1
# Copyright (C) 2005, 2006, 2007, 2008 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
from bzrlib.lazy_import import lazy_import
18
lazy_import(globals(), """
19
from itertools import izip
30
from bzrlib.index import (
35
GraphIndexPrefixAdapter,
37
from bzrlib.knit import KnitGraphIndex, _PackAccess, _KnitData
38
from bzrlib.osutils import rand_chars
39
from bzrlib.pack import ContainerWriter
40
from bzrlib.store import revision
41
from bzrlib import tsort
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,
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 (
77
class PackCommitBuilder(CommitBuilder):
78
"""A subclass of CommitBuilder to add texts with pack semantics.
80
Specifically this uses one knit object rather than one knit object per
81
added text, reducing memory and object pressure.
84
def __init__(self, repository, parents, config, timestamp=None,
85
timezone=None, committer=None, revprops=None,
87
CommitBuilder.__init__(self, repository, parents, config,
88
timestamp=timestamp, timezone=timezone, committer=committer,
89
revprops=revprops, revision_id=revision_id)
90
self._file_graph = graph.Graph(
91
repository._pack_collection.text_index.combined_index)
93
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
94
return self.repository._pack_collection._add_text_to_weave(file_id,
95
self._new_revision_id, new_lines, parents, nostore_sha,
98
def _heads(self, file_id, revision_ids):
99
keys = [(file_id, revision_id) for revision_id in revision_ids]
100
return set([key[1] for key in self._file_graph.heads(keys)])
103
class PackRootCommitBuilder(RootCommitBuilder):
104
"""A subclass of RootCommitBuilder to add texts with pack semantics.
106
Specifically this uses one knit object rather than one knit object per
107
added text, reducing memory and object pressure.
110
def __init__(self, repository, parents, config, timestamp=None,
111
timezone=None, committer=None, revprops=None,
113
CommitBuilder.__init__(self, repository, parents, config,
114
timestamp=timestamp, timezone=timezone, committer=committer,
115
revprops=revprops, revision_id=revision_id)
116
self._file_graph = graph.Graph(
117
repository._pack_collection.text_index.combined_index)
119
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
120
return self.repository._pack_collection._add_text_to_weave(file_id,
121
self._new_revision_id, new_lines, parents, nostore_sha,
124
def _heads(self, file_id, revision_ids):
125
keys = [(file_id, revision_id) for revision_id in revision_ids]
126
return set([key[1] for key in self._file_graph.heads(keys)])
130
"""An in memory proxy for a pack and its indices.
132
This is a base class that is not directly used, instead the classes
133
ExistingPack and NewPack are used.
136
def __init__(self, revision_index, inventory_index, text_index,
138
"""Create a pack instance.
140
:param revision_index: A GraphIndex for determining what revisions are
141
present in the Pack and accessing the locations of their texts.
142
:param inventory_index: A GraphIndex for determining what inventories are
143
present in the Pack and accessing the locations of their
145
:param text_index: A GraphIndex for determining what file texts
146
are present in the pack and accessing the locations of their
147
texts/deltas (via (fileid, revisionid) tuples).
148
:param revision_index: A GraphIndex for determining what signatures are
149
present in the Pack and accessing the locations of their texts.
151
self.revision_index = revision_index
152
self.inventory_index = inventory_index
153
self.text_index = text_index
154
self.signature_index = signature_index
156
def access_tuple(self):
157
"""Return a tuple (transport, name) for the pack content."""
158
return self.pack_transport, self.file_name()
161
"""Get the file name for the pack on disk."""
162
return self.name + '.pack'
164
def get_revision_count(self):
165
return self.revision_index.key_count()
167
def inventory_index_name(self, name):
168
"""The inv index is the name + .iix."""
169
return self.index_name('inventory', name)
171
def revision_index_name(self, name):
172
"""The revision index is the name + .rix."""
173
return self.index_name('revision', name)
175
def signature_index_name(self, name):
176
"""The signature index is the name + .six."""
177
return self.index_name('signature', name)
179
def text_index_name(self, name):
180
"""The text index is the name + .tix."""
181
return self.index_name('text', name)
183
def _external_compression_parents_of_texts(self):
186
for node in self.text_index.iter_all_entries():
188
refs.update(node[3][1])
192
class ExistingPack(Pack):
193
"""An in memory proxy for an existing .pack and its disk indices."""
195
def __init__(self, pack_transport, name, revision_index, inventory_index,
196
text_index, signature_index):
197
"""Create an ExistingPack object.
199
:param pack_transport: The transport where the pack file resides.
200
:param name: The name of the pack on disk in the pack_transport.
202
Pack.__init__(self, revision_index, inventory_index, text_index,
205
self.pack_transport = pack_transport
206
if None in (revision_index, inventory_index, text_index,
207
signature_index, name, pack_transport):
208
raise AssertionError()
210
def __eq__(self, other):
211
return self.__dict__ == other.__dict__
213
def __ne__(self, other):
214
return not self.__eq__(other)
217
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
218
id(self), self.transport, self.name)
222
"""An in memory proxy for a pack which is being created."""
224
# A map of index 'type' to the file extension and position in the
226
index_definitions = {
227
'revision': ('.rix', 0),
228
'inventory': ('.iix', 1),
230
'signature': ('.six', 3),
233
def __init__(self, upload_transport, index_transport, pack_transport,
234
upload_suffix='', file_mode=None):
235
"""Create a NewPack instance.
237
:param upload_transport: A writable transport for the pack to be
238
incrementally uploaded to.
239
:param index_transport: A writable transport for the pack's indices to
240
be written to when the pack is finished.
241
:param pack_transport: A writable transport for the pack to be renamed
242
to when the upload is complete. This *must* be the same as
243
upload_transport.clone('../packs').
244
:param upload_suffix: An optional suffix to be given to any temporary
245
files created during the pack creation. e.g '.autopack'
246
:param file_mode: An optional file mode to create the new files with.
248
# The relative locations of the packs are constrained, but all are
249
# passed in because the caller has them, so as to avoid object churn.
251
# Revisions: parents list, no text compression.
252
InMemoryGraphIndex(reference_lists=1),
253
# Inventory: We want to map compression only, but currently the
254
# knit code hasn't been updated enough to understand that, so we
255
# have a regular 2-list index giving parents and compression
257
InMemoryGraphIndex(reference_lists=2),
258
# Texts: compression and per file graph, for all fileids - so two
259
# reference lists and two elements in the key tuple.
260
InMemoryGraphIndex(reference_lists=2, key_elements=2),
261
# Signatures: Just blobs to store, no compression, no parents
263
InMemoryGraphIndex(reference_lists=0),
265
# where should the new pack be opened
266
self.upload_transport = upload_transport
267
# where are indices written out to
268
self.index_transport = index_transport
269
# where is the pack renamed to when it is finished?
270
self.pack_transport = pack_transport
271
# What file mode to upload the pack and indices with.
272
self._file_mode = file_mode
273
# tracks the content written to the .pack file.
274
self._hash = md5.new()
275
# a four-tuple with the length in bytes of the indices, once the pack
276
# is finalised. (rev, inv, text, sigs)
277
self.index_sizes = None
278
# How much data to cache when writing packs. Note that this is not
279
# synchronised with reads, because it's not in the transport layer, so
280
# is not safe unless the client knows it won't be reading from the pack
282
self._cache_limit = 0
283
# the temporary pack file name.
284
self.random_name = rand_chars(20) + upload_suffix
285
# when was this pack started ?
286
self.start_time = time.time()
287
# open an output stream for the data added to the pack.
288
self.write_stream = self.upload_transport.open_write_stream(
289
self.random_name, mode=self._file_mode)
290
if 'pack' in debug.debug_flags:
291
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
292
time.ctime(), self.upload_transport.base, self.random_name,
293
time.time() - self.start_time)
294
# A list of byte sequences to be written to the new pack, and the
295
# aggregate size of them. Stored as a list rather than separate
296
# variables so that the _write_data closure below can update them.
297
self._buffer = [[], 0]
298
# create a callable for adding data
300
# robertc says- this is a closure rather than a method on the object
301
# so that the variables are locals, and faster than accessing object
303
def _write_data(bytes, flush=False, _buffer=self._buffer,
304
_write=self.write_stream.write, _update=self._hash.update):
305
_buffer[0].append(bytes)
306
_buffer[1] += len(bytes)
308
if _buffer[1] > self._cache_limit or flush:
309
bytes = ''.join(_buffer[0])
313
# expose this on self, for the occasion when clients want to add data.
314
self._write_data = _write_data
315
# a pack writer object to serialise pack records.
316
self._writer = pack.ContainerWriter(self._write_data)
318
# what state is the pack in? (open, finished, aborted)
322
"""Cancel creating this pack."""
323
self._state = 'aborted'
324
self.write_stream.close()
325
# Remove the temporary pack file.
326
self.upload_transport.delete(self.random_name)
327
# The indices have no state on disk.
329
def access_tuple(self):
330
"""Return a tuple (transport, name) for the pack content."""
331
if self._state == 'finished':
332
return Pack.access_tuple(self)
333
elif self._state == 'open':
334
return self.upload_transport, self.random_name
336
raise AssertionError(self._state)
338
def data_inserted(self):
339
"""True if data has been added to this pack."""
340
return bool(self.get_revision_count() or
341
self.inventory_index.key_count() or
342
self.text_index.key_count() or
343
self.signature_index.key_count())
346
"""Finish the new pack.
349
- finalises the content
350
- assigns a name (the md5 of the content, currently)
351
- writes out the associated indices
352
- renames the pack into place.
