1
# Copyright (C) 2005, 2006, 2007 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,
66
import bzrlib.revision as _mod_revision
67
from bzrlib.store.revision.knit import KnitRevisionStore
68
from bzrlib.store.versioned import VersionedFileStore
69
from bzrlib.trace import mutter, note, warning
72
class PackCommitBuilder(CommitBuilder):
73
"""A subclass of CommitBuilder to add texts with pack semantics.
75
Specifically this uses one knit object rather than one knit object per
76
added text, reducing memory and object pressure.
79
def __init__(self, repository, parents, config, timestamp=None,
80
timezone=None, committer=None, revprops=None,
82
CommitBuilder.__init__(self, repository, parents, config,
83
timestamp=timestamp, timezone=timezone, committer=committer,
84
revprops=revprops, revision_id=revision_id)
85
self._file_graph = graph.Graph(
86
repository._pack_collection.text_index.combined_index)
88
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
89
return self.repository._pack_collection._add_text_to_weave(file_id,
90
self._new_revision_id, new_lines, parents, nostore_sha,
93
def _heads(self, file_id, revision_ids):
94
keys = [(file_id, revision_id) for revision_id in revision_ids]
95
return set([key[1] for key in self._file_graph.heads(keys)])
98
class PackRootCommitBuilder(RootCommitBuilder):
99
"""A subclass of RootCommitBuilder to add texts with pack semantics.
101
Specifically this uses one knit object rather than one knit object per
102
added text, reducing memory and object pressure.
105
def __init__(self, repository, parents, config, timestamp=None,
106
timezone=None, committer=None, revprops=None,
108
CommitBuilder.__init__(self, repository, parents, config,
109
timestamp=timestamp, timezone=timezone, committer=committer,
110
revprops=revprops, revision_id=revision_id)
111
self._file_graph = graph.Graph(
112
repository._pack_collection.text_index.combined_index)
114
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
115
return self.repository._pack_collection._add_text_to_weave(file_id,
116
self._new_revision_id, new_lines, parents, nostore_sha,
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
def __init__(self, revision_index, inventory_index, text_index,
133
"""Create a pack instance.
135
:param revision_index: A GraphIndex for determining what revisions are
136
present in the Pack and accessing the locations of their texts.
137
:param inventory_index: A GraphIndex for determining what inventories are
138
present in the Pack and accessing the locations of their
140
:param text_index: A GraphIndex for determining what file texts
141
are present in the pack and accessing the locations of their
142
texts/deltas (via (fileid, revisionid) tuples).
143
:param revision_index: A GraphIndex for determining what signatures are
144
present in the Pack and accessing the locations of their texts.
146
self.revision_index = revision_index
147
self.inventory_index = inventory_index
148
self.text_index = text_index
149
self.signature_index = signature_index
151
def access_tuple(self):
152
"""Return a tuple (transport, name) for the pack content."""
153
return self.pack_transport, self.file_name()
156
"""Get the file name for the pack on disk."""
157
return self.name + '.pack'
159
def get_revision_count(self):
160
return self.revision_index.key_count()
162
def inventory_index_name(self, name):
163
"""The inv index is the name + .iix."""
164
return self.index_name('inventory', name)
166
def revision_index_name(self, name):
167
"""The revision index is the name + .rix."""
168
return self.index_name('revision', name)
170
def signature_index_name(self, name):
171
"""The signature index is the name + .six."""
172
return self.index_name('signature', name)
174
def text_index_name(self, name):
175
"""The text index is the name + .tix."""
176
return self.index_name('text', name)
178
def _external_compression_parents_of_texts(self):
181
for node in self.text_index.iter_all_entries():
183
refs.update(node[3][1])
187
class ExistingPack(Pack):
188
"""An in memory proxy for an existing .pack and its disk indices."""
190
def __init__(self, pack_transport, name, revision_index, inventory_index,
191
text_index, signature_index):
192
"""Create an ExistingPack object.
194
:param pack_transport: The transport where the pack file resides.
195
:param name: The name of the pack on disk in the pack_transport.
197
Pack.__init__(self, revision_index, inventory_index, text_index,
200
self.pack_transport = pack_transport
201
if None in (revision_index, inventory_index, text_index,
202
signature_index, name, pack_transport):
203
raise AssertionError()
205
def __eq__(self, other):
206
return self.__dict__ == other.__dict__
208
def __ne__(self, other):
209
return not self.__eq__(other)
212
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
213
id(self), self.transport, self.name)
217
"""An in memory proxy for a pack which is being created."""
219
# A map of index 'type' to the file extension and position in the
221
index_definitions = {
222
'revision': ('.rix', 0),
223
'inventory': ('.iix', 1),
225
'signature': ('.six', 3),
228
def __init__(self, upload_transport, index_transport, pack_transport,
229
upload_suffix='', file_mode=None):
230
"""Create a NewPack instance.
232
:param upload_transport: A writable transport for the pack to be
233
incrementally uploaded to.
234
:param index_transport: A writable transport for the pack's indices to
235
be written to when the pack is finished.
236
:param pack_transport: A writable transport for the pack to be renamed
237
to when the upload is complete. This *must* be the same as
238
upload_transport.clone('../packs').
239
:param upload_suffix: An optional suffix to be given to any temporary
240
files created during the pack creation. e.g '.autopack'
241
:param file_mode: An optional file mode to create the new files with.
243
# The relative locations of the packs are constrained, but all are
244
# passed in because the caller has them, so as to avoid object churn.
246
# Revisions: parents list, no text compression.
247
InMemoryGraphIndex(reference_lists=1),
248
# Inventory: We want to map compression only, but currently the
249
# knit code hasn't been updated enough to understand that, so we
250
# have a regular 2-list index giving parents and compression
252
InMemoryGraphIndex(reference_lists=2),
253
# Texts: compression and per file graph, for all fileids - so two
254
# reference lists and two elements in the key tuple.
255
InMemoryGraphIndex(reference_lists=2, key_elements=2),
256
# Signatures: Just blobs to store, no compression, no parents
258
InMemoryGraphIndex(reference_lists=0),
260
# where should the new pack be opened
261
self.upload_transport = upload_transport
262
# where are indices written out to
263
self.index_transport = index_transport
264
# where is the pack renamed to when it is finished?
265
self.pack_transport = pack_transport
266
# What file mode to upload the pack and indices with.
267
self._file_mode = file_mode
268
# tracks the content written to the .pack file.
269
self._hash = md5.new()
270
# a four-tuple with the length in bytes of the indices, once the pack
271
# is finalised. (rev, inv, text, sigs)
272
self.index_sizes = None
273
# How much data to cache when writing packs. Note that this is not
274
# synchronised with reads, because it's not in the transport layer, so
275
# is not safe unless the client knows it won't be reading from the pack
277
self._cache_limit = 0
278
# the temporary pack file name.
279
self.random_name = rand_chars(20) + upload_suffix
280
# when was this pack started ?
281
self.start_time = time.time()
282
# open an output stream for the data added to the pack.
283
self.write_stream = self.upload_transport.open_write_stream(
284
self.random_name, mode=self._file_mode)
285
if 'pack' in debug.debug_flags:
286
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
287
time.ctime(), self.upload_transport.base, self.random_name,
288
time.time() - self.start_time)
289
# A list of byte sequences to be written to the new pack, and the
290
# aggregate size of them. Stored as a list rather than separate
291
# variables so that the _write_data closure below can update them.
292
self._buffer = [[], 0]
293
# create a callable for adding data
295
# robertc says- this is a closure rather than a method on the object
296
# so that the variables are locals, and faster than accessing object
298
def _write_data(bytes, flush=False, _buffer=self._buffer,
299
_write=self.write_stream.write, _update=self._hash.update):
300
_buffer[0].append(bytes)
301
_buffer[1] += len(bytes)
303
if _buffer[1] > self._cache_limit or flush:
304
bytes = ''.join(_buffer[0])
308
# expose this on self, for the occasion when clients want to add data.
309
self._write_data = _write_data
310
# a pack writer object to serialise pack records.
311
self._writer = pack.ContainerWriter(self._write_data)
313
# what state is the pack in? (open, finished, aborted)
317
"""Cancel creating this pack."""
318
self._state = 'aborted'
319
self.write_stream.close()
320
# Remove the temporary pack file.
321
self.upload_transport.delete(self.random_name)
322
# The indices have no state on disk.
324
def access_tuple(self):
325
"""Return a tuple (transport, name) for the pack content."""
326
if self._state == 'finished':
327
return Pack.access_tuple(self)
328
elif self._state == 'open':
329
return self.upload_transport, self.random_name
331
raise AssertionError(self._state)
333
def data_inserted(self):
334
"""True if data has been added to this pack."""
335
return bool(self.get_revision_count() or
336
self.inventory_index.key_count() or
337
self.text_index.key_count() or
338
self.signature_index.key_count())
341
"""Finish the new pack.
344
- finalises the content
345
- assigns a name (the md5 of the content, currently)
346
- writes out the associated indices
347
- renames the pack into place.
