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
33
from bzrlib.index import (
38
GraphIndexPrefixAdapter,
40
from bzrlib.knit import KnitGraphIndex, _PackAccess, _KnitData
41
from bzrlib.osutils import rand_chars
42
from bzrlib.pack import ContainerWriter
43
from bzrlib.store import revision
44
from bzrlib import tsort
57
from bzrlib.decorators import needs_read_lock, needs_write_lock
58
from bzrlib.repofmt.knitrepo import KnitRepository
59
from bzrlib.repository import (
62
MetaDirRepositoryFormat,
65
import bzrlib.revision as _mod_revision
66
from bzrlib.store.revision.knit import KnitRevisionStore
67
from bzrlib.store.versioned import VersionedFileStore
68
from bzrlib.trace import mutter, note, warning
71
class PackCommitBuilder(CommitBuilder):
72
"""A subclass of CommitBuilder to add texts with pack semantics.
74
Specifically this uses one knit object rather than one knit object per
75
added text, reducing memory and object pressure.
78
def __init__(self, repository, parents, config, timestamp=None,
79
timezone=None, committer=None, revprops=None,
81
CommitBuilder.__init__(self, repository, parents, config,
82
timestamp=timestamp, timezone=timezone, committer=committer,
83
revprops=revprops, revision_id=revision_id)
84
self._file_graph = graph.Graph(
85
repository._pack_collection.text_index.combined_index)
87
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
88
return self.repository._pack_collection._add_text_to_weave(file_id,
89
self._new_revision_id, new_lines, parents, nostore_sha,
92
def _heads(self, file_id, revision_ids):
93
keys = [(file_id, revision_id) for revision_id in revision_ids]
94
return set([key[1] for key in self._file_graph.heads(keys)])
97
class PackRootCommitBuilder(RootCommitBuilder):
98
"""A subclass of RootCommitBuilder to add texts with pack semantics.
100
Specifically this uses one knit object rather than one knit object per
101
added text, reducing memory and object pressure.
104
def __init__(self, repository, parents, config, timestamp=None,
105
timezone=None, committer=None, revprops=None,
107
CommitBuilder.__init__(self, repository, parents, config,
108
timestamp=timestamp, timezone=timezone, committer=committer,
109
revprops=revprops, revision_id=revision_id)
110
self._file_graph = graph.Graph(
111
repository._pack_collection.text_index.combined_index)
113
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
114
return self.repository._pack_collection._add_text_to_weave(file_id,
115
self._new_revision_id, new_lines, parents, nostore_sha,
118
def _heads(self, file_id, revision_ids):
119
keys = [(file_id, revision_id) for revision_id in revision_ids]
120
return set([key[1] for key in self._file_graph.heads(keys)])
124
"""An in memory proxy for a pack and its indices.
126
This is a base class that is not directly used, instead the classes
127
ExistingPack and NewPack are used.
130
def __init__(self, revision_index, inventory_index, text_index,
132
"""Create a pack instance.
134
:param revision_index: A GraphIndex for determining what revisions are
135
present in the Pack and accessing the locations of their texts.
136
:param inventory_index: A GraphIndex for determining what inventories are
137
present in the Pack and accessing the locations of their
139
:param text_index: A GraphIndex for determining what file texts
140
are present in the pack and accessing the locations of their
141
texts/deltas (via (fileid, revisionid) tuples).
142
:param revision_index: A GraphIndex for determining what signatures are
143
present in the Pack and accessing the locations of their texts.
145
self.revision_index = revision_index
146
self.inventory_index = inventory_index
147
self.text_index = text_index
148
self.signature_index = signature_index
150
def access_tuple(self):
151
"""Return a tuple (transport, name) for the pack content."""
152
return self.pack_transport, self.file_name()
155
"""Get the file name for the pack on disk."""
156
return self.name + '.pack'
158
def get_revision_count(self):
159
return self.revision_index.key_count()
161
def inventory_index_name(self, name):
162
"""The inv index is the name + .iix."""
163
return self.index_name('inventory', name)
165
def revision_index_name(self, name):
166
"""The revision index is the name + .rix."""
167
return self.index_name('revision', name)
169
def signature_index_name(self, name):
170
"""The signature index is the name + .six."""
171
return self.index_name('signature', name)
173
def text_index_name(self, name):
174
"""The text index is the name + .tix."""
175
return self.index_name('text', name)
177
def _external_compression_parents_of_texts(self):
180
for node in self.text_index.iter_all_entries():
182
refs.update(node[3][1])
186
class ExistingPack(Pack):
187
"""An in memory proxy for an existing .pack and its disk indices."""
189
def __init__(self, pack_transport, name, revision_index, inventory_index,
190
text_index, signature_index):
191
"""Create an ExistingPack object.
193
:param pack_transport: The transport where the pack file resides.
194
:param name: The name of the pack on disk in the pack_transport.
196
Pack.__init__(self, revision_index, inventory_index, text_index,
199
self.pack_transport = pack_transport
200
assert None not in (revision_index, inventory_index, text_index,
201
signature_index, name, pack_transport)
203
def __eq__(self, other):
204
return self.__dict__ == other.__dict__
206
def __ne__(self, other):
207
return not self.__eq__(other)
210
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
211
id(self), self.transport, self.name)
215
"""An in memory proxy for a pack which is being created."""
217
# A map of index 'type' to the file extension and position in the
219
index_definitions = {
220
'revision': ('.rix', 0),
221
'inventory': ('.iix', 1),
223
'signature': ('.six', 3),
226
def __init__(self, upload_transport, index_transport, pack_transport,
227
upload_suffix='', file_mode=None):
228
"""Create a NewPack instance.
230
:param upload_transport: A writable transport for the pack to be
231
incrementally uploaded to.
232
:param index_transport: A writable transport for the pack's indices to
233
be written to when the pack is finished.
234
:param pack_transport: A writable transport for the pack to be renamed
235
to when the upload is complete. This *must* be the same as
236
upload_transport.clone('../packs').
237
:param upload_suffix: An optional suffix to be given to any temporary
238
files created during the pack creation. e.g '.autopack'
239
:param file_mode: An optional file mode to create the new files with.
241
# The relative locations of the packs are constrained, but all are
242
# passed in because the caller has them, so as to avoid object churn.
244
# Revisions: parents list, no text compression.
245
InMemoryGraphIndex(reference_lists=1),
246
# Inventory: We want to map compression only, but currently the
247
# knit code hasn't been updated enough to understand that, so we
248
# have a regular 2-list index giving parents and compression
250
InMemoryGraphIndex(reference_lists=2),
251
# Texts: compression and per file graph, for all fileids - so two
252
# reference lists and two elements in the key tuple.
253
InMemoryGraphIndex(reference_lists=2, key_elements=2),
254
# Signatures: Just blobs to store, no compression, no parents
256
InMemoryGraphIndex(reference_lists=0),
258
# where should the new pack be opened
259
self.upload_transport = upload_transport
260
# where are indices written out to
261
self.index_transport = index_transport
262
# where is the pack renamed to when it is finished?
263
self.pack_transport = pack_transport
264
# What file mode to upload the pack and indices with.
265
self._file_mode = file_mode
266
# tracks the content written to the .pack file.
267
self._hash = md5.new()
268
# a four-tuple with the length in bytes of the indices, once the pack
269
# is finalised. (rev, inv, text, sigs)
270
self.index_sizes = None
271
# How much data to cache when writing packs. Note that this is not
272
# synchronised with reads, because it's not in the transport layer, so
273
# is not safe unless the client knows it won't be reading from the pack
275
self._cache_limit = 0
276
# the temporary pack file name.
277
self.random_name = rand_chars(20) + upload_suffix
278
# when was this pack started ?
279
self.start_time = time.time()
280
# open an output stream for the data added to the pack.
281
self.write_stream = self.upload_transport.open_write_stream(
282
self.random_name, mode=self._file_mode)
283
if 'pack' in debug.debug_flags:
284
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
285
time.ctime(), self.upload_transport.base, self.random_name,
286
time.time() - self.start_time)
287
# A list of byte sequences to be written to the new pack, and the
288
# aggregate size of them. Stored as a list rather than separate
289
# variables so that the _write_data closure below can update them.
290
self._buffer = [[], 0]
291
# create a callable for adding data
293
# robertc says- this is a closure rather than a method on the object
294
# so that the variables are locals, and faster than accessing object
296
def _write_data(bytes, flush=False, _buffer=self._buffer,
297
_write=self.write_stream.write, _update=self._hash.update):
298
_buffer[0].append(bytes)
299
_buffer[1] += len(bytes)
301
if _buffer[1] > self._cache_limit or flush:
302
bytes = ''.join(_buffer[0])
306
# expose this on self, for the occasion when clients want to add data.
307
self._write_data = _write_data
308
# a pack writer object to serialise pack records.
309
self._writer = pack.ContainerWriter(self._write_data)
311
# what state is the pack in? (open, finished, aborted)
315
"""Cancel creating this pack."""
316
self._state = 'aborted'
317
self.write_stream.close()
318
# Remove the temporary pack file.
319
self.upload_transport.delete(self.random_name)
320
# The indices have no state on disk.
322
def access_tuple(self):
323
"""Return a tuple (transport, name) for the pack content."""
324
assert self._state in ('open', 'finished')
325
if self._state == 'finished':
326
return Pack.access_tuple(self)
328
return self.upload_transport, self.random_name
330
def data_inserted(self):
331
"""True if data has been added to this pack."""
332
return bool(self.get_revision_count() or
333
self.inventory_index.key_count() or
334
self.text_index.key_count() or
335
self.signature_index.key_count())
338
"""Finish the new pack.
341
- finalises the content
342
- assigns a name (the md5 of the content, currently)
343
- writes out the associated indices
344
- renames the pack into place.
