« back to all changes in this revision
Viewing changes to bzrlib/repofmt/pack_repo.py
- browse files at revision 2685.2.2
- compare with another revision
- download diff
- download tarball
- view history from revision 2685.2.2
- Committer: Alexander Belchenko
- Date: 2007-08-10 09:04:38 UTC
- mto: This revision was merged to the branch mainline in revision 2694.
- Revision ID: bialix@ukr.net-20070810090438-0835xdz0rl8825qv
fixes after Ian's review
- files added:
- bzrlib/plugins/multiparent.py
- files removed:
- bzrlib/_patiencediff_c.c
- bzrlib/help_topics
- bzrlib/help_topics/en
- bzrlib/tests/inventory_implementations
- doc/en
- doc/en/admin-guide
- doc/en/developer-guide
- doc/en/mini-tutorial
- doc/en/quick-reference
- doc/en/release-notes
- doc/en/tutorials
- doc/en/user-guide
- doc/en/user-guide/images
- doc/en/user-reference
- files renamed:
- bzrlib/help_topics/__init__.py => bzrlib/help_topics.py
- bzrlib/_patiencediff_py.py => bzrlib/patiencediff.py
- bzrlib/plugins/launchpad/lp_directory.py => bzrlib/plugins/launchpad/lp_indirect.py
- bzrlib/plugins/launchpad/test_lp_directory.py => bzrlib/plugins/launchpad/test_lp_indirect.py
- bzrlib/tests/http_utils.py => bzrlib/tests/HTTPTestUtil.py
- bzrlib/tests/http_server.py => bzrlib/tests/HttpServer.py
- doc/en/user-guide/bug_trackers.txt => doc/bug_trackers.txt
- doc/en/tutorials/centralized_workflow.txt => doc/centralized_workflow.txt
- bzrlib/help_topics/en/configuration.txt => doc/configuration.txt
- bzrlib/help_topics/en/conflicts.txt => doc/conflicts.txt
- doc/en/user-guide/http_smart_server.txt => doc/http_smart_server.txt
- doc/en/user-guide/index.txt => doc/index.txt
- doc/en/user-guide/plugins.txt => doc/plugins.txt
- doc/en/user-guide/server.txt => doc/server.txt
- doc/en/user-guide/setting_up_email.txt => doc/setting_up_email.txt
- doc/en/user-guide/shared_repository_layouts.txt => doc/shared_repository_layouts.txt
- doc/en/user-guide/specifying_revisions.txt => doc/specifying_revisions.txt
- doc/en/tutorials/tutorial.txt => doc/tutorial.txt
- doc/en/user-guide/using_aliases.txt => doc/using_aliases.txt
- doc/en/user-guide/version_info.txt => doc/version_info.txt
- files modified:
- .bzrignore
added
removed
bzrlib/repofmt/pack_repo.py
1
# Copyright (C) 2005, 2006, 2007 Canonical Ltd
2
#
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.
7
#
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.
12
#
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
16
17
from bzrlib.lazy_import import lazy_import
18
lazy_import(globals(), """
19
from itertools import izip
20
import math
21
import md5
22
import time
23
24
from bzrlib import (
25
debug,
26
graph,
27
pack,
28
ui,
29
)
30
from bzrlib.index import (
31
GraphIndex,
32
GraphIndexBuilder,
33
InMemoryGraphIndex,
34
CombinedGraphIndex,
35
GraphIndexPrefixAdapter,
36
)
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
42
""")
43
from bzrlib import (
44
bzrdir,
45
deprecated_graph,
46
errors,
47
knit,
48
lockable_files,
49
lockdir,
50
osutils,
51
symbol_versioning,
52
transactions,
53
xml5,
54
xml6,
55
xml7,
56
)
57
58
from bzrlib.decorators import needs_read_lock, needs_write_lock
59
from bzrlib.repofmt.knitrepo import KnitRepository
60
from bzrlib.repository import (
61
CommitBuilder,
62
MetaDirRepository,
63
MetaDirRepositoryFormat,
64
RootCommitBuilder,
65
)
66
import bzrlib.revision as _mod_revision
67
from bzrlib.store.revision.knit import KnitRevisionStore
68
from bzrlib.store.versioned import VersionedFileStore
69
from bzrlib.trace import mutter, note, warning
70
71
72
class PackCommitBuilder(CommitBuilder):
73
"""A subclass of CommitBuilder to add texts with pack semantics.
74
75
Specifically this uses one knit object rather than one knit object per
76
added text, reducing memory and object pressure.
77
"""
78
79
def __init__(self, repository, parents, config, timestamp=None,
80
timezone=None, committer=None, revprops=None,
81
revision_id=None):
82
CommitBuilder.__init__(self, repository, parents, config,
83
timestamp=timestamp, timezone=timezone, committer=committer,
84
revprops=revprops, revision_id=revision_id)
85
self._file_graph = graph.Graph(
86
repository._pack_collection.text_index.combined_index)
87
88
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
89
return self.repository._pack_collection._add_text_to_weave(file_id,
90
self._new_revision_id, new_lines, parents, nostore_sha,
91
self.random_revid)
92
93
def _heads(self, file_id, revision_ids):
94
keys = [(file_id, revision_id) for revision_id in revision_ids]
95
return set([key[1] for key in self._file_graph.heads(keys)])
96
97
98
class PackRootCommitBuilder(RootCommitBuilder):
99
"""A subclass of RootCommitBuilder to add texts with pack semantics.
100
101
Specifically this uses one knit object rather than one knit object per
102
added text, reducing memory and object pressure.
103
"""
104
105
def __init__(self, repository, parents, config, timestamp=None,
106
timezone=None, committer=None, revprops=None,
107
revision_id=None):
108
CommitBuilder.__init__(self, repository, parents, config,
109
timestamp=timestamp, timezone=timezone, committer=committer,
110
revprops=revprops, revision_id=revision_id)
111
self._file_graph = graph.Graph(
112
repository._pack_collection.text_index.combined_index)
113
114
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
115
return self.repository._pack_collection._add_text_to_weave(file_id,
116
self._new_revision_id, new_lines, parents, nostore_sha,
117
self.random_revid)
118
119
def _heads(self, file_id, revision_ids):
120
keys = [(file_id, revision_id) for revision_id in revision_ids]
121
return set([key[1] for key in self._file_graph.heads(keys)])
122
123
124
class Pack(object):
125
"""An in memory proxy for a pack and its indices.
126
127
This is a base class that is not directly used, instead the classes
128
ExistingPack and NewPack are used.
129
"""
130
131
def __init__(self, revision_index, inventory_index, text_index,
132
signature_index):
133
"""Create a pack instance.
134
135
:param revision_index: A GraphIndex for determining what revisions are
136
present in the Pack and accessing the locations of their texts.
137
:param inventory_index: A GraphIndex for determining what inventories are
138
present in the Pack and accessing the locations of their
139
texts/deltas.
140
:param text_index: A GraphIndex for determining what file texts
141
are present in the pack and accessing the locations of their
142
texts/deltas (via (fileid, revisionid) tuples).
143
:param revision_index: A GraphIndex for determining what signatures are
144
present in the Pack and accessing the locations of their texts.
145
"""
146
self.revision_index = revision_index
147
self.inventory_index = inventory_index
148
self.text_index = text_index
149
self.signature_index = signature_index
150
151
def access_tuple(self):
152
"""Return a tuple (transport, name) for the pack content."""
153
return self.pack_transport, self.file_name()
154
155
def file_name(self):
156
"""Get the file name for the pack on disk."""
157
return self.name + '.pack'
158
159
def get_revision_count(self):
160
return self.revision_index.key_count()
161
162
def inventory_index_name(self, name):
163
"""The inv index is the name + .iix."""
164
return self.index_name('inventory', name)
165
166
def revision_index_name(self, name):
167
"""The revision index is the name + .rix."""
168
return self.index_name('revision', name)
169
170
def signature_index_name(self, name):
171
"""The signature index is the name + .six."""
172
return self.index_name('signature', name)
173
174
def text_index_name(self, name):
175
"""The text index is the name + .tix."""
176
return self.index_name('text', name)
177
178
def _external_compression_parents_of_texts(self):
179
keys = set()
180
refs = set()
181
for node in self.text_index.iter_all_entries():
182
keys.add(node[1])
183
refs.update(node[3][1])
184
return refs - keys
185
186
187
class ExistingPack(Pack):
188
"""An in memory proxy for an existing .pack and its disk indices."""
189
190
def __init__(self, pack_transport, name, revision_index, inventory_index,
191
text_index, signature_index):
192
"""Create an ExistingPack object.
193
194
:param pack_transport: The transport where the pack file resides.
195
:param name: The name of the pack on disk in the pack_transport.
196
"""
197
Pack.__init__(self, revision_index, inventory_index, text_index,
198
signature_index)
199
self.name = name
200
self.pack_transport = pack_transport
201
assert None not in (revision_index, inventory_index, text_index,
202
signature_index, name, pack_transport)
203
204
def __eq__(self, other):
205
return self.__dict__ == other.__dict__
206
207
def __ne__(self, other):
208
return not self.__eq__(other)
209
210
def __repr__(self):
211
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
212
id(self), self.transport, self.name)
213
214
215
class NewPack(Pack):
216
"""An in memory proxy for a pack which is being created."""
217
218
# A map of index 'type' to the file extension and position in the
219
# index_sizes array.
220
index_definitions = {
221
'revision': ('.rix', 0),
222
'inventory': ('.iix', 1),
223
'text': ('.tix', 2),
224
'signature': ('.six', 3),
225
}
226
227
def __init__(self, upload_transport, index_transport, pack_transport,
228
upload_suffix='', file_mode=None):
229
"""Create a NewPack instance.
230
231
:param upload_transport: A writable transport for the pack to be
232
incrementally uploaded to.
233
:param index_transport: A writable transport for the pack's indices to
234
be written to when the pack is finished.
235
:param pack_transport: A writable transport for the pack to be renamed
236
to when the upload is complete. This *must* be the same as
237
upload_transport.clone('../packs').
238
:param upload_suffix: An optional suffix to be given to any temporary
239
files created during the pack creation. e.g '.autopack'
240
:param file_mode: An optional file mode to create the new files with.