348
- stores the index size tuple for the pack in the index_sizes
353
self._write_data('', flush=True)
354
self.name = self._hash.hexdigest()
356
# XXX: It'd be better to write them all to temporary names, then
357
# rename them all into place, so that the window when only some are
358
# visible is smaller. On the other hand none will be seen until
359
# they're in the names list.
360
self.index_sizes = [None, None, None, None]
361
self._write_index('revision', self.revision_index, 'revision')
362
self._write_index('inventory', self.inventory_index, 'inventory')
363
self._write_index('text', self.text_index, 'file texts')
364
self._write_index('signature', self.signature_index,
365
'revision signatures')
366
self.write_stream.close()
367
# Note that this will clobber an existing pack with the same name,
368
# without checking for hash collisions. While this is undesirable this
369
# is something that can be rectified in a subsequent release. One way
370
# to rectify it may be to leave the pack at the original name, writing
371
# its pack-names entry as something like 'HASH: index-sizes
372
# temporary-name'. Allocate that and check for collisions, if it is
373
# collision free then rename it into place. If clients know this scheme
374
# they can handle missing-file errors by:
375
# - try for HASH.pack
376
# - try for temporary-name
377
# - refresh the pack-list to see if the pack is now absent
378
self.upload_transport.rename(self.random_name,
379
'../packs/' + self.name + '.pack')
380
self._state = 'finished'
381
if 'pack' in debug.debug_flags:
382
# XXX: size might be interesting?
383
mutter('%s: create_pack: pack renamed into place: %s%s->%s%s t+%6.3fs',
384
time.ctime(), self.upload_transport.base, self.random_name,
385
self.pack_transport, self.name,
386
time.time() - self.start_time)
389
"""Flush any current data."""
391
bytes = ''.join(self._buffer[0])
392
self.write_stream.write(bytes)
393
self._hash.update(bytes)
394
self._buffer[:] = [[], 0]
396
def index_name(self, index_type, name):
397
"""Get the disk name of an index type for pack name 'name'."""
398
return name + NewPack.index_definitions[index_type][0]
400
def index_offset(self, index_type):
401
"""Get the position in a index_size array for a given index type."""
402
return NewPack.index_definitions[index_type][1]
404
def _replace_index_with_readonly(self, index_type):
405
setattr(self, index_type + '_index',
406
GraphIndex(self.index_transport,
407
self.index_name(index_type, self.name),
408
self.index_sizes[self.index_offset(index_type)]))
410
def set_write_cache_size(self, size):
411
self._cache_limit = size
413
def _write_index(self, index_type, index, label):
414
"""Write out an index.
416
:param index_type: The type of index to write - e.g. 'revision'.
417
:param index: The index object to serialise.
418
:param label: What label to give the index e.g. 'revision'.
420
index_name = self.index_name(index_type, self.name)
421
self.index_sizes[self.index_offset(index_type)] = \
422
self.index_transport.put_file(index_name, index.finish(),
423
mode=self._file_mode)
424
if 'pack' in debug.debug_flags:
425
# XXX: size might be interesting?
426
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
427
time.ctime(), label, self.upload_transport.base,
428
self.random_name, time.time() - self.start_time)
429
# Replace the writable index on this object with a readonly,
430
# presently unloaded index. We should alter
431
# the index layer to make its finish() error if add_node is
432
# subsequently used. RBC
433
self._replace_index_with_readonly(index_type)
436
class AggregateIndex(object):
437
"""An aggregated index for the RepositoryPackCollection.
439
AggregateIndex is reponsible for managing the PackAccess object,
440
Index-To-Pack mapping, and all indices list for a specific type of index
441
such as 'revision index'.
443
A CombinedIndex provides an index on a single key space built up
444
from several on-disk indices. The AggregateIndex builds on this
445
to provide a knit access layer, and allows having up to one writable
446
index within the collection.
448
# XXX: Probably 'can be written to' could/should be separated from 'acts
449
# like a knit index' -- mbp 20071024
452
"""Create an AggregateIndex."""
453
self.index_to_pack = {}
454
self.combined_index = CombinedGraphIndex([])
455
self.knit_access = _PackAccess(self.index_to_pack)
457
def replace_indices(self, index_to_pack, indices):
458
"""Replace the current mappings with fresh ones.
460
This should probably not be used eventually, rather incremental add and
461
removal of indices. It has been added during refactoring of existing
464
:param index_to_pack: A mapping from index objects to
465
(transport, name) tuples for the pack file data.
466
:param indices: A list of indices.
468
# refresh the revision pack map dict without replacing the instance.
469
self.index_to_pack.clear()
470
self.index_to_pack.update(index_to_pack)
471
# XXX: API break - clearly a 'replace' method would be good?
472
self.combined_index._indices[:] = indices
473
# the current add nodes callback for the current writable index if
475
self.add_callback = None
477
def add_index(self, index, pack):
478
"""Add index to the aggregate, which is an index for Pack pack.
480
Future searches on the aggregate index will seach this new index
481
before all previously inserted indices.
483
:param index: An Index for the pack.
484
:param pack: A Pack instance.
486
# expose it to the index map
487
self.index_to_pack[index] = pack.access_tuple()
488
# put it at the front of the linear index list
489
self.combined_index.insert_index(0, index)
491
def add_writable_index(self, index, pack):
492
"""Add an index which is able to have data added to it.
494
There can be at most one writable index at any time. Any
495
modifications made to the knit are put into this index.
497
:param index: An index from the pack parameter.
498
:param pack: A Pack instance.
500
if self.add_callback is not None:
501
raise AssertionError(
502
"%s already has a writable index through %s" % \
503
(self, self.add_callback))
504
# allow writing: queue writes to a new index
505
self.add_index(index, pack)
506
# Updates the index to packs mapping as a side effect,
507
self.knit_access.set_writer(pack._writer, index, pack.access_tuple())
508
self.add_callback = index.add_nodes
511
"""Reset all the aggregate data to nothing."""
512
self.knit_access.set_writer(None, None, (None, None))
513
self.index_to_pack.clear()
514
del self.combined_index._indices[:]
515
self.add_callback = None
517
def remove_index(self, index, pack):
518
"""Remove index from the indices used to answer queries.
520
:param index: An index from the pack parameter.
521
:param pack: A Pack instance.
523
del self.index_to_pack[index]
524
self.combined_index._indices.remove(index)
525
if (self.add_callback is not None and
526
getattr(index, 'add_nodes', None) == self.add_callback):
527
self.add_callback = None
528
self.knit_access.set_writer(None, None, (None, None))
531
class Packer(object):
532
"""Create a pack from packs."""
534
def __init__(self, pack_collection, packs, suffix, revision_ids=None):
537
:param pack_collection: A RepositoryPackCollection object where the
538
new pack is being written to.
539
:param packs: The packs to combine.
540
:param suffix: The suffix to use on the temporary files for the pack.
541
:param revision_ids: Revision ids to limit the pack to.
545
self.revision_ids = revision_ids
546
# The pack object we are creating.
548
self._pack_collection = pack_collection
549
# The index layer keys for the revisions being copied. None for 'all
551
self._revision_keys = None
552
# What text keys to copy. None for 'all texts'. This is set by
553
# _copy_inventory_texts
554
self._text_filter = None
557
def _extra_init(self):
558
"""A template hook to allow extending the constructor trivially."""
560
def pack(self, pb=None):
561
"""Create a new pack by reading data from other packs.
563
This does little more than a bulk copy of data. One key difference
564
is that data with the same item key across multiple packs is elided
565
from the output. The new pack is written into the current pack store
566
along with its indices, and the name added to the pack names. The
567
source packs are not altered and are not required to be in the current
570
:param pb: An optional progress bar to use. A nested bar is created if
572
:return: A Pack object, or None if nothing was copied.
574
# open a pack - using the same name as the last temporary file
575
# - which has already been flushed, so its safe.
576
# XXX: - duplicate code warning with start_write_group; fix before
577
# considering 'done'.
578
if self._pack_collection._new_pack is not None:
579
raise errors.BzrError('call to create_pack_from_packs while '
580
'another pack is being written.')
581
if self.revision_ids is not None:
582
if len(self.revision_ids) == 0:
583
# silly fetch request.
586
self.revision_ids = frozenset(self.revision_ids)
588
self.pb = ui.ui_factory.nested_progress_bar()
592
return self._create_pack_from_packs()
598
"""Open a pack for the pack we are creating."""
599
return NewPack(self._pack_collection._upload_transport,
600
self._pack_collection._index_transport,
601
self._pack_collection._pack_transport, upload_suffix=self.suffix,
602
file_mode=self._pack_collection.repo.control_files._file_mode)
604
def _copy_revision_texts(self):
605
"""Copy revision data to the new pack."""
607
if self.revision_ids:
608
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
611
# select revision keys
612
revision_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
613
self.packs, 'revision_index')[0]
614
revision_nodes = self._pack_collection._index_contents(revision_index_map, revision_keys)
615
# copy revision keys and adjust values
616
self.pb.update("Copying revision texts", 1)
617
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
618
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
619
self.new_pack.revision_index, readv_group_iter, total_items))
620
if 'pack' in debug.debug_flags:
621
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
622
time.ctime(), self._pack_collection._upload_transport.base,
623
self.new_pack.random_name,
624
self.new_pack.revision_index.key_count(),
625
time.time() - self.new_pack.start_time)
626
self._revision_keys = revision_keys
628
def _copy_inventory_texts(self):
629
"""Copy the inventory texts to the new pack.
631
self._revision_keys is used to determine what inventories to copy.
633
Sets self._text_filter appropriately.