345
- stores the index size tuple for the pack in the index_sizes
350
self._write_data('', flush=True)
351
self.name = self._hash.hexdigest()
353
# XXX: It'd be better to write them all to temporary names, then
354
# rename them all into place, so that the window when only some are
355
# visible is smaller. On the other hand none will be seen until
356
# they're in the names list.
357
self.index_sizes = [None, None, None, None]
358
self._write_index('revision', self.revision_index, 'revision')
359
self._write_index('inventory', self.inventory_index, 'inventory')
360
self._write_index('text', self.text_index, 'file texts')
361
self._write_index('signature', self.signature_index,
362
'revision signatures')
363
self.write_stream.close()
364
# Note that this will clobber an existing pack with the same name,
365
# without checking for hash collisions. While this is undesirable this
366
# is something that can be rectified in a subsequent release. One way
367
# to rectify it may be to leave the pack at the original name, writing
368
# its pack-names entry as something like 'HASH: index-sizes
369
# temporary-name'. Allocate that and check for collisions, if it is
370
# collision free then rename it into place. If clients know this scheme
371
# they can handle missing-file errors by:
372
# - try for HASH.pack
373
# - try for temporary-name
374
# - refresh the pack-list to see if the pack is now absent
375
self.upload_transport.rename(self.random_name,
376
'../packs/' + self.name + '.pack')
377
self._state = 'finished'
378
if 'pack' in debug.debug_flags:
379
# XXX: size might be interesting?
380
mutter('%s: create_pack: pack renamed into place: %s%s->%s%s t+%6.3fs',
381
time.ctime(), self.upload_transport.base, self.random_name,
382
self.pack_transport, self.name,
383
time.time() - self.start_time)
386
"""Flush any current data."""
388
bytes = ''.join(self._buffer[0])
389
self.write_stream.write(bytes)
390
self._hash.update(bytes)
391
self._buffer[:] = [[], 0]
393
def index_name(self, index_type, name):
394
"""Get the disk name of an index type for pack name 'name'."""
395
return name + NewPack.index_definitions[index_type][0]
397
def index_offset(self, index_type):
398
"""Get the position in a index_size array for a given index type."""
399
return NewPack.index_definitions[index_type][1]
401
def _replace_index_with_readonly(self, index_type):
402
setattr(self, index_type + '_index',
403
GraphIndex(self.index_transport,
404
self.index_name(index_type, self.name),
405
self.index_sizes[self.index_offset(index_type)]))
407
def set_write_cache_size(self, size):
408
self._cache_limit = size
410
def _write_index(self, index_type, index, label):
411
"""Write out an index.
413
:param index_type: The type of index to write - e.g. 'revision'.
414
:param index: The index object to serialise.
415
:param label: What label to give the index e.g. 'revision'.
417
index_name = self.index_name(index_type, self.name)
418
self.index_sizes[self.index_offset(index_type)] = \
419
self.index_transport.put_file(index_name, index.finish(),
420
mode=self._file_mode)
421
if 'pack' in debug.debug_flags:
422
# XXX: size might be interesting?
423
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
424
time.ctime(), label, self.upload_transport.base,
425
self.random_name, time.time() - self.start_time)
426
# Replace the writable index on this object with a readonly,
427
# presently unloaded index. We should alter
428
# the index layer to make its finish() error if add_node is
429
# subsequently used. RBC
430
self._replace_index_with_readonly(index_type)
433
class AggregateIndex(object):
434
"""An aggregated index for the RepositoryPackCollection.
436
AggregateIndex is reponsible for managing the PackAccess object,
437
Index-To-Pack mapping, and all indices list for a specific type of index
438
such as 'revision index'.
440
A CombinedIndex provides an index on a single key space built up
441
from several on-disk indices. The AggregateIndex builds on this
442
to provide a knit access layer, and allows having up to one writable
443
index within the collection.
445
# XXX: Probably 'can be written to' could/should be separated from 'acts
446
# like a knit index' -- mbp 20071024
449
"""Create an AggregateIndex."""
450
self.index_to_pack = {}
451
self.combined_index = CombinedGraphIndex([])
452
self.knit_access = _PackAccess(self.index_to_pack)
454
def replace_indices(self, index_to_pack, indices):
455
"""Replace the current mappings with fresh ones.
457
This should probably not be used eventually, rather incremental add and
458
removal of indices. It has been added during refactoring of existing
461
:param index_to_pack: A mapping from index objects to
462
(transport, name) tuples for the pack file data.
463
:param indices: A list of indices.
465
# refresh the revision pack map dict without replacing the instance.
466
self.index_to_pack.clear()
467
self.index_to_pack.update(index_to_pack)
468
# XXX: API break - clearly a 'replace' method would be good?
469
self.combined_index._indices[:] = indices
470
# the current add nodes callback for the current writable index if
472
self.add_callback = None
474
def add_index(self, index, pack):
475
"""Add index to the aggregate, which is an index for Pack pack.
477
Future searches on the aggregate index will seach this new index
478
before all previously inserted indices.
480
:param index: An Index for the pack.
481
:param pack: A Pack instance.
483
# expose it to the index map
484
self.index_to_pack[index] = pack.access_tuple()
485
# put it at the front of the linear index list
486
self.combined_index.insert_index(0, index)
488
def add_writable_index(self, index, pack):
489
"""Add an index which is able to have data added to it.
491
There can be at most one writable index at any time. Any
492
modifications made to the knit are put into this index.
494
:param index: An index from the pack parameter.
495
:param pack: A Pack instance.
497
assert self.add_callback is None, \
498
"%s already has a writable index through %s" % \
499
(self, self.add_callback)
500
# allow writing: queue writes to a new index
501
self.add_index(index, pack)
502
# Updates the index to packs mapping as a side effect,
503
self.knit_access.set_writer(pack._writer, index, pack.access_tuple())
504
self.add_callback = index.add_nodes
507
"""Reset all the aggregate data to nothing."""
508
self.knit_access.set_writer(None, None, (None, None))
509
self.index_to_pack.clear()
510
del self.combined_index._indices[:]
511
self.add_callback = None
513
def remove_index(self, index, pack):
514
"""Remove index from the indices used to answer queries.
516
:param index: An index from the pack parameter.
517
:param pack: A Pack instance.
519
del self.index_to_pack[index]
520
self.combined_index._indices.remove(index)
521
if (self.add_callback is not None and
522
getattr(index, 'add_nodes', None) == self.add_callback):
523
self.add_callback = None
524
self.knit_access.set_writer(None, None, (None, None))
527
class Packer(object):
528
"""Create a pack from packs."""
530
def __init__(self, pack_collection, packs, suffix, revision_ids=None):
533
:param pack_collection: A RepositoryPackCollection object where the
534
new pack is being written to.
535
:param packs: The packs to combine.
536
:param suffix: The suffix to use on the temporary files for the pack.
537
:param revision_ids: Revision ids to limit the pack to.
541
self.revision_ids = revision_ids
542
# The pack object we are creating.
544
self._pack_collection = pack_collection
545
# The index layer keys for the revisions being copied. None for 'all
547
self._revision_keys = None
548
# What text keys to copy. None for 'all texts'. This is set by
549
# _copy_inventory_texts
550
self._text_filter = None
553
def _extra_init(self):
554
"""A template hook to allow extending the constructor trivially."""
556
def pack(self, pb=None):
557
"""Create a new pack by reading data from other packs.
559
This does little more than a bulk copy of data. One key difference
560
is that data with the same item key across multiple packs is elided
561
from the output. The new pack is written into the current pack store
562
along with its indices, and the name added to the pack names. The
563
source packs are not altered and are not required to be in the current
566
:param pb: An optional progress bar to use. A nested bar is created if
568
:return: A Pack object, or None if nothing was copied.
570
# open a pack - using the same name as the last temporary file
571
# - which has already been flushed, so its safe.
572
# XXX: - duplicate code warning with start_write_group; fix before
573
# considering 'done'.
574
if self._pack_collection._new_pack is not None:
575
raise errors.BzrError('call to create_pack_from_packs while '
576
'another pack is being written.')
577
if self.revision_ids is not None:
578
if len(self.revision_ids) == 0:
579
# silly fetch request.
582
self.revision_ids = frozenset(self.revision_ids)
584
self.pb = ui.ui_factory.nested_progress_bar()
588
return self._create_pack_from_packs()
594
"""Open a pack for the pack we are creating."""
595
return NewPack(self._pack_collection._upload_transport,
596
self._pack_collection._index_transport,
597
self._pack_collection._pack_transport, upload_suffix=self.suffix,
598
file_mode=self._pack_collection.repo.control_files._file_mode)
600
def _copy_revision_texts(self):
601
"""Copy revision data to the new pack."""
603
if self.revision_ids:
604
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
607
# select revision keys
608
revision_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
609
self.packs, 'revision_index')[0]
610
revision_nodes = self._pack_collection._index_contents(revision_index_map, revision_keys)
611
# copy revision keys and adjust values
612
self.pb.update("Copying revision texts", 1)
613
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
614
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
615
self.new_pack.revision_index, readv_group_iter, total_items))
616
if 'pack' in debug.debug_flags:
617
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
618
time.ctime(), self._pack_collection._upload_transport.base,
619
self.new_pack.random_name,
620
self.new_pack.revision_index.key_count(),
621
time.time() - self.new_pack.start_time)
622
self._revision_keys = revision_keys
624
def _copy_inventory_texts(self):
625
"""Copy the inventory texts to the new pack.
627
self._revision_keys is used to determine what inventories to copy.
629
Sets self._text_filter appropriately.