241
"""
242
# The relative locations of the packs are constrained, but all are
243
# passed in because the caller has them, so as to avoid object churn.
244
Pack.__init__(self,
245
# Revisions: parents list, no text compression.
246
InMemoryGraphIndex(reference_lists=1),
247
# Inventory: We want to map compression only, but currently the
248
# knit code hasn't been updated enough to understand that, so we
249
# have a regular 2-list index giving parents and compression
250
# source.
251
InMemoryGraphIndex(reference_lists=2),
252
# Texts: compression and per file graph, for all fileids - so two
253
# reference lists and two elements in the key tuple.
254
InMemoryGraphIndex(reference_lists=2, key_elements=2),
255
# Signatures: Just blobs to store, no compression, no parents
256
# listing.
257
InMemoryGraphIndex(reference_lists=0),
258
)
259
# where should the new pack be opened
260
self.upload_transport = upload_transport
261
# where are indices written out to
262
self.index_transport = index_transport
263
# where is the pack renamed to when it is finished?
264
self.pack_transport = pack_transport
265
# What file mode to upload the pack and indices with.
266
self._file_mode = file_mode
267
# tracks the content written to the .pack file.
268
self._hash = md5.new()
269
# a four-tuple with the length in bytes of the indices, once the pack
270
# is finalised. (rev, inv, text, sigs)
271
self.index_sizes = None
272
# How much data to cache when writing packs. Note that this is not
273
# synchronised with reads, because it's not in the transport layer, so
274
# is not safe unless the client knows it won't be reading from the pack
275
# under creation.
276
self._cache_limit = 0
277
# the temporary pack file name.
278
self.random_name = rand_chars(20) + upload_suffix
279
# when was this pack started ?
280
self.start_time = time.time()
281
# open an output stream for the data added to the pack.
282
self.write_stream = self.upload_transport.open_write_stream(
283
self.random_name, mode=self._file_mode)
284
if 'pack' in debug.debug_flags:
285
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
286
time.ctime(), self.upload_transport.base, self.random_name,
287
time.time() - self.start_time)
288
# A list of byte sequences to be written to the new pack, and the
289
# aggregate size of them. Stored as a list rather than separate
290
# variables so that the _write_data closure below can update them.
291
self._buffer = [[], 0]
292
# create a callable for adding data
293
#
294
# robertc says- this is a closure rather than a method on the object
295
# so that the variables are locals, and faster than accessing object
296
# members.
297
def _write_data(bytes, flush=False, _buffer=self._buffer,
298
_write=self.write_stream.write, _update=self._hash.update):
299
_buffer[0].append(bytes)
300
_buffer[1] += len(bytes)
301
# buffer cap
302
if _buffer[1] > self._cache_limit or flush:
303
bytes = ''.join(_buffer[0])
304
_write(bytes)
305
_update(bytes)
306
_buffer[:] = [[], 0]
307
# expose this on self, for the occasion when clients want to add data.
308
self._write_data = _write_data
309
# a pack writer object to serialise pack records.
310
self._writer = pack.ContainerWriter(self._write_data)
311
self._writer.begin()
312
# what state is the pack in? (open, finished, aborted)
313
self._state = 'open'
314
315
def abort(self):
316
"""Cancel creating this pack."""
317
self._state = 'aborted'
318
self.write_stream.close()
319
# Remove the temporary pack file.
320
self.upload_transport.delete(self.random_name)
321
# The indices have no state on disk.
322
323
def access_tuple(self):
324
"""Return a tuple (transport, name) for the pack content."""
325
assert self._state in ('open', 'finished')
326
if self._state == 'finished':
327
return Pack.access_tuple(self)
328
else:
329
return self.upload_transport, self.random_name
330
331
def data_inserted(self):
332
"""True if data has been added to this pack."""
333
return bool(self.get_revision_count() or
334
self.inventory_index.key_count() or
335
self.text_index.key_count() or
336
self.signature_index.key_count())
337
338
def finish(self):
339
"""Finish the new pack.
340
341
This:
342
- finalises the content
343
- assigns a name (the md5 of the content, currently)
344
- writes out the associated indices
345
- renames the pack into place.
346
- stores the index size tuple for the pack in the index_sizes
347
attribute.
348
"""
349
self._writer.end()
350
if self._buffer[1]:
351
self._write_data('', flush=True)
352
self.name = self._hash.hexdigest()
353
# write indices
354
# XXX: It'd be better to write them all to temporary names, then
355
# rename them all into place, so that the window when only some are
356
# visible is smaller. On the other hand none will be seen until
357
# they're in the names list.
358
self.index_sizes = [None, None, None, None]
359
self._write_index('revision', self.revision_index, 'revision')
360
self._write_index('inventory', self.inventory_index, 'inventory')
361
self._write_index('text', self.text_index, 'file texts')
362
self._write_index('signature', self.signature_index,
363
'revision signatures')
364
self.write_stream.close()
365
# Note that this will clobber an existing pack with the same name,
366
# without checking for hash collisions. While this is undesirable this
367
# is something that can be rectified in a subsequent release. One way
368
# to rectify it may be to leave the pack at the original name, writing
369
# its pack-names entry as something like 'HASH: index-sizes
370
# temporary-name'. Allocate that and check for collisions, if it is
371
# collision free then rename it into place. If clients know this scheme
372
# they can handle missing-file errors by:
373
# - try for HASH.pack
374
# - try for temporary-name
375
# - refresh the pack-list to see if the pack is now absent
376
self.upload_transport.rename(self.random_name,
377
'../packs/' + self.name + '.pack')
378
self._state = 'finished'
379
if 'pack' in debug.debug_flags:
380
# XXX: size might be interesting?
381
mutter('%s: create_pack: pack renamed into place: %s%s->%s%s t+%6.3fs',
382
time.ctime(), self.upload_transport.base, self.random_name,
383
self.pack_transport, self.name,
384
time.time() - self.start_time)
385
386
def flush(self):
387
"""Flush any current data."""
388
if self._buffer[1]:
389
bytes = ''.join(self._buffer[0])
390
self.write_stream.write(bytes)
391
self._hash.update(bytes)
392
self._buffer[:] = [[], 0]
393
394
def index_name(self, index_type, name):
395
"""Get the disk name of an index type for pack name 'name'."""
396
return name + NewPack.index_definitions[index_type][0]
397
398
def index_offset(self, index_type):
399
"""Get the position in a index_size array for a given index type."""
400
return NewPack.index_definitions[index_type][1]
401
402
def _replace_index_with_readonly(self, index_type):
403
setattr(self, index_type + '_index',
404
GraphIndex(self.index_transport,
405
self.index_name(index_type, self.name),
406
self.index_sizes[self.index_offset(index_type)]))
407
408
def set_write_cache_size(self, size):
409
self._cache_limit = size
410
411
def _write_index(self, index_type, index, label):
412
"""Write out an index.
413
414
:param index_type: The type of index to write - e.g. 'revision'.
415
:param index: The index object to serialise.
416
:param label: What label to give the index e.g. 'revision'.
417
"""
418
index_name = self.index_name(index_type, self.name)
419
self.index_sizes[self.index_offset(index_type)] = \
420
self.index_transport.put_file(index_name, index.finish(),
421
mode=self._file_mode)
422
if 'pack' in debug.debug_flags:
423
# XXX: size might be interesting?
424
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
425
time.ctime(), label, self.upload_transport.base,
426
self.random_name, time.time() - self.start_time)
427
# Replace the writable index on this object with a readonly,
428
# presently unloaded index. We should alter
429
# the index layer to make its finish() error if add_node is
430
# subsequently used. RBC
431
self._replace_index_with_readonly(index_type)
432
433
434
class AggregateIndex(object):
435
"""An aggregated index for the RepositoryPackCollection.
436
437
AggregateIndex is reponsible for managing the PackAccess object,
438
Index-To-Pack mapping, and all indices list for a specific type of index
439
such as 'revision index'.
440
441
A CombinedIndex provides an index on a single key space built up
442
from several on-disk indices. The AggregateIndex builds on this
443
to provide a knit access layer, and allows having up to one writable
444
index within the collection.
445
"""
446
# XXX: Probably 'can be written to' could/should be separated from 'acts
447
# like a knit index' -- mbp 20071024
448
449
def __init__(self):
450
"""Create an AggregateIndex."""
451
self.index_to_pack = {}
452
self.combined_index = CombinedGraphIndex([])
453
self.knit_access = _PackAccess(self.index_to_pack)
454
455
def replace_indices(self, index_to_pack, indices):
456
"""Replace the current mappings with fresh ones.
457
458
This should probably not be used eventually, rather incremental add and
459
removal of indices. It has been added during refactoring of existing
460
code.
461
462
:param index_to_pack: A mapping from index objects to
463
(transport, name) tuples for the pack file data.
464
:param indices: A list of indices.
465
"""
466
# refresh the revision pack map dict without replacing the instance.
467
self.index_to_pack.clear()
468
self.index_to_pack.update(index_to_pack)
469
# XXX: API break - clearly a 'replace' method would be good?
470
self.combined_index._indices[:] = indices
471
# the current add nodes callback for the current writable index if
472
# there is one.
473
self.add_callback = None
474
475
def add_index(self, index, pack):
476
"""Add index to the aggregate, which is an index for Pack pack.
477
478
Future searches on the aggregate index will seach this new index
479
before all previously inserted indices.
480
481
:param index: An Index for the pack.
482
:param pack: A Pack instance.
483
"""
484
# expose it to the index map
485
self.index_to_pack[index] = pack.access_tuple()
486
# put it at the front of the linear index list
487
self.combined_index.insert_index(0, index)
488
489
def add_writable_index(self, index, pack):
490
"""Add an index which is able to have data added to it.
491
492
There can be at most one writable index at any time. Any
493
modifications made to the knit are put into this index.
494
495
:param index: An index from the pack parameter.
496
:param pack: A Pack instance.
497
"""
498
assert self.add_callback is None, \
499
"%s already has a writable index through %s" % \
500
(self, self.add_callback)
501
# allow writing: queue writes to a new index
502
self.add_index(index, pack)
503
# Updates the index to packs mapping as a side effect,
504
self.knit_access.set_writer(pack._writer, index, pack.access_tuple())
505
self.add_callback = index.add_nodes
506
507
def clear(self):
508
"""Reset all the aggregate data to nothing."""
509
self.knit_access.set_writer(None, None, (None, None))
510
self.index_to_pack.clear()
511
del self.combined_index._indices[:]
512
self.add_callback = None
513
514
def remove_index(self, index, pack):
515
"""Remove index from the indices used to answer queries.
516
517
:param index: An index from the pack parameter.
518
:param pack: A Pack instance.
519
"""
520
del self.index_to_pack[index]
521
self.combined_index._indices.remove(index)
522
if (self.add_callback is not None and
523
getattr(index, 'add_nodes', None) == self.add_callback):
524
self.add_callback = None
525
self.knit_access.set_writer(None, None, (None, None))
526
527
528
class Packer(object):
529
"""Create a pack from packs."""
530
531
def __init__(self, pack_collection, packs, suffix, revision_ids=None):
532
"""Create a Packer.
533
534
:param pack_collection: A RepositoryPackCollection object where the
535
new pack is being written to.
536
:param packs: The packs to combine.
537
:param suffix: The suffix to use on the temporary files for the pack.
538
:param revision_ids: Revision ids to limit the pack to.
539
"""
540
self.packs = packs
541
self.suffix = suffix
542
self.revision_ids = revision_ids
543
# The pack object we are creating.
544
self.new_pack = None
545
self._pack_collection = pack_collection
546
# The index layer keys for the revisions being copied. None for 'all
547
# objects'.
548
self._revision_keys = None
549
# What text keys to copy. None for 'all texts'. This is set by
550
# _copy_inventory_texts
551
self._text_filter = None
552
self._extra_init()
553
554
def _extra_init(self):
555
"""A template hook to allow extending the constructor trivially."""
556
557
def pack(self, pb=None):
558
"""Create a new pack by reading data from other packs.
559
560
This does little more than a bulk copy of data. One key difference
561
is that data with the same item key across multiple packs is elided
562
from the output. The new pack is written into the current pack store
563
along with its indices, and the name added to the pack names. The
564
source packs are not altered and are not required to be in the current
565
pack collection.
566
567
:param pb: An optional progress bar to use. A nested bar is created if
568
this is None.
569
:return: A Pack object, or None if nothing was copied.
570
"""
571
# open a pack - using the same name as the last temporary file
572
# - which has already been flushed, so its safe.
573
# XXX: - duplicate code warning with start_write_group; fix before
574
# considering 'done'.
575
if self._pack_collection._new_pack is not None:
576
raise errors.BzrError('call to create_pack_from_packs while '
577
'another pack is being written.')