353
- stores the index size tuple for the pack in the index_sizes
358
self._write_data('', flush=True)
359
self.name = self._hash.hexdigest()
361
# XXX: It'd be better to write them all to temporary names, then
362
# rename them all into place, so that the window when only some are
363
# visible is smaller. On the other hand none will be seen until
364
# they're in the names list.
365
self.index_sizes = [None, None, None, None]
366
self._write_index('revision', self.revision_index, 'revision')
367
self._write_index('inventory', self.inventory_index, 'inventory')
368
self._write_index('text', self.text_index, 'file texts')
369
self._write_index('signature', self.signature_index,
370
'revision signatures')
371
self.write_stream.close()
372
# Note that this will clobber an existing pack with the same name,
373
# without checking for hash collisions. While this is undesirable this
374
# is something that can be rectified in a subsequent release. One way
375
# to rectify it may be to leave the pack at the original name, writing
376
# its pack-names entry as something like 'HASH: index-sizes
377
# temporary-name'. Allocate that and check for collisions, if it is
378
# collision free then rename it into place. If clients know this scheme
379
# they can handle missing-file errors by:
380
# - try for HASH.pack
381
# - try for temporary-name
382
# - refresh the pack-list to see if the pack is now absent
383
self.upload_transport.rename(self.random_name,
384
'../packs/' + self.name + '.pack')
385
self._state = 'finished'
386
if 'pack' in debug.debug_flags:
387
# XXX: size might be interesting?
388
mutter('%s: create_pack: pack renamed into place: %s%s->%s%s t+%6.3fs',
389
time.ctime(), self.upload_transport.base, self.random_name,
390
self.pack_transport, self.name,
391
time.time() - self.start_time)
394
"""Flush any current data."""
396
bytes = ''.join(self._buffer[0])
397
self.write_stream.write(bytes)
398
self._hash.update(bytes)
399
self._buffer[:] = [[], 0]
401
def index_name(self, index_type, name):
402
"""Get the disk name of an index type for pack name 'name'."""
403
return name + NewPack.index_definitions[index_type][0]
405
def index_offset(self, index_type):
406
"""Get the position in a index_size array for a given index type."""
407
return NewPack.index_definitions[index_type][1]
409
def _replace_index_with_readonly(self, index_type):
410
setattr(self, index_type + '_index',
411
GraphIndex(self.index_transport,
412
self.index_name(index_type, self.name),
413
self.index_sizes[self.index_offset(index_type)]))
415
def set_write_cache_size(self, size):
416
self._cache_limit = size
418
def _write_index(self, index_type, index, label):
419
"""Write out an index.
421
:param index_type: The type of index to write - e.g. 'revision'.
422
:param index: The index object to serialise.
423
:param label: What label to give the index e.g. 'revision'.
425
index_name = self.index_name(index_type, self.name)
426
self.index_sizes[self.index_offset(index_type)] = \
427
self.index_transport.put_file(index_name, index.finish(),
428
mode=self._file_mode)
429
if 'pack' in debug.debug_flags:
430
# XXX: size might be interesting?
431
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
432
time.ctime(), label, self.upload_transport.base,
433
self.random_name, time.time() - self.start_time)
434
# Replace the writable index on this object with a readonly,
435
# presently unloaded index. We should alter
436
# the index layer to make its finish() error if add_node is
437
# subsequently used. RBC
438
self._replace_index_with_readonly(index_type)
441
class AggregateIndex(object):
442
"""An aggregated index for the RepositoryPackCollection.
444
AggregateIndex is reponsible for managing the PackAccess object,
445
Index-To-Pack mapping, and all indices list for a specific type of index
446
such as 'revision index'.
448
A CombinedIndex provides an index on a single key space built up
449
from several on-disk indices. The AggregateIndex builds on this
450
to provide a knit access layer, and allows having up to one writable
451
index within the collection.
453
# XXX: Probably 'can be written to' could/should be separated from 'acts
454
# like a knit index' -- mbp 20071024
457
"""Create an AggregateIndex."""
458
self.index_to_pack = {}
459
self.combined_index = CombinedGraphIndex([])
460
self.knit_access = _PackAccess(self.index_to_pack)
462
def replace_indices(self, index_to_pack, indices):
463
"""Replace the current mappings with fresh ones.
465
This should probably not be used eventually, rather incremental add and
466
removal of indices. It has been added during refactoring of existing
469
:param index_to_pack: A mapping from index objects to
470
(transport, name) tuples for the pack file data.
471
:param indices: A list of indices.
473
# refresh the revision pack map dict without replacing the instance.
474
self.index_to_pack.clear()
475
self.index_to_pack.update(index_to_pack)
476
# XXX: API break - clearly a 'replace' method would be good?
477
self.combined_index._indices[:] = indices
478
# the current add nodes callback for the current writable index if
480
self.add_callback = None
482
def add_index(self, index, pack):
483
"""Add index to the aggregate, which is an index for Pack pack.
485
Future searches on the aggregate index will seach this new index
486
before all previously inserted indices.
488
:param index: An Index for the pack.
489
:param pack: A Pack instance.
491
# expose it to the index map
492
self.index_to_pack[index] = pack.access_tuple()
493
# put it at the front of the linear index list
494
self.combined_index.insert_index(0, index)
496
def add_writable_index(self, index, pack):
497
"""Add an index which is able to have data added to it.
499
There can be at most one writable index at any time. Any
500
modifications made to the knit are put into this index.
502
:param index: An index from the pack parameter.
503
:param pack: A Pack instance.
505
if self.add_callback is not None:
506
raise AssertionError(
507
"%s already has a writable index through %s" % \
508
(self, self.add_callback))
509
# allow writing: queue writes to a new index
510
self.add_index(index, pack)
511
# Updates the index to packs mapping as a side effect,
512
self.knit_access.set_writer(pack._writer, index, pack.access_tuple())
513
self.add_callback = index.add_nodes
516
"""Reset all the aggregate data to nothing."""
517
self.knit_access.set_writer(None, None, (None, None))
518
self.index_to_pack.clear()
519
del self.combined_index._indices[:]
520
self.add_callback = None
522
def remove_index(self, index, pack):
523
"""Remove index from the indices used to answer queries.
525
:param index: An index from the pack parameter.
526
:param pack: A Pack instance.
528
del self.index_to_pack[index]
529
self.combined_index._indices.remove(index)
530
if (self.add_callback is not None and
531
getattr(index, 'add_nodes', None) == self.add_callback):
532
self.add_callback = None
533
self.knit_access.set_writer(None, None, (None, None))
536
class Packer(object):
537
"""Create a pack from packs."""
539
def __init__(self, pack_collection, packs, suffix, revision_ids=None):
542
:param pack_collection: A RepositoryPackCollection object where the
543
new pack is being written to.
544
:param packs: The packs to combine.
545
:param suffix: The suffix to use on the temporary files for the pack.
546
:param revision_ids: Revision ids to limit the pack to.
550
self.revision_ids = revision_ids
551
# The pack object we are creating.
553
self._pack_collection = pack_collection
554
# The index layer keys for the revisions being copied. None for 'all
556
self._revision_keys = None
557
# What text keys to copy. None for 'all texts'. This is set by
558
# _copy_inventory_texts
559
self._text_filter = None
562
def _extra_init(self):
563
"""A template hook to allow extending the constructor trivially."""
565
def pack(self, pb=None):
566
"""Create a new pack by reading data from other packs.
568
This does little more than a bulk copy of data. One key difference
569
is that data with the same item key across multiple packs is elided
570
from the output. The new pack is written into the current pack store
571
along with its indices, and the name added to the pack names. The
572
source packs are not altered and are not required to be in the current
575
:param pb: An optional progress bar to use. A nested bar is created if
577
:return: A Pack object, or None if nothing was copied.
579
# open a pack - using the same name as the last temporary file
580
# - which has already been flushed, so its safe.
581
# XXX: - duplicate code warning with start_write_group; fix before
582
# considering 'done'.
583
if self._pack_collection._new_pack is not None:
584
raise errors.BzrError('call to create_pack_from_packs while '
585
'another pack is being written.')
586
if self.revision_ids is not None:
587
if len(self.revision_ids) == 0:
588
# silly fetch request.
591
self.revision_ids = frozenset(self.revision_ids)
593
self.pb = ui.ui_factory.nested_progress_bar()
597
return self._create_pack_from_packs()
603
"""Open a pack for the pack we are creating."""
604
return NewPack(self._pack_collection._upload_transport,
605
self._pack_collection._index_transport,
606
self._pack_collection._pack_transport, upload_suffix=self.suffix,
607
file_mode=self._pack_collection.repo.bzrdir._get_file_mode())
609
def _copy_revision_texts(self):
610
"""Copy revision data to the new pack."""
612
if self.revision_ids:
613
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
616
# select revision keys
617
revision_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
618
self.packs, 'revision_index')[0]
619
revision_nodes = self._pack_collection._index_contents(revision_index_map, revision_keys)
620
# copy revision keys and adjust values
621
self.pb.update("Copying revision texts", 1)
622
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
623
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
624
self.new_pack.revision_index, readv_group_iter, total_items))
625
if 'pack' in debug.debug_flags:
626
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
627
time.ctime(), self._pack_collection._upload_transport.base,
628
self.new_pack.random_name,
629
self.new_pack.revision_index.key_count(),
630
time.time() - self.new_pack.start_time)
631
self._revision_keys = revision_keys
633
def _copy_inventory_texts(self):
634
"""Copy the inventory texts to the new pack.
636
self._revision_keys is used to determine what inventories to copy.
638
Sets self._text_filter appropriately.