635
# select inventory keys
636
inv_keys = self._revision_keys # currently the same keyspace, and note that
637
# querying for keys here could introduce a bug where an inventory item
638
# is missed, so do not change it to query separately without cross
639
# checking like the text key check below.
640
inventory_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
641
self.packs, 'inventory_index')[0]
642
inv_nodes = self._pack_collection._index_contents(inventory_index_map, inv_keys)
643
# copy inventory keys and adjust values
644
# XXX: Should be a helper function to allow different inv representation
646
self.pb.update("Copying inventory texts", 2)
647
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
648
# Only grab the output lines if we will be processing them
649
output_lines = bool(self.revision_ids)
650
inv_lines = self._copy_nodes_graph(inventory_index_map,
651
self.new_pack._writer, self.new_pack.inventory_index,
652
readv_group_iter, total_items, output_lines=output_lines)
653
if self.revision_ids:
654
self._process_inventory_lines(inv_lines)
656
# eat the iterator to cause it to execute.
658
self._text_filter = None
659
if 'pack' in debug.debug_flags:
660
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
661
time.ctime(), self._pack_collection._upload_transport.base,
662
self.new_pack.random_name,
663
self.new_pack.inventory_index.key_count(),
664
time.time() - self.new_pack.start_time)
666
def _copy_text_texts(self):
668
text_index_map, text_nodes = self._get_text_nodes()
669
if self._text_filter is not None:
670
# We could return the keys copied as part of the return value from
671
# _copy_nodes_graph but this doesn't work all that well with the
672
# need to get line output too, so we check separately, and as we're
673
# going to buffer everything anyway, we check beforehand, which
674
# saves reading knit data over the wire when we know there are
676
text_nodes = set(text_nodes)
677
present_text_keys = set(_node[1] for _node in text_nodes)
678
missing_text_keys = set(self._text_filter) - present_text_keys
679
if missing_text_keys:
680
# TODO: raise a specific error that can handle many missing
682
a_missing_key = missing_text_keys.pop()
683
raise errors.RevisionNotPresent(a_missing_key[1],
685
# copy text keys and adjust values
686
self.pb.update("Copying content texts", 3)
687
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
688
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
689
self.new_pack.text_index, readv_group_iter, total_items))
690
self._log_copied_texts()
692
def _check_references(self):
693
"""Make sure our external refereneces are present."""
694
external_refs = self.new_pack._external_compression_parents_of_texts()
696
index = self._pack_collection.text_index.combined_index
697
found_items = list(index.iter_entries(external_refs))
698
if len(found_items) != len(external_refs):
699
found_keys = set(k for idx, k, refs, value in found_items)
700
missing_items = external_refs - found_keys
701
missing_file_id, missing_revision_id = missing_items.pop()
702
raise errors.RevisionNotPresent(missing_revision_id,
705
def _create_pack_from_packs(self):
706
self.pb.update("Opening pack", 0, 5)
707
self.new_pack = self.open_pack()
708
new_pack = self.new_pack
709
# buffer data - we won't be reading-back during the pack creation and
710
# this makes a significant difference on sftp pushes.
711
new_pack.set_write_cache_size(1024*1024)
712
if 'pack' in debug.debug_flags:
713
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
714
for a_pack in self.packs]
715
if self.revision_ids is not None:
716
rev_count = len(self.revision_ids)
719
mutter('%s: create_pack: creating pack from source packs: '
720
'%s%s %s revisions wanted %s t=0',
721
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
722
plain_pack_list, rev_count)
723
self._copy_revision_texts()
724
self._copy_inventory_texts()
725
self._copy_text_texts()
726
# select signature keys
727
signature_filter = self._revision_keys # same keyspace
728
signature_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
729
self.packs, 'signature_index')[0]
730
signature_nodes = self._pack_collection._index_contents(signature_index_map,
732
# copy signature keys and adjust values
733
self.pb.update("Copying signature texts", 4)
734
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
735
new_pack.signature_index)
736
if 'pack' in debug.debug_flags:
737
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
738
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
739
new_pack.signature_index.key_count(),
740
time.time() - new_pack.start_time)
741
self._check_references()
742
if not self._use_pack(new_pack):
745
self.pb.update("Finishing pack", 5)
747
self._pack_collection.allocate(new_pack)
750
def _copy_nodes(self, nodes, index_map, writer, write_index):
751
"""Copy knit nodes between packs with no graph references."""
752
pb = ui.ui_factory.nested_progress_bar()
754
return self._do_copy_nodes(nodes, index_map, writer,
759
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb):
760
# for record verification
761
knit_data = _KnitData(None)
762
# plan a readv on each source pack:
764
nodes = sorted(nodes)
765
# how to map this into knit.py - or knit.py into this?
766
# we don't want the typical knit logic, we want grouping by pack
767
# at this point - perhaps a helper library for the following code
768
# duplication points?
770
for index, key, value in nodes:
771
if index not in request_groups:
772
request_groups[index] = []
773
request_groups[index].append((key, value))
775
pb.update("Copied record", record_index, len(nodes))
776
for index, items in request_groups.iteritems():
777
pack_readv_requests = []
778
for key, value in items:
779
# ---- KnitGraphIndex.get_position
780
bits = value[1:].split(' ')
781
offset, length = int(bits[0]), int(bits[1])
782
pack_readv_requests.append((offset, length, (key, value[0])))
783
# linear scan up the pack
784
pack_readv_requests.sort()
786
transport, path = index_map[index]
787
reader = pack.make_readv_reader(transport, path,
788
[offset[0:2] for offset in pack_readv_requests])
789
for (names, read_func), (_1, _2, (key, eol_flag)) in \
790
izip(reader.iter_records(), pack_readv_requests):
791
raw_data = read_func(None)
792
# check the header only
793
df, _ = knit_data._parse_record_header(key[-1], raw_data)
795
pos, size = writer.add_bytes_record(raw_data, names)
796
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
797
pb.update("Copied record", record_index)
800
def _copy_nodes_graph(self, index_map, writer, write_index,
801
readv_group_iter, total_items, output_lines=False):
802
"""Copy knit nodes between packs.
804
:param output_lines: Return lines present in the copied data as
805
an iterator of line,version_id.
807
pb = ui.ui_factory.nested_progress_bar()
809
for result in self._do_copy_nodes_graph(index_map, writer,
810
write_index, output_lines, pb, readv_group_iter, total_items):
813
# Python 2.4 does not permit try:finally: in a generator.
819
def _do_copy_nodes_graph(self, index_map, writer, write_index,
820
output_lines, pb, readv_group_iter, total_items):
821
# for record verification
822
knit_data = _KnitData(None)
823
# for line extraction when requested (inventories only)
825
factory = knit.KnitPlainFactory()
827
pb.update("Copied record", record_index, total_items)
828
for index, readv_vector, node_vector in readv_group_iter:
830
transport, path = index_map[index]
831
reader = pack.make_readv_reader(transport, path, readv_vector)
832
for (names, read_func), (key, eol_flag, references) in \
833
izip(reader.iter_records(), node_vector):
834
raw_data = read_func(None)
837
# read the entire thing
838
content, _ = knit_data._parse_record(version_id, raw_data)
839
if len(references[-1]) == 0:
840
line_iterator = factory.get_fulltext_content(content)
842
line_iterator = factory.get_linedelta_content(content)
843
for line in line_iterator:
844
yield line, version_id
846
# check the header only
847
df, _ = knit_data._parse_record_header(version_id, raw_data)
849
pos, size = writer.add_bytes_record(raw_data, names)
850
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
851
pb.update("Copied record", record_index)
854
def _get_text_nodes(self):
855
text_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
856
self.packs, 'text_index')[0]
857
return text_index_map, self._pack_collection._index_contents(text_index_map,
860
def _least_readv_node_readv(self, nodes):
861
"""Generate request groups for nodes using the least readv's.
863
:param nodes: An iterable of graph index nodes.
864
:return: Total node count and an iterator of the data needed to perform
865
readvs to obtain the data for nodes. Each item yielded by the
866
iterator is a tuple with:
867
index, readv_vector, node_vector. readv_vector is a list ready to
868
hand to the transport readv method, and node_vector is a list of
869
(key, eol_flag, references) for the the node retrieved by the
870
matching readv_vector.
872
# group by pack so we do one readv per pack
873
nodes = sorted(nodes)
876
for index, key, value, references in nodes:
877
if index not in request_groups:
878
request_groups[index] = []
879
request_groups[index].append((key, value, references))
881
for index, items in request_groups.iteritems():
882
pack_readv_requests = []
883
for key, value, references in items:
884
# ---- KnitGraphIndex.get_position
885
bits = value[1:].split(' ')
886
offset, length = int(bits[0]), int(bits[1])
887
pack_readv_requests.append(
888
((offset, length), (key, value[0], references)))
889
# linear scan up the pack to maximum range combining.
890
pack_readv_requests.sort()
891
# split out the readv and the node data.
892
pack_readv = [readv for readv, node in pack_readv_requests]
893
node_vector = [node for readv, node in pack_readv_requests]
894
result.append((index, pack_readv, node_vector))
897
def _log_copied_texts(self):
898
if 'pack' in debug.debug_flags:
899
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
900
time.ctime(), self._pack_collection._upload_transport.base,
901
self.new_pack.random_name,
902
self.new_pack.text_index.key_count(),
903
time.time() - self.new_pack.start_time)
905
def _process_inventory_lines(self, inv_lines):
906
"""Use up the inv_lines generator and setup a text key filter."""
907
repo = self._pack_collection.repo
908
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
909
inv_lines, self.revision_ids)
911
for fileid, file_revids in fileid_revisions.iteritems():
912
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
913
self._text_filter = text_filter
915
def _revision_node_readv(self, revision_nodes):
916
"""Return the total revisions and the readv's to issue.
918
:param revision_nodes: The revision index contents for the packs being
919
incorporated into the new pack.
920
:return: As per _least_readv_node_readv.
922
return self._least_readv_node_readv(revision_nodes)
924
def _use_pack(self, new_pack):
925
"""Return True if new_pack should be used.
927
:param new_pack: The pack that has just been created.
928
:return: True if the pack should be used.
930
return new_pack.data_inserted()
933
class OptimisingPacker(Packer):
934
"""A packer which spends more time to create better disk layouts."""
936
def _revision_node_readv(self, revision_nodes):
937
"""Return the total revisions and the readv's to issue.
939
This sort places revisions in topological order with the ancestors
942
:param revision_nodes: The revision index contents for the packs being
943
incorporated into the new pack.
944
:return: As per _least_readv_node_readv.
