("Checking cached revision graph.", offset)
1887
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1890
rev_ids = [item[2] for item in to_query]
1891
revs = self.get_revisions(rev_ids)
1892
for revision, item in zip(revs, to_query):
1893
index_parents = item[3]
1894
rev_parents = tuple(revision.parent_ids)
1895
if index_parents != rev_parents:
1896
result.append((revision.revision_id, index_parents, rev_parents))
1901
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
1902
def get_parents(self, revision_ids):
1903
"""See graph._StackedParentsProvider.get_parents."""
1904
parent_map = self.get_parent_map(revision_ids)
1905
return [parent_map.get(r, None) for r in revision_ids]
1907
def get_parent_map(self, keys):
1908
"""See graph._StackedParentsProvider.get_parent_map
1910
This implementation accesses the combined revision index to provide
1913
self._pack_collection.ensure_loaded()
1914
index = self._pack_collection.revision_index.combined_index
1916
if _mod_revision.NULL_REVISION in keys:
1917
keys.discard(_mod_revision.NULL_REVISION)
1918
found_parents = {_mod_revision.NULL_REVISION:()}
1921
search_keys = set((revision_id,) for revision_id in keys)
1922
for index, key, value, refs in index.iter_entries(search_keys):
1925
parents = (_mod_revision.NULL_REVISION,)
1927
parents = tuple(parent[0] for parent in parents)
1928
found_parents[key[0]] = parents
1929
return found_parents
1931
@symbol_versioning.deprecated_method(symbol_versioning.one_four)
1933
def get_revision_graph(self, revision_id=None):
1934
"""Return a dictionary containing the revision graph.
1936
:param revision_id: The revision_id to get a graph from. If None, then
1937
the entire revision graph is returned. This is a deprecated mode of
1938
operation and will be removed in the future.
1939
:return: a dictionary of revision_id->revision_parents_list.
1941
if 'evil' in debug.debug_flags:
1943
"get_revision_graph scales with size of history.")
1944
# special case NULL_REVISION
1945
if revision_id == _mod_revision.NULL_REVISION:
1947
if revision_id is None:
1948
revision_vf = self._get_revision_vf()
1949
return revision_vf.get_graph()
1950
g = self.get_graph()
1951
first = g.get_parent_map([revision_id])
1952
if revision_id not in first:
1953
raise errors.NoSuchRevision(self, revision_id)
1957
NULL_REVISION = _mod_revision.NULL_REVISION
1958
ghosts = set([NULL_REVISION])
1959
for rev_id, parent_ids in g.iter_ancestry([revision_id]):
1960
if parent_ids is None: # This is a ghost
1963
ancestry[rev_id] = parent_ids
1964
for p in parent_ids:
1966
children[p].append(rev_id)
1968
children[p] = [rev_id]
1970
if NULL_REVISION in ancestry:
1971
del ancestry[NULL_REVISION]
1973
# Find all nodes that reference a ghost, and filter the ghosts out
1974
# of their parent lists. To preserve the order of parents, and
1975
# avoid double filtering nodes, we just find all children first,
1977
children_of_ghosts = set()
1978
for ghost in ghosts:
1979
children_of_ghosts.update(children[ghost])
1981
for child in children_of_ghosts:
1982
ancestry[child] = tuple(p for p in ancestry[child]
1986
def has_revisions(self, revision_ids):
1987
"""See Repository.has_revisions()."""