578
if self.revision_ids is not None:
579
if len(self.revision_ids) == 0:
580
# silly fetch request.
581
return None
582
else:
583
self.revision_ids = frozenset(self.revision_ids)
584
if pb is None:
585
self.pb = ui.ui_factory.nested_progress_bar()
586
else:
587
self.pb = pb
588
try:
589
return self._create_pack_from_packs()
590
finally:
591
if pb is None:
592
self.pb.finished()
593
594
def open_pack(self):
595
"""Open a pack for the pack we are creating."""
596
return NewPack(self._pack_collection._upload_transport,
597
self._pack_collection._index_transport,
598
self._pack_collection._pack_transport, upload_suffix=self.suffix,
599
file_mode=self._pack_collection.repo.control_files._file_mode)
600
601
def _copy_revision_texts(self):
602
"""Copy revision data to the new pack."""
603
# select revisions
604
if self.revision_ids:
605
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
606
else:
607
revision_keys = None
608
# select revision keys
609
revision_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
610
self.packs, 'revision_index')[0]
611
revision_nodes = self._pack_collection._index_contents(revision_index_map, revision_keys)
612
# copy revision keys and adjust values
613
self.pb.update("Copying revision texts", 1)
614
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
615
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
616
self.new_pack.revision_index, readv_group_iter, total_items))
617
if 'pack' in debug.debug_flags:
618
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
619
time.ctime(), self._pack_collection._upload_transport.base,
620
self.new_pack.random_name,
621
self.new_pack.revision_index.key_count(),
622
time.time() - self.new_pack.start_time)
623
self._revision_keys = revision_keys
624
625
def _copy_inventory_texts(self):
626
"""Copy the inventory texts to the new pack.
627
628
self._revision_keys is used to determine what inventories to copy.
629
630
Sets self._text_filter appropriately.
631
"""
632
# select inventory keys
633
inv_keys = self._revision_keys # currently the same keyspace, and note that
634
# querying for keys here could introduce a bug where an inventory item
635
# is missed, so do not change it to query separately without cross
636
# checking like the text key check below.
637
inventory_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
638
self.packs, 'inventory_index')[0]
639
inv_nodes = self._pack_collection._index_contents(inventory_index_map, inv_keys)
640
# copy inventory keys and adjust values
641
# XXX: Should be a helper function to allow different inv representation
642
# at this point.
643
self.pb.update("Copying inventory texts", 2)
644
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
645
# Only grab the output lines if we will be processing them
646
output_lines = bool(self.revision_ids)
647
inv_lines = self._copy_nodes_graph(inventory_index_map,
648
self.new_pack._writer, self.new_pack.inventory_index,
649
readv_group_iter, total_items, output_lines=output_lines)
650
if self.revision_ids:
651
self._process_inventory_lines(inv_lines)
652
else:
653
# eat the iterator to cause it to execute.
654
list(inv_lines)
655
self._text_filter = None
656
if 'pack' in debug.debug_flags:
657
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
658
time.ctime(), self._pack_collection._upload_transport.base,
659
self.new_pack.random_name,
660
self.new_pack.inventory_index.key_count(),
661
time.time() - self.new_pack.start_time)
662
663
def _copy_text_texts(self):
664
# select text keys
665
text_index_map, text_nodes = self._get_text_nodes()
666
if self._text_filter is not None:
667
# We could return the keys copied as part of the return value from
668
# _copy_nodes_graph but this doesn't work all that well with the
669
# need to get line output too, so we check separately, and as we're
670
# going to buffer everything anyway, we check beforehand, which
671
# saves reading knit data over the wire when we know there are
672
# mising records.
673
text_nodes = set(text_nodes)
674
present_text_keys = set(_node[1] for _node in text_nodes)
675
missing_text_keys = set(self._text_filter) - present_text_keys
676
if missing_text_keys:
677
# TODO: raise a specific error that can handle many missing
678
# keys.
679
a_missing_key = missing_text_keys.pop()
680
raise errors.RevisionNotPresent(a_missing_key[1],
681
a_missing_key[0])
682
# copy text keys and adjust values
683
self.pb.update("Copying content texts", 3)
684
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
685
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
686
self.new_pack.text_index, readv_group_iter, total_items))
687
self._log_copied_texts()
688
689
def _check_references(self):
690
"""Make sure our external refereneces are present."""
691
external_refs = self.new_pack._external_compression_parents_of_texts()
692
if external_refs:
693
index = self._pack_collection.text_index.combined_index
694
found_items = list(index.iter_entries(external_refs))
695
if len(found_items) != len(external_refs):
696
found_keys = set(k for idx, k, refs, value in found_items)
697
missing_items = external_refs - found_keys
698
missing_file_id, missing_revision_id = missing_items.pop()
699
raise errors.RevisionNotPresent(missing_revision_id,
700
missing_file_id)
701
702
def _create_pack_from_packs(self):
703
self.pb.update("Opening pack", 0, 5)
704
self.new_pack = self.open_pack()
705
new_pack = self.new_pack
706
# buffer data - we won't be reading-back during the pack creation and
707
# this makes a significant difference on sftp pushes.
708
new_pack.set_write_cache_size(1024*1024)
709
if 'pack' in debug.debug_flags:
710
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
711
for a_pack in self.packs]
712
if self.revision_ids is not None:
713
rev_count = len(self.revision_ids)
714
else:
715
rev_count = 'all'
716
mutter('%s: create_pack: creating pack from source packs: '
717
'%s%s %s revisions wanted %s t=0',
718
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
719
plain_pack_list, rev_count)
720
self._copy_revision_texts()
721
self._copy_inventory_texts()
722
self._copy_text_texts()
723
# select signature keys
724
signature_filter = self._revision_keys # same keyspace
725
signature_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
726
self.packs, 'signature_index')[0]
727
signature_nodes = self._pack_collection._index_contents(signature_index_map,
728
signature_filter)
729
# copy signature keys and adjust values
730
self.pb.update("Copying signature texts", 4)
731
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
732
new_pack.signature_index)
733
if 'pack' in debug.debug_flags:
734
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
735
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
736
new_pack.signature_index.key_count(),
737
time.time() - new_pack.start_time)
738
self._check_references()
739
if not self._use_pack(new_pack):
740
new_pack.abort()
741
return None
742
self.pb.update("Finishing pack", 5)
743
new_pack.finish()
744
self._pack_collection.allocate(new_pack)
745
return new_pack
746
747
def _copy_nodes(self, nodes, index_map, writer, write_index):
748
"""Copy knit nodes between packs with no graph references."""
749
pb = ui.ui_factory.nested_progress_bar()
750
try:
751
return self._do_copy_nodes(nodes, index_map, writer,
752
write_index, pb)
753
finally:
754
pb.finished()
755
756
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb):
757
# for record verification
758
knit_data = _KnitData(None)
759
# plan a readv on each source pack:
760
# group by pack
761
nodes = sorted(nodes)
762
# how to map this into knit.py - or knit.py into this?
763
# we don't want the typical knit logic, we want grouping by pack
764
# at this point - perhaps a helper library for the following code
765
# duplication points?
766
request_groups = {}
767
for index, key, value in nodes:
768
if index not in request_groups:
769
request_groups[index] = []
770
request_groups[index].append((key, value))
771
record_index = 0
772
pb.update("Copied record", record_index, len(nodes))
773
for index, items in request_groups.iteritems():
774
pack_readv_requests = []
775
for key, value in items:
776
# ---- KnitGraphIndex.get_position
777
bits = value[1:].split(' ')
778
offset, length = int(bits[0]), int(bits[1])
779
pack_readv_requests.append((offset, length, (key, value[0])))
780
# linear scan up the pack
781
pack_readv_requests.sort()
782
# copy the data
783
transport, path = index_map[index]
784
reader = pack.make_readv_reader(transport, path,
785
[offset[0:2] for offset in pack_readv_requests])
786
for (names, read_func), (_1, _2, (key, eol_flag)) in \
787
izip(reader.iter_records(), pack_readv_requests):
788
raw_data = read_func(None)
789
# check the header only
790
df, _ = knit_data._parse_record_header(key[-1], raw_data)
791
df.close()
792
pos, size = writer.add_bytes_record(raw_data, names)
793
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
794
pb.update("Copied record", record_index)
795
record_index += 1
796
797
def _copy_nodes_graph(self, index_map, writer, write_index,
798
readv_group_iter, total_items, output_lines=False):
799
"""Copy knit nodes between packs.
800
801
:param output_lines: Return lines present in the copied data as
802
an iterator of line,version_id.
803
"""
804
pb = ui.ui_factory.nested_progress_bar()
805
try:
806
for result in self._do_copy_nodes_graph(index_map, writer,
807
write_index, output_lines, pb, readv_group_iter, total_items):
808
yield result
809
except Exception:
810
# Python 2.4 does not permit try:finally: in a generator.
811
pb.finished()
812
raise
813
else:
814
pb.finished()
815
816
def _do_copy_nodes_graph(self, index_map, writer, write_index,
817
output_lines, pb, readv_group_iter, total_items):
818
# for record verification
819
knit_data = _KnitData(None)
820
# for line extraction when requested (inventories only)
821
if output_lines:
822
factory = knit.KnitPlainFactory()
823
record_index = 0
824
pb.update("Copied record", record_index, total_items)
825
for index, readv_vector, node_vector in readv_group_iter:
826
# copy the data
827
transport, path = index_map[index]
828
reader = pack.make_readv_reader(transport, path, readv_vector)
829
for (names, read_func), (key, eol_flag, references) in \
830
izip(reader.iter_records(), node_vector):
831
raw_data = read_func(None)
832
version_id = key[-1]
833
if output_lines:
834
# read the entire thing
835
content, _ = knit_data._parse_record(version_id, raw_data)
836
if len(references[-1]) == 0:
837
line_iterator = factory.get_fulltext_content(content)
838
else:
839
line_iterator = factory.get_linedelta_content(content)
840
for line in line_iterator:
841
yield line, version_id
842
else:
843
# check the header only
844
df, _ = knit_data._parse_record_header(version_id, raw_data)
845
df.close()
846
pos, size = writer.add_bytes_record(raw_data, names)
847
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
848
pb.update("Copied record", record_index)
849
record_index += 1
850
851
def _get_text_nodes(self):
852
text_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
853
self.packs, 'text_index')[0]
854
return text_index_map, self._pack_collection._index_contents(text_index_map,
855
self._text_filter)
856
857
def _least_readv_node_readv(self, nodes):
858
"""Generate request groups for nodes using the least readv's.
859
860
:param nodes: An iterable of graph index nodes.
861
:return: Total node count and an iterator of the data needed to perform
862
readvs to obtain the data for nodes. Each item yielded by the
863
iterator is a tuple with:
864
index, readv_vector, node_vector. readv_vector is a list ready to
865
hand to the transport readv method, and node_vector is a list of
866
(key, eol_flag, references) for the the node retrieved by the
867
matching readv_vector.