946
# build an ancestors dict
949
for index, key, value, references in revision_nodes:
950
ancestors[key] = references[0]
951
by_key[key] = (index, value, references)
952
order = tsort.topo_sort(ancestors)
954
# Single IO is pathological, but it will work as a starting point.
956
for key in reversed(order):
957
index, value, references = by_key[key]
958
# ---- KnitGraphIndex.get_position
959
bits = value[1:].split(' ')
960
offset, length = int(bits[0]), int(bits[1])
962
(index, [(offset, length)], [(key, value[0], references)]))
963
# TODO: combine requests in the same index that are in ascending order.
964
return total, requests
967
class ReconcilePacker(Packer):
968
"""A packer which regenerates indices etc as it copies.
970
This is used by ``bzr reconcile`` to cause parent text pointers to be
974
def _extra_init(self):
975
self._data_changed = False
977
def _process_inventory_lines(self, inv_lines):
978
"""Generate a text key reference map rather for reconciling with."""
979
repo = self._pack_collection.repo
980
refs = repo._find_text_key_references_from_xml_inventory_lines(
982
self._text_refs = refs
983
# during reconcile we:
984
# - convert unreferenced texts to full texts
985
# - correct texts which reference a text not copied to be full texts
986
# - copy all others as-is but with corrected parents.
987
# - so at this point we don't know enough to decide what becomes a full
989
self._text_filter = None
991
def _copy_text_texts(self):
992
"""generate what texts we should have and then copy."""
993
self.pb.update("Copying content texts", 3)
994
# we have three major tasks here:
995
# 1) generate the ideal index
996
repo = self._pack_collection.repo
997
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
999
self.new_pack.revision_index.iter_all_entries()])
1000
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1001
# 2) generate a text_nodes list that contains all the deltas that can
1002
# be used as-is, with corrected parents.
1005
discarded_nodes = []
1006
NULL_REVISION = _mod_revision.NULL_REVISION
1007
text_index_map, text_nodes = self._get_text_nodes()
1008
for node in text_nodes:
1014
ideal_parents = tuple(ideal_index[node[1]])
1016
discarded_nodes.append(node)
1017
self._data_changed = True
1019
if ideal_parents == (NULL_REVISION,):
1021
if ideal_parents == node[3][0]:
1023
ok_nodes.append(node)
1024
elif ideal_parents[0:1] == node[3][0][0:1]:
1025
# the left most parent is the same, or there are no parents
1026
# today. Either way, we can preserve the representation as
1027
# long as we change the refs to be inserted.
1028
self._data_changed = True
1029
ok_nodes.append((node[0], node[1], node[2],
1030
(ideal_parents, node[3][1])))
1031
self._data_changed = True
1033
# Reinsert this text completely
1034
bad_texts.append((node[1], ideal_parents))
1035
self._data_changed = True
1036
# we're finished with some data.
1039
# 3) bulk copy the ok data
1040
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1041
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1042
self.new_pack.text_index, readv_group_iter, total_items))
1043
# 4) adhoc copy all the other texts.
1044
# We have to topologically insert all texts otherwise we can fail to
1045
# reconcile when parts of a single delta chain are preserved intact,
1046
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1047
# reinserted, and if d3 has incorrect parents it will also be
1048
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1049
# copied), so we will try to delta, but d2 is not currently able to be
1050
# extracted because it's basis d1 is not present. Topologically sorting
1051
# addresses this. The following generates a sort for all the texts that
1052
# are being inserted without having to reference the entire text key
1053
# space (we only topo sort the revisions, which is smaller).
1054
topo_order = tsort.topo_sort(ancestors)
1055
rev_order = dict(zip(topo_order, range(len(topo_order))))
1056
bad_texts.sort(key=lambda key:rev_order[key[0][1]])
1057
transaction = repo.get_transaction()
1058
file_id_index = GraphIndexPrefixAdapter(
1059
self.new_pack.text_index,
1061
add_nodes_callback=self.new_pack.text_index.add_nodes)
1062
knit_index = KnitGraphIndex(file_id_index,
1063
add_callback=file_id_index.add_nodes,
1064
deltas=True, parents=True)
1065
output_knit = knit.KnitVersionedFile('reconcile-texts',
1066
self._pack_collection.transport,
1068
access_method=_PackAccess(
1069
{self.new_pack.text_index:self.new_pack.access_tuple()},
1070
(self.new_pack._writer, self.new_pack.text_index)),
1071
factory=knit.KnitPlainFactory())
1072
for key, parent_keys in bad_texts:
1073
# We refer to the new pack to delta data being output.
1074
# A possible improvement would be to catch errors on short reads
1075
# and only flush then.
1076
self.new_pack.flush()
1078
for parent_key in parent_keys:
1079
if parent_key[0] != key[0]:
1080
# Graph parents must match the fileid
1081
raise errors.BzrError('Mismatched key parent %r:%r' %
1083
parents.append(parent_key[1])
1084
source_weave = repo.weave_store.get_weave(key[0], transaction)
1085
text_lines = source_weave.get_lines(key[1])
1086
# adapt the 'knit' to the current file_id.
1087
file_id_index = GraphIndexPrefixAdapter(
1088
self.new_pack.text_index,
1090
add_nodes_callback=self.new_pack.text_index.add_nodes)
1091
knit_index._graph_index = file_id_index
1092
knit_index._add_callback = file_id_index.add_nodes
1093
output_knit.add_lines_with_ghosts(
1094
key[1], parents, text_lines, random_id=True, check_content=False)
1095
# 5) check that nothing inserted has a reference outside the keyspace.
1096
missing_text_keys = self.new_pack._external_compression_parents_of_texts()
1097
if missing_text_keys:
1098
raise errors.BzrError('Reference to missing compression parents %r'
1100
self._log_copied_texts()
1102
def _use_pack(self, new_pack):
1103
"""Override _use_pack to check for reconcile having changed content."""
1104
# XXX: we might be better checking this at the copy time.
1105
original_inventory_keys = set()
1106
inv_index = self._pack_collection.inventory_index.combined_index
1107
for entry in inv_index.iter_all_entries():
1108
original_inventory_keys.add(entry[1])
1109
new_inventory_keys = set()
1110
for entry in new_pack.inventory_index.iter_all_entries():
1111
new_inventory_keys.add(entry[1])
1112
if new_inventory_keys != original_inventory_keys:
1113
self._data_changed = True
1114
return new_pack.data_inserted() and self._data_changed
1117
class RepositoryPackCollection(object):
1118
"""Management of packs within a repository."""
1120
def __init__(self, repo, transport, index_transport, upload_transport,
1122
"""Create a new RepositoryPackCollection.
1124
:param transport: Addresses the repository base directory
1125
(typically .bzr/repository/).
1126
:param index_transport: Addresses the directory containing indices.
1127
:param upload_transport: Addresses the directory into which packs are written
1128
while they're being created.
1129
:param pack_transport: Addresses the directory of existing complete packs.
1132
self.transport = transport
1133
self._index_transport = index_transport
1134
self._upload_transport = upload_transport
1135
self._pack_transport = pack_transport
1136
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
1139
self._packs_by_name = {}
1140
# the previous pack-names content
1141
self._packs_at_load = None
1142
# when a pack is being created by this object, the state of that pack.
1143
self._new_pack = None
1144
# aggregated revision index data
1145
self.revision_index = AggregateIndex()
1146
self.inventory_index = AggregateIndex()
1147
self.text_index = AggregateIndex()
1148
self.signature_index = AggregateIndex()
1150
def add_pack_to_memory(self, pack):
1151
"""Make a Pack object available to the repository to satisfy queries.
1153
:param pack: A Pack object.
1155
if pack.name in self._packs_by_name:
1156
raise AssertionError()
1157
self.packs.append(pack)
1158
self._packs_by_name[pack.name] = pack
1159
self.revision_index.add_index(pack.revision_index, pack)
1160
self.inventory_index.add_index(pack.inventory_index, pack)
1161
self.text_index.add_index(pack.text_index, pack)
1162
self.signature_index.add_index(pack.signature_index, pack)
1164
def _add_text_to_weave(self, file_id, revision_id, new_lines, parents,
1165
nostore_sha, random_revid):
1166
file_id_index = GraphIndexPrefixAdapter(
1167
self.text_index.combined_index,
1169
add_nodes_callback=self.text_index.add_callback)
1170
self.repo._text_knit._index._graph_index = file_id_index
1171
self.repo._text_knit._index._add_callback = file_id_index.add_nodes
1172
return self.repo._text_knit.add_lines_with_ghosts(
1173
revision_id, parents, new_lines, nostore_sha=nostore_sha,
1174
random_id=random_revid, check_content=False)[0:2]
1176
def all_packs(self):
1177
"""Return a list of all the Pack objects this repository has.
1179
Note that an in-progress pack being created is not returned.
1181
:return: A list of Pack objects for all the packs in the repository.
1184
for name in self.names():
1185
result.append(self.get_pack_by_name(name))
1189
"""Pack the pack collection incrementally.
1191
This will not attempt global reorganisation or recompression,
1192
rather it will just ensure that the total number of packs does
1193
not grow without bound. It uses the _max_pack_count method to
1194
determine if autopacking is needed, and the pack_distribution
1195
method to determine the number of revisions in each pack.
1197
If autopacking takes place then the packs name collection will have
1198
been flushed to disk - packing requires updating the name collection
1199
in synchronisation with certain steps. Otherwise the names collection
1202
:return: True if packing took place.