1988
revision_ids = set(revision_ids)
1989
result = revision_ids.intersection(
1990
set([None, _mod_revision.NULL_REVISION]))
1991
revision_ids.difference_update(result)
1992
index = self._pack_collection.revision_index.combined_index
1993
keys = [(revision_id,) for revision_id in revision_ids]
1994
result.update(node[1][0] for node in index.iter_entries(keys))
1997
def _make_parents_provider(self):
1998
return graph.CachingParentsProvider(self)
2000
def _refresh_data(self):
2001
if self._write_lock_count == 1 or (
2002
self.control_files._lock_count == 1 and
2003
self.control_files._lock_mode == 'r'):
2004
# forget what names there are
2005
self._pack_collection.reset()
2006
# XXX: Better to do an in-memory merge when acquiring a new lock -
2007
# factor out code from _save_pack_names.
2008
self._pack_collection.ensure_loaded()
2010
def _start_write_group(self):
2011
self._pack_collection._start_write_group()
2013
def _commit_write_group(self):
2014
return self._pack_collection._commit_write_group()
2016
def get_inventory_weave(self):
2017
return self._inv_thunk.get_weave()
2019
def get_transaction(self):
2020
if self._write_lock_count:
2021
return self._transaction
2023
return self.control_files.get_transaction()
2025
def is_locked(self):
2026
return self._write_lock_count or self.control_files.is_locked()
2028
def is_write_locked(self):
2029
return self._write_lock_count
2031
def lock_write(self, token=None):
2032
if not self._write_lock_count and self.is_locked():
2033
raise errors.ReadOnlyError(self)
2034
self._write_lock_count += 1
2035
if self._write_lock_count == 1:
2036
from bzrlib import transactions
2037
self._transaction = transactions.WriteTransaction()
2038
self._refresh_data()
2040
def lock_read(self):
2041
if self._write_lock_count:
2042
self._write_lock_count += 1
2044
self.control_files.lock_read()
2045
self._refresh_data()
2047
def leave_lock_in_place(self):
2048
# not supported - raise an error
2049
raise NotImplementedError(self.leave_lock_in_place)
2051
def dont_leave_lock_in_place(self):
2052
# not supported - raise an error
2053
raise NotImplementedError(self.dont_leave_lock_in_place)
2057
"""Compress the data within the repository.
2059
This will pack all the data to a single pack. In future it may
2060
recompress deltas or do other such expensive operations.
2062
self._pack_collection.pack()
2065
def reconcile(self, other=None, thorough=False):
2066
"""Reconcile this repository."""
2067
from bzrlib.reconcile import PackReconciler
2068
reconciler = PackReconciler(self, thorough=thorough)
2069
reconciler.reconcile()
2073
if self._write_lock_count == 1 and self._write_group is not None:
2074
self.abort_write_group()
2075
self._transaction = None
2076
self._write_lock_count = 0
2077
raise errors.BzrError(
2078
'Must end write group before releasing write lock on %s'
2080
if self._write_lock_count:
2081
self._write_lock_count -= 1
2082
if not self._write_lock_count:
2083
transaction = self._transaction
2084
self._transaction = None
2085
transaction.finish()
2087
self.control_files.unlock()
2090
class RepositoryFormatPack(MetaDirRepositoryFormat):
2091
"""Format logic for pack structured repositories.
2093
This repository format has:
2094
- a list of packs in pack-names
2095
- packs in packs/NAME.pack
2096
- indices in indices/NAME.{iix,six,tix,rix}
2097
- knit deltas in the packs, knit indices mapped to the indices.
2098
- thunk objects to support the knits programming API.
2099
- a format marker of its own
2100
- an optional 'shared-storage' flag
2101
- an optional 'no-working-trees' flag
2105
# Set this attribute in derived classes to control the repository class
2106
# created by open and initialize.
2107
repository_class = None
2108
# Set this attribute in derived classes to control the
2109
# _commit_builder_class that the repository objects will have passed to
2110
# their constructor.
2111
_commit_builder_class = None
2112
# Set this attribute in derived clases to control the _serializer that the
2113
# repository objects will have passed to their constructor.
2115
# External references are not supported in pack repositories yet.
2116
supports_external_lookups = False
2118
def _get_control_store(self, repo_transport, control_files):
2119
"""Return the control store for this repository."""
2120
return VersionedFileStore(
2123
file_mode=control_files._file_mode,
2124
versionedfile_class=knit.make_file_knit,
2125
versionedfile_kwargs={'factory': knit.KnitPlainFactory()},
2128
def _get_revision_store(self, repo_transport, control_files):
2129
"""See RepositoryFormat._get_revision_store()."""