868
"""
869
# group by pack so we do one readv per pack
870
nodes = sorted(nodes)
871
total = len(nodes)
872
request_groups = {}
873
for index, key, value, references in nodes:
874
if index not in request_groups:
875
request_groups[index] = []
876
request_groups[index].append((key, value, references))
877
result = []
878
for index, items in request_groups.iteritems():
879
pack_readv_requests = []
880
for key, value, references in items:
881
# ---- KnitGraphIndex.get_position
882
bits = value[1:].split(' ')
883
offset, length = int(bits[0]), int(bits[1])
884
pack_readv_requests.append(
885
((offset, length), (key, value[0], references)))
886
# linear scan up the pack to maximum range combining.
887
pack_readv_requests.sort()
888
# split out the readv and the node data.
889
pack_readv = [readv for readv, node in pack_readv_requests]
890
node_vector = [node for readv, node in pack_readv_requests]
891
result.append((index, pack_readv, node_vector))
892
return total, result
893
894
def _log_copied_texts(self):
895
if 'pack' in debug.debug_flags:
896
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
897
time.ctime(), self._pack_collection._upload_transport.base,
898
self.new_pack.random_name,
899
self.new_pack.text_index.key_count(),
900
time.time() - self.new_pack.start_time)
901
902
def _process_inventory_lines(self, inv_lines):
903
"""Use up the inv_lines generator and setup a text key filter."""
904
repo = self._pack_collection.repo
905
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
906
inv_lines, self.revision_ids)
907
text_filter = []
908
for fileid, file_revids in fileid_revisions.iteritems():
909
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
910
self._text_filter = text_filter
911
912
def _revision_node_readv(self, revision_nodes):
913
"""Return the total revisions and the readv's to issue.
914
915
:param revision_nodes: The revision index contents for the packs being
916
incorporated into the new pack.
917
:return: As per _least_readv_node_readv.
918
"""
919
return self._least_readv_node_readv(revision_nodes)
920
921
def _use_pack(self, new_pack):
922
"""Return True if new_pack should be used.
923
924
:param new_pack: The pack that has just been created.
925
:return: True if the pack should be used.
926
"""
927
return new_pack.data_inserted()
928
929
930
class OptimisingPacker(Packer):
931
"""A packer which spends more time to create better disk layouts."""
932
933
def _revision_node_readv(self, revision_nodes):
934
"""Return the total revisions and the readv's to issue.
935
936
This sort places revisions in topological order with the ancestors
937
after the children.
938
939
:param revision_nodes: The revision index contents for the packs being
940
incorporated into the new pack.
941
:return: As per _least_readv_node_readv.
942
"""
943
# build an ancestors dict
944
ancestors = {}
945
by_key = {}
946
for index, key, value, references in revision_nodes:
947
ancestors[key] = references[0]
948
by_key[key] = (index, value, references)
949
order = tsort.topo_sort(ancestors)
950
total = len(order)
951
# Single IO is pathological, but it will work as a starting point.
952
requests = []
953
for key in reversed(order):
954
index, value, references = by_key[key]
955
# ---- KnitGraphIndex.get_position
956
bits = value[1:].split(' ')
957
offset, length = int(bits[0]), int(bits[1])
958
requests.append(
959
(index, [(offset, length)], [(key, value[0], references)]))
960
# TODO: combine requests in the same index that are in ascending order.
961
return total, requests
962
963
964
class ReconcilePacker(Packer):
965
"""A packer which regenerates indices etc as it copies.
966
967
This is used by ``bzr reconcile`` to cause parent text pointers to be
968
regenerated.
969
"""
970
971
def _extra_init(self):
972
self._data_changed = False
973
974
def _process_inventory_lines(self, inv_lines):
975
"""Generate a text key reference map rather for reconciling with."""
976
repo = self._pack_collection.repo
977
refs = repo._find_text_key_references_from_xml_inventory_lines(
978
inv_lines)
979
self._text_refs = refs
980
# during reconcile we:
981
# - convert unreferenced texts to full texts
982
# - correct texts which reference a text not copied to be full texts
983
# - copy all others as-is but with corrected parents.
984
# - so at this point we don't know enough to decide what becomes a full
985
# text.
986
self._text_filter = None
987
988
def _copy_text_texts(self):
989
"""generate what texts we should have and then copy."""
990
self.pb.update("Copying content texts", 3)
991
# we have three major tasks here:
992
# 1) generate the ideal index
993
repo = self._pack_collection.repo
994
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
995
_1, key, _2, refs in
996
self.new_pack.revision_index.iter_all_entries()])
997
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
998
# 2) generate a text_nodes list that contains all the deltas that can
999
# be used as-is, with corrected parents.
1000
ok_nodes = []
1001
bad_texts = []
1002
discarded_nodes = []
1003
NULL_REVISION = _mod_revision.NULL_REVISION
1004
text_index_map, text_nodes = self._get_text_nodes()
1005
for node in text_nodes:
1006
# 0 - index
1007
# 1 - key
1008
# 2 - value
1009
# 3 - refs
1010
try:
1011
ideal_parents = tuple(ideal_index[node[1]])
1012
except KeyError:
1013
discarded_nodes.append(node)
1014
self._data_changed = True
1015
else:
1016
if ideal_parents == (NULL_REVISION,):
1017
ideal_parents = ()
1018
if ideal_parents == node[3][0]:
1019
# no change needed.
1020
ok_nodes.append(node)
1021
elif ideal_parents[0:1] == node[3][0][0:1]:
1022
# the left most parent is the same, or there are no parents
1023
# today. Either way, we can preserve the representation as
1024
# long as we change the refs to be inserted.
1025
self._data_changed = True
1026
ok_nodes.append((node[0], node[1], node[2],
1027
(ideal_parents, node[3][1])))
1028
self._data_changed = True
1029
else:
1030
# Reinsert this text completely
1031
bad_texts.append((node[1], ideal_parents))
1032
self._data_changed = True
1033
# we're finished with some data.
1034
del ideal_index
1035
del text_nodes
1036
# 3) bulk copy the ok data
1037
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1038
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1039
self.new_pack.text_index, readv_group_iter, total_items))
1040
# 4) adhoc copy all the other texts.
1041
# We have to topologically insert all texts otherwise we can fail to
1042
# reconcile when parts of a single delta chain are preserved intact,
1043
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1044
# reinserted, and if d3 has incorrect parents it will also be
1045
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1046
# copied), so we will try to delta, but d2 is not currently able to be
1047
# extracted because it's basis d1 is not present. Topologically sorting
1048
# addresses this. The following generates a sort for all the texts that
1049
# are being inserted without having to reference the entire text key
1050
# space (we only topo sort the revisions, which is smaller).
1051
topo_order = tsort.topo_sort(ancestors)
1052
rev_order = dict(zip(topo_order, range(len(topo_order))))
1053
bad_texts.sort(key=lambda key:rev_order[key[0][1]])
1054
transaction = repo.get_transaction()
1055
file_id_index = GraphIndexPrefixAdapter(
1056
self.new_pack.text_index,
1057
('blank', ), 1,
1058
add_nodes_callback=self.new_pack.text_index.add_nodes)
1059
knit_index = KnitGraphIndex(file_id_index,
1060
add_callback=file_id_index.add_nodes,
1061
deltas=True, parents=True)
1062
output_knit = knit.KnitVersionedFile('reconcile-texts',
1063
self._pack_collection.transport,
1064
None,
1065
index=knit_index,
1066
access_method=_PackAccess(
1067
{self.new_pack.text_index:self.new_pack.access_tuple()},
1068
(self.new_pack._writer, self.new_pack.text_index)),
1069
factory=knit.KnitPlainFactory())
1070
for key, parent_keys in bad_texts:
1071
# We refer to the new pack to delta data being output.
1072
# A possible improvement would be to catch errors on short reads
1073
# and only flush then.
1074
self.new_pack.flush()
1075
parents = []
1076
for parent_key in parent_keys:
1077
if parent_key[0] != key[0]:
1078
# Graph parents must match the fileid
1079
raise errors.BzrError('Mismatched key parent %r:%r' %
1080
(key, parent_keys))
1081
parents.append(parent_key[1])
1082
source_weave = repo.weave_store.get_weave(key[0], transaction)
1083
text_lines = source_weave.get_lines(key[1])
1084
# adapt the 'knit' to the current file_id.
1085
file_id_index = GraphIndexPrefixAdapter(
1086
self.new_pack.text_index,
1087
(key[0], ), 1,
1088
add_nodes_callback=self.new_pack.text_index.add_nodes)
1089
knit_index._graph_index = file_id_index
1090
knit_index._add_callback = file_id_index.add_nodes
1091
output_knit.add_lines_with_ghosts(
1092
key[1], parents, text_lines, random_id=True, check_content=False)
1093
# 5) check that nothing inserted has a reference outside the keyspace.
1094
missing_text_keys = self.new_pack._external_compression_parents_of_texts()
1095
if missing_text_keys:
1096
raise errors.BzrError('Reference to missing compression parents %r'
1097
% (refs - keys,))
1098
self._log_copied_texts()
1099
1100
def _use_pack(self, new_pack):
1101
"""Override _use_pack to check for reconcile having changed content."""
1102
# XXX: we might be better checking this at the copy time.
1103
original_inventory_keys = set()
1104
inv_index = self._pack_collection.inventory_index.combined_index
1105
for entry in inv_index.iter_all_entries():
1106
original_inventory_keys.add(entry[1])
1107
new_inventory_keys = set()
1108
for entry in new_pack.inventory_index.iter_all_entries():
1109
new_inventory_keys.add(entry[1])
1110
if new_inventory_keys != original_inventory_keys:
1111
self._data_changed = True
1112
return new_pack.data_inserted() and self._data_changed
1113
1114
1115
class RepositoryPackCollection(object):
1116
"""Management of packs within a repository."""
1117
1118
def __init__(self, repo, transport, index_transport, upload_transport,
1119
pack_transport):
1120
"""Create a new RepositoryPackCollection.
1121
1122
:param transport: Addresses the repository base directory
1123
(typically .bzr/repository/).
1124
:param index_transport: Addresses the directory containing indices.
1125
:param upload_transport: Addresses the directory into which packs are written
1126
while they're being created.
1127
:param pack_transport: Addresses the directory of existing complete packs.
1128
"""
1129
self.repo = repo
1130
self.transport = transport
1131
self._index_transport = index_transport
1132
self._upload_transport = upload_transport
1133
self._pack_transport = pack_transport
1134
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
1135
self.packs = []
1136
# name:Pack mapping
1137
self._packs_by_name = {}
1138
# the previous pack-names content
1139
self._packs_at_load = None
1140
# when a pack is being created by this object, the state of that pack.
1141
self._new_pack = None
1142
# aggregated revision index data
1143
self.revision_index = AggregateIndex()
1144
self.inventory_index = AggregateIndex()
1145
self.text_index = AggregateIndex()
1146
self.signature_index = AggregateIndex()
1147
1148
def add_pack_to_memory(self, pack):
1149
"""Make a Pack object available to the repository to satisfy queries.
1150
1151
:param pack: A Pack object.
1152
"""
1153
assert pack.name not in self._packs_by_name
1154
self.packs.append(pack)
1155
self._packs_by_name[pack.name] = pack
1156
self.revision_index.add_index(pack.revision_index, pack)
1157
self.inventory_index.add_index(pack.inventory_index, pack)
1158
self.text_index.add_index(pack.text_index, pack)
1159
self.signature_index.add_index(pack.signature_index, pack)
1160
1161
def _add_text_to_weave(self, file_id, revision_id, new_lines, parents,
1162
nostore_sha, random_revid):
1163
file_id_index = GraphIndexPrefixAdapter(
1164
self.text_index.combined_index,
1165
(file_id, ), 1,
1166
add_nodes_callback=self.text_index.add_callback)
1167
self.repo._text_knit._index._graph_index = file_id_index
1168
self.repo._text_knit._index._add_callback = file_id_index.add_nodes
1169
return self.repo._text_knit.add_lines_with_ghosts(
1170
revision_id, parents, new_lines, nostore_sha=nostore_sha,
1171
random_id=random_revid, check_content=False)[0:2]
1172
1173
def all_packs(self):
1174
"""Return a list of all the Pack objects this repository has.