1204
# XXX: Should not be needed when the management of indices is sane.
1205
total_revisions = self.revision_index.combined_index.key_count()
1206
total_packs = len(self._names)
1207
if self._max_pack_count(total_revisions) >= total_packs:
1209
# XXX: the following may want to be a class, to pack with a given
1211
mutter('Auto-packing repository %s, which has %d pack files, '
1212
'containing %d revisions into %d packs.', self, total_packs,
1213
total_revisions, self._max_pack_count(total_revisions))
1214
# determine which packs need changing
1215
pack_distribution = self.pack_distribution(total_revisions)
1217
for pack in self.all_packs():
1218
revision_count = pack.get_revision_count()
1219
if revision_count == 0:
1220
# revision less packs are not generated by normal operation,
1221
# only by operations like sign-my-commits, and thus will not
1222
# tend to grow rapdily or without bound like commit containing
1223
# packs do - leave them alone as packing them really should
1224
# group their data with the relevant commit, and that may
1225
# involve rewriting ancient history - which autopack tries to
1226
# avoid. Alternatively we could not group the data but treat
1227
# each of these as having a single revision, and thus add
1228
# one revision for each to the total revision count, to get
1229
# a matching distribution.
1231
existing_packs.append((revision_count, pack))
1232
pack_operations = self.plan_autopack_combinations(
1233
existing_packs, pack_distribution)
1234
self._execute_pack_operations(pack_operations)
1237
def _execute_pack_operations(self, pack_operations, _packer_class=Packer):
1238
"""Execute a series of pack operations.
1240
:param pack_operations: A list of [revision_count, packs_to_combine].
1241
:param _packer_class: The class of packer to use (default: Packer).
1244
for revision_count, packs in pack_operations:
1245
# we may have no-ops from the setup logic
1248
_packer_class(self, packs, '.autopack').pack()
1250
self._remove_pack_from_memory(pack)
1251
# record the newly available packs and stop advertising the old
1253
self._save_pack_names(clear_obsolete_packs=True)
1254
# Move the old packs out of the way now they are no longer referenced.
1255
for revision_count, packs in pack_operations:
1256
self._obsolete_packs(packs)
1258
def lock_names(self):
1259
"""Acquire the mutex around the pack-names index.
1261
This cannot be used in the middle of a read-only transaction on the
1264
self.repo.control_files.lock_write()
1267
"""Pack the pack collection totally."""
1268
self.ensure_loaded()
1269
total_packs = len(self._names)
1271
# This is arguably wrong because we might not be optimal, but for
1272
# now lets leave it in. (e.g. reconcile -> one pack. But not
1275
total_revisions = self.revision_index.combined_index.key_count()
1276
# XXX: the following may want to be a class, to pack with a given
1278
mutter('Packing repository %s, which has %d pack files, '
1279
'containing %d revisions into 1 packs.', self, total_packs,
1281
# determine which packs need changing
1282
pack_distribution = [1]
1283
pack_operations = [[0, []]]
1284
for pack in self.all_packs():
1285
pack_operations[-1][0] += pack.get_revision_count()
1286
pack_operations[-1][1].append(pack)
1287
self._execute_pack_operations(pack_operations, OptimisingPacker)
1289
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1290
"""Plan a pack operation.
1292
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1294
:param pack_distribution: A list with the number of revisions desired
1297
if len(existing_packs) <= len(pack_distribution):
1299
existing_packs.sort(reverse=True)
1300
pack_operations = [[0, []]]
1301
# plan out what packs to keep, and what to reorganise
1302
while len(existing_packs):
1303
# take the largest pack, and if its less than the head of the
1304
# distribution chart we will include its contents in the new pack for
1305
# that position. If its larger, we remove its size from the
1306
# distribution chart
1307
next_pack_rev_count, next_pack = existing_packs.pop(0)
1308
if next_pack_rev_count >= pack_distribution[0]:
1309
# this is already packed 'better' than this, so we can
1310
# not waste time packing it.
1311
while next_pack_rev_count > 0:
1312
next_pack_rev_count -= pack_distribution[0]
1313
if next_pack_rev_count >= 0:
1315
del pack_distribution[0]
1317
# didn't use that entire bucket up
1318
pack_distribution[0] = -next_pack_rev_count
1320
# add the revisions we're going to add to the next output pack
1321
pack_operations[-1][0] += next_pack_rev_count
1322
# allocate this pack to the next pack sub operation
1323
pack_operations[-1][1].append(next_pack)
1324
if pack_operations[-1][0] >= pack_distribution[0]:
1325
# this pack is used up, shift left.
1326
del pack_distribution[0]
1327
pack_operations.append([0, []])
1329
return pack_operations
1331
def ensure_loaded(self):
1332
# NB: if you see an assertion error here, its probably access against
1333
# an unlocked repo. Naughty.
1334
if not self.repo.is_locked():
1335
raise errors.ObjectNotLocked(self.repo)
1336
if self._names is None:
1338
self._packs_at_load = set()
1339
for index, key, value in self._iter_disk_pack_index():
1341
self._names[name] = self._parse_index_sizes(value)
1342
self._packs_at_load.add((key, value))
1343
# populate all the metadata.
1346
def _parse_index_sizes(self, value):
1347
"""Parse a string of index sizes."""
1348
return tuple([int(digits) for digits in value.split(' ')])
1350
def get_pack_by_name(self, name):
1351
"""Get a Pack object by name.
1353
:param name: The name of the pack - e.g. '123456'
1354
:return: A Pack object.
1357
return self._packs_by_name[name]
1359
rev_index = self._make_index(name, '.rix')
1360
inv_index = self._make_index(name, '.iix')
1361
txt_index = self._make_index(name, '.tix')
1362
sig_index = self._make_index(name, '.six')
1363
result = ExistingPack(self._pack_transport, name, rev_index,
1364
inv_index, txt_index, sig_index)
1365
self.add_pack_to_memory(result)
1368
def allocate(self, a_new_pack):
1369
"""Allocate name in the list of packs.
1371
:param a_new_pack: A NewPack instance to be added to the collection of
1372
packs for this repository.
1374
self.ensure_loaded()
1375
if a_new_pack.name in self._names:
1376
raise errors.BzrError(
1377
'Pack %r already exists in %s' % (a_new_pack.name, self))
1378
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1379
self.add_pack_to_memory(a_new_pack)
1381
def _iter_disk_pack_index(self):
1382
"""Iterate over the contents of the pack-names index.
1384
This is used when loading the list from disk, and before writing to
1385
detect updates from others during our write operation.
1386
:return: An iterator of the index contents.
1388
return GraphIndex(self.transport, 'pack-names', None
1389
).iter_all_entries()
1391
def _make_index(self, name, suffix):
1392
size_offset = self._suffix_offsets[suffix]
1393
index_name = name + suffix
1394
index_size = self._names[name][size_offset]
1396
self._index_transport, index_name, index_size)
1398
def _max_pack_count(self, total_revisions):
1399
"""Return the maximum number of packs to use for total revisions.
1401
:param total_revisions: The total number of revisions in the
1404
if not total_revisions:
1406
digits = str(total_revisions)
1408
for digit in digits:
1409
result += int(digit)
1413
"""Provide an order to the underlying names."""
1414
return sorted(self._names.keys())
1416
def _obsolete_packs(self, packs):
1417
"""Move a number of packs which have been obsoleted out of the way.
1419
Each pack and its associated indices are moved out of the way.
1421
Note: for correctness this function should only be called after a new
1422
pack names index has been written without these pack names, and with
1423
the names of packs that contain the data previously available via these
1426
:param packs: The packs to obsolete.
1427
:param return: None.
1430
pack.pack_transport.rename(pack.file_name(),
1431
'../obsolete_packs/' + pack.file_name())
1432
# TODO: Probably needs to know all possible indices for this pack
1433
# - or maybe list the directory and move all indices matching this
1434
# name whether we recognize it or not?
1435
for suffix in ('.iix', '.six', '.tix', '.rix'):
1436
self._index_transport.rename(pack.name + suffix,
1437
'../obsolete_packs/' + pack.name + suffix)
1439
def pack_distribution(self, total_revisions):
1440
"""Generate a list of the number of revisions to put in each pack.
1442
:param total_revisions: The total number of revisions in the
1445
if total_revisions == 0:
1447
digits = reversed(str(total_revisions))
1449
for exponent, count in enumerate(digits):
1450
size = 10 ** exponent
1451
for pos in range(int(count)):
1453
return list(reversed(result))
1455
def _pack_tuple(self, name):
1456
"""Return a tuple with the transport and file name for a pack name."""
1457
return self._pack_transport, name + '.pack'
1459
def _remove_pack_from_memory(self, pack):
1460
"""Remove pack from the packs accessed by this repository.
1462
Only affects memory state, until self._save_pack_names() is invoked.
1464
self._names.pop(pack.name)
1465
self._packs_by_name.pop(pack.name)
1466
self._remove_pack_indices(pack)
1468
def _remove_pack_indices(self, pack):
1469
"""Remove the indices for pack from the aggregated indices."""
1470
self.revision_index.remove_index(pack.revision_index, pack)
1471
self.inventory_index.remove_index(pack.inventory_index, pack)
1472
self.text_index.remove_index(pack.text_index, pack)
1473
self.signature_index.remove_index(pack.signature_index, pack)
1476
"""Clear all cached data."""
1477
# cached revision data
1478
self.repo._revision_knit = None
1479
self.revision_index.clear()
1480
# cached signature data
1481
self.repo._signature_knit = None
1482
self.signature_index.clear()
1483
# cached file text data
1484
self.text_index.clear()
1485
self.repo._text_knit = None
1486
# cached inventory data
1487
self.inventory_index.clear()
1488
# remove the open pack
1489
self._new_pack = None
1490
# information about packs.
1493
self._packs_by_name = {}
1494
self._packs_at_load = None
1496
def _make_index_map(self, index_suffix):
1497
"""Return information on existing indices.
1499
:param suffix: Index suffix added to pack name.
1501
:returns: (pack_map, indices) where indices is a list of GraphIndex
1502
objects, and pack_map is a mapping from those objects to the
1503
pack tuple they describe.