640
# select inventory keys
641
inv_keys = self._revision_keys # currently the same keyspace, and note that
642
# querying for keys here could introduce a bug where an inventory item
643
# is missed, so do not change it to query separately without cross
644
# checking like the text key check below.
645
inventory_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
646
self.packs, 'inventory_index')[0]
647
inv_nodes = self._pack_collection._index_contents(inventory_index_map, inv_keys)
648
# copy inventory keys and adjust values
649
# XXX: Should be a helper function to allow different inv representation
651
self.pb.update("Copying inventory texts", 2)
652
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
653
# Only grab the output lines if we will be processing them
654
output_lines = bool(self.revision_ids)
655
inv_lines = self._copy_nodes_graph(inventory_index_map,
656
self.new_pack._writer, self.new_pack.inventory_index,
657
readv_group_iter, total_items, output_lines=output_lines)
658
if self.revision_ids:
659
self._process_inventory_lines(inv_lines)
661
# eat the iterator to cause it to execute.
663
self._text_filter = None
664
if 'pack' in debug.debug_flags:
665
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
666
time.ctime(), self._pack_collection._upload_transport.base,
667
self.new_pack.random_name,
668
self.new_pack.inventory_index.key_count(),
669
time.time() - self.new_pack.start_time)
671
def _copy_text_texts(self):
673
text_index_map, text_nodes = self._get_text_nodes()
674
if self._text_filter is not None:
675
# We could return the keys copied as part of the return value from
676
# _copy_nodes_graph but this doesn't work all that well with the
677
# need to get line output too, so we check separately, and as we're
678
# going to buffer everything anyway, we check beforehand, which
679
# saves reading knit data over the wire when we know there are
681
text_nodes = set(text_nodes)
682
present_text_keys = set(_node[1] for _node in text_nodes)
683
missing_text_keys = set(self._text_filter) - present_text_keys
684
if missing_text_keys:
685
# TODO: raise a specific error that can handle many missing
687
a_missing_key = missing_text_keys.pop()
688
raise errors.RevisionNotPresent(a_missing_key[1],
690
# copy text keys and adjust values
691
self.pb.update("Copying content texts", 3)
692
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
693
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
694
self.new_pack.text_index, readv_group_iter, total_items))
695
self._log_copied_texts()
697
def _check_references(self):
698
"""Make sure our external refereneces are present."""
699
external_refs = self.new_pack._external_compression_parents_of_texts()
701
index = self._pack_collection.text_index.combined_index
702
found_items = list(index.iter_entries(external_refs))
703
if len(found_items) != len(external_refs):
704
found_keys = set(k for idx, k, refs, value in found_items)
705
missing_items = external_refs - found_keys
706
missing_file_id, missing_revision_id = missing_items.pop()
707
raise errors.RevisionNotPresent(missing_revision_id,
710
def _create_pack_from_packs(self):
711
self.pb.update("Opening pack", 0, 5)
712
self.new_pack = self.open_pack()
713
new_pack = self.new_pack
714
# buffer data - we won't be reading-back during the pack creation and
715
# this makes a significant difference on sftp pushes.
716
new_pack.set_write_cache_size(1024*1024)
717
if 'pack' in debug.debug_flags:
718
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
719
for a_pack in self.packs]
720
if self.revision_ids is not None:
721
rev_count = len(self.revision_ids)
724
mutter('%s: create_pack: creating pack from source packs: '
725
'%s%s %s revisions wanted %s t=0',
726
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
727
plain_pack_list, rev_count)
728
self._copy_revision_texts()
729
self._copy_inventory_texts()
730
self._copy_text_texts()
731
# select signature keys
732
signature_filter = self._revision_keys # same keyspace
733
signature_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
734
self.packs, 'signature_index')[0]
735
signature_nodes = self._pack_collection._index_contents(signature_index_map,
737
# copy signature keys and adjust values
738
self.pb.update("Copying signature texts", 4)
739
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
740
new_pack.signature_index)
741
if 'pack' in debug.debug_flags:
742
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
743
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
744
new_pack.signature_index.key_count(),
745
time.time() - new_pack.start_time)
746
self._check_references()
747
if not self._use_pack(new_pack):
750
self.pb.update("Finishing pack", 5)
752
self._pack_collection.allocate(new_pack)
755
def _copy_nodes(self, nodes, index_map, writer, write_index):
756
"""Copy knit nodes between packs with no graph references."""
757
pb = ui.ui_factory.nested_progress_bar()
759
return self._do_copy_nodes(nodes, index_map, writer,
764
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb):
765
# for record verification
766
knit_data = _KnitData(None)
767
# plan a readv on each source pack:
769
nodes = sorted(nodes)
770
# how to map this into knit.py - or knit.py into this?
771
# we don't want the typical knit logic, we want grouping by pack
772
# at this point - perhaps a helper library for the following code
773
# duplication points?
775
for index, key, value in nodes:
776
if index not in request_groups:
777
request_groups[index] = []
778
request_groups[index].append((key, value))
780
pb.update("Copied record", record_index, len(nodes))
781
for index, items in request_groups.iteritems():
782
pack_readv_requests = []
783
for key, value in items:
784
# ---- KnitGraphIndex.get_position
785
bits = value[1:].split(' ')
786
offset, length = int(bits[0]), int(bits[1])
787
pack_readv_requests.append((offset, length, (key, value[0])))
788
# linear scan up the pack
789
pack_readv_requests.sort()
791
transport, path = index_map[index]
792
reader = pack.make_readv_reader(transport, path,
793
[offset[0:2] for offset in pack_readv_requests])
794
for (names, read_func), (_1, _2, (key, eol_flag)) in \
795
izip(reader.iter_records(), pack_readv_requests):
796
raw_data = read_func(None)
797
# check the header only
798
df, _ = knit_data._parse_record_header(key[-1], raw_data)
800
pos, size = writer.add_bytes_record(raw_data, names)
801
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
802
pb.update("Copied record", record_index)
805
def _copy_nodes_graph(self, index_map, writer, write_index,
806
readv_group_iter, total_items, output_lines=False):
807
"""Copy knit nodes between packs.
809
:param output_lines: Return lines present in the copied data as
810
an iterator of line,version_id.
812
pb = ui.ui_factory.nested_progress_bar()
814
for result in self._do_copy_nodes_graph(index_map, writer,
815
write_index, output_lines, pb, readv_group_iter, total_items):
818
# Python 2.4 does not permit try:finally: in a generator.
824
def _do_copy_nodes_graph(self, index_map, writer, write_index,
825
output_lines, pb, readv_group_iter, total_items):
826
# for record verification
827
knit_data = _KnitData(None)
828
# for line extraction when requested (inventories only)
830
factory = knit.KnitPlainFactory()
832
pb.update("Copied record", record_index, total_items)
833
for index, readv_vector, node_vector in readv_group_iter:
835
transport, path = index_map[index]
836
reader = pack.make_readv_reader(transport, path, readv_vector)
837
for (names, read_func), (key, eol_flag, references) in \
838
izip(reader.iter_records(), node_vector):
839
raw_data = read_func(None)
842
# read the entire thing
843
content, _ = knit_data._parse_record(version_id, raw_data)
844
if len(references[-1]) == 0:
845
line_iterator = factory.get_fulltext_content(content)
847
line_iterator = factory.get_linedelta_content(content)
848
for line in line_iterator:
849
yield line, version_id
851
# check the header only
852
df, _ = knit_data._parse_record_header(version_id, raw_data)
854
pos, size = writer.add_bytes_record(raw_data, names)
855
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
856
pb.update("Copied record", record_index)
859
def _get_text_nodes(self):
860
text_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
861
self.packs, 'text_index')[0]
862
return text_index_map, self._pack_collection._index_contents(text_index_map,
865
def _least_readv_node_readv(self, nodes):
866
"""Generate request groups for nodes using the least readv's.
868
:param nodes: An iterable of graph index nodes.
869
:return: Total node count and an iterator of the data needed to perform
870
readvs to obtain the data for nodes. Each item yielded by the
871
iterator is a tuple with:
872
index, readv_vector, node_vector. readv_vector is a list ready to
873
hand to the transport readv method, and node_vector is a list of
874
(key, eol_flag, references) for the the node retrieved by the
875
matching readv_vector.
877
# group by pack so we do one readv per pack
878
nodes = sorted(nodes)
881
for index, key, value, references in nodes:
882
if index not in request_groups:
883
request_groups[index] = []
884
request_groups[index].append((key, value, references))
886
for index, items in request_groups.iteritems():
887
pack_readv_requests = []
888
for key, value, references in items:
889
# ---- KnitGraphIndex.get_position
890
bits = value[1:].split(' ')
891
offset, length = int(bits[0]), int(bits[1])
892
pack_readv_requests.append(
893
((offset, length), (key, value[0], references)))
894
# linear scan up the pack to maximum range combining.
895
pack_readv_requests.sort()
896
# split out the readv and the node data.
897
pack_readv = [readv for readv, node in pack_readv_requests]
898
node_vector = [node for readv, node in pack_readv_requests]
899
result.append((index, pack_readv, node_vector))
902
def _log_copied_texts(self):
903
if 'pack' in debug.debug_flags:
904
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
905
time.ctime(), self._pack_collection._upload_transport.base,
906
self.new_pack.random_name,
907
self.new_pack.text_index.key_count(),
908
time.time() - self.new_pack.start_time)
910
def _process_inventory_lines(self, inv_lines):
911
"""Use up the inv_lines generator and setup a text key filter."""
912
repo = self._pack_collection.repo
913
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
914
inv_lines, self.revision_ids)
916
for fileid, file_revids in fileid_revisions.iteritems():
917
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
918
self._text_filter = text_filter
920
def _revision_node_readv(self, revision_nodes):
921
"""Return the total revisions and the readv's to issue.
923
:param revision_nodes: The revision index contents for the packs being
924
incorporated into the new pack.
925
:return: As per _least_readv_node_readv.