631
# select inventory keys
632
inv_keys = self._revision_keys # currently the same keyspace, and note that
633
# querying for keys here could introduce a bug where an inventory item
634
# is missed, so do not change it to query separately without cross
635
# checking like the text key check below.
636
inventory_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
637
self.packs, 'inventory_index')[0]
638
inv_nodes = self._pack_collection._index_contents(inventory_index_map, inv_keys)
639
# copy inventory keys and adjust values
640
# XXX: Should be a helper function to allow different inv representation
642
self.pb.update("Copying inventory texts", 2)
643
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
644
# Only grab the output lines if we will be processing them
645
output_lines = bool(self.revision_ids)
646
inv_lines = self._copy_nodes_graph(inventory_index_map,
647
self.new_pack._writer, self.new_pack.inventory_index,
648
readv_group_iter, total_items, output_lines=output_lines)
649
if self.revision_ids:
650
self._process_inventory_lines(inv_lines)
652
# eat the iterator to cause it to execute.
654
self._text_filter = None
655
if 'pack' in debug.debug_flags:
656
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
657
time.ctime(), self._pack_collection._upload_transport.base,
658
self.new_pack.random_name,
659
self.new_pack.inventory_index.key_count(),
660
time.time() - self.new_pack.start_time)
662
def _copy_text_texts(self):
664
text_index_map, text_nodes = self._get_text_nodes()
665
if self._text_filter is not None:
666
# We could return the keys copied as part of the return value from
667
# _copy_nodes_graph but this doesn't work all that well with the
668
# need to get line output too, so we check separately, and as we're
669
# going to buffer everything anyway, we check beforehand, which
670
# saves reading knit data over the wire when we know there are
672
text_nodes = set(text_nodes)
673
present_text_keys = set(_node[1] for _node in text_nodes)
674
missing_text_keys = set(self._text_filter) - present_text_keys
675
if missing_text_keys:
676
# TODO: raise a specific error that can handle many missing
678
a_missing_key = missing_text_keys.pop()
679
raise errors.RevisionNotPresent(a_missing_key[1],
681
# copy text keys and adjust values
682
self.pb.update("Copying content texts", 3)
683
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
684
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
685
self.new_pack.text_index, readv_group_iter, total_items))
686
self._log_copied_texts()
688
def _check_references(self):
689
"""Make sure our external refereneces are present."""
690
external_refs = self.new_pack._external_compression_parents_of_texts()
692
index = self._pack_collection.text_index.combined_index
693
found_items = list(index.iter_entries(external_refs))
694
if len(found_items) != len(external_refs):
695
found_keys = set(k for idx, k, refs, value in found_items)
696
missing_items = external_refs - found_keys
697
missing_file_id, missing_revision_id = missing_items.pop()
698
raise errors.RevisionNotPresent(missing_revision_id,
701
def _create_pack_from_packs(self):
702
self.pb.update("Opening pack", 0, 5)
703
self.new_pack = self.open_pack()
704
new_pack = self.new_pack
705
# buffer data - we won't be reading-back during the pack creation and
706
# this makes a significant difference on sftp pushes.
707
new_pack.set_write_cache_size(1024*1024)
708
if 'pack' in debug.debug_flags:
709
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
710
for a_pack in self.packs]
711
if self.revision_ids is not None:
712
rev_count = len(self.revision_ids)
715
mutter('%s: create_pack: creating pack from source packs: '
716
'%s%s %s revisions wanted %s t=0',
717
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
718
plain_pack_list, rev_count)
719
self._copy_revision_texts()
720
self._copy_inventory_texts()
721
self._copy_text_texts()
722
# select signature keys
723
signature_filter = self._revision_keys # same keyspace
724
signature_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
725
self.packs, 'signature_index')[0]
726
signature_nodes = self._pack_collection._index_contents(signature_index_map,
728
# copy signature keys and adjust values
729
self.pb.update("Copying signature texts", 4)
730
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
731
new_pack.signature_index)
732
if 'pack' in debug.debug_flags:
733
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
734
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
735
new_pack.signature_index.key_count(),
736
time.time() - new_pack.start_time)
737
self._check_references()
738
if not self._use_pack(new_pack):
741
self.pb.update("Finishing pack", 5)
743
self._pack_collection.allocate(new_pack)
746
def _copy_nodes(self, nodes, index_map, writer, write_index):
747
"""Copy knit nodes between packs with no graph references."""
748
pb = ui.ui_factory.nested_progress_bar()
750
return self._do_copy_nodes(nodes, index_map, writer,
755
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb):
756
# for record verification
757
knit_data = _KnitData(None)
758
# plan a readv on each source pack:
760
nodes = sorted(nodes)
761
# how to map this into knit.py - or knit.py into this?
762
# we don't want the typical knit logic, we want grouping by pack
763
# at this point - perhaps a helper library for the following code
764
# duplication points?
766
for index, key, value in nodes:
767
if index not in request_groups:
768
request_groups[index] = []
769
request_groups[index].append((key, value))
771
pb.update("Copied record", record_index, len(nodes))
772
for index, items in request_groups.iteritems():
773
pack_readv_requests = []
774
for key, value in items:
775
# ---- KnitGraphIndex.get_position
776
bits = value[1:].split(' ')
777
offset, length = int(bits[0]), int(bits[1])
778
pack_readv_requests.append((offset, length, (key, value[0])))
779
# linear scan up the pack
780
pack_readv_requests.sort()
782
transport, path = index_map[index]
783
reader = pack.make_readv_reader(transport, path,
784
[offset[0:2] for offset in pack_readv_requests])
785
for (names, read_func), (_1, _2, (key, eol_flag)) in \
786
izip(reader.iter_records(), pack_readv_requests):
787
raw_data = read_func(None)
788
# check the header only
789
df, _ = knit_data._parse_record_header(key[-1], raw_data)
791
pos, size = writer.add_bytes_record(raw_data, names)
792
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
793
pb.update("Copied record", record_index)
796
def _copy_nodes_graph(self, index_map, writer, write_index,
797
readv_group_iter, total_items, output_lines=False):
798
"""Copy knit nodes between packs.
800
:param output_lines: Return lines present in the copied data as
801
an iterator of line,version_id.
803
pb = ui.ui_factory.nested_progress_bar()
805
for result in self._do_copy_nodes_graph(index_map, writer,
806
write_index, output_lines, pb, readv_group_iter, total_items):
809
# Python 2.4 does not permit try:finally: in a generator.
815
def _do_copy_nodes_graph(self, index_map, writer, write_index,
816
output_lines, pb, readv_group_iter, total_items):
817
# for record verification
818
knit_data = _KnitData(None)
819
# for line extraction when requested (inventories only)
821
factory = knit.KnitPlainFactory()
823
pb.update("Copied record", record_index, total_items)
824
for index, readv_vector, node_vector in readv_group_iter:
826
transport, path = index_map[index]
827
reader = pack.make_readv_reader(transport, path, readv_vector)
828
for (names, read_func), (key, eol_flag, references) in \
829
izip(reader.iter_records(), node_vector):
830
raw_data = read_func(None)
833
# read the entire thing
834
content, _ = knit_data._parse_record(version_id, raw_data)
835
if len(references[-1]) == 0:
836
line_iterator = factory.get_fulltext_content(content)
838
line_iterator = factory.get_linedelta_content(content)
839
for line in line_iterator:
840
yield line, version_id
842
# check the header only
843
df, _ = knit_data._parse_record_header(version_id, raw_data)
845
pos, size = writer.add_bytes_record(raw_data, names)
846
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
847
pb.update("Copied record", record_index)
850
def _get_text_nodes(self):
851
text_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
852
self.packs, 'text_index')[0]
853
return text_index_map, self._pack_collection._index_contents(text_index_map,
856
def _least_readv_node_readv(self, nodes):
857
"""Generate request groups for nodes using the least readv's.
859
:param nodes: An iterable of graph index nodes.
860
:return: Total node count and an iterator of the data needed to perform
861
readvs to obtain the data for nodes. Each item yielded by the
862
iterator is a tuple with:
863
index, readv_vector, node_vector. readv_vector is a list ready to
864
hand to the transport readv method, and node_vector is a list of
865
(key, eol_flag, references) for the the node retrieved by the
866
matching readv_vector.
868
# group by pack so we do one readv per pack
869
nodes = sorted(nodes)
872
for index, key, value, references in nodes:
873
if index not in request_groups:
874
request_groups[index] = []
875
request_groups[index].append((key, value, references))
877
for index, items in request_groups.iteritems():
878
pack_readv_requests = []
879
for key, value, references in items:
880
# ---- KnitGraphIndex.get_position
881
bits = value[1:].split(' ')
882
offset, length = int(bits[0]), int(bits[1])
883
pack_readv_requests.append(
884
((offset, length), (key, value[0], references)))
885
# linear scan up the pack to maximum range combining.
886
pack_readv_requests.sort()
887
# split out the readv and the node data.
888
pack_readv = [readv for readv, node in pack_readv_requests]
889
node_vector = [node for readv, node in pack_readv_requests]
890
result.append((index, pack_readv, node_vector))
893
def _log_copied_texts(self):
894
if 'pack' in debug.debug_flags:
895
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
896
time.ctime(), self._pack_collection._upload_transport.base,
897
self.new_pack.random_name,
898
self.new_pack.text_index.key_count(),
899
time.time() - self.new_pack.start_time)
901
def _process_inventory_lines(self, inv_lines):
902
"""Use up the inv_lines generator and setup a text key filter."""
903
repo = self._pack_collection.repo
904
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
905
inv_lines, self.revision_ids)
907
for fileid, file_revids in fileid_revisions.iteritems():
908
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
909
self._text_filter = text_filter
911
def _revision_node_readv(self, revision_nodes):
912
"""Return the total revisions and the readv's to issue.