1505
# TODO: stop using this; it creates new indices unnecessarily.
1506
self.ensure_loaded()
1507
suffix_map = {'.rix': 'revision_index',
1508
'.six': 'signature_index',
1509
'.iix': 'inventory_index',
1510
'.tix': 'text_index',
1512
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1513
suffix_map[index_suffix])
1515
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1516
"""Convert a list of packs to an index pack map and index list.
1518
:param packs: The packs list to process.
1519
:param index_attribute: The attribute that the desired index is found
1521
:return: A tuple (map, list) where map contains the dict from
1522
index:pack_tuple, and lsit contains the indices in the same order
1528
index = getattr(pack, index_attribute)
1529
indices.append(index)
1530
pack_map[index] = (pack.pack_transport, pack.file_name())
1531
return pack_map, indices
1533
def _index_contents(self, pack_map, key_filter=None):
1534
"""Get an iterable of the index contents from a pack_map.
1536
:param pack_map: A map from indices to pack details.
1537
:param key_filter: An optional filter to limit the
1540
indices = [index for index in pack_map.iterkeys()]
1541
all_index = CombinedGraphIndex(indices)
1542
if key_filter is None:
1543
return all_index.iter_all_entries()
1545
return all_index.iter_entries(key_filter)
1547
def _unlock_names(self):
1548
"""Release the mutex around the pack-names index."""
1549
self.repo.control_files.unlock()
1551
def _save_pack_names(self, clear_obsolete_packs=False):
1552
"""Save the list of packs.
1554
This will take out the mutex around the pack names list for the
1555
duration of the method call. If concurrent updates have been made, a
1556
three-way merge between the current list and the current in memory list
1559
:param clear_obsolete_packs: If True, clear out the contents of the
1560
obsolete_packs directory.
1564
builder = GraphIndexBuilder()
1565
# load the disk nodes across
1567
for index, key, value in self._iter_disk_pack_index():
1568
disk_nodes.add((key, value))
1569
# do a two-way diff against our original content
1570
current_nodes = set()
1571
for name, sizes in self._names.iteritems():
1573
((name, ), ' '.join(str(size) for size in sizes)))
1574
deleted_nodes = self._packs_at_load - current_nodes
1575
new_nodes = current_nodes - self._packs_at_load
1576
disk_nodes.difference_update(deleted_nodes)
1577
disk_nodes.update(new_nodes)
1578
# TODO: handle same-name, index-size-changes here -
1579
# e.g. use the value from disk, not ours, *unless* we're the one
1581
for key, value in disk_nodes:
1582
builder.add_node(key, value)
1583
self.transport.put_file('pack-names', builder.finish(),
1584
mode=self.repo.control_files._file_mode)
1585
# move the baseline forward
1586
self._packs_at_load = disk_nodes
1587
# now clear out the obsolete packs directory
1588
if clear_obsolete_packs:
1589
self.transport.clone('obsolete_packs').delete_multi(
1590
self.transport.list_dir('obsolete_packs'))
1592
self._unlock_names()
1593
# synchronise the memory packs list with what we just wrote:
1594
new_names = dict(disk_nodes)
1595
# drop no longer present nodes
1596
for pack in self.all_packs():
1597
if (pack.name,) not in new_names:
1598
self._remove_pack_from_memory(pack)
1599
# add new nodes/refresh existing ones
1600
for key, value in disk_nodes:
1602
sizes = self._parse_index_sizes(value)
1603
if name in self._names:
1605
if sizes != self._names[name]:
1606
# the pack for name has had its indices replaced - rare but
1607
# important to handle. XXX: probably can never happen today
1608
# because the three-way merge code above does not handle it
1609
# - you may end up adding the same key twice to the new
1610
# disk index because the set values are the same, unless
1611
# the only index shows up as deleted by the set difference
1612
# - which it may. Until there is a specific test for this,
1613
# assume its broken. RBC 20071017.
1614
self._remove_pack_from_memory(self.get_pack_by_name(name))
1615
self._names[name] = sizes
1616
self.get_pack_by_name(name)
1619
self._names[name] = sizes
1620
self.get_pack_by_name(name)
1622
def _start_write_group(self):
1623
# Do not permit preparation for writing if we're not in a 'write lock'.
1624
if not self.repo.is_write_locked():
1625
raise errors.NotWriteLocked(self)
1626
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1627
self._pack_transport, upload_suffix='.pack',
1628
file_mode=self.repo.control_files._file_mode)
1629
# allow writing: queue writes to a new index
1630
self.revision_index.add_writable_index(self._new_pack.revision_index,
1632
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
1634
self.text_index.add_writable_index(self._new_pack.text_index,
1636
self.signature_index.add_writable_index(self._new_pack.signature_index,
1639
# reused revision and signature knits may need updating
1641
# "Hysterical raisins. client code in bzrlib grabs those knits outside
1642
# of write groups and then mutates it inside the write group."
1643
if self.repo._revision_knit is not None:
1644
self.repo._revision_knit._index._add_callback = \
1645
self.revision_index.add_callback
1646
if self.repo._signature_knit is not None:
1647
self.repo._signature_knit._index._add_callback = \
1648
self.signature_index.add_callback
1649
# create a reused knit object for text addition in commit.
1650
self.repo._text_knit = self.repo.weave_store.get_weave_or_empty(
1653
def _abort_write_group(self):
1654
# FIXME: just drop the transient index.
1655
# forget what names there are
1656
if self._new_pack is not None:
1657
self._new_pack.abort()
1658
self._remove_pack_indices(self._new_pack)
1659
self._new_pack = None
1660
self.repo._text_knit = None
1662
def _commit_write_group(self):
1663
self._remove_pack_indices(self._new_pack)
1664
if self._new_pack.data_inserted():
1665
# get all the data to disk and read to use
1666
self._new_pack.finish()
1667
self.allocate(self._new_pack)
1668
self._new_pack = None
1669
if not self.autopack():
1670
# when autopack takes no steps, the names list is still
1672
self._save_pack_names()
1674
self._new_pack.abort()
1675
self._new_pack = None
1676
self.repo._text_knit = None
1679
class KnitPackRevisionStore(KnitRevisionStore):
1680
"""An object to adapt access from RevisionStore's to use KnitPacks.
1682
This class works by replacing the original RevisionStore.
1683
We need to do this because the KnitPackRevisionStore is less
1684
isolated in its layering - it uses services from the repo.
1687
def __init__(self, repo, transport, revisionstore):
1688
"""Create a KnitPackRevisionStore on repo with revisionstore.
1690
This will store its state in the Repository, use the
1691
indices to provide a KnitGraphIndex,
1692
and at the end of transactions write new indices.
1694
KnitRevisionStore.__init__(self, revisionstore.versioned_file_store)
1696
self._serializer = revisionstore._serializer
1697
self.transport = transport
1699
def get_revision_file(self, transaction):
1700
"""Get the revision versioned file object."""
1701
if getattr(self.repo, '_revision_knit', None) is not None:
1702
return self.repo._revision_knit
1703
self.repo._pack_collection.ensure_loaded()
1704
add_callback = self.repo._pack_collection.revision_index.add_callback
1705
# setup knit specific objects
1706
knit_index = KnitGraphIndex(
1707
self.repo._pack_collection.revision_index.combined_index,
1708
add_callback=add_callback)
1709
self.repo._revision_knit = knit.KnitVersionedFile(
1710
'revisions', self.transport.clone('..'),
1711
self.repo.control_files._file_mode,
1713
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1714
access_method=self.repo._pack_collection.revision_index.knit_access)
1715
return self.repo._revision_knit
1717
def get_signature_file(self, transaction):
1718
"""Get the signature versioned file object."""
1719
if getattr(self.repo, '_signature_knit', None) is not None:
1720
return self.repo._signature_knit
1721
self.repo._pack_collection.ensure_loaded()
1722
add_callback = self.repo._pack_collection.signature_index.add_callback
1723
# setup knit specific objects
1724
knit_index = KnitGraphIndex(
1725
self.repo._pack_collection.signature_index.combined_index,
1726
add_callback=add_callback, parents=False)
1727
self.repo._signature_knit = knit.KnitVersionedFile(
1728
'signatures', self.transport.clone('..'),
1729
self.repo.control_files._file_mode,
1731
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1732
access_method=self.repo._pack_collection.signature_index.knit_access)
1733
return self.repo._signature_knit
1736
class KnitPackTextStore(VersionedFileStore):
1737
"""Presents a TextStore abstraction on top of packs.
1739
This class works by replacing the original VersionedFileStore.
1740
We need to do this because the KnitPackRevisionStore is less
1741
isolated in its layering - it uses services from the repo and shares them
1742
with all the data written in a single write group.
1745
def __init__(self, repo, transport, weavestore):
1746
"""Create a KnitPackTextStore on repo with weavestore.
1748
This will store its state in the Repository, use the
1749
indices FileNames to provide a KnitGraphIndex,
1750
and at the end of transactions write new indices.
1752
# don't call base class constructor - it's not suitable.
1753
# no transient data stored in the transaction
1755
self._precious = False
1757
self.transport = transport
1758
self.weavestore = weavestore
1759
# XXX for check() which isn't updated yet
1760
self._transport = weavestore._transport
1762
def get_weave_or_empty(self, file_id, transaction):
1763
"""Get a 'Knit' backed by the .tix indices.
1765
The transaction parameter is ignored.
1767
self.repo._pack_collection.ensure_loaded()
1768
add_callback = self.repo._pack_collection.text_index.add_callback
1769
# setup knit specific objects
1770
file_id_index = GraphIndexPrefixAdapter(
1771
self.repo._pack_collection.text_index.combined_index,
1772
(file_id, ), 1, add_nodes_callback=add_callback)
1773
knit_index = KnitGraphIndex(file_id_index,
1774
add_callback=file_id_index.add_nodes,
1775
deltas=True, parents=True)
1776
return knit.KnitVersionedFile('text:' + file_id,
1777
self.transport.clone('..'),
1780
access_method=self.repo._pack_collection.text_index.knit_access,
1781
factory=knit.KnitPlainFactory())
1783
get_weave = get_weave_or_empty
1786
"""Generate a list of the fileids inserted, for use by check."""