1175
1176
Note that an in-progress pack being created is not returned.
1177
1178
:return: A list of Pack objects for all the packs in the repository.
1179
"""
1180
result = []
1181
for name in self.names():
1182
result.append(self.get_pack_by_name(name))
1183
return result
1184
1185
def autopack(self):
1186
"""Pack the pack collection incrementally.
1187
1188
This will not attempt global reorganisation or recompression,
1189
rather it will just ensure that the total number of packs does
1190
not grow without bound. It uses the _max_pack_count method to
1191
determine if autopacking is needed, and the pack_distribution
1192
method to determine the number of revisions in each pack.
1193
1194
If autopacking takes place then the packs name collection will have
1195
been flushed to disk - packing requires updating the name collection
1196
in synchronisation with certain steps. Otherwise the names collection
1197
is not flushed.
1198
1199
:return: True if packing took place.
1200
"""
1201
# XXX: Should not be needed when the management of indices is sane.
1202
total_revisions = self.revision_index.combined_index.key_count()
1203
total_packs = len(self._names)
1204
if self._max_pack_count(total_revisions) >= total_packs:
1205
return False
1206
# XXX: the following may want to be a class, to pack with a given
1207
# policy.
1208
mutter('Auto-packing repository %s, which has %d pack files, '
1209
'containing %d revisions into %d packs.', self, total_packs,
1210
total_revisions, self._max_pack_count(total_revisions))
1211
# determine which packs need changing
1212
pack_distribution = self.pack_distribution(total_revisions)
1213
existing_packs = []
1214
for pack in self.all_packs():
1215
revision_count = pack.get_revision_count()
1216
if revision_count == 0:
1217
# revision less packs are not generated by normal operation,
1218
# only by operations like sign-my-commits, and thus will not
1219
# tend to grow rapdily or without bound like commit containing
1220
# packs do - leave them alone as packing them really should
1221
# group their data with the relevant commit, and that may
1222
# involve rewriting ancient history - which autopack tries to
1223
# avoid. Alternatively we could not group the data but treat
1224
# each of these as having a single revision, and thus add
1225
# one revision for each to the total revision count, to get
1226
# a matching distribution.
1227
continue
1228
existing_packs.append((revision_count, pack))
1229
pack_operations = self.plan_autopack_combinations(
1230
existing_packs, pack_distribution)
1231
self._execute_pack_operations(pack_operations)
1232
return True
1233
1234
def _execute_pack_operations(self, pack_operations, _packer_class=Packer):
1235
"""Execute a series of pack operations.
1236
1237
:param pack_operations: A list of [revision_count, packs_to_combine].
1238
:param _packer_class: The class of packer to use (default: Packer).
1239
:return: None.
1240
"""
1241
for revision_count, packs in pack_operations:
1242
# we may have no-ops from the setup logic
1243
if len(packs) == 0:
1244
continue
1245
_packer_class(self, packs, '.autopack').pack()
1246
for pack in packs:
1247
self._remove_pack_from_memory(pack)
1248
# record the newly available packs and stop advertising the old
1249
# packs
1250
self._save_pack_names(clear_obsolete_packs=True)
1251
# Move the old packs out of the way now they are no longer referenced.
1252
for revision_count, packs in pack_operations:
1253
self._obsolete_packs(packs)
1254
1255
def lock_names(self):
1256
"""Acquire the mutex around the pack-names index.
1257
1258
This cannot be used in the middle of a read-only transaction on the
1259
repository.
1260
"""
1261
self.repo.control_files.lock_write()
1262
1263
def pack(self):
1264
"""Pack the pack collection totally."""
1265
self.ensure_loaded()
1266
total_packs = len(self._names)
1267
if total_packs < 2:
1268
# This is arguably wrong because we might not be optimal, but for
1269
# now lets leave it in. (e.g. reconcile -> one pack. But not
1270
# optimal.
1271
return
1272
total_revisions = self.revision_index.combined_index.key_count()
1273
# XXX: the following may want to be a class, to pack with a given
1274
# policy.
1275
mutter('Packing repository %s, which has %d pack files, '
1276
'containing %d revisions into 1 packs.', self, total_packs,
1277
total_revisions)
1278
# determine which packs need changing
1279
pack_distribution = [1]
1280
pack_operations = [[0, []]]
1281
for pack in self.all_packs():
1282
pack_operations[-1][0] += pack.get_revision_count()
1283
pack_operations[-1][1].append(pack)
1284
self._execute_pack_operations(pack_operations, OptimisingPacker)
1285
1286
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1287
"""Plan a pack operation.
1288
1289
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1290
tuples).
1291
:param pack_distribution: A list with the number of revisions desired
1292
in each pack.
1293
"""
1294
if len(existing_packs) <= len(pack_distribution):
1295
return []
1296
existing_packs.sort(reverse=True)
1297
pack_operations = [[0, []]]
1298
# plan out what packs to keep, and what to reorganise
1299
while len(existing_packs):
1300
# take the largest pack, and if its less than the head of the
1301
# distribution chart we will include its contents in the new pack for
1302
# that position. If its larger, we remove its size from the
1303
# distribution chart
1304
next_pack_rev_count, next_pack = existing_packs.pop(0)
1305
if next_pack_rev_count >= pack_distribution[0]:
1306
# this is already packed 'better' than this, so we can
1307
# not waste time packing it.
1308
while next_pack_rev_count > 0:
1309
next_pack_rev_count -= pack_distribution[0]
1310
if next_pack_rev_count >= 0:
1311
# more to go
1312
del pack_distribution[0]
1313
else:
1314
# didn't use that entire bucket up
1315
pack_distribution[0] = -next_pack_rev_count
1316
else:
1317
# add the revisions we're going to add to the next output pack
1318
pack_operations[-1][0] += next_pack_rev_count
1319
# allocate this pack to the next pack sub operation
1320
pack_operations[-1][1].append(next_pack)
1321
if pack_operations[-1][0] >= pack_distribution[0]:
1322
# this pack is used up, shift left.
1323
del pack_distribution[0]
1324
pack_operations.append([0, []])
1325
1326
return pack_operations
1327
1328
def ensure_loaded(self):
1329
# NB: if you see an assertion error here, its probably access against
1330
# an unlocked repo. Naughty.
1331
if not self.repo.is_locked():
1332
raise errors.ObjectNotLocked(self.repo)
1333
if self._names is None:
1334
self._names = {}
1335
self._packs_at_load = set()
1336
for index, key, value in self._iter_disk_pack_index():
1337
name = key[0]
1338
self._names[name] = self._parse_index_sizes(value)
1339
self._packs_at_load.add((key, value))
1340
# populate all the metadata.
1341
self.all_packs()
1342
1343
def _parse_index_sizes(self, value):
1344
"""Parse a string of index sizes."""
1345
return tuple([int(digits) for digits in value.split(' ')])
1346
1347
def get_pack_by_name(self, name):
1348
"""Get a Pack object by name.
1349
1350
:param name: The name of the pack - e.g. '123456'
1351
:return: A Pack object.
1352
"""
1353
try:
1354
return self._packs_by_name[name]
1355
except KeyError:
1356
rev_index = self._make_index(name, '.rix')
1357
inv_index = self._make_index(name, '.iix')
1358
txt_index = self._make_index(name, '.tix')
1359
sig_index = self._make_index(name, '.six')
1360
result = ExistingPack(self._pack_transport, name, rev_index,
1361
inv_index, txt_index, sig_index)
1362
self.add_pack_to_memory(result)
1363
return result
1364
1365
def allocate(self, a_new_pack):
1366
"""Allocate name in the list of packs.
1367
1368
:param a_new_pack: A NewPack instance to be added to the collection of
1369
packs for this repository.
1370
"""
1371
self.ensure_loaded()
1372
if a_new_pack.name in self._names:
1373
raise errors.BzrError(
1374
'Pack %r already exists in %s' % (a_new_pack.name, self))
1375
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1376
self.add_pack_to_memory(a_new_pack)
1377
1378
def _iter_disk_pack_index(self):
1379
"""Iterate over the contents of the pack-names index.
1380
1381
This is used when loading the list from disk, and before writing to
1382
detect updates from others during our write operation.
1383
:return: An iterator of the index contents.
1384
"""
1385
return GraphIndex(self.transport, 'pack-names', None
1386
).iter_all_entries()
1387
1388
def _make_index(self, name, suffix):
1389
size_offset = self._suffix_offsets[suffix]
1390
index_name = name + suffix
1391
index_size = self._names[name][size_offset]
1392
return GraphIndex(
1393
self._index_transport, index_name, index_size)
1394
1395
def _max_pack_count(self, total_revisions):
1396
"""Return the maximum number of packs to use for total revisions.
1397
1398
:param total_revisions: The total number of revisions in the
1399
repository.
1400
"""
1401
if not total_revisions:
1402
return 1
1403
digits = str(total_revisions)
1404
result = 0
1405
for digit in digits:
1406
result += int(digit)
1407
return result
1408
1409
def names(self):
1410
"""Provide an order to the underlying names."""
1411
return sorted(self._names.keys())
1412
1413
def _obsolete_packs(self, packs):
1414
"""Move a number of packs which have been obsoleted out of the way.
1415
1416
Each pack and its associated indices are moved out of the way.
1417
1418
Note: for correctness this function should only be called after a new
1419
pack names index has been written without these pack names, and with
1420
the names of packs that contain the data previously available via these
1421
packs.
1422
1423
:param packs: The packs to obsolete.
1424
:param return: None.
1425
"""
1426
for pack in packs:
1427
pack.pack_transport.rename(pack.file_name(),
1428
'../obsolete_packs/' + pack.file_name())
1429
# TODO: Probably needs to know all possible indices for this pack
1430
# - or maybe list the directory and move all indices matching this
1431
# name whether we recognize it or not?
1432
for suffix in ('.iix', '.six', '.tix', '.rix'):
1433
self._index_transport.rename(pack.name + suffix,
1434
'../obsolete_packs/' + pack.name + suffix)
1435
1436
def pack_distribution(self, total_revisions):
1437
"""Generate a list of the number of revisions to put in each pack.
1438
1439
:param total_revisions: The total number of revisions in the
1440
repository.
1441
"""
1442
if total_revisions == 0:
1443
return [0]
1444
digits = reversed(str(total_revisions))
1445
result = []
1446
for exponent, count in enumerate(digits):
1447
size = 10 ** exponent
1448
for pos in range(int(count)):
1449
result.append(size)
1450
return list(reversed(result))
1451
1452
def _pack_tuple(self, name):
1453
"""Return a tuple with the transport and file name for a pack name."""
1454
return self._pack_transport, name + '.pack'
1455
1456
def _remove_pack_from_memory(self, pack):
1457
"""Remove pack from the packs accessed by this repository.
1458
1459
Only affects memory state, until self._save_pack_names() is invoked.
1460
"""
1461
self._names.pop(pack.name)
1462
self._packs_by_name.pop(pack.name)
1463
self._remove_pack_indices(pack)
1464
1465
def _remove_pack_indices(self, pack):
1466
"""Remove the indices for pack from the aggregated indices."""