914
:param revision_nodes: The revision index contents for the packs being
915
incorporated into the new pack.
916
:return: As per _least_readv_node_readv.
918
return self._least_readv_node_readv(revision_nodes)
920
def _use_pack(self, new_pack):
921
"""Return True if new_pack should be used.
923
:param new_pack: The pack that has just been created.
924
:return: True if the pack should be used.
926
return new_pack.data_inserted()
929
class OptimisingPacker(Packer):
930
"""A packer which spends more time to create better disk layouts."""
932
def _revision_node_readv(self, revision_nodes):
933
"""Return the total revisions and the readv's to issue.
935
This sort places revisions in topological order with the ancestors
938
:param revision_nodes: The revision index contents for the packs being
939
incorporated into the new pack.
940
:return: As per _least_readv_node_readv.
942
# build an ancestors dict
945
for index, key, value, references in revision_nodes:
946
ancestors[key] = references[0]
947
by_key[key] = (index, value, references)
948
order = tsort.topo_sort(ancestors)
950
# Single IO is pathological, but it will work as a starting point.
952
for key in reversed(order):
953
index, value, references = by_key[key]
954
# ---- KnitGraphIndex.get_position
955
bits = value[1:].split(' ')
956
offset, length = int(bits[0]), int(bits[1])
958
(index, [(offset, length)], [(key, value[0], references)]))
959
# TODO: combine requests in the same index that are in ascending order.
960
return total, requests
963
class ReconcilePacker(Packer):
964
"""A packer which regenerates indices etc as it copies.
966
This is used by ``bzr reconcile`` to cause parent text pointers to be
970
def _extra_init(self):
971
self._data_changed = False
973
def _process_inventory_lines(self, inv_lines):
974
"""Generate a text key reference map rather for reconciling with."""
975
repo = self._pack_collection.repo
976
refs = repo._find_text_key_references_from_xml_inventory_lines(
978
self._text_refs = refs
979
# during reconcile we:
980
# - convert unreferenced texts to full texts
981
# - correct texts which reference a text not copied to be full texts
982
# - copy all others as-is but with corrected parents.
983
# - so at this point we don't know enough to decide what becomes a full
985
self._text_filter = None
987
def _copy_text_texts(self):
988
"""generate what texts we should have and then copy."""
989
self.pb.update("Copying content texts", 3)
990
# we have three major tasks here:
991
# 1) generate the ideal index
992
repo = self._pack_collection.repo
993
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
995
self.new_pack.revision_index.iter_all_entries()])
996
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
997
# 2) generate a text_nodes list that contains all the deltas that can
998
# be used as-is, with corrected parents.
1001
discarded_nodes = []
1002
NULL_REVISION = _mod_revision.NULL_REVISION
1003
text_index_map, text_nodes = self._get_text_nodes()
1004
for node in text_nodes:
1010
ideal_parents = tuple(ideal_index[node[1]])
1012
discarded_nodes.append(node)
1013
self._data_changed = True
1015
if ideal_parents == (NULL_REVISION,):
1017
if ideal_parents == node[3][0]:
1019
ok_nodes.append(node)
1020
elif ideal_parents[0:1] == node[3][0][0:1]:
1021
# the left most parent is the same, or there are no parents
1022
# today. Either way, we can preserve the representation as
1023
# long as we change the refs to be inserted.
1024
self._data_changed = True
1025
ok_nodes.append((node[0], node[1], node[2],
1026
(ideal_parents, node[3][1])))
1027
self._data_changed = True
1029
# Reinsert this text completely
1030
bad_texts.append((node[1], ideal_parents))
1031
self._data_changed = True
1032
# we're finished with some data.
1035
# 3) bulk copy the ok data
1036
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1037
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1038
self.new_pack.text_index, readv_group_iter, total_items))
1039
# 4) adhoc copy all the other texts.
1040
# We have to topologically insert all texts otherwise we can fail to
1041
# reconcile when parts of a single delta chain are preserved intact,
1042
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1043
# reinserted, and if d3 has incorrect parents it will also be
1044
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1045
# copied), so we will try to delta, but d2 is not currently able to be
1046
# extracted because it's basis d1 is not present. Topologically sorting
1047
# addresses this. The following generates a sort for all the texts that
1048
# are being inserted without having to reference the entire text key
1049
# space (we only topo sort the revisions, which is smaller).
1050
topo_order = tsort.topo_sort(ancestors)
1051
rev_order = dict(zip(topo_order, range(len(topo_order))))
1052
bad_texts.sort(key=lambda key:rev_order[key[0][1]])
1053
transaction = repo.get_transaction()
1054
file_id_index = GraphIndexPrefixAdapter(
1055
self.new_pack.text_index,
1057
add_nodes_callback=self.new_pack.text_index.add_nodes)
1058
knit_index = KnitGraphIndex(file_id_index,
1059
add_callback=file_id_index.add_nodes,
1060
deltas=True, parents=True)
1061
output_knit = knit.KnitVersionedFile('reconcile-texts',
1062
self._pack_collection.transport,
1065
access_method=_PackAccess(
1066
{self.new_pack.text_index:self.new_pack.access_tuple()},
1067
(self.new_pack._writer, self.new_pack.text_index)),
1068
factory=knit.KnitPlainFactory())
1069
for key, parent_keys in bad_texts:
1070
# We refer to the new pack to delta data being output.
1071
# A possible improvement would be to catch errors on short reads
1072
# and only flush then.
1073
self.new_pack.flush()
1075
for parent_key in parent_keys:
1076
if parent_key[0] != key[0]:
1077
# Graph parents must match the fileid
1078
raise errors.BzrError('Mismatched key parent %r:%r' %
1080
parents.append(parent_key[1])
1081
source_weave = repo.weave_store.get_weave(key[0], transaction)
1082
text_lines = source_weave.get_lines(key[1])
1083
# adapt the 'knit' to the current file_id.
1084
file_id_index = GraphIndexPrefixAdapter(
1085
self.new_pack.text_index,
1087
add_nodes_callback=self.new_pack.text_index.add_nodes)
1088
knit_index._graph_index = file_id_index
1089
knit_index._add_callback = file_id_index.add_nodes
1090
output_knit.add_lines_with_ghosts(
1091
key[1], parents, text_lines, random_id=True, check_content=False)
1092
# 5) check that nothing inserted has a reference outside the keyspace.
1093
missing_text_keys = self.new_pack._external_compression_parents_of_texts()
1094
if missing_text_keys:
1095
raise errors.BzrError('Reference to missing compression parents %r'
1097
self._log_copied_texts()
1099
def _use_pack(self, new_pack):
1100
"""Override _use_pack to check for reconcile having changed content."""
1101
# XXX: we might be better checking this at the copy time.
1102
original_inventory_keys = set()
1103
inv_index = self._pack_collection.inventory_index.combined_index
1104
for entry in inv_index.iter_all_entries():
1105
original_inventory_keys.add(entry[1])
1106
new_inventory_keys = set()
1107
for entry in new_pack.inventory_index.iter_all_entries():
1108
new_inventory_keys.add(entry[1])
1109
if new_inventory_keys != original_inventory_keys:
1110
self._data_changed = True
1111
return new_pack.data_inserted() and self._data_changed
1114
class RepositoryPackCollection(object):
1115
"""Management of packs within a repository."""
1117
def __init__(self, repo, transport, index_transport, upload_transport,
1119
"""Create a new RepositoryPackCollection.
1121
:param transport: Addresses the repository base directory
1122
(typically .bzr/repository/).
1123
:param index_transport: Addresses the directory containing indices.
1124
:param upload_transport: Addresses the directory into which packs are written
1125
while they're being created.
1126
:param pack_transport: Addresses the directory of existing complete packs.
1129
self.transport = transport
1130
self._index_transport = index_transport
1131
self._upload_transport = upload_transport
1132
self._pack_transport = pack_transport
1133
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
1136
self._packs_by_name = {}
1137
# the previous pack-names content
1138
self._packs_at_load = None
1139
# when a pack is being created by this object, the state of that pack.
1140
self._new_pack = None
1141
# aggregated revision index data
1142
self.revision_index = AggregateIndex()
1143
self.inventory_index = AggregateIndex()
1144
self.text_index = AggregateIndex()
1145
self.signature_index = AggregateIndex()
1147
def add_pack_to_memory(self, pack):
1148
"""Make a Pack object available to the repository to satisfy queries.
1150
:param pack: A Pack object.
1152
assert pack.name not in self._packs_by_name
1153
self.packs.append(pack)
1154
self._packs_by_name[pack.name] = pack
1155
self.revision_index.add_index(pack.revision_index, pack)
1156
self.inventory_index.add_index(pack.inventory_index, pack)
1157
self.text_index.add_index(pack.text_index, pack)
1158
self.signature_index.add_index(pack.signature_index, pack)
1160
def _add_text_to_weave(self, file_id, revision_id, new_lines, parents,
1161
nostore_sha, random_revid):
1162
file_id_index = GraphIndexPrefixAdapter(
1163
self.text_index.combined_index,
1165
add_nodes_callback=self.text_index.add_callback)
1166
self.repo._text_knit._index._graph_index = file_id_index
1167
self.repo._text_knit._index._add_callback = file_id_index.add_nodes
1168
return self.repo._text_knit.add_lines_with_ghosts(
1169
revision_id, parents, new_lines, nostore_sha=nostore_sha,
1170
random_id=random_revid, check_content=False)[0:2]
1172
def all_packs(self):
1173
"""Return a list of all the Pack objects this repository has.
1175
Note that an in-progress pack being created is not returned.
1177
:return: A list of Pack objects for all the packs in the repository.
1180
for name in self.names():
1181
result.append(self.get_pack_by_name(name))
1185
"""Pack the pack collection incrementally.