1787
self.repo._pack_collection.ensure_loaded()
1789
for index, key, value, refs in \
1790
self.repo._pack_collection.text_index.combined_index.iter_all_entries():
1795
class InventoryKnitThunk(object):
1796
"""An object to manage thunking get_inventory_weave to pack based knits."""
1798
def __init__(self, repo, transport):
1799
"""Create an InventoryKnitThunk for repo at transport.
1801
This will store its state in the Repository, use the
1802
indices FileNames to provide a KnitGraphIndex,
1803
and at the end of transactions write a new index..
1806
self.transport = transport
1808
def get_weave(self):
1809
"""Get a 'Knit' that contains inventory data."""
1810
self.repo._pack_collection.ensure_loaded()
1811
add_callback = self.repo._pack_collection.inventory_index.add_callback
1812
# setup knit specific objects
1813
knit_index = KnitGraphIndex(
1814
self.repo._pack_collection.inventory_index.combined_index,
1815
add_callback=add_callback, deltas=True, parents=True)
1816
return knit.KnitVersionedFile(
1817
'inventory', self.transport.clone('..'),
1818
self.repo.control_files._file_mode,
1820
index=knit_index, delta=True, factory=knit.KnitPlainFactory(),
1821
access_method=self.repo._pack_collection.inventory_index.knit_access)
1824
class KnitPackRepository(KnitRepository):
1825
"""Experimental graph-knit using repository."""
1827
def __init__(self, _format, a_bzrdir, control_files, _revision_store,
1828
control_store, text_store, _commit_builder_class, _serializer):
1829
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
1830
_revision_store, control_store, text_store, _commit_builder_class,
1832
index_transport = control_files._transport.clone('indices')
1833
self._pack_collection = RepositoryPackCollection(self, control_files._transport,
1835
control_files._transport.clone('upload'),
1836
control_files._transport.clone('packs'))
1837
self._revision_store = KnitPackRevisionStore(self, index_transport, self._revision_store)
1838
self.weave_store = KnitPackTextStore(self, index_transport, self.weave_store)
1839
self._inv_thunk = InventoryKnitThunk(self, index_transport)
1840
# True when the repository object is 'write locked' (as opposed to the
1841
# physical lock only taken out around changes to the pack-names list.)
1842
# Another way to represent this would be a decorator around the control
1843
# files object that presents logical locks as physical ones - if this
1844
# gets ugly consider that alternative design. RBC 20071011
1845
self._write_lock_count = 0
1846
self._transaction = None
1848
self._reconcile_does_inventory_gc = True
1849
self._reconcile_fixes_text_parents = True
1850
self._reconcile_backsup_inventory = False
1852
def _abort_write_group(self):
1853
self._pack_collection._abort_write_group()
1855
def _find_inconsistent_revision_parents(self):
1856
"""Find revisions with incorrectly cached parents.
1858
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1859
parents-in-revision).
1861
if not self.is_locked():
1862
raise errors.ObjectNotLocked(self)
1863
pb = ui.ui_factory.nested_progress_bar()
1866
revision_nodes = self._pack_collection.revision_index \
1867
.combined_index.iter_all_entries()
1868
index_positions = []
1869
# Get the cached index values for all revisions, and also the location
1870
# in each index of the revision text so we can perform linear IO.
1871
for index, key, value, refs in revision_nodes:
1872
pos, length = value[1:].split(' ')
1873
index_positions.append((index, int(pos), key[0],
1874
tuple(parent[0] for parent in refs[0])))
1875
pb.update("Reading revision index.", 0, 0)
1876
index_positions.sort()
1877
batch_count = len(index_positions) / 1000 + 1
1878
pb.update("Checking cached revision graph.", 0, batch_count)
1879
for offset in xrange(batch_count):
1880
pb.update("Checking cached revision graph.", offset)
1881
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1884
rev_ids = [item[2] for item in to_query]
1885
revs = self.get_revisions(rev_ids)
1886
for revision, item in zip(revs, to_query):
1887
index_parents = item[3]
1888
rev_parents = tuple(revision.parent_ids)
1889
if index_parents != rev_parents:
1890
result.append((revision.revision_id, index_parents, rev_parents))
1895
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
1896
def get_parents(self, revision_ids):
1897
"""See graph._StackedParentsProvider.get_parents."""
1898
parent_map = self.get_parent_map(revision_ids)
1899
return [parent_map.get(r, None) for r in revision_ids]
1901
def get_parent_map(self, keys):
1902
"""See graph._StackedParentsProvider.get_parent_map
1904
This implementation accesses the combined revision index to provide
1907
self._pack_collection.ensure_loaded()
1908
index = self._pack_collection.revision_index.combined_index
1910
if _mod_revision.NULL_REVISION in keys:
1911
keys.discard(_mod_revision.NULL_REVISION)
1912
found_parents = {_mod_revision.NULL_REVISION:()}
1915
search_keys = set((revision_id,) for revision_id in keys)
1916
for index, key, value, refs in index.iter_entries(search_keys):
1919
parents = (_mod_revision.NULL_REVISION,)
1921
parents = tuple(parent[0] for parent in parents)
1922
found_parents[key[0]] = parents
1923
return found_parents
1925
@symbol_versioning.deprecated_method(symbol_versioning.one_four)
1927
def get_revision_graph(self, revision_id=None):
1928
"""Return a dictionary containing the revision graph.
1930
:param revision_id: The revision_id to get a graph from. If None, then
1931
the entire revision graph is returned. This is a deprecated mode of
1932
operation and will be removed in the future.
1933
:return: a dictionary of revision_id->revision_parents_list.
1935
if 'evil' in debug.debug_flags:
1937
"get_revision_graph scales with size of history.")
1938
# special case NULL_REVISION
1939
if revision_id == _mod_revision.NULL_REVISION:
1941
if revision_id is None:
1942
revision_vf = self._get_revision_vf()
1943
return revision_vf.get_graph()
1944
g = self.get_graph()
1945
first = g.get_parent_map([revision_id])
1946
if revision_id not in first:
1947
raise errors.NoSuchRevision(self, revision_id)
1951
NULL_REVISION = _mod_revision.NULL_REVISION
1952
ghosts = set([NULL_REVISION])
1953
for rev_id, parent_ids in g.iter_ancestry([revision_id]):
1954
if parent_ids is None: # This is a ghost
1957
ancestry[rev_id] = parent_ids
1958
for p in parent_ids:
1960
children[p].append(rev_id)
1962
children[p] = [rev_id]
1964
if NULL_REVISION in ancestry:
1965
del ancestry[NULL_REVISION]
1967
# Find all nodes that reference a ghost, and filter the ghosts out
1968
# of their parent lists. To preserve the order of parents, and
1969
# avoid double filtering nodes, we just find all children first,
1971
children_of_ghosts = set()
1972
for ghost in ghosts:
1973
children_of_ghosts.update(children[ghost])
1975
for child in children_of_ghosts:
1976
ancestry[child] = tuple(p for p in ancestry[child]
1980
def has_revisions(self, revision_ids):
1981
"""See Repository.has_revisions()."""
1982
revision_ids = set(revision_ids)
1983
result = revision_ids.intersection(
1984
set([None, _mod_revision.NULL_REVISION]))
1985
revision_ids.difference_update(result)
1986
index = self._pack_collection.revision_index.combined_index
1987
keys = [(revision_id,) for revision_id in revision_ids]
1988
result.update(node[1][0] for node in index.iter_entries(keys))
1991
def _make_parents_provider(self):
1992
return graph.CachingParentsProvider(self)
1994
def _refresh_data(self):
1995
if self._write_lock_count == 1 or (
1996
self.control_files._lock_count == 1 and
1997
self.control_files._lock_mode == 'r'):
1998
# forget what names there are
1999
self._pack_collection.reset()
2000
# XXX: Better to do an in-memory merge when acquiring a new lock -
2001
# factor out code from _save_pack_names.
2002
self._pack_collection.ensure_loaded()
2004
def _start_write_group(self):
2005
self._pack_collection._start_write_group()
2007
def _commit_write_group(self):
2008
return self._pack_collection._commit_write_group()
2010
def get_inventory_weave(self):
2011
return self._inv_thunk.get_weave()
2013
def get_transaction(self):
2014
if self._write_lock_count:
2015
return self._transaction
2017
return self.control_files.get_transaction()
2019
def is_locked(self):
2020
return self._write_lock_count or self.control_files.is_locked()
2022
def is_write_locked(self):
2023
return self._write_lock_count
2025
def lock_write(self, token=None):
2026
if not self._write_lock_count and self.is_locked():
2027
raise errors.ReadOnlyError(self)
2028
self._write_lock_count += 1
2029
if self._write_lock_count == 1:
2030
from bzrlib import transactions
2031
self._transaction = transactions.WriteTransaction()
2032
self._refresh_data()
2034
def lock_read(self):
2035
if self._write_lock_count:
2036
self._write_lock_count += 1
2038
self.control_files.lock_read()
2039
self._refresh_data()
2041
def leave_lock_in_place(self):
2042
# not supported - raise an error
2043
raise NotImplementedError(self.leave_lock_in_place)
2045
def dont_leave_lock_in_place(self):
2046
# not supported - raise an error
2047
raise NotImplementedError(self.dont_leave_lock_in_place)
2051
"""Compress the data within the repository.
2053
This will pack all the data to a single pack. In future it may
2054
recompress deltas or do other such expensive operations.
2056
self._pack_collection.pack()
2059
def reconcile(self, other=None, thorough=False):
2060
"""Reconcile this repository."""