2130
versioned_file_store = VersionedFileStore(
2132
file_mode=control_files._file_mode,
2135
versionedfile_class=knit.make_file_knit,
2136
versionedfile_kwargs={'delta': False,
2137
'factory': knit.KnitPlainFactory(),
2141
return KnitRevisionStore(versioned_file_store)
2143
def _get_text_store(self, transport, control_files):
2144
"""See RepositoryFormat._get_text_store()."""
2145
return self._get_versioned_file_store('knits',
2148
versionedfile_class=knit.make_file_knit,
2149
versionedfile_kwargs={
2150
'create_parent_dir': True,
2151
'delay_create': True,
2152
'dir_mode': control_files._dir_mode,
2156
def initialize(self, a_bzrdir, shared=False):
2157
"""Create a pack based repository.
2159
:param a_bzrdir: bzrdir to contain the new repository; must already
2161
:param shared: If true the repository will be initialized as a shared
2164
mutter('creating repository in %s.', a_bzrdir.transport.base)
2165
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
2166
builder = GraphIndexBuilder()
2167
files = [('pack-names', builder.finish())]
2168
utf8_files = [('format', self.get_format_string())]
2170
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
2171
return self.open(a_bzrdir=a_bzrdir, _found=True)
2173
def open(self, a_bzrdir, _found=False, _override_transport=None):
2174
"""See RepositoryFormat.open().
2176
:param _override_transport: INTERNAL USE ONLY. Allows opening the
2177
repository at a slightly different url
2178
than normal. I.e. during 'upgrade'.
2181
format = RepositoryFormat.find_format(a_bzrdir)
2182
if _override_transport is not None:
2183
repo_transport = _override_transport
2185
repo_transport = a_bzrdir.get_repository_transport(None)
2186
control_files = lockable_files.LockableFiles(repo_transport,
2187
'lock', lockdir.LockDir)
2188
text_store = self._get_text_store(repo_transport, control_files)
2189
control_store = self._get_control_store(repo_transport, control_files)
2190
_revision_store = self._get_revision_store(repo_transport, control_files)
2191
return self.repository_class(_format=self,
2193
control_files=control_files,
2194
_revision_store=_revision_store,
2195
control_store=control_store,
2196
text_store=text_store,
2197
_commit_builder_class=self._commit_builder_class,
2198
_serializer=self._serializer)
2201
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2202
"""A no-subtrees parameterized Pack repository.
2204
This format was introduced in 0.92.
2207
repository_class = KnitPackRepository
2208
_commit_builder_class = PackCommitBuilder
2209
_serializer = xml5.serializer_v5
2211
def _get_matching_bzrdir(self):
2212
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2214
def _ignore_setting_bzrdir(self, format):
2217
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2219
def get_format_string(self):
2220
"""See RepositoryFormat.get_format_string()."""
2221
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2223
def get_format_description(self):
2224
"""See RepositoryFormat.get_format_description()."""
2225
return "Packs containing knits without subtree support"
2227
def check_conversion_target(self, target_format):
2231
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2232
"""A subtrees parameterized Pack repository.
2234
This repository format uses the xml7 serializer to get:
2235
- support for recording full info about the tree root
2236
- support for recording tree-references
2238
This format was introduced in 0.92.
2241
repository_class = KnitPackRepository
2242
_commit_builder_class = PackRootCommitBuilder
2243
rich_root_data = True
2244
supports_tree_reference = True
2245
_serializer = xml7.serializer_v7
2247
def _get_matching_bzrdir(self):
2248
return bzrdir.format_registry.make_bzrdir(
2249
'pack-0.92-subtree')
2251
def _ignore_setting_bzrdir(self, format):
2254
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2256
def check_conversion_target(self, target_format):
2257
if not target_format.rich_root_data:
2258
raise errors.BadConversionTarget(
2259
'Does not support rich root data.', target_format)
2260
if not getattr(target_format, 'supports_tree_reference', False):
2261
raise errors.BadConversionTarget(
2262
'Does not support nested trees', target_format)
2264
def get_format_string(self):
2265
"""See RepositoryFormat.get_format_string()."""