927
return self._least_readv_node_readv(revision_nodes)
929
def _use_pack(self, new_pack):
930
"""Return True if new_pack should be used.
932
:param new_pack: The pack that has just been created.
933
:return: True if the pack should be used.
935
return new_pack.data_inserted()
938
class OptimisingPacker(Packer):
939
"""A packer which spends more time to create better disk layouts."""
941
def _revision_node_readv(self, revision_nodes):
942
"""Return the total revisions and the readv's to issue.
944
This sort places revisions in topological order with the ancestors
947
:param revision_nodes: The revision index contents for the packs being
948
incorporated into the new pack.
949
:return: As per _least_readv_node_readv.
951
# build an ancestors dict
954
for index, key, value, references in revision_nodes:
955
ancestors[key] = references[0]
956
by_key[key] = (index, value, references)
957
order = tsort.topo_sort(ancestors)
959
# Single IO is pathological, but it will work as a starting point.
961
for key in reversed(order):
962
index, value, references = by_key[key]
963
# ---- KnitGraphIndex.get_position
964
bits = value[1:].split(' ')
965
offset, length = int(bits[0]), int(bits[1])
967
(index, [(offset, length)], [(key, value[0], references)]))
968
# TODO: combine requests in the same index that are in ascending order.
969
return total, requests
972
class ReconcilePacker(Packer):
973
"""A packer which regenerates indices etc as it copies.
975
This is used by ``bzr reconcile`` to cause parent text pointers to be
979
def _extra_init(self):
980
self._data_changed = False
982
def _process_inventory_lines(self, inv_lines):
983
"""Generate a text key reference map rather for reconciling with."""
984
repo = self._pack_collection.repo
985
refs = repo._find_text_key_references_from_xml_inventory_lines(
987
self._text_refs = refs
988
# during reconcile we:
989
# - convert unreferenced texts to full texts
990
# - correct texts which reference a text not copied to be full texts
991
# - copy all others as-is but with corrected parents.
992
# - so at this point we don't know enough to decide what becomes a full
994
self._text_filter = None
996
def _copy_text_texts(self):
997
"""generate what texts we should have and then copy."""
998
self.pb.update("Copying content texts", 3)
999
# we have three major tasks here:
1000
# 1) generate the ideal index
1001
repo = self._pack_collection.repo
1002
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1003
_1, key, _2, refs in
1004
self.new_pack.revision_index.iter_all_entries()])
1005
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1006
# 2) generate a text_nodes list that contains all the deltas that can
1007
# be used as-is, with corrected parents.
1010
discarded_nodes = []
1011
NULL_REVISION = _mod_revision.NULL_REVISION
1012
text_index_map, text_nodes = self._get_text_nodes()
1013
for node in text_nodes:
1019
ideal_parents = tuple(ideal_index[node[1]])
1021
discarded_nodes.append(node)
1022
self._data_changed = True
1024
if ideal_parents == (NULL_REVISION,):
1026
if ideal_parents == node[3][0]:
1028
ok_nodes.append(node)
1029
elif ideal_parents[0:1] == node[3][0][0:1]:
1030
# the left most parent is the same, or there are no parents
1031
# today. Either way, we can preserve the representation as
1032
# long as we change the refs to be inserted.
1033
self._data_changed = True
1034
ok_nodes.append((node[0], node[1], node[2],
1035
(ideal_parents, node[3][1])))
1036
self._data_changed = True
1038
# Reinsert this text completely
1039
bad_texts.append((node[1], ideal_parents))
1040
self._data_changed = True
1041
# we're finished with some data.
1044
# 3) bulk copy the ok data
1045
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1046
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1047
self.new_pack.text_index, readv_group_iter, total_items))
1048
# 4) adhoc copy all the other texts.
1049
# We have to topologically insert all texts otherwise we can fail to
1050
# reconcile when parts of a single delta chain are preserved intact,
1051
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1052
# reinserted, and if d3 has incorrect parents it will also be
1053
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1054
# copied), so we will try to delta, but d2 is not currently able to be
1055
# extracted because it's basis d1 is not present. Topologically sorting
1056
# addresses this. The following generates a sort for all the texts that
1057
# are being inserted without having to reference the entire text key
1058
# space (we only topo sort the revisions, which is smaller).
1059
topo_order = tsort.topo_sort(ancestors)
1060
rev_order = dict(zip(topo_order, range(len(topo_order))))
1061
bad_texts.sort(key=lambda key:rev_order[key[0][1]])
1062
transaction = repo.get_transaction()
1063
file_id_index = GraphIndexPrefixAdapter(
1064
self.new_pack.text_index,
1066
add_nodes_callback=self.new_pack.text_index.add_nodes)
1067
knit_index = KnitGraphIndex(file_id_index,
1068
add_callback=file_id_index.add_nodes,
1069
deltas=True, parents=True)
1070
output_knit = knit.KnitVersionedFile('reconcile-texts',
1071
self._pack_collection.transport,
1073
access_method=_PackAccess(
1074
{self.new_pack.text_index:self.new_pack.access_tuple()},
1075
(self.new_pack._writer, self.new_pack.text_index)),
1076
factory=knit.KnitPlainFactory())
1077
for key, parent_keys in bad_texts:
1078
# We refer to the new pack to delta data being output.
1079
# A possible improvement would be to catch errors on short reads
1080
# and only flush then.
1081
self.new_pack.flush()
1083
for parent_key in parent_keys:
1084
if parent_key[0] != key[0]:
1085
# Graph parents must match the fileid
1086
raise errors.BzrError('Mismatched key parent %r:%r' %
1088
parents.append(parent_key[1])
1089
source_weave = repo.weave_store.get_weave(key[0], transaction)
1090
text_lines = source_weave.get_lines(key[1])
1091
# adapt the 'knit' to the current file_id.
1092
file_id_index = GraphIndexPrefixAdapter(
1093
self.new_pack.text_index,
1095
add_nodes_callback=self.new_pack.text_index.add_nodes)
1096
knit_index._graph_index = file_id_index
1097
knit_index._add_callback = file_id_index.add_nodes
1098
output_knit.add_lines_with_ghosts(
1099
key[1], parents, text_lines, random_id=True, check_content=False)
1100
# 5) check that nothing inserted has a reference outside the keyspace.
1101
missing_text_keys = self.new_pack._external_compression_parents_of_texts()
1102
if missing_text_keys:
1103
raise errors.BzrError('Reference to missing compression parents %r'
1104
% (missing_text_keys,))
1105
self._log_copied_texts()
1107
def _use_pack(self, new_pack):
1108
"""Override _use_pack to check for reconcile having changed content."""
1109
# XXX: we might be better checking this at the copy time.
1110
original_inventory_keys = set()
1111
inv_index = self._pack_collection.inventory_index.combined_index
1112
for entry in inv_index.iter_all_entries():
1113
original_inventory_keys.add(entry[1])
1114
new_inventory_keys = set()
1115
for entry in new_pack.inventory_index.iter_all_entries():
1116
new_inventory_keys.add(entry[1])
1117
if new_inventory_keys != original_inventory_keys:
1118
self._data_changed = True
1119
return new_pack.data_inserted() and self._data_changed
1122
class RepositoryPackCollection(object):
1123
"""Management of packs within a repository."""
1125
def __init__(self, repo, transport, index_transport, upload_transport,
1127
"""Create a new RepositoryPackCollection.
1129
:param transport: Addresses the repository base directory
1130
(typically .bzr/repository/).
1131
:param index_transport: Addresses the directory containing indices.
1132
:param upload_transport: Addresses the directory into which packs are written
1133
while they're being created.
1134
:param pack_transport: Addresses the directory of existing complete packs.
1137
self.transport = transport
1138
self._index_transport = index_transport
1139
self._upload_transport = upload_transport
1140
self._pack_transport = pack_transport
1141
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
1144
self._packs_by_name = {}
1145
# the previous pack-names content
1146
self._packs_at_load = None
1147
# when a pack is being created by this object, the state of that pack.
1148
self._new_pack = None
1149
# aggregated revision index data
1150
self.revision_index = AggregateIndex()
1151
self.inventory_index = AggregateIndex()
1152
self.text_index = AggregateIndex()
1153
self.signature_index = AggregateIndex()
1155
def add_pack_to_memory(self, pack):
1156
"""Make a Pack object available to the repository to satisfy queries.
1158
:param pack: A Pack object.
1160
if pack.name in self._packs_by_name:
1161
raise AssertionError()
1162
self.packs.append(pack)
1163
self._packs_by_name[pack.name] = pack
1164
self.revision_index.add_index(pack.revision_index, pack)
1165
self.inventory_index.add_index(pack.inventory_index, pack)
1166
self.text_index.add_index(pack.text_index, pack)
1167
self.signature_index.add_index(pack.signature_index, pack)
1169
def _add_text_to_weave(self, file_id, revision_id, new_lines, parents,
1170
nostore_sha, random_revid):
1171
file_id_index = GraphIndexPrefixAdapter(
1172
self.text_index.combined_index,
1174
add_nodes_callback=self.text_index.add_callback)
1175
self.repo._text_knit._index._graph_index = file_id_index
1176
self.repo._text_knit._index._add_callback = file_id_index.add_nodes
1177
return self.repo._text_knit.add_lines_with_ghosts(
1178
revision_id, parents, new_lines, nostore_sha=nostore_sha,
1179
random_id=random_revid, check_content=False)[0:2]
1181
def all_packs(self):
1182
"""Return a list of all the Pack objects this repository has.
1184
Note that an in-progress pack being created is not returned.
1186
:return: A list of Pack objects for all the packs in the repository.
1189
for name in self.names():
1190
result.append(self.get_pack_by_name(name))
1194
"""Pack the pack collection incrementally.