1187
This will not attempt global reorganisation or recompression,
1188
rather it will just ensure that the total number of packs does
1189
not grow without bound. It uses the _max_pack_count method to
1190
determine if autopacking is needed, and the pack_distribution
1191
method to determine the number of revisions in each pack.
1193
If autopacking takes place then the packs name collection will have
1194
been flushed to disk - packing requires updating the name collection
1195
in synchronisation with certain steps. Otherwise the names collection
1198
:return: True if packing took place.
1200
# XXX: Should not be needed when the management of indices is sane.
1201
total_revisions = self.revision_index.combined_index.key_count()
1202
total_packs = len(self._names)
1203
if self._max_pack_count(total_revisions) >= total_packs:
1205
# XXX: the following may want to be a class, to pack with a given
1207
mutter('Auto-packing repository %s, which has %d pack files, '
1208
'containing %d revisions into %d packs.', self, total_packs,
1209
total_revisions, self._max_pack_count(total_revisions))
1210
# determine which packs need changing
1211
pack_distribution = self.pack_distribution(total_revisions)
1213
for pack in self.all_packs():
1214
revision_count = pack.get_revision_count()
1215
if revision_count == 0:
1216
# revision less packs are not generated by normal operation,
1217
# only by operations like sign-my-commits, and thus will not
1218
# tend to grow rapdily or without bound like commit containing
1219
# packs do - leave them alone as packing them really should
1220
# group their data with the relevant commit, and that may
1221
# involve rewriting ancient history - which autopack tries to
1222
# avoid. Alternatively we could not group the data but treat
1223
# each of these as having a single revision, and thus add
1224
# one revision for each to the total revision count, to get
1225
# a matching distribution.
1227
existing_packs.append((revision_count, pack))
1228
pack_operations = self.plan_autopack_combinations(
1229
existing_packs, pack_distribution)
1230
self._execute_pack_operations(pack_operations)
1233
def _execute_pack_operations(self, pack_operations, _packer_class=Packer):
1234
"""Execute a series of pack operations.
1236
:param pack_operations: A list of [revision_count, packs_to_combine].
1237
:param _packer_class: The class of packer to use (default: Packer).
1240
for revision_count, packs in pack_operations:
1241
# we may have no-ops from the setup logic
1244
_packer_class(self, packs, '.autopack').pack()
1246
self._remove_pack_from_memory(pack)
1247
# record the newly available packs and stop advertising the old
1249
self._save_pack_names(clear_obsolete_packs=True)
1250
# Move the old packs out of the way now they are no longer referenced.
1251
for revision_count, packs in pack_operations:
1252
self._obsolete_packs(packs)
1254
def lock_names(self):
1255
"""Acquire the mutex around the pack-names index.
1257
This cannot be used in the middle of a read-only transaction on the
1260
self.repo.control_files.lock_write()
1263
"""Pack the pack collection totally."""
1264
self.ensure_loaded()
1265
total_packs = len(self._names)
1267
# This is arguably wrong because we might not be optimal, but for
1268
# now lets leave it in. (e.g. reconcile -> one pack. But not
1271
total_revisions = self.revision_index.combined_index.key_count()
1272
# XXX: the following may want to be a class, to pack with a given
1274
mutter('Packing repository %s, which has %d pack files, '
1275
'containing %d revisions into 1 packs.', self, total_packs,
1277
# determine which packs need changing
1278
pack_distribution = [1]
1279
pack_operations = [[0, []]]
1280
for pack in self.all_packs():
1281
pack_operations[-1][0] += pack.get_revision_count()
1282
pack_operations[-1][1].append(pack)
1283
self._execute_pack_operations(pack_operations, OptimisingPacker)
1285
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1286
"""Plan a pack operation.
1288
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1290
:param pack_distribution: A list with the number of revisions desired
1293
if len(existing_packs) <= len(pack_distribution):
1295
existing_packs.sort(reverse=True)
1296
pack_operations = [[0, []]]
1297
# plan out what packs to keep, and what to reorganise
1298
while len(existing_packs):
1299
# take the largest pack, and if its less than the head of the
1300
# distribution chart we will include its contents in the new pack for
1301
# that position. If its larger, we remove its size from the
1302
# distribution chart
1303
next_pack_rev_count, next_pack = existing_packs.pop(0)
1304
if next_pack_rev_count >= pack_distribution[0]:
1305
# this is already packed 'better' than this, so we can
1306
# not waste time packing it.
1307
while next_pack_rev_count > 0:
1308
next_pack_rev_count -= pack_distribution[0]
1309
if next_pack_rev_count >= 0:
1311
del pack_distribution[0]
1313
# didn't use that entire bucket up
1314
pack_distribution[0] = -next_pack_rev_count
1316
# add the revisions we're going to add to the next output pack
1317
pack_operations[-1][0] += next_pack_rev_count
1318
# allocate this pack to the next pack sub operation
1319
pack_operations[-1][1].append(next_pack)
1320
if pack_operations[-1][0] >= pack_distribution[0]:
1321
# this pack is used up, shift left.
1322
del pack_distribution[0]
1323
pack_operations.append([0, []])
1325
return pack_operations
1327
def ensure_loaded(self):
1328
# NB: if you see an assertion error here, its probably access against
1329
# an unlocked repo. Naughty.
1330
if not self.repo.is_locked():
1331
raise errors.ObjectNotLocked(self.repo)
1332
if self._names is None:
1334
self._packs_at_load = set()
1335
for index, key, value in self._iter_disk_pack_index():
1337
self._names[name] = self._parse_index_sizes(value)
1338
self._packs_at_load.add((key, value))
1339
# populate all the metadata.
1342
def _parse_index_sizes(self, value):
1343
"""Parse a string of index sizes."""
1344
return tuple([int(digits) for digits in value.split(' ')])
1346
def get_pack_by_name(self, name):
1347
"""Get a Pack object by name.
1349
:param name: The name of the pack - e.g. '123456'
1350
:return: A Pack object.
1353
return self._packs_by_name[name]
1355
rev_index = self._make_index(name, '.rix')
1356
inv_index = self._make_index(name, '.iix')
1357
txt_index = self._make_index(name, '.tix')
1358
sig_index = self._make_index(name, '.six')
1359
result = ExistingPack(self._pack_transport, name, rev_index,
1360
inv_index, txt_index, sig_index)
1361
self.add_pack_to_memory(result)
1364
def allocate(self, a_new_pack):
1365
"""Allocate name in the list of packs.
1367
:param a_new_pack: A NewPack instance to be added to the collection of
1368
packs for this repository.
1370
self.ensure_loaded()
1371
if a_new_pack.name in self._names:
1372
raise errors.BzrError(
1373
'Pack %r already exists in %s' % (a_new_pack.name, self))
1374
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1375
self.add_pack_to_memory(a_new_pack)
1377
def _iter_disk_pack_index(self):
1378
"""Iterate over the contents of the pack-names index.
1380
This is used when loading the list from disk, and before writing to
1381
detect updates from others during our write operation.
1382
:return: An iterator of the index contents.
1384
return GraphIndex(self.transport, 'pack-names', None
1385
).iter_all_entries()
1387
def _make_index(self, name, suffix):
1388
size_offset = self._suffix_offsets[suffix]
1389
index_name = name + suffix
1390
index_size = self._names[name][size_offset]
1392
self._index_transport, index_name, index_size)
1394
def _max_pack_count(self, total_revisions):
1395
"""Return the maximum number of packs to use for total revisions.
1397
:param total_revisions: The total number of revisions in the
1400
if not total_revisions:
1402
digits = str(total_revisions)
1404
for digit in digits:
1405
result += int(digit)
1409
"""Provide an order to the underlying names."""
1410
return sorted(self._names.keys())
1412
def _obsolete_packs(self, packs):
1413
"""Move a number of packs which have been obsoleted out of the way.
1415
Each pack and its associated indices are moved out of the way.
1417
Note: for correctness this function should only be called after a new
1418
pack names index has been written without these pack names, and with
1419
the names of packs that contain the data previously available via these
1422
:param packs: The packs to obsolete.
1423
:param return: None.
1426
pack.pack_transport.rename(pack.file_name(),
1427
'../obsolete_packs/' + pack.file_name())
1428
# TODO: Probably needs to know all possible indices for this pack
1429
# - or maybe list the directory and move all indices matching this
1430
# name whether we recognize it or not?
1431
for suffix in ('.iix', '.six', '.tix', '.rix'):
1432
self._index_transport.rename(pack.name + suffix,
1433
'../obsolete_packs/' + pack.name + suffix)
1435
def pack_distribution(self, total_revisions):
1436
"""Generate a list of the number of revisions to put in each pack.
1438
:param total_revisions: The total number of revisions in the
1441
if total_revisions == 0:
1443
digits = reversed(str(total_revisions))
1445
for exponent, count in enumerate(digits):
1446
size = 10 ** exponent
1447
for pos in range(int(count)):
1449
return list(reversed(result))
1451
def _pack_tuple(self, name):
1452
"""Return a tuple with the transport and file name for a pack name."""
1453
return self._pack_transport, name + '.pack'
1455
def _remove_pack_from_memory(self, pack):
1456
"""Remove pack from the packs accessed by this repository.
1458
Only affects memory state, until self._save_pack_names() is invoked.
1460
self._names.pop(pack.name)
1461
self._packs_by_name.pop(pack.name)
1462
self._remove_pack_indices(pack)
1464
def _remove_pack_indices(self, pack):
1465
"""Remove the indices for pack from the aggregated indices."""
1466
self.revision_index.remove_index(pack.revision_index, pack)
1467
self.inventory_index.remove_index(pack.inventory_index, pack)
1468
self.text_index.remove_index(pack.text_index, pack)
1469
self.signature_index.remove_index(pack.signature_index, pack)
1472
"""Clear all cached data."""