1467
self.revision_index.remove_index(pack.revision_index, pack)
1468
self.inventory_index.remove_index(pack.inventory_index, pack)
1469
self.text_index.remove_index(pack.text_index, pack)
1470
self.signature_index.remove_index(pack.signature_index, pack)
1471
1472
def reset(self):
1473
"""Clear all cached data."""
1474
# cached revision data
1475
self.repo._revision_knit = None
1476
self.revision_index.clear()
1477
# cached signature data
1478
self.repo._signature_knit = None
1479
self.signature_index.clear()
1480
# cached file text data
1481
self.text_index.clear()
1482
self.repo._text_knit = None
1483
# cached inventory data
1484
self.inventory_index.clear()
1485
# remove the open pack
1486
self._new_pack = None
1487
# information about packs.
1488
self._names = None
1489
self.packs = []
1490
self._packs_by_name = {}
1491
self._packs_at_load = None
1492
1493
def _make_index_map(self, index_suffix):
1494
"""Return information on existing indices.
1495
1496
:param suffix: Index suffix added to pack name.
1497
1498
:returns: (pack_map, indices) where indices is a list of GraphIndex
1499
objects, and pack_map is a mapping from those objects to the
1500
pack tuple they describe.
1501
"""
1502
# TODO: stop using this; it creates new indices unnecessarily.
1503
self.ensure_loaded()
1504
suffix_map = {'.rix': 'revision_index',
1505
'.six': 'signature_index',
1506
'.iix': 'inventory_index',
1507
'.tix': 'text_index',
1508
}
1509
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1510
suffix_map[index_suffix])
1511
1512
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1513
"""Convert a list of packs to an index pack map and index list.
1514
1515
:param packs: The packs list to process.
1516
:param index_attribute: The attribute that the desired index is found
1517
on.
1518
:return: A tuple (map, list) where map contains the dict from
1519
index:pack_tuple, and lsit contains the indices in the same order
1520
as the packs list.
1521
"""
1522
indices = []
1523
pack_map = {}
1524
for pack in packs:
1525
index = getattr(pack, index_attribute)
1526
indices.append(index)
1527
pack_map[index] = (pack.pack_transport, pack.file_name())
1528
return pack_map, indices
1529
1530
def _index_contents(self, pack_map, key_filter=None):
1531
"""Get an iterable of the index contents from a pack_map.
1532
1533
:param pack_map: A map from indices to pack details.
1534
:param key_filter: An optional filter to limit the
1535
keys returned.
1536
"""
1537
indices = [index for index in pack_map.iterkeys()]
1538
all_index = CombinedGraphIndex(indices)
1539
if key_filter is None:
1540
return all_index.iter_all_entries()
1541
else:
1542
return all_index.iter_entries(key_filter)
1543
1544
def _unlock_names(self):
1545
"""Release the mutex around the pack-names index."""
1546
self.repo.control_files.unlock()
1547
1548
def _save_pack_names(self, clear_obsolete_packs=False):
1549
"""Save the list of packs.
1550
1551
This will take out the mutex around the pack names list for the
1552
duration of the method call. If concurrent updates have been made, a
1553
three-way merge between the current list and the current in memory list
1554
is performed.
1555
1556
:param clear_obsolete_packs: If True, clear out the contents of the
1557
obsolete_packs directory.
1558
"""
1559
self.lock_names()
1560
try:
1561
builder = GraphIndexBuilder()
1562
# load the disk nodes across
1563
disk_nodes = set()
1564
for index, key, value in self._iter_disk_pack_index():
1565
disk_nodes.add((key, value))
1566
# do a two-way diff against our original content
1567
current_nodes = set()
1568
for name, sizes in self._names.iteritems():
1569
current_nodes.add(
1570
((name, ), ' '.join(str(size) for size in sizes)))
1571
deleted_nodes = self._packs_at_load - current_nodes
1572
new_nodes = current_nodes - self._packs_at_load
1573
disk_nodes.difference_update(deleted_nodes)
1574
disk_nodes.update(new_nodes)
1575
# TODO: handle same-name, index-size-changes here -
1576
# e.g. use the value from disk, not ours, *unless* we're the one
1577
# changing it.
1578
for key, value in disk_nodes:
1579
builder.add_node(key, value)
1580
self.transport.put_file('pack-names', builder.finish(),
1581
mode=self.repo.control_files._file_mode)
1582
# move the baseline forward
1583
self._packs_at_load = disk_nodes
1584
# now clear out the obsolete packs directory
1585
if clear_obsolete_packs:
1586
self.transport.clone('obsolete_packs').delete_multi(
1587
self.transport.list_dir('obsolete_packs'))
1588
finally:
1589
self._unlock_names()
1590
# synchronise the memory packs list with what we just wrote:
1591
new_names = dict(disk_nodes)
1592
# drop no longer present nodes
1593
for pack in self.all_packs():
1594
if (pack.name,) not in new_names:
1595
self._remove_pack_from_memory(pack)
1596
# add new nodes/refresh existing ones
1597
for key, value in disk_nodes:
1598
name = key[0]
1599
sizes = self._parse_index_sizes(value)
1600
if name in self._names:
1601
# existing
1602
if sizes != self._names[name]:
1603
# the pack for name has had its indices replaced - rare but
1604
# important to handle. XXX: probably can never happen today
1605
# because the three-way merge code above does not handle it
1606
# - you may end up adding the same key twice to the new
1607
# disk index because the set values are the same, unless
1608
# the only index shows up as deleted by the set difference
1609
# - which it may. Until there is a specific test for this,
1610
# assume its broken. RBC 20071017.
1611
self._remove_pack_from_memory(self.get_pack_by_name(name))
1612
self._names[name] = sizes
1613
self.get_pack_by_name(name)
1614
else:
1615
# new
1616
self._names[name] = sizes
1617
self.get_pack_by_name(name)
1618
1619
def _start_write_group(self):
1620
# Do not permit preparation for writing if we're not in a 'write lock'.
1621
if not self.repo.is_write_locked():
1622
raise errors.NotWriteLocked(self)
1623
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1624
self._pack_transport, upload_suffix='.pack',
1625
file_mode=self.repo.control_files._file_mode)
1626
# allow writing: queue writes to a new index
1627
self.revision_index.add_writable_index(self._new_pack.revision_index,
1628
self._new_pack)
1629
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
1630
self._new_pack)
1631
self.text_index.add_writable_index(self._new_pack.text_index,
1632
self._new_pack)
1633
self.signature_index.add_writable_index(self._new_pack.signature_index,
1634
self._new_pack)
1635
1636
# reused revision and signature knits may need updating
1637
#
1638
# "Hysterical raisins. client code in bzrlib grabs those knits outside
1639
# of write groups and then mutates it inside the write group."
1640
if self.repo._revision_knit is not None:
1641
self.repo._revision_knit._index._add_callback = \
1642
self.revision_index.add_callback
1643
if self.repo._signature_knit is not None:
1644
self.repo._signature_knit._index._add_callback = \
1645
self.signature_index.add_callback
1646
# create a reused knit object for text addition in commit.
1647
self.repo._text_knit = self.repo.weave_store.get_weave_or_empty(
1648
'all-texts', None)
1649
1650
def _abort_write_group(self):
1651
# FIXME: just drop the transient index.
1652
# forget what names there are
1653
if self._new_pack is not None:
1654
self._new_pack.abort()
1655
self._remove_pack_indices(self._new_pack)
1656
self._new_pack = None
1657
self.repo._text_knit = None
1658
1659
def _commit_write_group(self):
1660
self._remove_pack_indices(self._new_pack)
1661
if self._new_pack.data_inserted():
1662
# get all the data to disk and read to use
1663
self._new_pack.finish()
1664
self.allocate(self._new_pack)
1665
self._new_pack = None
1666
if not self.autopack():
1667
# when autopack takes no steps, the names list is still
1668
# unsaved.
1669
self._save_pack_names()
1670
else:
1671
self._new_pack.abort()
1672
self._new_pack = None
1673
self.repo._text_knit = None
1674
1675
1676
class KnitPackRevisionStore(KnitRevisionStore):
1677
"""An object to adapt access from RevisionStore's to use KnitPacks.
1678
1679
This class works by replacing the original RevisionStore.
1680
We need to do this because the KnitPackRevisionStore is less
1681
isolated in its layering - it uses services from the repo.
1682
"""
1683
1684
def __init__(self, repo, transport, revisionstore):
1685
"""Create a KnitPackRevisionStore on repo with revisionstore.
1686
1687
This will store its state in the Repository, use the
1688
indices to provide a KnitGraphIndex,
1689
and at the end of transactions write new indices.
1690
"""
1691
KnitRevisionStore.__init__(self, revisionstore.versioned_file_store)
1692
self.repo = repo
1693
self._serializer = revisionstore._serializer
1694
self.transport = transport
1695
1696
def get_revision_file(self, transaction):
1697
"""Get the revision versioned file object."""
1698
if getattr(self.repo, '_revision_knit', None) is not None:
1699
return self.repo._revision_knit
1700
self.repo._pack_collection.ensure_loaded()
1701
add_callback = self.repo._pack_collection.revision_index.add_callback
1702
# setup knit specific objects
1703
knit_index = KnitGraphIndex(
1704
self.repo._pack_collection.revision_index.combined_index,
1705
add_callback=add_callback)
1706
self.repo._revision_knit = knit.KnitVersionedFile(
1707
'revisions', self.transport.clone('..'),
1708
self.repo.control_files._file_mode,
1709
create=False, access_mode=self.repo._access_mode(),
1710
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1711
access_method=self.repo._pack_collection.revision_index.knit_access)
1712
return self.repo._revision_knit
1713
1714
def get_signature_file(self, transaction):
1715
"""Get the signature versioned file object."""
1716
if getattr(self.repo, '_signature_knit', None) is not None:
1717
return self.repo._signature_knit
1718
self.repo._pack_collection.ensure_loaded()
1719
add_callback = self.repo._pack_collection.signature_index.add_callback
1720
# setup knit specific objects
1721
knit_index = KnitGraphIndex(
1722
self.repo._pack_collection.signature_index.combined_index,
1723
add_callback=add_callback, parents=False)
1724
self.repo._signature_knit = knit.KnitVersionedFile(
1725
'signatures', self.transport.clone('..'),
1726
self.repo.control_files._file_mode,
1727
create=False, access_mode=self.repo._access_mode(),
1728
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1729
access_method=self.repo._pack_collection.signature_index.knit_access)
1730
return self.repo._signature_knit
1731
1732
1733
class KnitPackTextStore(VersionedFileStore):
1734
"""Presents a TextStore abstraction on top of packs.
1735
1736
This class works by replacing the original VersionedFileStore.
1737
We need to do this because the KnitPackRevisionStore is less
1738
isolated in its layering - it uses services from the repo and shares them
1739
with all the data written in a single write group.
1740
"""
1741
1742
def __init__(self, repo, transport, weavestore):
1743
"""Create a KnitPackTextStore on repo with weavestore.
1744
1745
This will store its state in the Repository, use the
1746
indices FileNames to provide a KnitGraphIndex,
1747
and at the end of transactions write new indices.
1748
"""
1749
# don't call base class constructor - it's not suitable.
1750
# no transient data stored in the transaction
1751
# cache.
1752
self._precious = False
1753
self.repo = repo
1754
self.transport = transport
1755
self.weavestore = weavestore
1756
# XXX for check() which isn't updated yet
1757
self._transport = weavestore._transport
1758
1759
def get_weave_or_empty(self, file_id, transaction):
1760
"""Get a 'Knit' backed by the .tix indices.