1196
This will not attempt global reorganisation or recompression,
1197
rather it will just ensure that the total number of packs does
1198
not grow without bound. It uses the _max_pack_count method to
1199
determine if autopacking is needed, and the pack_distribution
1200
method to determine the number of revisions in each pack.
1202
If autopacking takes place then the packs name collection will have
1203
been flushed to disk - packing requires updating the name collection
1204
in synchronisation with certain steps. Otherwise the names collection
1207
:return: True if packing took place.
1209
# XXX: Should not be needed when the management of indices is sane.
1210
total_revisions = self.revision_index.combined_index.key_count()
1211
total_packs = len(self._names)
1212
if self._max_pack_count(total_revisions) >= total_packs:
1214
# XXX: the following may want to be a class, to pack with a given
1216
mutter('Auto-packing repository %s, which has %d pack files, '
1217
'containing %d revisions into %d packs.', self, total_packs,
1218
total_revisions, self._max_pack_count(total_revisions))
1219
# determine which packs need changing
1220
pack_distribution = self.pack_distribution(total_revisions)
1222
for pack in self.all_packs():
1223
revision_count = pack.get_revision_count()
1224
if revision_count == 0:
1225
# revision less packs are not generated by normal operation,
1226
# only by operations like sign-my-commits, and thus will not
1227
# tend to grow rapdily or without bound like commit containing
1228
# packs do - leave them alone as packing them really should
1229
# group their data with the relevant commit, and that may
1230
# involve rewriting ancient history - which autopack tries to
1231
# avoid. Alternatively we could not group the data but treat
1232
# each of these as having a single revision, and thus add
1233
# one revision for each to the total revision count, to get
1234
# a matching distribution.
1236
existing_packs.append((revision_count, pack))
1237
pack_operations = self.plan_autopack_combinations(
1238
existing_packs, pack_distribution)
1239
self._execute_pack_operations(pack_operations)
1242
def _execute_pack_operations(self, pack_operations, _packer_class=Packer):
1243
"""Execute a series of pack operations.
1245
:param pack_operations: A list of [revision_count, packs_to_combine].
1246
:param _packer_class: The class of packer to use (default: Packer).
1249
for revision_count, packs in pack_operations:
1250
# we may have no-ops from the setup logic
1253
_packer_class(self, packs, '.autopack').pack()
1255
self._remove_pack_from_memory(pack)
1256
# record the newly available packs and stop advertising the old
1258
self._save_pack_names(clear_obsolete_packs=True)
1259
# Move the old packs out of the way now they are no longer referenced.
1260
for revision_count, packs in pack_operations:
1261
self._obsolete_packs(packs)
1263
def lock_names(self):
1264
"""Acquire the mutex around the pack-names index.
1266
This cannot be used in the middle of a read-only transaction on the
1269
self.repo.control_files.lock_write()
1272
"""Pack the pack collection totally."""
1273
self.ensure_loaded()
1274
total_packs = len(self._names)
1276
# This is arguably wrong because we might not be optimal, but for
1277
# now lets leave it in. (e.g. reconcile -> one pack. But not
1280
total_revisions = self.revision_index.combined_index.key_count()
1281
# XXX: the following may want to be a class, to pack with a given
1283
mutter('Packing repository %s, which has %d pack files, '
1284
'containing %d revisions into 1 packs.', self, total_packs,
1286
# determine which packs need changing
1287
pack_distribution = [1]
1288
pack_operations = [[0, []]]
1289
for pack in self.all_packs():
1290
pack_operations[-1][0] += pack.get_revision_count()
1291
pack_operations[-1][1].append(pack)
1292
self._execute_pack_operations(pack_operations, OptimisingPacker)
1294
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1295
"""Plan a pack operation.
1297
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1299
:param pack_distribution: A list with the number of revisions desired
1302
if len(existing_packs) <= len(pack_distribution):
1304
existing_packs.sort(reverse=True)
1305
pack_operations = [[0, []]]
1306
# plan out what packs to keep, and what to reorganise
1307
while len(existing_packs):
1308
# take the largest pack, and if its less than the head of the
1309
# distribution chart we will include its contents in the new pack for
1310
# that position. If its larger, we remove its size from the
1311
# distribution chart
1312
next_pack_rev_count, next_pack = existing_packs.pop(0)
1313
if next_pack_rev_count >= pack_distribution[0]:
1314
# this is already packed 'better' than this, so we can
1315
# not waste time packing it.
1316
while next_pack_rev_count > 0:
1317
next_pack_rev_count -= pack_distribution[0]
1318
if next_pack_rev_count >= 0:
1320
del pack_distribution[0]
1322
# didn't use that entire bucket up
1323
pack_distribution[0] = -next_pack_rev_count
1325
# add the revisions we're going to add to the next output pack
1326
pack_operations[-1][0] += next_pack_rev_count
1327
# allocate this pack to the next pack sub operation
1328
pack_operations[-1][1].append(next_pack)
1329
if pack_operations[-1][0] >= pack_distribution[0]:
1330
# this pack is used up, shift left.
1331
del pack_distribution[0]
1332
pack_operations.append([0, []])
1334
return pack_operations
1336
def ensure_loaded(self):
1337
# NB: if you see an assertion error here, its probably access against
1338
# an unlocked repo. Naughty.
1339
if not self.repo.is_locked():
1340
raise errors.ObjectNotLocked(self.repo)
1341
if self._names is None:
1343
self._packs_at_load = set()
1344
for index, key, value in self._iter_disk_pack_index():
1346
self._names[name] = self._parse_index_sizes(value)
1347
self._packs_at_load.add((key, value))
1348
# populate all the metadata.
1351
def _parse_index_sizes(self, value):
1352
"""Parse a string of index sizes."""
1353
return tuple([int(digits) for digits in value.split(' ')])
1355
def get_pack_by_name(self, name):
1356
"""Get a Pack object by name.
1358
:param name: The name of the pack - e.g. '123456'
1359
:return: A Pack object.
1362
return self._packs_by_name[name]
1364
rev_index = self._make_index(name, '.rix')
1365
inv_index = self._make_index(name, '.iix')
1366
txt_index = self._make_index(name, '.tix')
1367
sig_index = self._make_index(name, '.six')
1368
result = ExistingPack(self._pack_transport, name, rev_index,
1369
inv_index, txt_index, sig_index)
1370
self.add_pack_to_memory(result)
1373
def allocate(self, a_new_pack):
1374
"""Allocate name in the list of packs.
1376
:param a_new_pack: A NewPack instance to be added to the collection of
1377
packs for this repository.
1379
self.ensure_loaded()
1380
if a_new_pack.name in self._names:
1381
raise errors.BzrError(
1382
'Pack %r already exists in %s' % (a_new_pack.name, self))
1383
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1384
self.add_pack_to_memory(a_new_pack)
1386
def _iter_disk_pack_index(self):
1387
"""Iterate over the contents of the pack-names index.
1389
This is used when loading the list from disk, and before writing to
1390
detect updates from others during our write operation.
1391
:return: An iterator of the index contents.
1393
return GraphIndex(self.transport, 'pack-names', None
1394
).iter_all_entries()
1396
def _make_index(self, name, suffix):
1397
size_offset = self._suffix_offsets[suffix]
1398
index_name = name + suffix
1399
index_size = self._names[name][size_offset]
1401
self._index_transport, index_name, index_size)
1403
def _max_pack_count(self, total_revisions):
1404
"""Return the maximum number of packs to use for total revisions.
1406
:param total_revisions: The total number of revisions in the
1409
if not total_revisions:
1411
digits = str(total_revisions)
1413
for digit in digits:
1414
result += int(digit)
1418
"""Provide an order to the underlying names."""
1419
return sorted(self._names.keys())
1421
def _obsolete_packs(self, packs):
1422
"""Move a number of packs which have been obsoleted out of the way.
1424
Each pack and its associated indices are moved out of the way.
1426
Note: for correctness this function should only be called after a new
1427
pack names index has been written without these pack names, and with
1428
the names of packs that contain the data previously available via these
1431
:param packs: The packs to obsolete.
1432
:param return: None.
1435
pack.pack_transport.rename(pack.file_name(),
1436
'../obsolete_packs/' + pack.file_name())
1437
# TODO: Probably needs to know all possible indices for this pack
1438
# - or maybe list the directory and move all indices matching this
1439
# name whether we recognize it or not?
1440
for suffix in ('.iix', '.six', '.tix', '.rix'):
1441
self._index_transport.rename(pack.name + suffix,
1442
'../obsolete_packs/' + pack.name + suffix)
1444
def pack_distribution(self, total_revisions):
1445
"""Generate a list of the number of revisions to put in each pack.
1447
:param total_revisions: The total number of revisions in the
1450
if total_revisions == 0:
1452
digits = reversed(str(total_revisions))
1454
for exponent, count in enumerate(digits):
1455
size = 10 ** exponent
1456
for pos in range(int(count)):
1458
return list(reversed(result))
1460
def _pack_tuple(self, name):
1461
"""Return a tuple with the transport and file name for a pack name."""
1462
return self._pack_transport, name + '.pack'
1464
def _remove_pack_from_memory(self, pack):
1465
"""Remove pack from the packs accessed by this repository.
1467
Only affects memory state, until self._save_pack_names() is invoked.
1469
self._names.pop(pack.name)
1470
self._packs_by_name.pop(pack.name)
1471
self._remove_pack_indices(pack)
1473
def _remove_pack_indices(self, pack):
1474
"""Remove the indices for pack from the aggregated indices."""