1473
# cached revision data
1474
self.repo._revision_knit = None
1475
self.revision_index.clear()
1476
# cached signature data
1477
self.repo._signature_knit = None
1478
self.signature_index.clear()
1479
# cached file text data
1480
self.text_index.clear()
1481
self.repo._text_knit = None
1482
# cached inventory data
1483
self.inventory_index.clear()
1484
# remove the open pack
1485
self._new_pack = None
1486
# information about packs.
1489
self._packs_by_name = {}
1490
self._packs_at_load = None
1492
def _make_index_map(self, index_suffix):
1493
"""Return information on existing indices.
1495
:param suffix: Index suffix added to pack name.
1497
:returns: (pack_map, indices) where indices is a list of GraphIndex
1498
objects, and pack_map is a mapping from those objects to the
1499
pack tuple they describe.
1501
# TODO: stop using this; it creates new indices unnecessarily.
1502
self.ensure_loaded()
1503
suffix_map = {'.rix': 'revision_index',
1504
'.six': 'signature_index',
1505
'.iix': 'inventory_index',
1506
'.tix': 'text_index',
1508
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1509
suffix_map[index_suffix])
1511
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1512
"""Convert a list of packs to an index pack map and index list.
1514
:param packs: The packs list to process.
1515
:param index_attribute: The attribute that the desired index is found
1517
:return: A tuple (map, list) where map contains the dict from
1518
index:pack_tuple, and lsit contains the indices in the same order
1524
index = getattr(pack, index_attribute)
1525
indices.append(index)
1526
pack_map[index] = (pack.pack_transport, pack.file_name())
1527
return pack_map, indices
1529
def _index_contents(self, pack_map, key_filter=None):
1530
"""Get an iterable of the index contents from a pack_map.
1532
:param pack_map: A map from indices to pack details.
1533
:param key_filter: An optional filter to limit the
1536
indices = [index for index in pack_map.iterkeys()]
1537
all_index = CombinedGraphIndex(indices)
1538
if key_filter is None:
1539
return all_index.iter_all_entries()
1541
return all_index.iter_entries(key_filter)
1543
def _unlock_names(self):
1544
"""Release the mutex around the pack-names index."""
1545
self.repo.control_files.unlock()
1547
def _save_pack_names(self, clear_obsolete_packs=False):
1548
"""Save the list of packs.
1550
This will take out the mutex around the pack names list for the
1551
duration of the method call. If concurrent updates have been made, a
1552
three-way merge between the current list and the current in memory list
1555
:param clear_obsolete_packs: If True, clear out the contents of the
1556
obsolete_packs directory.
1560
builder = GraphIndexBuilder()
1561
# load the disk nodes across
1563
for index, key, value in self._iter_disk_pack_index():
1564
disk_nodes.add((key, value))
1565
# do a two-way diff against our original content
1566
current_nodes = set()
1567
for name, sizes in self._names.iteritems():
1569
((name, ), ' '.join(str(size) for size in sizes)))
1570
deleted_nodes = self._packs_at_load - current_nodes
1571
new_nodes = current_nodes - self._packs_at_load
1572
disk_nodes.difference_update(deleted_nodes)
1573
disk_nodes.update(new_nodes)
1574
# TODO: handle same-name, index-size-changes here -
1575
# e.g. use the value from disk, not ours, *unless* we're the one
1577
for key, value in disk_nodes:
1578
builder.add_node(key, value)
1579
self.transport.put_file('pack-names', builder.finish(),
1580
mode=self.repo.control_files._file_mode)
1581
# move the baseline forward
1582
self._packs_at_load = disk_nodes
1583
# now clear out the obsolete packs directory
1584
if clear_obsolete_packs:
1585
self.transport.clone('obsolete_packs').delete_multi(
1586
self.transport.list_dir('obsolete_packs'))
1588
self._unlock_names()
1589
# synchronise the memory packs list with what we just wrote:
1590
new_names = dict(disk_nodes)
1591
# drop no longer present nodes
1592
for pack in self.all_packs():
1593
if (pack.name,) not in new_names:
1594
self._remove_pack_from_memory(pack)
1595
# add new nodes/refresh existing ones
1596
for key, value in disk_nodes:
1598
sizes = self._parse_index_sizes(value)
1599
if name in self._names:
1601
if sizes != self._names[name]:
1602
# the pack for name has had its indices replaced - rare but
1603
# important to handle. XXX: probably can never happen today
1604
# because the three-way merge code above does not handle it
1605
# - you may end up adding the same key twice to the new
1606
# disk index because the set values are the same, unless
1607
# the only index shows up as deleted by the set difference
1608
# - which it may. Until there is a specific test for this,
1609
# assume its broken. RBC 20071017.
1610
self._remove_pack_from_memory(self.get_pack_by_name(name))
1611
self._names[name] = sizes
1612
self.get_pack_by_name(name)
1615
self._names[name] = sizes
1616
self.get_pack_by_name(name)
1618
def _start_write_group(self):
1619
# Do not permit preparation for writing if we're not in a 'write lock'.
1620
if not self.repo.is_write_locked():
1621
raise errors.NotWriteLocked(self)
1622
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1623
self._pack_transport, upload_suffix='.pack',
1624
file_mode=self.repo.control_files._file_mode)
1625
# allow writing: queue writes to a new index
1626
self.revision_index.add_writable_index(self._new_pack.revision_index,
1628
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
1630
self.text_index.add_writable_index(self._new_pack.text_index,
1632
self.signature_index.add_writable_index(self._new_pack.signature_index,
1635
# reused revision and signature knits may need updating
1637
# "Hysterical raisins. client code in bzrlib grabs those knits outside
1638
# of write groups and then mutates it inside the write group."
1639
if self.repo._revision_knit is not None:
1640
self.repo._revision_knit._index._add_callback = \
1641
self.revision_index.add_callback
1642
if self.repo._signature_knit is not None:
1643
self.repo._signature_knit._index._add_callback = \
1644
self.signature_index.add_callback
1645
# create a reused knit object for text addition in commit.
1646
self.repo._text_knit = self.repo.weave_store.get_weave_or_empty(
1649
def _abort_write_group(self):
1650
# FIXME: just drop the transient index.
1651
# forget what names there are
1652
if self._new_pack is not None:
1653
self._new_pack.abort()
1654
self._remove_pack_indices(self._new_pack)
1655
self._new_pack = None
1656
self.repo._text_knit = None
1658
def _commit_write_group(self):
1659
self._remove_pack_indices(self._new_pack)
1660
if self._new_pack.data_inserted():
1661
# get all the data to disk and read to use
1662
self._new_pack.finish()
1663
self.allocate(self._new_pack)
1664
self._new_pack = None
1665
if not self.autopack():
1666
# when autopack takes no steps, the names list is still
1668
self._save_pack_names()
1670
self._new_pack.abort()
1671
self._new_pack = None
1672
self.repo._text_knit = None
1675
class KnitPackRevisionStore(KnitRevisionStore):
1676
"""An object to adapt access from RevisionStore's to use KnitPacks.
1678
This class works by replacing the original RevisionStore.
1679
We need to do this because the KnitPackRevisionStore is less
1680
isolated in its layering - it uses services from the repo.
1683
def __init__(self, repo, transport, revisionstore):
1684
"""Create a KnitPackRevisionStore on repo with revisionstore.
1686
This will store its state in the Repository, use the
1687
indices to provide a KnitGraphIndex,
1688
and at the end of transactions write new indices.
1690
KnitRevisionStore.__init__(self, revisionstore.versioned_file_store)
1692
self._serializer = revisionstore._serializer
1693
self.transport = transport
1695
def get_revision_file(self, transaction):
1696
"""Get the revision versioned file object."""
1697
if getattr(self.repo, '_revision_knit', None) is not None:
1698
return self.repo._revision_knit
1699
self.repo._pack_collection.ensure_loaded()
1700
add_callback = self.repo._pack_collection.revision_index.add_callback
1701
# setup knit specific objects
1702
knit_index = KnitGraphIndex(
1703
self.repo._pack_collection.revision_index.combined_index,
1704
add_callback=add_callback)
1705
self.repo._revision_knit = knit.KnitVersionedFile(
1706
'revisions', self.transport.clone('..'),
1707
self.repo.control_files._file_mode,
1708
create=False, access_mode=self.repo._access_mode(),
1709
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1710
access_method=self.repo._pack_collection.revision_index.knit_access)
1711
return self.repo._revision_knit
1713
def get_signature_file(self, transaction):
1714
"""Get the signature versioned file object."""
1715
if getattr(self.repo, '_signature_knit', None) is not None:
1716
return self.repo._signature_knit
1717
self.repo._pack_collection.ensure_loaded()
1718
add_callback = self.repo._pack_collection.signature_index.add_callback
1719
# setup knit specific objects
1720
knit_index = KnitGraphIndex(
1721
self.repo._pack_collection.signature_index.combined_index,
1722
add_callback=add_callback, parents=False)
1723
self.repo._signature_knit = knit.KnitVersionedFile(
1724
'signatures', self.transport.clone('..'),
1725
self.repo.control_files._file_mode,
1726
create=False, access_mode=self.repo._access_mode(),
1727
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1728
access_method=self.repo._pack_collection.signature_index.knit_access)
1729
return self.repo._signature_knit
1732
class KnitPackTextStore(VersionedFileStore):
1733
"""Presents a TextStore abstraction on top of packs.
1735
This class works by replacing the original VersionedFileStore.
1736
We need to do this because the KnitPackRevisionStore is less
1737
isolated in its layering - it uses services from the repo and shares them
1738
with all the data written in a single write group.