2061
from bzrlib.reconcile import PackReconciler
2062
reconciler = PackReconciler(self, thorough=thorough)
2063
reconciler.reconcile()
2067
if self._write_lock_count == 1 and self._write_group is not None:
2068
self.abort_write_group()
2069
self._transaction = None
2070
self._write_lock_count = 0
2071
raise errors.BzrError(
2072
'Must end write group before releasing write lock on %s'
2074
if self._write_lock_count:
2075
self._write_lock_count -= 1
2076
if not self._write_lock_count:
2077
transaction = self._transaction
2078
self._transaction = None
2079
transaction.finish()
2081
self.control_files.unlock()
2084
class RepositoryFormatPack(MetaDirRepositoryFormat):
2085
"""Format logic for pack structured repositories.
2087
This repository format has:
2088
- a list of packs in pack-names
2089
- packs in packs/NAME.pack
2090
- indices in indices/NAME.{iix,six,tix,rix}
2091
- knit deltas in the packs, knit indices mapped to the indices.
2092
- thunk objects to support the knits programming API.
2093
- a format marker of its own
2094
- an optional 'shared-storage' flag
2095
- an optional 'no-working-trees' flag
2099
# Set this attribute in derived classes to control the repository class
2100
# created by open and initialize.
2101
repository_class = None
2102
# Set this attribute in derived classes to control the
2103
# _commit_builder_class that the repository objects will have passed to
2104
# their constructor.
2105
_commit_builder_class = None
2106
# Set this attribute in derived clases to control the _serializer that the
2107
# repository objects will have passed to their constructor.
2109
# External references are not supported in pack repositories yet.
2110
supports_external_lookups = False
2112
def _get_control_store(self, repo_transport, control_files):
2113
"""Return the control store for this repository."""
2114
return VersionedFileStore(
2117
file_mode=control_files._file_mode,
2118
versionedfile_class=knit.make_file_knit,
2119
versionedfile_kwargs={'factory': knit.KnitPlainFactory()},
2122
def _get_revision_store(self, repo_transport, control_files):
2123
"""See RepositoryFormat._get_revision_store()."""
2124
versioned_file_store = VersionedFileStore(
2126
file_mode=control_files._file_mode,
2129
versionedfile_class=knit.make_file_knit,
2130
versionedfile_kwargs={'delta': False,
2131
'factory': knit.KnitPlainFactory(),
2135
return KnitRevisionStore(versioned_file_store)
2137
def _get_text_store(self, transport, control_files):
2138
"""See RepositoryFormat._get_text_store()."""
2139
return self._get_versioned_file_store('knits',
2142
versionedfile_class=knit.make_file_knit,
2143
versionedfile_kwargs={
2144
'create_parent_dir': True,
2145
'delay_create': True,
2146
'dir_mode': control_files._dir_mode,
2150
def initialize(self, a_bzrdir, shared=False):
2151
"""Create a pack based repository.
2153
:param a_bzrdir: bzrdir to contain the new repository; must already
2155
:param shared: If true the repository will be initialized as a shared
2158
mutter('creating repository in %s.', a_bzrdir.transport.base)
2159
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
2160
builder = GraphIndexBuilder()
2161
files = [('pack-names', builder.finish())]
2162
utf8_files = [('format', self.get_format_string())]
2164
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
2165
return self.open(a_bzrdir=a_bzrdir, _found=True)
2167
def open(self, a_bzrdir, _found=False, _override_transport=None):
2168
"""See RepositoryFormat.open().
2170
:param _override_transport: INTERNAL USE ONLY. Allows opening the
2171
repository at a slightly different url
2172
than normal. I.e. during 'upgrade'.
2175
format = RepositoryFormat.find_format(a_bzrdir)
2176
if _override_transport is not None:
2177
repo_transport = _override_transport
2179
repo_transport = a_bzrdir.get_repository_transport(None)
2180
control_files = lockable_files.LockableFiles(repo_transport,
2181
'lock', lockdir.LockDir)
2182
text_store = self._get_text_store(repo_transport, control_files)
2183
control_store = self._get_control_store(repo_transport, control_files)
2184
_revision_store = self._get_revision_store(repo_transport, control_files)
2185
return self.repository_class(_format=self,
2187
control_files=control_files,
2188
_revision_store=_revision_store,
2189
control_store=control_store,
2190
text_store=text_store,
2191
_commit_builder_class=self._commit_builder_class,
2192
_serializer=self._serializer)
2195
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2196
"""A no-subtrees parameterized Pack repository.
2198
This format was introduced in 0.92.
2201
repository_class = KnitPackRepository
2202
_commit_builder_class = PackCommitBuilder
2203
_serializer = xml5.serializer_v5
2205
def _get_matching_bzrdir(self):
2206
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2208
def _ignore_setting_bzrdir(self, format):
2211
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2213
def get_format_string(self):
2214
"""See RepositoryFormat.get_format_string()."""
2215
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2217
def get_format_description(self):
2218
"""See RepositoryFormat.get_format_description()."""
2219
return "Packs containing knits without subtree support"
2221
def check_conversion_target(self, target_format):
2225
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2226
"""A subtrees parameterized Pack repository.
2228
This repository format uses the xml7 serializer to get:
2229
- support for recording full info about the tree root
2230
- support for recording tree-references
2232
This format was introduced in 0.92.
2235
repository_class = KnitPackRepository
2236
_commit_builder_class = PackRootCommitBuilder
2237
rich_root_data = True
2238
supports_tree_reference = True
2239
_serializer = xml7.serializer_v7
2241
def _get_matching_bzrdir(self):
2242
return bzrdir.format_registry.make_bzrdir(
2243
'pack-0.92-subtree')
2245
def _ignore_setting_bzrdir(self, format):
2248
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2250
def check_conversion_target(self, target_format):
2251
if not target_format.rich_root_data:
2252
raise errors.BadConversionTarget(
2253
'Does not support rich root data.', target_format)
2254
if not getattr(target_format, 'supports_tree_reference', False):
2255
raise errors.BadConversionTarget(
2256
'Does not support nested trees', target_format)
2258
def get_format_string(self):
2259
"""See RepositoryFormat.get_format_string()."""
2260
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2262
def get_format_description(self):
2263
"""See RepositoryFormat.get_format_description()."""
2264
return "Packs containing knits with subtree support\n"
2267
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2268
"""A rich-root, no subtrees parameterized Pack repository.
2270
This repository format uses the xml6 serializer to get:
2271
- support for recording full info about the tree root
2273
This format was introduced in 1.0.
2276
repository_class = KnitPackRepository
2277
_commit_builder_class = PackRootCommitBuilder
2278
rich_root_data = True
2279
supports_tree_reference = False
2280
_serializer = xml6.serializer_v6
2282
def _get_matching_bzrdir(self):
2283
return bzrdir.format_registry.make_bzrdir(
2286
def _ignore_setting_bzrdir(self, format):
2289
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2291
def check_conversion_target(self, target_format):
2292
if not target_format.rich_root_data:
2293
raise errors.BadConversionTarget(
2294
'Does not support rich root data.', target_format)
2296
def get_format_string(self):
2297
"""See RepositoryFormat.get_format_string()."""
2298
return ("Bazaar pack repository format 1 with rich root"
2299
" (needs bzr 1.0)\n")
2301
def get_format_description(self):
2302
"""See RepositoryFormat.get_format_description()."""
2303
return "Packs containing knits with rich root support\n"
2306
class RepositoryFormatPackDevelopment0(RepositoryFormatPack):
2307
"""A no-subtrees development repository.
2309
This format should be retained until the second release after bzr 1.0.
2311
No changes to the disk behaviour from pack-0.92.
2314
repository_class = KnitPackRepository
2315
_commit_builder_class = PackCommitBuilder
2316
_serializer = xml5.serializer_v5
2318
def _get_matching_bzrdir(self):
2319
return bzrdir.format_registry.make_bzrdir('development0')
2321
def _ignore_setting_bzrdir(self, format):
2324
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2326
def get_format_string(self):
2327
"""See RepositoryFormat.get_format_string()."""
2328
return "Bazaar development format 0 (needs bzr.dev from before 1.3)\n"
2330
def get_format_description(self):
2331
"""See RepositoryFormat.get_format_description()."""
2332
return ("Development repository format, currently the same as "
2335
def check_conversion_target(self, target_format):
2339
class RepositoryFormatPackDevelopment0Subtree(RepositoryFormatPack):
2340
"""A subtrees development repository.
2342
This format should be retained until the second release after bzr 1.0.
2344
No changes to the disk behaviour from pack-0.92-subtree.
2347
repository_class = KnitPackRepository
2348
_commit_builder_class = PackRootCommitBuilder
2349
rich_root_data = True
2350
supports_tree_reference = True
2351
_serializer = xml7.serializer_v7
2353
def _get_matching_bzrdir(self):
2354
return bzrdir.format_registry.make_bzrdir(
2355
'development0-subtree')
2357
def _ignore_setting_bzrdir(self, format):
2360
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2362
def check_conversion_target(self, target_format):
2363
if not target_format.rich_root_data:
2364
raise errors.BadConversionTarget(
2365
'Does not support rich root data.', target_format)
2366
if not getattr(target_format, 'supports_tree_reference', False):
2367
raise errors.BadConversionTarget(
2368
'Does not support nested trees', target_format)
2370
def get_format_string(self):
2371
"""See RepositoryFormat.get_format_string()."""
2372
return ("Bazaar development format 0 with subtree support "
2373
"(needs bzr.dev from before 1.3)\n")
2375
def get_format_description(self):
2376
"""See RepositoryFormat.get_format_description()."""
2377
return ("Development repository format, currently the same as "
2378
"pack-0.92-subtree\n")
added
removed
bzrlib/repofmt/pack_repo.py