1475
self.revision_index.remove_index(pack.revision_index, pack)
1476
self.inventory_index.remove_index(pack.inventory_index, pack)
1477
self.text_index.remove_index(pack.text_index, pack)
1478
self.signature_index.remove_index(pack.signature_index, pack)
1481
"""Clear all cached data."""
1482
# cached revision data
1483
self.repo._revision_knit = None
1484
self.revision_index.clear()
1485
# cached signature data
1486
self.repo._signature_knit = None
1487
self.signature_index.clear()
1488
# cached file text data
1489
self.text_index.clear()
1490
self.repo._text_knit = None
1491
# cached inventory data
1492
self.inventory_index.clear()
1493
# remove the open pack
1494
self._new_pack = None
1495
# information about packs.
1498
self._packs_by_name = {}
1499
self._packs_at_load = None
1501
def _make_index_map(self, index_suffix):
1502
"""Return information on existing indices.
1504
:param suffix: Index suffix added to pack name.
1506
:returns: (pack_map, indices) where indices is a list of GraphIndex
1507
objects, and pack_map is a mapping from those objects to the
1508
pack tuple they describe.
1510
# TODO: stop using this; it creates new indices unnecessarily.
1511
self.ensure_loaded()
1512
suffix_map = {'.rix': 'revision_index',
1513
'.six': 'signature_index',
1514
'.iix': 'inventory_index',
1515
'.tix': 'text_index',
1517
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1518
suffix_map[index_suffix])
1520
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1521
"""Convert a list of packs to an index pack map and index list.
1523
:param packs: The packs list to process.
1524
:param index_attribute: The attribute that the desired index is found
1526
:return: A tuple (map, list) where map contains the dict from
1527
index:pack_tuple, and lsit contains the indices in the same order
1533
index = getattr(pack, index_attribute)
1534
indices.append(index)
1535
pack_map[index] = (pack.pack_transport, pack.file_name())
1536
return pack_map, indices
1538
def _index_contents(self, pack_map, key_filter=None):
1539
"""Get an iterable of the index contents from a pack_map.
1541
:param pack_map: A map from indices to pack details.
1542
:param key_filter: An optional filter to limit the
1545
indices = [index for index in pack_map.iterkeys()]
1546
all_index = CombinedGraphIndex(indices)
1547
if key_filter is None:
1548
return all_index.iter_all_entries()
1550
return all_index.iter_entries(key_filter)
1552
def _unlock_names(self):
1553
"""Release the mutex around the pack-names index."""
1554
self.repo.control_files.unlock()
1556
def _save_pack_names(self, clear_obsolete_packs=False):
1557
"""Save the list of packs.
1559
This will take out the mutex around the pack names list for the
1560
duration of the method call. If concurrent updates have been made, a
1561
three-way merge between the current list and the current in memory list
1564
:param clear_obsolete_packs: If True, clear out the contents of the
1565
obsolete_packs directory.
1569
builder = GraphIndexBuilder()
1570
# load the disk nodes across
1572
for index, key, value in self._iter_disk_pack_index():
1573
disk_nodes.add((key, value))
1574
# do a two-way diff against our original content
1575
current_nodes = set()
1576
for name, sizes in self._names.iteritems():
1578
((name, ), ' '.join(str(size) for size in sizes)))
1579
deleted_nodes = self._packs_at_load - current_nodes
1580
new_nodes = current_nodes - self._packs_at_load
1581
disk_nodes.difference_update(deleted_nodes)
1582
disk_nodes.update(new_nodes)
1583
# TODO: handle same-name, index-size-changes here -
1584
# e.g. use the value from disk, not ours, *unless* we're the one
1586
for key, value in disk_nodes:
1587
builder.add_node(key, value)
1588
self.transport.put_file('pack-names', builder.finish(),
1589
mode=self.repo.bzrdir._get_file_mode())
1590
# move the baseline forward
1591
self._packs_at_load = disk_nodes
1592
if clear_obsolete_packs:
1593
self._clear_obsolete_packs()
1595
self._unlock_names()
1596
# synchronise the memory packs list with what we just wrote:
1597
new_names = dict(disk_nodes)
1598
# drop no longer present nodes
1599
for pack in self.all_packs():
1600
if (pack.name,) not in new_names:
1601
self._remove_pack_from_memory(pack)
1602
# add new nodes/refresh existing ones
1603
for key, value in disk_nodes:
1605
sizes = self._parse_index_sizes(value)
1606
if name in self._names:
1608
if sizes != self._names[name]:
1609
# the pack for name has had its indices replaced - rare but
1610
# important to handle. XXX: probably can never happen today
1611
# because the three-way merge code above does not handle it
1612
# - you may end up adding the same key twice to the new
1613
# disk index because the set values are the same, unless
1614
# the only index shows up as deleted by the set difference
1615
# - which it may. Until there is a specific test for this,
1616
# assume its broken. RBC 20071017.
1617
self._remove_pack_from_memory(self.get_pack_by_name(name))
1618
self._names[name] = sizes
1619
self.get_pack_by_name(name)
1622
self._names[name] = sizes
1623
self.get_pack_by_name(name)
1625
def _clear_obsolete_packs(self):
1626
"""Delete everything from the obsolete-packs directory.
1628
obsolete_pack_transport = self.transport.clone('obsolete_packs')
1629
for filename in obsolete_pack_transport.list_dir('.'):
1631
obsolete_pack_transport.delete(filename)
1632
except (errors.PathError, errors.TransportError), e:
1633
warning("couldn't delete obsolete pack, skipping it:\n%s" % (e,))
1635
def _start_write_group(self):
1636
# Do not permit preparation for writing if we're not in a 'write lock'.
1637
if not self.repo.is_write_locked():
1638
raise errors.NotWriteLocked(self)
1639
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1640
self._pack_transport, upload_suffix='.pack',
1641
file_mode=self.repo.bzrdir._get_file_mode())
1642
# allow writing: queue writes to a new index
1643
self.revision_index.add_writable_index(self._new_pack.revision_index,
1645
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
1647
self.text_index.add_writable_index(self._new_pack.text_index,
1649
self.signature_index.add_writable_index(self._new_pack.signature_index,
1652
# reused revision and signature knits may need updating
1654
# "Hysterical raisins. client code in bzrlib grabs those knits outside
1655
# of write groups and then mutates it inside the write group."
1656
if self.repo._revision_knit is not None:
1657
self.repo._revision_knit._index._add_callback = \
1658
self.revision_index.add_callback
1659
if self.repo._signature_knit is not None:
1660
self.repo._signature_knit._index._add_callback = \
1661
self.signature_index.add_callback
1662
# create a reused knit object for text addition in commit.
1663
self.repo._text_knit = self.repo.weave_store.get_weave_or_empty(
1666
def _abort_write_group(self):
1667
# FIXME: just drop the transient index.
1668
# forget what names there are
1669
if self._new_pack is not None:
1670
self._new_pack.abort()
1671
self._remove_pack_indices(self._new_pack)
1672
self._new_pack = None
1673
self.repo._text_knit = None
1675
def _commit_write_group(self):
1676
self._remove_pack_indices(self._new_pack)
1677
if self._new_pack.data_inserted():
1678
# get all the data to disk and read to use
1679
self._new_pack.finish()
1680
self.allocate(self._new_pack)
1681
self._new_pack = None
1682
if not self.autopack():
1683
# when autopack takes no steps, the names list is still
1685
self._save_pack_names()
1687
self._new_pack.abort()
1688
self._new_pack = None
1689
self.repo._text_knit = None
1692
class KnitPackRevisionStore(KnitRevisionStore):
1693
"""An object to adapt access from RevisionStore's to use KnitPacks.
1695
This class works by replacing the original RevisionStore.
1696
We need to do this because the KnitPackRevisionStore is less
1697
isolated in its layering - it uses services from the repo.
1700
def __init__(self, repo, transport, revisionstore):
1701
"""Create a KnitPackRevisionStore on repo with revisionstore.
1703
This will store its state in the Repository, use the
1704
indices to provide a KnitGraphIndex,
1705
and at the end of transactions write new indices.
1707
KnitRevisionStore.__init__(self, revisionstore.versioned_file_store)
1709
self._serializer = revisionstore._serializer
1710
self.transport = transport
1712
def get_revision_file(self, transaction):
1713
"""Get the revision versioned file object."""
1714
if getattr(self.repo, '_revision_knit', None) is not None:
1715
return self.repo._revision_knit
1716
self.repo._pack_collection.ensure_loaded()
1717
add_callback = self.repo._pack_collection.revision_index.add_callback
1718
# setup knit specific objects
1719
knit_index = KnitGraphIndex(
1720
self.repo._pack_collection.revision_index.combined_index,
1721
add_callback=add_callback)
1722
self.repo._revision_knit = knit.KnitVersionedFile(
1723
'revisions', self.transport.clone('..'),
1724
self.repo.bzrdir._get_file_mode(),
1726
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1727
access_method=self.repo._pack_collection.revision_index.knit_access)
1728
return self.repo._revision_knit
1730
def get_signature_file(self, transaction):
1731
"""Get the signature versioned file object."""
1732
if getattr(self.repo, '_signature_knit', None) is not None:
1733
return self.repo._signature_knit
1734
self.repo._pack_collection.ensure_loaded()
1735
add_callback = self.repo._pack_collection.signature_index.add_callback
1736
# setup knit specific objects
1737
knit_index = KnitGraphIndex(
1738
self.repo._pack_collection.signature_index.combined_index,
1739
add_callback=add_callback, parents=False)
1740
self.repo._signature_knit = knit.KnitVersionedFile(
1741
'signatures', self.transport.clone('..'),
1742
self.repo.bzrdir._get_file_mode(),
1744
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1745
access_method=self.repo._pack_collection.signature_index.knit_access)
1746
return self.repo._signature_knit
1749
class KnitPackTextStore(VersionedFileStore):
1750
"""Presents a TextStore abstraction on top of packs.