1741
def __init__(self, repo, transport, weavestore):
1742
"""Create a KnitPackTextStore on repo with weavestore.
1744
This will store its state in the Repository, use the
1745
indices FileNames to provide a KnitGraphIndex,
1746
and at the end of transactions write new indices.
1748
# don't call base class constructor - it's not suitable.
1749
# no transient data stored in the transaction
1751
self._precious = False
1753
self.transport = transport
1754
self.weavestore = weavestore
1755
# XXX for check() which isn't updated yet
1756
self._transport = weavestore._transport
1758
def get_weave_or_empty(self, file_id, transaction):
1759
"""Get a 'Knit' backed by the .tix indices.
1761
The transaction parameter is ignored.
1763
self.repo._pack_collection.ensure_loaded()
1764
add_callback = self.repo._pack_collection.text_index.add_callback
1765
# setup knit specific objects
1766
file_id_index = GraphIndexPrefixAdapter(
1767
self.repo._pack_collection.text_index.combined_index,
1768
(file_id, ), 1, add_nodes_callback=add_callback)
1769
knit_index = KnitGraphIndex(file_id_index,
1770
add_callback=file_id_index.add_nodes,
1771
deltas=True, parents=True)
1772
return knit.KnitVersionedFile('text:' + file_id,
1773
self.transport.clone('..'),
1776
access_method=self.repo._pack_collection.text_index.knit_access,
1777
factory=knit.KnitPlainFactory())
1779
get_weave = get_weave_or_empty
1782
"""Generate a list of the fileids inserted, for use by check."""
1783
self.repo._pack_collection.ensure_loaded()
1785
for index, key, value, refs in \
1786
self.repo._pack_collection.text_index.combined_index.iter_all_entries():
1791
class InventoryKnitThunk(object):
1792
"""An object to manage thunking get_inventory_weave to pack based knits."""
1794
def __init__(self, repo, transport):
1795
"""Create an InventoryKnitThunk for repo at transport.
1797
This will store its state in the Repository, use the
1798
indices FileNames to provide a KnitGraphIndex,
1799
and at the end of transactions write a new index..
1802
self.transport = transport
1804
def get_weave(self):
1805
"""Get a 'Knit' that contains inventory data."""
1806
self.repo._pack_collection.ensure_loaded()
1807
add_callback = self.repo._pack_collection.inventory_index.add_callback
1808
# setup knit specific objects
1809
knit_index = KnitGraphIndex(
1810
self.repo._pack_collection.inventory_index.combined_index,
1811
add_callback=add_callback, deltas=True, parents=True)
1812
return knit.KnitVersionedFile(
1813
'inventory', self.transport.clone('..'),
1814
self.repo.control_files._file_mode,
1815
create=False, access_mode=self.repo._access_mode(),
1816
index=knit_index, delta=True, factory=knit.KnitPlainFactory(),
1817
access_method=self.repo._pack_collection.inventory_index.knit_access)
1820
class KnitPackRepository(KnitRepository):
1821
"""Experimental graph-knit using repository."""
1823
def __init__(self, _format, a_bzrdir, control_files, _revision_store,
1824
control_store, text_store, _commit_builder_class, _serializer):
1825
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
1826
_revision_store, control_store, text_store, _commit_builder_class,
1828
index_transport = control_files._transport.clone('indices')
1829
self._pack_collection = RepositoryPackCollection(self, control_files._transport,
1831
control_files._transport.clone('upload'),
1832
control_files._transport.clone('packs'))
1833
self._revision_store = KnitPackRevisionStore(self, index_transport, self._revision_store)
1834
self.weave_store = KnitPackTextStore(self, index_transport, self.weave_store)
1835
self._inv_thunk = InventoryKnitThunk(self, index_transport)
1836
# True when the repository object is 'write locked' (as opposed to the
1837
# physical lock only taken out around changes to the pack-names list.)
1838
# Another way to represent this would be a decorator around the control
1839
# files object that presents logical locks as physical ones - if this
1840
# gets ugly consider that alternative design. RBC 20071011
1841
self._write_lock_count = 0
1842
self._transaction = None
1844
self._reconcile_does_inventory_gc = True
1845
self._reconcile_fixes_text_parents = True
1846
self._reconcile_backsup_inventory = False
1848
def _abort_write_group(self):
1849
self._pack_collection._abort_write_group()
1851
def _access_mode(self):
1852
"""Return 'w' or 'r' for depending on whether a write lock is active.
1854
This method is a helper for the Knit-thunking support objects.
1856
if self.is_write_locked():
1860
def _find_inconsistent_revision_parents(self):
1861
"""Find revisions with incorrectly cached parents.
1863
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1864
parents-in-revision).
1866
if not self.is_locked():
1867
raise errors.ObjectNotLocked(self)
1868
pb = ui.ui_factory.nested_progress_bar()
1871
revision_nodes = self._pack_collection.revision_index \
1872
.combined_index.iter_all_entries()
1873
index_positions = []
1874
# Get the cached index values for all revisions, and also the location
1875
# in each index of the revision text so we can perform linear IO.
1876
for index, key, value, refs in revision_nodes:
1877
pos, length = value[1:].split(' ')
1878
index_positions.append((index, int(pos), key[0],
1879
tuple(parent[0] for parent in refs[0])))
1880
pb.update("Reading revision index.", 0, 0)
1881
index_positions.sort()
1882
batch_count = len(index_positions) / 1000 + 1
1883
pb.update("Checking cached revision graph.", 0, batch_count)
1884
for offset in xrange(batch_count):
1885
pb.update("Checking cached revision graph.", offset)
1886
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1889
rev_ids = [item[2] for item in to_query]
1890
revs = self.get_revisions(rev_ids)
1891
for revision, item in zip(revs, to_query):
1892
index_parents = item[3]
1893
rev_parents = tuple(revision.parent_ids)
1894
if index_parents != rev_parents:
1895
result.append((revision.revision_id, index_parents, rev_parents))
1900
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
1901
def get_parents(self, revision_ids):
1902
"""See graph._StackedParentsProvider.get_parents."""
1903
parent_map = self.get_parent_map(revision_ids)
1904
return [parent_map.get(r, None) for r in revision_ids]
1906
def get_parent_map(self, keys):
1907
"""See graph._StackedParentsProvider.get_parent_map
1909
This implementation accesses the combined revision index to provide
1912
self._pack_collection.ensure_loaded()
1913
index = self._pack_collection.revision_index.combined_index
1915
if _mod_revision.NULL_REVISION in keys:
1916
keys.discard(_mod_revision.NULL_REVISION)
1917
found_parents = {_mod_revision.NULL_REVISION:()}
1920
search_keys = set((revision_id,) for revision_id in keys)
1921
for index, key, value, refs in index.iter_entries(search_keys):
1924
parents = (_mod_revision.NULL_REVISION,)
1926
parents = tuple(parent[0] for parent in parents)
1927
found_parents[key[0]] = parents
1928
return found_parents
1930
def has_revisions(self, revision_ids):
1931
"""See Repository.has_revisions()."""
1932
revision_ids = set(revision_ids)
1933
result = revision_ids.intersection(
1934
set([None, _mod_revision.NULL_REVISION]))
1935
revision_ids.difference_update(result)
1936
index = self._pack_collection.revision_index.combined_index
1937
keys = [(revision_id,) for revision_id in revision_ids]
1938
result.update(node[1][0] for node in index.iter_entries(keys))
1941
def _make_parents_provider(self):
1942
return graph.CachingParentsProvider(self)
1944
def _refresh_data(self):
1945
if self._write_lock_count == 1 or (
1946
self.control_files._lock_count == 1 and
1947
self.control_files._lock_mode == 'r'):
1948
# forget what names there are
1949
self._pack_collection.reset()
1950
# XXX: Better to do an in-memory merge when acquiring a new lock -
1951
# factor out code from _save_pack_names.
1952
self._pack_collection.ensure_loaded()
1954
def _start_write_group(self):
1955
self._pack_collection._start_write_group()
1957
def _commit_write_group(self):
1958
return self._pack_collection._commit_write_group()
1960
def get_inventory_weave(self):
1961
return self._inv_thunk.get_weave()
1963
def get_transaction(self):
1964
if self._write_lock_count:
1965
return self._transaction
1967
return self.control_files.get_transaction()
1969
def is_locked(self):
1970
return self._write_lock_count or self.control_files.is_locked()
1972
def is_write_locked(self):
1973
return self._write_lock_count
1975
def lock_write(self, token=None):
1976
if not self._write_lock_count and self.is_locked():
1977
raise errors.ReadOnlyError(self)
1978
self._write_lock_count += 1
1979
if self._write_lock_count == 1:
1980
from bzrlib import transactions
1981
self._transaction = transactions.WriteTransaction()
1982
self._refresh_data()
1984
def lock_read(self):
1985
if self._write_lock_count:
1986
self._write_lock_count += 1
1988
self.control_files.lock_read()
1989
self._refresh_data()
1991
def leave_lock_in_place(self):
1992
# not supported - raise an error
1993
raise NotImplementedError(self.leave_lock_in_place)
1995
def dont_leave_lock_in_place(self):
1996
# not supported - raise an error
1997
raise NotImplementedError(self.dont_leave_lock_in_place)
2001
"""Compress the data within the repository.
2003
This will pack all the data to a single pack. In future it may
2004
recompress deltas or do other such expensive operations.
2006
self._pack_collection.pack()
2009
def reconcile(self, other=None, thorough=False):
2010
"""Reconcile this repository."""