1761
1762
The transaction parameter is ignored.
1763
"""
1764
self.repo._pack_collection.ensure_loaded()
1765
add_callback = self.repo._pack_collection.text_index.add_callback
1766
# setup knit specific objects
1767
file_id_index = GraphIndexPrefixAdapter(
1768
self.repo._pack_collection.text_index.combined_index,
1769
(file_id, ), 1, add_nodes_callback=add_callback)
1770
knit_index = KnitGraphIndex(file_id_index,
1771
add_callback=file_id_index.add_nodes,
1772
deltas=True, parents=True)
1773
return knit.KnitVersionedFile('text:' + file_id,
1774
self.transport.clone('..'),
1775
None,
1776
index=knit_index,
1777
access_method=self.repo._pack_collection.text_index.knit_access,
1778
factory=knit.KnitPlainFactory())
1779
1780
get_weave = get_weave_or_empty
1781
1782
def __iter__(self):
1783
"""Generate a list of the fileids inserted, for use by check."""
1784
self.repo._pack_collection.ensure_loaded()
1785
ids = set()
1786
for index, key, value, refs in \
1787
self.repo._pack_collection.text_index.combined_index.iter_all_entries():
1788
ids.add(key[0])
1789
return iter(ids)
1790
1791
1792
class InventoryKnitThunk(object):
1793
"""An object to manage thunking get_inventory_weave to pack based knits."""
1794
1795
def __init__(self, repo, transport):
1796
"""Create an InventoryKnitThunk for repo at transport.
1797
1798
This will store its state in the Repository, use the
1799
indices FileNames to provide a KnitGraphIndex,
1800
and at the end of transactions write a new index..
1801
"""
1802
self.repo = repo
1803
self.transport = transport
1804
1805
def get_weave(self):
1806
"""Get a 'Knit' that contains inventory data."""
1807
self.repo._pack_collection.ensure_loaded()
1808
add_callback = self.repo._pack_collection.inventory_index.add_callback
1809
# setup knit specific objects
1810
knit_index = KnitGraphIndex(
1811
self.repo._pack_collection.inventory_index.combined_index,
1812
add_callback=add_callback, deltas=True, parents=True)
1813
return knit.KnitVersionedFile(
1814
'inventory', self.transport.clone('..'),
1815
self.repo.control_files._file_mode,
1816
create=False, access_mode=self.repo._access_mode(),
1817
index=knit_index, delta=True, factory=knit.KnitPlainFactory(),
1818
access_method=self.repo._pack_collection.inventory_index.knit_access)
1819
1820
1821
class KnitPackRepository(KnitRepository):
1822
"""Experimental graph-knit using repository."""
1823
1824
def __init__(self, _format, a_bzrdir, control_files, _revision_store,
1825
control_store, text_store, _commit_builder_class, _serializer):
1826
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
1827
_revision_store, control_store, text_store, _commit_builder_class,
1828
_serializer)
1829
index_transport = control_files._transport.clone('indices')
1830
self._pack_collection = RepositoryPackCollection(self, control_files._transport,
1831
index_transport,
1832
control_files._transport.clone('upload'),
1833
control_files._transport.clone('packs'))
1834
self._revision_store = KnitPackRevisionStore(self, index_transport, self._revision_store)
1835
self.weave_store = KnitPackTextStore(self, index_transport, self.weave_store)
1836
self._inv_thunk = InventoryKnitThunk(self, index_transport)
1837
# True when the repository object is 'write locked' (as opposed to the
1838
# physical lock only taken out around changes to the pack-names list.)
1839
# Another way to represent this would be a decorator around the control
1840
# files object that presents logical locks as physical ones - if this
1841
# gets ugly consider that alternative design. RBC 20071011
1842
self._write_lock_count = 0
1843
self._transaction = None
1844
# for tests
1845
self._reconcile_does_inventory_gc = True
1846
self._reconcile_fixes_text_parents = True
1847
self._reconcile_backsup_inventory = False
1848
1849
def _abort_write_group(self):
1850
self._pack_collection._abort_write_group()
1851
1852
def _access_mode(self):
1853
"""Return 'w' or 'r' for depending on whether a write lock is active.
1854
1855
This method is a helper for the Knit-thunking support objects.
1856
"""
1857
if self.is_write_locked():
1858
return 'w'
1859
return 'r'
1860
1861
def _find_inconsistent_revision_parents(self):
1862
"""Find revisions with incorrectly cached parents.
1863
1864
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1865
parents-in-revision).
1866
"""
1867
if not self.is_locked():
1868
raise errors.ObjectNotLocked(self)
1869
pb = ui.ui_factory.nested_progress_bar()
1870
result = []
1871
try:
1872
revision_nodes = self._pack_collection.revision_index \
1873
.combined_index.iter_all_entries()
1874
index_positions = []
1875
# Get the cached index values for all revisions, and also the location
1876
# in each index of the revision text so we can perform linear IO.
1877
for index, key, value, refs in revision_nodes:
1878
pos, length = value[1:].split(' ')
1879
index_positions.append((index, int(pos), key[0],
1880
tuple(parent[0] for parent in refs[0])))
1881
pb.update("Reading revision index.", 0, 0)
1882
index_positions.sort()
1883
batch_count = len(index_positions) / 1000 + 1
1884
pb.update("Checking cached revision graph.", 0, batch_count)
1885
for offset in xrange(batch_count):
1886
pb.update("Checking cached revision graph.", offset)
1887
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1888
if not to_query:
1889
break
1890
rev_ids = [item[2] for item in to_query]
1891
revs = self.get_revisions(rev_ids)
1892
for revision, item in zip(revs, to_query):
1893
index_parents = item[3]
1894
rev_parents = tuple(revision.parent_ids)
1895
if index_parents != rev_parents:
1896
result.append((revision.revision_id, index_parents, rev_parents))
1897
finally:
1898
pb.finished()
1899
return result
1900
1901
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
1902
def get_parents(self, revision_ids):
1903
"""See graph._StackedParentsProvider.get_parents."""
1904
parent_map = self.get_parent_map(revision_ids)
1905
return [parent_map.get(r, None) for r in revision_ids]
1906
1907
def get_parent_map(self, keys):
1908
"""See graph._StackedParentsProvider.get_parent_map
1909
1910
This implementation accesses the combined revision index to provide
1911
answers.
1912
"""
1913
self._pack_collection.ensure_loaded()
1914
index = self._pack_collection.revision_index.combined_index
1915
keys = set(keys)
1916
if _mod_revision.NULL_REVISION in keys:
1917
keys.discard(_mod_revision.NULL_REVISION)
1918
found_parents = {_mod_revision.NULL_REVISION:()}
1919
else:
1920
found_parents = {}
1921
search_keys = set((revision_id,) for revision_id in keys)
1922
for index, key, value, refs in index.iter_entries(search_keys):
1923
parents = refs[0]
1924
if not parents:
1925
parents = (_mod_revision.NULL_REVISION,)
1926
else:
1927
parents = tuple(parent[0] for parent in parents)
1928
found_parents[key[0]] = parents
1929
return found_parents
1930
1931
@needs_read_lock
1932
def get_revision_graph(self, revision_id=None):
1933
"""Return a dictionary containing the revision graph.
1934
1935
:param revision_id: The revision_id to get a graph from. If None, then
1936
the entire revision graph is returned. This is a deprecated mode of
1937
operation and will be removed in the future.
1938
:return: a dictionary of revision_id->revision_parents_list.
1939
"""
1940
if 'evil' in debug.debug_flags:
1941
mutter_callsite(3,
1942
"get_revision_graph scales with size of history.")
1943
# special case NULL_REVISION
1944
if revision_id == _mod_revision.NULL_REVISION:
1945
return {}
1946
if revision_id is None:
1947
revision_vf = self._get_revision_vf()
1948
return revision_vf.get_graph()
1949
g = self.get_graph()
1950
first = g.get_parent_map([revision_id])
1951
if revision_id not in first:
1952
raise errors.NoSuchRevision(self, revision_id)
1953
else:
1954
ancestry = {}
1955
children = {}
1956
NULL_REVISION = _mod_revision.NULL_REVISION
1957
ghosts = set([NULL_REVISION])
1958
for rev_id, parent_ids in g.iter_ancestry([revision_id]):
1959
if parent_ids is None: # This is a ghost
1960
ghosts.add(rev_id)
1961
continue
1962
ancestry[rev_id] = parent_ids
1963
for p in parent_ids:
1964
if p in children:
1965
children[p].append(rev_id)
1966
else:
1967
children[p] = [rev_id]
1968
1969
if NULL_REVISION in ancestry:
1970
del ancestry[NULL_REVISION]
1971
1972
# Find all nodes that reference a ghost, and filter the ghosts out
1973
# of their parent lists. To preserve the order of parents, and
1974
# avoid double filtering nodes, we just find all children first,
1975
# and then filter.
1976
children_of_ghosts = set()
1977
for ghost in ghosts:
1978
children_of_ghosts.update(children[ghost])
1979
1980
for child in children_of_ghosts:
1981
ancestry[child] = tuple(p for p in ancestry[child]
1982
if p not in ghosts)
1983
return ancestry
1984
1985
def has_revisions(self, revision_ids):
1986
"""See Repository.has_revisions()."""
1987
revision_ids = set(revision_ids)
1988
result = revision_ids.intersection(
1989
set([None, _mod_revision.NULL_REVISION]))
1990
revision_ids.difference_update(result)
1991
index = self._pack_collection.revision_index.combined_index
1992
keys = [(revision_id,) for revision_id in revision_ids]
1993
result.update(node[1][0] for node in index.iter_entries(keys))
1994
return result
1995
1996
def _make_parents_provider(self):
1997
return graph.CachingParentsProvider(self)
1998
1999
def _refresh_data(self):
2000
if self._write_lock_count == 1 or (
2001
self.control_files._lock_count == 1 and
2002
self.control_files._lock_mode == 'r'):
2003
# forget what names there are
2004
self._pack_collection.reset()
2005
# XXX: Better to do an in-memory merge when acquiring a new lock -
2006
# factor out code from _save_pack_names.
2007
self._pack_collection.ensure_loaded()
2008
2009
def _start_write_group(self):
2010
self._pack_collection._start_write_group()
2011
2012
def _commit_write_group(self):
2013
return self._pack_collection._commit_write_group()
2014
2015
def get_inventory_weave(self):
2016
return self._inv_thunk.get_weave()
2017
2018
def get_transaction(self):
2019
if self._write_lock_count:
2020
return self._transaction
2021
else:
2022
return self.control_files.get_transaction()
2023
2024
def is_locked(self):
2025
return self._write_lock_count or self.control_files.is_locked()
2026
2027
def is_write_locked(self):
2028
return self._write_lock_count
2029
2030
def lock_write(self, token=None):
2031
if not self._write_lock_count and self.is_locked():
2032
raise errors.ReadOnlyError(self)
2033
self._write_lock_count += 1
2034
if self._write_lock_count == 1:
2035
from bzrlib import transactions
2036
self._transaction = transactions.WriteTransaction()
2037
self._refresh_data()
2038
2039
def lock_read(self):
2040
if self._write_lock_count:
2041
self._write_lock_count += 1
2042
else:
2043
self.control_files.lock_read()
2044
self._refresh_data()
2045
2046
def leave_lock_in_place(self):
2047
# not supported - raise an error
2048
raise NotImplementedError(self.leave_lock_in_place)
2049
2050
def dont_leave_lock_in_place(self):
2051
# not supported - raise an error
2052
raise NotImplementedError(self.dont_leave_lock_in_place)
2053
2054
@needs_write_lock
2055
def pack(self):
2056
"""Compress the data within the repository.