1752
This class works by replacing the original VersionedFileStore.
1753
We need to do this because the KnitPackRevisionStore is less
1754
isolated in its layering - it uses services from the repo and shares them
1755
with all the data written in a single write group.
1758
def __init__(self, repo, transport, weavestore):
1759
"""Create a KnitPackTextStore on repo with weavestore.
1761
This will store its state in the Repository, use the
1762
indices FileNames to provide a KnitGraphIndex,
1763
and at the end of transactions write new indices.
1765
# don't call base class constructor - it's not suitable.
1766
# no transient data stored in the transaction
1768
self._precious = False
1770
self.transport = transport
1771
self.weavestore = weavestore
1772
# XXX for check() which isn't updated yet
1773
self._transport = weavestore._transport
1775
def get_weave_or_empty(self, file_id, transaction):
1776
"""Get a 'Knit' backed by the .tix indices.
1778
The transaction parameter is ignored.
1780
self.repo._pack_collection.ensure_loaded()
1781
add_callback = self.repo._pack_collection.text_index.add_callback
1782
# setup knit specific objects
1783
file_id_index = GraphIndexPrefixAdapter(
1784
self.repo._pack_collection.text_index.combined_index,
1785
(file_id, ), 1, add_nodes_callback=add_callback)
1786
knit_index = KnitGraphIndex(file_id_index,
1787
add_callback=file_id_index.add_nodes,
1788
deltas=True, parents=True)
1789
return knit.KnitVersionedFile('text:' + file_id,
1790
self.transport.clone('..'),
1793
access_method=self.repo._pack_collection.text_index.knit_access,
1794
factory=knit.KnitPlainFactory())
1796
get_weave = get_weave_or_empty
1799
"""Generate a list of the fileids inserted, for use by check."""
1800
self.repo._pack_collection.ensure_loaded()
1802
for index, key, value, refs in \
1803
self.repo._pack_collection.text_index.combined_index.iter_all_entries():
1808
class InventoryKnitThunk(object):
1809
"""An object to manage thunking get_inventory_weave to pack based knits."""
1811
def __init__(self, repo, transport):
1812
"""Create an InventoryKnitThunk for repo at transport.
1814
This will store its state in the Repository, use the
1815
indices FileNames to provide a KnitGraphIndex,
1816
and at the end of transactions write a new index..
1819
self.transport = transport
1821
def get_weave(self):
1822
"""Get a 'Knit' that contains inventory data."""
1823
self.repo._pack_collection.ensure_loaded()
1824
add_callback = self.repo._pack_collection.inventory_index.add_callback
1825
# setup knit specific objects
1826
knit_index = KnitGraphIndex(
1827
self.repo._pack_collection.inventory_index.combined_index,
1828
add_callback=add_callback, deltas=True, parents=True)
1829
return knit.KnitVersionedFile(
1830
'inventory', self.transport.clone('..'),
1831
self.repo.bzrdir._get_file_mode(),
1833
index=knit_index, delta=True, factory=knit.KnitPlainFactory(),
1834
access_method=self.repo._pack_collection.inventory_index.knit_access)
1837
class KnitPackRepository(KnitRepository):
1838
"""Repository with knit objects stored inside pack containers."""
1840
def __init__(self, _format, a_bzrdir, control_files, _revision_store,
1841
control_store, text_store, _commit_builder_class, _serializer):
1842
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
1843
_revision_store, control_store, text_store, _commit_builder_class,
1845
index_transport = self._transport.clone('indices')
1846
self._pack_collection = RepositoryPackCollection(self,
1849
self._transport.clone('upload'),
1850
self._transport.clone('packs'))
1851
self._revision_store = KnitPackRevisionStore(self, index_transport, self._revision_store)
1852
self.weave_store = KnitPackTextStore(self, index_transport, self.weave_store)
1853
self._inv_thunk = InventoryKnitThunk(self, index_transport)
1854
# True when the repository object is 'write locked' (as opposed to the
1855
# physical lock only taken out around changes to the pack-names list.)
1856
# Another way to represent this would be a decorator around the control
1857
# files object that presents logical locks as physical ones - if this
1858
# gets ugly consider that alternative design. RBC 20071011
1859
self._write_lock_count = 0
1860
self._transaction = None
1862
self._reconcile_does_inventory_gc = True
1863
self._reconcile_fixes_text_parents = True
1864
self._reconcile_backsup_inventory = False
1866
def _abort_write_group(self):
1867
self._pack_collection._abort_write_group()
1869
def _find_inconsistent_revision_parents(self):
1870
"""Find revisions with incorrectly cached parents.
1872
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1873
parents-in-revision).
1875
if not self.is_locked():
1876
raise errors.ObjectNotLocked(self)
1877
pb = ui.ui_factory.nested_progress_bar()
1880
revision_nodes = self._pack_collection.revision_index \
1881
.combined_index.iter_all_entries()
1882
index_positions = []
1883
# Get the cached index values for all revisions, and also the location
1884
# in each index of the revision text so we can perform linear IO.
1885
for index, key, value, refs in revision_nodes:
1886
pos, length = value[1:].split(' ')
1887
index_positions.append((index, int(pos), key[0],
1888
tuple(parent[0] for parent in refs[0])))
1889
pb.update("Reading revision index.", 0, 0)
1890
index_positions.sort()
1891
batch_count = len(index_positions) / 1000 + 1
1892
pb.update("Checking cached revision graph.", 0, batch_count)
1893
for offset in xrange(batch_count):
1894
pb.update("Checking cached revision graph.", offset)
1895
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1898
rev_ids = [item[2] for item in to_query]
1899
revs = self.get_revisions(rev_ids)
1900
for revision, item in zip(revs, to_query):
1901
index_parents = item[3]
1902
rev_parents = tuple(revision.parent_ids)
1903
if index_parents != rev_parents:
1904
result.append((revision.revision_id, index_parents, rev_parents))
1909
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
1910
def get_parents(self, revision_ids):
1911
"""See graph._StackedParentsProvider.get_parents."""
1912
parent_map = self.get_parent_map(revision_ids)
1913
return [parent_map.get(r, None) for r in revision_ids]
1915
def get_parent_map(self, keys):
1916
"""See graph._StackedParentsProvider.get_parent_map
1918
This implementation accesses the combined revision index to provide
1921
self._pack_collection.ensure_loaded()
1922
index = self._pack_collection.revision_index.combined_index
1925
raise ValueError('get_parent_map(None) is not valid')
1926
if _mod_revision.NULL_REVISION in keys:
1927
keys.discard(_mod_revision.NULL_REVISION)
1928
found_parents = {_mod_revision.NULL_REVISION:()}
1931
search_keys = set((revision_id,) for revision_id in keys)
1932
for index, key, value, refs in index.iter_entries(search_keys):
1935
parents = (_mod_revision.NULL_REVISION,)
1937
parents = tuple(parent[0] for parent in parents)
1938
found_parents[key[0]] = parents
1939
return found_parents
1941
def has_revisions(self, revision_ids):
1942
"""See Repository.has_revisions()."""
1943
revision_ids = set(revision_ids)
1944
result = revision_ids.intersection(
1945
set([None, _mod_revision.NULL_REVISION]))
1946
revision_ids.difference_update(result)
1947
index = self._pack_collection.revision_index.combined_index
1948
keys = [(revision_id,) for revision_id in revision_ids]
1949
result.update(node[1][0] for node in index.iter_entries(keys))
1952
def _make_parents_provider(self):
1953
return graph.CachingParentsProvider(self)
1955
def _refresh_data(self):
1956
if self._write_lock_count == 1 or (
1957
self.control_files._lock_count == 1 and
1958
self.control_files._lock_mode == 'r'):
1959
# forget what names there are
1960
self._pack_collection.reset()
1961
# XXX: Better to do an in-memory merge when acquiring a new lock -
1962
# factor out code from _save_pack_names.
1963
self._pack_collection.ensure_loaded()
1965
def _start_write_group(self):
1966
self._pack_collection._start_write_group()
1968
def _commit_write_group(self):
1969
return self._pack_collection._commit_write_group()
1971
def get_inventory_weave(self):
1972
return self._inv_thunk.get_weave()
1974
def get_transaction(self):
1975
if self._write_lock_count:
1976
return self._transaction
1978
return self.control_files.get_transaction()
1980
def is_locked(self):
1981
return self._write_lock_count or self.control_files.is_locked()
1983
def is_write_locked(self):
1984
return self._write_lock_count
1986
def lock_write(self, token=None):
1987
if not self._write_lock_count and self.is_locked():
1988
raise errors.ReadOnlyError(self)
1989
self._write_lock_count += 1
1990
if self._write_lock_count == 1:
1991
from bzrlib import transactions
1992
self._transaction = transactions.WriteTransaction()
1993
self._refresh_data()
1995
def lock_read(self):
1996
if self._write_lock_count:
1997
self._write_lock_count += 1
1999
self.control_files.lock_read()
2000
self._refresh_data()
2002
def leave_lock_in_place(self):
2003
# not supported - raise an error
2004
raise NotImplementedError(self.leave_lock_in_place)
2006
def dont_leave_lock_in_place(self):
2007
# not supported - raise an error
2008
raise NotImplementedError(self.dont_leave_lock_in_place)
2012
"""Compress the data within the repository.
2014
This will pack all the data to a single pack. In future it may
2015
recompress deltas or do other such expensive operations.
2017
self._pack_collection.pack()
2020
def reconcile(self, other=None, thorough=False):
2021
"""Reconcile this repository."""