2011
from bzrlib.reconcile import PackReconciler
2012
reconciler = PackReconciler(self, thorough=thorough)
2013
reconciler.reconcile()
2017
if self._write_lock_count == 1 and self._write_group is not None:
2018
self.abort_write_group()
2019
self._transaction = None
2020
self._write_lock_count = 0
2021
raise errors.BzrError(
2022
'Must end write group before releasing write lock on %s'
2024
if self._write_lock_count:
2025
self._write_lock_count -= 1
2026
if not self._write_lock_count:
2027
transaction = self._transaction
2028
self._transaction = None
2029
transaction.finish()
2031
self.control_files.unlock()
2034
class RepositoryFormatPack(MetaDirRepositoryFormat):
2035
"""Format logic for pack structured repositories.
2037
This repository format has:
2038
- a list of packs in pack-names
2039
- packs in packs/NAME.pack
2040
- indices in indices/NAME.{iix,six,tix,rix}
2041
- knit deltas in the packs, knit indices mapped to the indices.
2042
- thunk objects to support the knits programming API.
2043
- a format marker of its own
2044
- an optional 'shared-storage' flag
2045
- an optional 'no-working-trees' flag
2049
# Set this attribute in derived classes to control the repository class
2050
# created by open and initialize.
2051
repository_class = None
2052
# Set this attribute in derived classes to control the
2053
# _commit_builder_class that the repository objects will have passed to
2054
# their constructor.
2055
_commit_builder_class = None
2056
# Set this attribute in derived clases to control the _serializer that the
2057
# repository objects will have passed to their constructor.
2059
# External references are not supported in pack repositories yet.
2060
supports_external_lookups = False
2062
def _get_control_store(self, repo_transport, control_files):
2063
"""Return the control store for this repository."""
2064
return VersionedFileStore(
2067
file_mode=control_files._file_mode,
2068
versionedfile_class=knit.KnitVersionedFile,
2069
versionedfile_kwargs={'factory': knit.KnitPlainFactory()},
2072
def _get_revision_store(self, repo_transport, control_files):
2073
"""See RepositoryFormat._get_revision_store()."""
2074
versioned_file_store = VersionedFileStore(
2076
file_mode=control_files._file_mode,
2079
versionedfile_class=knit.KnitVersionedFile,
2080
versionedfile_kwargs={'delta': False,
2081
'factory': knit.KnitPlainFactory(),
2085
return KnitRevisionStore(versioned_file_store)
2087
def _get_text_store(self, transport, control_files):
2088
"""See RepositoryFormat._get_text_store()."""
2089
return self._get_versioned_file_store('knits',
2092
versionedfile_class=knit.KnitVersionedFile,
2093
versionedfile_kwargs={
2094
'create_parent_dir': True,
2095
'delay_create': True,
2096
'dir_mode': control_files._dir_mode,
2100
def initialize(self, a_bzrdir, shared=False):
2101
"""Create a pack based repository.
2103
:param a_bzrdir: bzrdir to contain the new repository; must already
2105
:param shared: If true the repository will be initialized as a shared
2108
mutter('creating repository in %s.', a_bzrdir.transport.base)
2109
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
2110
builder = GraphIndexBuilder()
2111
files = [('pack-names', builder.finish())]
2112
utf8_files = [('format', self.get_format_string())]
2114
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
2115
return self.open(a_bzrdir=a_bzrdir, _found=True)
2117
def open(self, a_bzrdir, _found=False, _override_transport=None):
2118
"""See RepositoryFormat.open().
2120
:param _override_transport: INTERNAL USE ONLY. Allows opening the
2121
repository at a slightly different url
2122
than normal. I.e. during 'upgrade'.
2125
format = RepositoryFormat.find_format(a_bzrdir)
2126
assert format.__class__ == self.__class__
2127
if _override_transport is not None:
2128
repo_transport = _override_transport
2130
repo_transport = a_bzrdir.get_repository_transport(None)
2131
control_files = lockable_files.LockableFiles(repo_transport,
2132
'lock', lockdir.LockDir)
2133
text_store = self._get_text_store(repo_transport, control_files)
2134
control_store = self._get_control_store(repo_transport, control_files)
2135
_revision_store = self._get_revision_store(repo_transport, control_files)
2136
return self.repository_class(_format=self,
2138
control_files=control_files,
2139
_revision_store=_revision_store,
2140
control_store=control_store,
2141
text_store=text_store,
2142
_commit_builder_class=self._commit_builder_class,
2143
_serializer=self._serializer)
2146
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2147
"""A no-subtrees parameterized Pack repository.
2149
This format was introduced in 0.92.
2152
repository_class = KnitPackRepository
2153
_commit_builder_class = PackCommitBuilder
2155
def _serializer(self):
2156
return xml5.serializer_v5
2158
def _get_matching_bzrdir(self):
2159
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2161
def _ignore_setting_bzrdir(self, format):
2164
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2166
def get_format_string(self):
2167
"""See RepositoryFormat.get_format_string()."""
2168
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2170
def get_format_description(self):
2171
"""See RepositoryFormat.get_format_description()."""
2172
return "Packs containing knits without subtree support"
2174
def check_conversion_target(self, target_format):
2178
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2179
"""A subtrees parameterized Pack repository.
2181
This repository format uses the xml7 serializer to get:
2182
- support for recording full info about the tree root
2183
- support for recording tree-references
2185
This format was introduced in 0.92.
2188
repository_class = KnitPackRepository
2189
_commit_builder_class = PackRootCommitBuilder
2190
rich_root_data = True
2191
supports_tree_reference = True
2193
def _serializer(self):
2194
return xml7.serializer_v7
2196
def _get_matching_bzrdir(self):
2197
return bzrdir.format_registry.make_bzrdir(
2198
'pack-0.92-subtree')
2200
def _ignore_setting_bzrdir(self, format):
2203
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2205
def check_conversion_target(self, target_format):
2206
if not target_format.rich_root_data:
2207
raise errors.BadConversionTarget(
2208
'Does not support rich root data.', target_format)
2209
if not getattr(target_format, 'supports_tree_reference', False):
2210
raise errors.BadConversionTarget(
2211
'Does not support nested trees', target_format)
2213
def get_format_string(self):
2214
"""See RepositoryFormat.get_format_string()."""
2215
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2217
def get_format_description(self):
2218
"""See RepositoryFormat.get_format_description()."""
2219
return "Packs containing knits with subtree support\n"
2222
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2223
"""A rich-root, no subtrees parameterized Pack repository.
2225
This repository format uses the xml6 serializer to get:
2226
- support for recording full info about the tree root
2228
This format was introduced in 1.0.
2231
repository_class = KnitPackRepository
2232
_commit_builder_class = PackRootCommitBuilder
2233
rich_root_data = True
2234
supports_tree_reference = False
2236
def _serializer(self):
2237
return xml6.serializer_v6
2239
def _get_matching_bzrdir(self):
2240
return bzrdir.format_registry.make_bzrdir(
2243
def _ignore_setting_bzrdir(self, format):
2246
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2248
def check_conversion_target(self, target_format):
2249
if not target_format.rich_root_data:
2250
raise errors.BadConversionTarget(
2251
'Does not support rich root data.', target_format)
2253
def get_format_string(self):
2254
"""See RepositoryFormat.get_format_string()."""
2255
return ("Bazaar pack repository format 1 with rich root"
2256
" (needs bzr 1.0)\n")
2258
def get_format_description(self):
2259
"""See RepositoryFormat.get_format_description()."""
2260
return "Packs containing knits with rich root support\n"
2263
class RepositoryFormatPackDevelopment0(RepositoryFormatPack):
2264
"""A no-subtrees development repository.
2266
This format should be retained until the second release after bzr 1.0.
2268
No changes to the disk behaviour from pack-0.92.
2271
repository_class = KnitPackRepository
2272
_commit_builder_class = PackCommitBuilder
2273
_serializer = xml5.serializer_v5
2275
def _get_matching_bzrdir(self):
2276
return bzrdir.format_registry.make_bzrdir('development0')
2278
def _ignore_setting_bzrdir(self, format):
2281
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2283
def get_format_string(self):
2284
"""See RepositoryFormat.get_format_string()."""
2285
return "Bazaar development format 0 (needs bzr.dev from before 1.3)\n"
2287
def get_format_description(self):
2288
"""See RepositoryFormat.get_format_description()."""
2289
return ("Development repository format, currently the same as "
2292
def check_conversion_target(self, target_format):
2296
class RepositoryFormatPackDevelopment0Subtree(RepositoryFormatPack):
2297
"""A subtrees development repository.
2299
This format should be retained until the second release after bzr 1.0.
2301
No changes to the disk behaviour from pack-0.92-subtree.
2304
repository_class = KnitPackRepository
2305
_commit_builder_class = PackRootCommitBuilder
2306
rich_root_data = True
2307
supports_tree_reference = True
2308
_serializer = xml7.serializer_v7
2310
def _get_matching_bzrdir(self):
2311
return bzrdir.format_registry.make_bzrdir(
2312
'development0-subtree')
2314
def _ignore_setting_bzrdir(self, format):
2317
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2319
def check_conversion_target(self, target_format):
2320
if not target_format.rich_root_data:
2321
raise errors.BadConversionTarget(
2322
'Does not support rich root data.', target_format)
2323
if not getattr(target_format, 'supports_tree_reference', False):
2324
raise errors.BadConversionTarget(
2325
'Does not support nested trees', target_format)
2327
def get_format_string(self):
2328
"""See RepositoryFormat.get_format_string()."""
2329
return ("Bazaar development format 0 with subtree support "
2330
"(needs bzr.dev from before 1.3)\n")
2332
def get_format_description(self):
2333
"""See RepositoryFormat.get_format_description()."""
2334
return ("Development repository format, currently the same as "
2335
"pack-0.92-subtree\n")
added
removed
bzrlib/repofmt/pack_repo.py