2266
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2268
def get_format_description(self):
2269
"""See RepositoryFormat.get_format_description()."""
2270
return "Packs containing knits with subtree support\n"
2273
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2274
"""A rich-root, no subtrees parameterized Pack repository.
2276
This repository format uses the xml6 serializer to get:
2277
- support for recording full info about the tree root
2279
This format was introduced in 1.0.
2282
repository_class = KnitPackRepository
2283
_commit_builder_class = PackRootCommitBuilder
2284
rich_root_data = True
2285
supports_tree_reference = False
2286
_serializer = xml6.serializer_v6
2288
def _get_matching_bzrdir(self):
2289
return bzrdir.format_registry.make_bzrdir(
2292
def _ignore_setting_bzrdir(self, format):
2295
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2297
def check_conversion_target(self, target_format):
2298
if not target_format.rich_root_data:
2299
raise errors.BadConversionTarget(
2300
'Does not support rich root data.', target_format)
2302
def get_format_string(self):
2303
"""See RepositoryFormat.get_format_string()."""
2304
return ("Bazaar pack repository format 1 with rich root"
2305
" (needs bzr 1.0)\n")
2307
def get_format_description(self):
2308
"""See RepositoryFormat.get_format_description()."""
2309
return "Packs containing knits with rich root support\n"
2312
class RepositoryFormatPackDevelopment0(RepositoryFormatPack):
2313
"""A no-subtrees development repository.
2315
This format should be retained until the second release after bzr 1.0.
2317
No changes to the disk behaviour from pack-0.92.
2320
repository_class = KnitPackRepository
2321
_commit_builder_class = PackCommitBuilder
2322
_serializer = xml5.serializer_v5
2324
def _get_matching_bzrdir(self):
2325
return bzrdir.format_registry.make_bzrdir('development0')
2327
def _ignore_setting_bzrdir(self, format):
2330
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2332
def get_format_string(self):
2333
"""See RepositoryFormat.get_format_string()."""
2334
return "Bazaar development format 0 (needs bzr.dev from before 1.3)\n"
2336
def get_format_description(self):
2337
"""See RepositoryFormat.get_format_description()."""
2338
return ("Development repository format, currently the same as "
2341
def check_conversion_target(self, target_format):
2345
class RepositoryFormatPackDevelopment0Subtree(RepositoryFormatPack):
2346
"""A subtrees development repository.
2348
This format should be retained until the second release after bzr 1.0.
2350
No changes to the disk behaviour from pack-0.92-subtree.
2353
repository_class = KnitPackRepository
2354
_commit_builder_class = PackRootCommitBuilder
2355
rich_root_data = True
2356
supports_tree_reference = True
2357
_serializer = xml7.serializer_v7
2359
def _get_matching_bzrdir(self):
2360
return bzrdir.format_registry.make_bzrdir(
2361
'development0-subtree')
2363
def _ignore_setting_bzrdir(self, format):
2366
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2368
def check_conversion_target(self, target_format):
2369
if not target_format.rich_root_data:
2370
raise errors.BadConversionTarget(
2371
'Does not support rich root data.', target_format)
2372
if not getattr(target_format, 'supports_tree_reference', False):
2373
raise errors.BadConversionTarget(
2374
'Does not support nested trees', target_format)
2376
def get_format_string(self):
2377
"""See RepositoryFormat.get_format_string()."""
2378
return ("Bazaar development format 0 with subtree support "
2379
"(needs bzr.dev from before 1.3)\n")
2381
def get_format_description(self):
2382
"""See RepositoryFormat.get_format_description()."""
2383
return ("Development repository format, currently the same as "
2384
"pack-0.92-subtree\n")
added
removed
bzrlib/repofmt/pack_repo.py