2057
2058
This will pack all the data to a single pack. In future it may
2059
recompress deltas or do other such expensive operations.
2060
"""
2061
self._pack_collection.pack()
2062
2063
@needs_write_lock
2064
def reconcile(self, other=None, thorough=False):
2065
"""Reconcile this repository."""
2066
from bzrlib.reconcile import PackReconciler
2067
reconciler = PackReconciler(self, thorough=thorough)
2068
reconciler.reconcile()
2069
return reconciler
2070
2071
def unlock(self):
2072
if self._write_lock_count == 1 and self._write_group is not None:
2073
self.abort_write_group()
2074
self._transaction = None
2075
self._write_lock_count = 0
2076
raise errors.BzrError(
2077
'Must end write group before releasing write lock on %s'
2078
% self)
2079
if self._write_lock_count:
2080
self._write_lock_count -= 1
2081
if not self._write_lock_count:
2082
transaction = self._transaction
2083
self._transaction = None
2084
transaction.finish()
2085
else:
2086
self.control_files.unlock()
2087
2088
2089
class RepositoryFormatPack(MetaDirRepositoryFormat):
2090
"""Format logic for pack structured repositories.
2091
2092
This repository format has:
2093
- a list of packs in pack-names
2094
- packs in packs/NAME.pack
2095
- indices in indices/NAME.{iix,six,tix,rix}
2096
- knit deltas in the packs, knit indices mapped to the indices.
2097
- thunk objects to support the knits programming API.
2098
- a format marker of its own
2099
- an optional 'shared-storage' flag
2100
- an optional 'no-working-trees' flag
2101
- a LockDir lock
2102
"""
2103
2104
# Set this attribute in derived classes to control the repository class
2105
# created by open and initialize.
2106
repository_class = None
2107
# Set this attribute in derived classes to control the
2108
# _commit_builder_class that the repository objects will have passed to
2109
# their constructor.
2110
_commit_builder_class = None
2111
# Set this attribute in derived clases to control the _serializer that the
2112
# repository objects will have passed to their constructor.
2113
_serializer = None
2114
# External references are not supported in pack repositories yet.
2115
supports_external_lookups = False
2116
2117
def _get_control_store(self, repo_transport, control_files):
2118
"""Return the control store for this repository."""
2119
return VersionedFileStore(
2120
repo_transport,
2121
prefixed=False,
2122
file_mode=control_files._file_mode,
2123
versionedfile_class=knit.KnitVersionedFile,
2124
versionedfile_kwargs={'factory': knit.KnitPlainFactory()},
2125
)
2126
2127
def _get_revision_store(self, repo_transport, control_files):
2128
"""See RepositoryFormat._get_revision_store()."""
2129
versioned_file_store = VersionedFileStore(
2130
repo_transport,
2131
file_mode=control_files._file_mode,
2132
prefixed=False,
2133
precious=True,
2134
versionedfile_class=knit.KnitVersionedFile,
2135
versionedfile_kwargs={'delta': False,
2136
'factory': knit.KnitPlainFactory(),
2137
},
2138
escaped=True,
2139
)
2140
return KnitRevisionStore(versioned_file_store)
2141
2142
def _get_text_store(self, transport, control_files):
2143
"""See RepositoryFormat._get_text_store()."""
2144
return self._get_versioned_file_store('knits',
2145
transport,
2146
control_files,
2147
versionedfile_class=knit.KnitVersionedFile,
2148
versionedfile_kwargs={
2149
'create_parent_dir': True,
2150
'delay_create': True,
2151
'dir_mode': control_files._dir_mode,
2152
},
2153
escaped=True)
2154
2155
def initialize(self, a_bzrdir, shared=False):
2156
"""Create a pack based repository.
2157
2158
:param a_bzrdir: bzrdir to contain the new repository; must already
2159
be initialized.
2160
:param shared: If true the repository will be initialized as a shared
2161
repository.
2162
"""
2163
mutter('creating repository in %s.', a_bzrdir.transport.base)
2164
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
2165
builder = GraphIndexBuilder()
2166
files = [('pack-names', builder.finish())]
2167
utf8_files = [('format', self.get_format_string())]
2168
2169
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
2170
return self.open(a_bzrdir=a_bzrdir, _found=True)
2171
2172
def open(self, a_bzrdir, _found=False, _override_transport=None):
2173
"""See RepositoryFormat.open().
2174
2175
:param _override_transport: INTERNAL USE ONLY. Allows opening the
2176
repository at a slightly different url
2177
than normal. I.e. during 'upgrade'.
2178
"""
2179
if not _found:
2180
format = RepositoryFormat.find_format(a_bzrdir)
2181
assert format.__class__ == self.__class__
2182
if _override_transport is not None:
2183
repo_transport = _override_transport
2184
else:
2185
repo_transport = a_bzrdir.get_repository_transport(None)
2186
control_files = lockable_files.LockableFiles(repo_transport,
2187
'lock', lockdir.LockDir)
2188
text_store = self._get_text_store(repo_transport, control_files)
2189
control_store = self._get_control_store(repo_transport, control_files)
2190
_revision_store = self._get_revision_store(repo_transport, control_files)
2191
return self.repository_class(_format=self,
2192
a_bzrdir=a_bzrdir,
2193
control_files=control_files,
2194
_revision_store=_revision_store,
2195
control_store=control_store,
2196
text_store=text_store,
2197
_commit_builder_class=self._commit_builder_class,
2198
_serializer=self._serializer)
2199
2200
2201
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2202
"""A no-subtrees parameterized Pack repository.
2203
2204
This format was introduced in 0.92.
2205
"""
2206
2207
repository_class = KnitPackRepository
2208
_commit_builder_class = PackCommitBuilder
2209
_serializer = xml5.serializer_v5
2210
2211
def _get_matching_bzrdir(self):
2212
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2213
2214
def _ignore_setting_bzrdir(self, format):
2215
pass
2216
2217
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2218
2219
def get_format_string(self):
2220
"""See RepositoryFormat.get_format_string()."""
2221
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2222
2223
def get_format_description(self):
2224
"""See RepositoryFormat.get_format_description()."""
2225
return "Packs containing knits without subtree support"
2226
2227
def check_conversion_target(self, target_format):
2228
pass
2229
2230
2231
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2232
"""A subtrees parameterized Pack repository.
2233
2234
This repository format uses the xml7 serializer to get:
2235
- support for recording full info about the tree root
2236
- support for recording tree-references
2237
2238
This format was introduced in 0.92.
2239
"""
2240
2241
repository_class = KnitPackRepository
2242
_commit_builder_class = PackRootCommitBuilder
2243
rich_root_data = True
2244
supports_tree_reference = True
2245
_serializer = xml7.serializer_v7
2246
2247
def _get_matching_bzrdir(self):
2248
return bzrdir.format_registry.make_bzrdir(
2249
'pack-0.92-subtree')
2250
2251
def _ignore_setting_bzrdir(self, format):
2252
pass
2253
2254
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2255
2256
def check_conversion_target(self, target_format):
2257
if not target_format.rich_root_data:
2258
raise errors.BadConversionTarget(
2259
'Does not support rich root data.', target_format)
2260
if not getattr(target_format, 'supports_tree_reference', False):
2261
raise errors.BadConversionTarget(
2262
'Does not support nested trees', target_format)
2263
2264
def get_format_string(self):
2265
"""See RepositoryFormat.get_format_string()."""
2266
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2267
2268
def get_format_description(self):
2269
"""See RepositoryFormat.get_format_description()."""
2270
return "Packs containing knits with subtree support\n"
2271
2272
2273
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2274
"""A rich-root, no subtrees parameterized Pack repository.
2275
2276
This repository format uses the xml6 serializer to get:
2277
- support for recording full info about the tree root
2278
2279
This format was introduced in 1.0.
2280
"""
2281
2282
repository_class = KnitPackRepository
2283
_commit_builder_class = PackRootCommitBuilder
2284
rich_root_data = True
2285
supports_tree_reference = False
2286
_serializer = xml6.serializer_v6
2287
2288
def _get_matching_bzrdir(self):
2289
return bzrdir.format_registry.make_bzrdir(
2290
'rich-root-pack')
2291
2292
def _ignore_setting_bzrdir(self, format):
2293
pass
2294
2295
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2296
2297
def check_conversion_target(self, target_format):
2298
if not target_format.rich_root_data:
2299
raise errors.BadConversionTarget(
2300
'Does not support rich root data.', target_format)
2301
2302
def get_format_string(self):
2303
"""See RepositoryFormat.get_format_string()."""
2304
return ("Bazaar pack repository format 1 with rich root"
2305
" (needs bzr 1.0)\n")
2306
2307
def get_format_description(self):
2308
"""See RepositoryFormat.get_format_description()."""
2309
return "Packs containing knits with rich root support\n"
2310
2311
2312
class RepositoryFormatPackDevelopment0(RepositoryFormatPack):
2313
"""A no-subtrees development repository.
2314
2315
This format should be retained until the second release after bzr 1.0.
2316
2317
No changes to the disk behaviour from pack-0.92.
2318
"""
2319
2320
repository_class = KnitPackRepository
2321
_commit_builder_class = PackCommitBuilder
2322
_serializer = xml5.serializer_v5
2323
2324
def _get_matching_bzrdir(self):
2325
return bzrdir.format_registry.make_bzrdir('development0')
2326
2327
def _ignore_setting_bzrdir(self, format):
2328
pass
2329
2330
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2331
2332
def get_format_string(self):
2333
"""See RepositoryFormat.get_format_string()."""
2334
return "Bazaar development format 0 (needs bzr.dev from before 1.3)\n"
2335
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\n")
2340
2341
def check_conversion_target(self, target_format):
2342
pass
2343
2344
2345
class RepositoryFormatPackDevelopment0Subtree(RepositoryFormatPack):
2346
"""A subtrees development repository.
2347
2348
This format should be retained until the second release after bzr 1.0.
2349
2350
No changes to the disk behaviour from pack-0.92-subtree.
2351
"""
2352
2353
repository_class = KnitPackRepository
2354
_commit_builder_class = PackRootCommitBuilder
2355
rich_root_data = True
2356
supports_tree_reference = True
2357
_serializer = xml7.serializer_v7
2358
2359
def _get_matching_bzrdir(self):
2360
return bzrdir.format_registry.make_bzrdir(
2361
'development0-subtree')
2362
2363
def _ignore_setting_bzrdir(self, format):
2364
pass
2365
2366
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2367
2368
def check_conversion_target(self, target_format):
2369
if not target_format.rich_root_data:
2370
raise errors.BadConversionTarget(
2371
'Does not support rich root data.', target_format)
2372
if not getattr(target_format, 'supports_tree_reference', False):
2373
raise errors.BadConversionTarget(
2374
'Does not support nested trees', target_format)
2375
2376
def get_format_string(self):
2377
"""See RepositoryFormat.get_format_string()."""
2378
return ("Bazaar development format 0 with subtree support "
2379
"(needs bzr.dev from before 1.3)\n")
2380
2381
def get_format_description(self):
2382
"""See RepositoryFormat.get_format_description()."""
2383
return ("Development repository format, currently the same as "
2384
"pack-0.92-subtree\n")
2385
2386
Loggerhead is a web-based interface for Breezy
Version: 2.0.1