2022
from bzrlib.reconcile import PackReconciler
2023
reconciler = PackReconciler(self, thorough=thorough)
2024
reconciler.reconcile()
2028
if self._write_lock_count == 1 and self._write_group is not None:
2029
self.abort_write_group()
2030
self._transaction = None
2031
self._write_lock_count = 0
2032
raise errors.BzrError(
2033
'Must end write group before releasing write lock on %s'
2035
if self._write_lock_count:
2036
self._write_lock_count -= 1
2037
if not self._write_lock_count:
2038
transaction = self._transaction
2039
self._transaction = None
2040
transaction.finish()
2042
self.control_files.unlock()
2045
class RepositoryFormatPack(MetaDirRepositoryFormat):
2046
"""Format logic for pack structured repositories.
2048
This repository format has:
2049
- a list of packs in pack-names
2050
- packs in packs/NAME.pack
2051
- indices in indices/NAME.{iix,six,tix,rix}
2052
- knit deltas in the packs, knit indices mapped to the indices.
2053
- thunk objects to support the knits programming API.
2054
- a format marker of its own
2055
- an optional 'shared-storage' flag
2056
- an optional 'no-working-trees' flag
2060
# Set this attribute in derived classes to control the repository class
2061
# created by open and initialize.
2062
repository_class = None
2063
# Set this attribute in derived classes to control the
2064
# _commit_builder_class that the repository objects will have passed to
2065
# their constructor.
2066
_commit_builder_class = None
2067
# Set this attribute in derived clases to control the _serializer that the
2068
# repository objects will have passed to their constructor.
2070
# External references are not supported in pack repositories yet.
2071
supports_external_lookups = False
2073
def _get_control_store(self, repo_transport, control_files):
2074
"""Return the control store for this repository."""
2075
return VersionedFileStore(
2078
file_mode=control_files._file_mode,
2079
versionedfile_class=knit.make_file_knit,
2080
versionedfile_kwargs={'factory': knit.KnitPlainFactory()},
2083
def _get_revision_store(self, repo_transport, control_files):
2084
"""See RepositoryFormat._get_revision_store()."""
2085
versioned_file_store = VersionedFileStore(
2087
file_mode=control_files._file_mode,
2090
versionedfile_class=knit.make_file_knit,
2091
versionedfile_kwargs={'delta': False,
2092
'factory': knit.KnitPlainFactory(),
2096
return KnitRevisionStore(versioned_file_store)
2098
def _get_text_store(self, transport, control_files):
2099
"""See RepositoryFormat._get_text_store()."""
2100
return self._get_versioned_file_store('knits',
2103
versionedfile_class=knit.make_file_knit,
2104
versionedfile_kwargs={
2105
'create_parent_dir': True,
2106
'delay_create': True,
2107
'dir_mode': control_files._dir_mode,
2111
def initialize(self, a_bzrdir, shared=False):
2112
"""Create a pack based repository.
2114
:param a_bzrdir: bzrdir to contain the new repository; must already
2116
:param shared: If true the repository will be initialized as a shared
2119
mutter('creating repository in %s.', a_bzrdir.transport.base)
2120
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
2121
builder = GraphIndexBuilder()
2122
files = [('pack-names', builder.finish())]
2123
utf8_files = [('format', self.get_format_string())]
2125
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
2126
return self.open(a_bzrdir=a_bzrdir, _found=True)
2128
def open(self, a_bzrdir, _found=False, _override_transport=None):
2129
"""See RepositoryFormat.open().
2131
:param _override_transport: INTERNAL USE ONLY. Allows opening the
2132
repository at a slightly different url
2133
than normal. I.e. during 'upgrade'.
2136
format = RepositoryFormat.find_format(a_bzrdir)
2137
if _override_transport is not None:
2138
repo_transport = _override_transport
2140
repo_transport = a_bzrdir.get_repository_transport(None)
2141
control_files = lockable_files.LockableFiles(repo_transport,
2142
'lock', lockdir.LockDir)
2143
text_store = self._get_text_store(repo_transport, control_files)
2144
control_store = self._get_control_store(repo_transport, control_files)
2145
_revision_store = self._get_revision_store(repo_transport, control_files)
2146
return self.repository_class(_format=self,
2148
control_files=control_files,
2149
_revision_store=_revision_store,
2150
control_store=control_store,
2151
text_store=text_store,
2152
_commit_builder_class=self._commit_builder_class,
2153
_serializer=self._serializer)
2156
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2157
"""A no-subtrees parameterized Pack repository.
2159
This format was introduced in 0.92.
2162
repository_class = KnitPackRepository
2163
_commit_builder_class = PackCommitBuilder
2164
_serializer = xml5.serializer_v5
2166
def _get_matching_bzrdir(self):
2167
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2169
def _ignore_setting_bzrdir(self, format):
2172
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2174
def get_format_string(self):
2175
"""See RepositoryFormat.get_format_string()."""
2176
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2178
def get_format_description(self):
2179
"""See RepositoryFormat.get_format_description()."""
2180
return "Packs containing knits without subtree support"
2182
def check_conversion_target(self, target_format):
2186
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2187
"""A subtrees parameterized Pack repository.
2189
This repository format uses the xml7 serializer to get:
2190
- support for recording full info about the tree root
2191
- support for recording tree-references
2193
This format was introduced in 0.92.
2196
repository_class = KnitPackRepository
2197
_commit_builder_class = PackRootCommitBuilder
2198
rich_root_data = True
2199
supports_tree_reference = True
2200
_serializer = xml7.serializer_v7
2202
def _get_matching_bzrdir(self):
2203
return bzrdir.format_registry.make_bzrdir(
2204
'pack-0.92-subtree')
2206
def _ignore_setting_bzrdir(self, format):
2209
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2211
def check_conversion_target(self, target_format):
2212
if not target_format.rich_root_data:
2213
raise errors.BadConversionTarget(
2214
'Does not support rich root data.', target_format)
2215
if not getattr(target_format, 'supports_tree_reference', False):
2216
raise errors.BadConversionTarget(
2217
'Does not support nested trees', target_format)
2219
def get_format_string(self):
2220
"""See RepositoryFormat.get_format_string()."""
2221
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2223
def get_format_description(self):
2224
"""See RepositoryFormat.get_format_description()."""
2225
return "Packs containing knits with subtree support\n"
2228
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2229
"""A rich-root, no subtrees parameterized Pack repository.
2231
This repository format uses the xml6 serializer to get:
2232
- support for recording full info about the tree root
2234
This format was introduced in 1.0.
2237
repository_class = KnitPackRepository
2238
_commit_builder_class = PackRootCommitBuilder
2239
rich_root_data = True
2240
supports_tree_reference = False
2241
_serializer = xml6.serializer_v6
2243
def _get_matching_bzrdir(self):
2244
return bzrdir.format_registry.make_bzrdir(
2247
def _ignore_setting_bzrdir(self, format):
2250
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2252
def check_conversion_target(self, target_format):
2253
if not target_format.rich_root_data:
2254
raise errors.BadConversionTarget(
2255
'Does not support rich root data.', target_format)
2257
def get_format_string(self):
2258
"""See RepositoryFormat.get_format_string()."""
2259
return ("Bazaar pack repository format 1 with rich root"
2260
" (needs bzr 1.0)\n")
2262
def get_format_description(self):
2263
"""See RepositoryFormat.get_format_description()."""
2264
return "Packs containing knits with rich root support\n"
2267
class RepositoryFormatPackDevelopment0(RepositoryFormatPack):
2268
"""A no-subtrees development repository.
2270
This format should be retained until the second release after bzr 1.0.
2272
No changes to the disk behaviour from pack-0.92.
2275
repository_class = KnitPackRepository
2276
_commit_builder_class = PackCommitBuilder
2277
_serializer = xml5.serializer_v5
2279
def _get_matching_bzrdir(self):
2280
return bzrdir.format_registry.make_bzrdir('development0')
2282
def _ignore_setting_bzrdir(self, format):
2285
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2287
def get_format_string(self):
2288
"""See RepositoryFormat.get_format_string()."""
2289
return "Bazaar development format 0 (needs bzr.dev from before 1.3)\n"
2291
def get_format_description(self):
2292
"""See RepositoryFormat.get_format_description()."""
2293
return ("Development repository format, currently the same as "
2296
def check_conversion_target(self, target_format):
2300
class RepositoryFormatPackDevelopment0Subtree(RepositoryFormatPack):
2301
"""A subtrees development repository.
2303
This format should be retained until the second release after bzr 1.0.
2305
No changes to the disk behaviour from pack-0.92-subtree.
2308
repository_class = KnitPackRepository
2309
_commit_builder_class = PackRootCommitBuilder
2310
rich_root_data = True
2311
supports_tree_reference = True
2312
_serializer = xml7.serializer_v7
2314
def _get_matching_bzrdir(self):
2315
return bzrdir.format_registry.make_bzrdir(
2316
'development0-subtree')
2318
def _ignore_setting_bzrdir(self, format):
2321
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2323
def check_conversion_target(self, target_format):
2324
if not target_format.rich_root_data:
2325
raise errors.BadConversionTarget(
2326
'Does not support rich root data.', target_format)
2327
if not getattr(target_format, 'supports_tree_reference', False):
2328
raise errors.BadConversionTarget(
2329
'Does not support nested trees', target_format)
2331
def get_format_string(self):
2332
"""See RepositoryFormat.get_format_string()."""
2333
return ("Bazaar development format 0 with subtree support "
2334
"(needs bzr.dev from before 1.3)\n")
2336
def get_format_description(self):
2337
"""See RepositoryFormat.get_format_description()."""
2338
return ("Development repository format, currently the same as "
2339
"pack-0.92-subtree\n")
added
removed
bzrlib/repofmt/pack_repo.py
