1
# Copyright (C) 2005, 2006, 2007 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
from bzrlib.lazy_import import lazy_import
18
lazy_import(globals(), """
19
from itertools import izip
29
from bzrlib.index import (
34
GraphIndexPrefixAdapter,
36
from bzrlib.knit import KnitGraphIndex, _PackAccess, _KnitData
37
from bzrlib.osutils import rand_chars
38
from bzrlib.pack import ContainerWriter
39
from bzrlib.store import revision
54
from bzrlib.decorators import needs_read_lock, needs_write_lock
55
from bzrlib.repofmt.knitrepo import KnitRepository
56
from bzrlib.repository import (
59
MetaDirRepositoryFormat,
62
import bzrlib.revision as _mod_revision
63
from bzrlib.store.revision.knit import KnitRevisionStore
64
from bzrlib.store.versioned import VersionedFileStore
65
from bzrlib.trace import mutter, note, warning
68
class PackCommitBuilder(CommitBuilder):
69
"""A subclass of CommitBuilder to add texts with pack semantics.
71
Specifically this uses one knit object rather than one knit object per
72
added text, reducing memory and object pressure.
75
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
76
return self.repository._pack_collection._add_text_to_weave(file_id,
77
self._new_revision_id, new_lines, parents, nostore_sha,
81
class PackRootCommitBuilder(RootCommitBuilder):
82
"""A subclass of RootCommitBuilder to add texts with pack semantics.
84
Specifically this uses one knit object rather than one knit object per
85
added text, reducing memory and object pressure.
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,
95
"""An in memory proxy for a pack and its indices.
97
This is a base class that is not directly used, instead the classes
98
ExistingPack and NewPack are used.
101
def __init__(self, revision_index, inventory_index, text_index,
103
"""Create a pack instance.
105
:param revision_index: A GraphIndex for determining what revisions are
106
present in the Pack and accessing the locations of their texts.
107
:param inventory_index: A GraphIndex for determining what inventories are
108
present in the Pack and accessing the locations of their
110
:param text_index: A GraphIndex for determining what file texts
111
are present in the pack and accessing the locations of their
112
texts/deltas (via (fileid, revisionid) tuples).
113
:param revision_index: A GraphIndex for determining what signatures are
114
present in the Pack and accessing the locations of their texts.
116
self.revision_index = revision_index
117
self.inventory_index = inventory_index
118
self.text_index = text_index
119
self.signature_index = signature_index
121
def access_tuple(self):
122
"""Return a tuple (transport, name) for the pack content."""
123
return self.pack_transport, self.file_name()
126
"""Get the file name for the pack on disk."""
127
return self.name + '.pack'
129
def get_revision_count(self):
130
return self.revision_index.key_count()
132
def inventory_index_name(self, name):
133
"""The inv index is the name + .iix."""
134
return self.index_name('inventory', name)
136
def revision_index_name(self, name):
137
"""The revision index is the name + .rix."""
138
return self.index_name('revision', name)
140
def signature_index_name(self, name):
141
"""The signature index is the name + .six."""
142
return self.index_name('signature', name)
144
def text_index_name(self, name):
145
"""The text index is the name + .tix."""
146
return self.index_name('text', name)
149
class ExistingPack(Pack):
150
"""An in memory proxy for an existing .pack and its disk indices."""
152
def __init__(self, pack_transport, name, revision_index, inventory_index,
153
text_index, signature_index):
154
"""Create an ExistingPack object.
156
:param pack_transport: The transport where the pack file resides.
157
:param name: The name of the pack on disk in the pack_transport.
159
Pack.__init__(self, revision_index, inventory_index, text_index,
162
self.pack_transport = pack_transport
163
assert None not in (revision_index, inventory_index, text_index,
164
signature_index, name, pack_transport)
166
def __eq__(self, other):
167
return self.__dict__ == other.__dict__
169
def __ne__(self, other):
170
return not self.__eq__(other)
173
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
174
id(self), self.transport, self.name)
178
"""An in memory proxy for a pack which is being created."""
180
# A map of index 'type' to the file extension and position in the
182
index_definitions = {
183
'revision': ('.rix', 0),
184
'inventory': ('.iix', 1),
186
'signature': ('.six', 3),
189
def __init__(self, upload_transport, index_transport, pack_transport,
191
"""Create a NewPack instance.
193
:param upload_transport: A writable transport for the pack to be
194
incrementally uploaded to.
195
:param index_transport: A writable transport for the pack's indices to
196
be written to when the pack is finished.
197
:param pack_transport: A writable transport for the pack to be renamed
198
to when the upload is complete. This *must* be the same as
199
upload_transport.clone('../packs').
200
:param upload_suffix: An optional suffix to be given to any temporary
201
files created during the pack creation. e.g '.autopack'
203
# The relative locations of the packs are constrained, but all are
204
# passed in because the caller has them, so as to avoid object churn.
206
# Revisions: parents list, no text compression.
207
InMemoryGraphIndex(reference_lists=1),
208
# Inventory: We want to map compression only, but currently the
209
# knit code hasn't been updated enough to understand that, so we
210
# have a regular 2-list index giving parents and compression
212
InMemoryGraphIndex(reference_lists=2),
213
# Texts: compression and per file graph, for all fileids - so two
214
# reference lists and two elements in the key tuple.
215
InMemoryGraphIndex(reference_lists=2, key_elements=2),
216
# Signatures: Just blobs to store, no compression, no parents
218
InMemoryGraphIndex(reference_lists=0),
220
# where should the new pack be opened
221
self.upload_transport = upload_transport
222
# where are indices written out to
223
self.index_transport = index_transport
224
# where is the pack renamed to when it is finished?
225
self.pack_transport = pack_transport
226
# tracks the content written to the .pack file.
227
self._hash = md5.new()
228
# a four-tuple with the length in bytes of the indices, once the pack
229
# is finalised. (rev, inv, text, sigs)
230
self.index_sizes = None
231
# How much data to cache when writing packs. Note that this is not
232
# synchronised with reads, because it's not in the transport layer, so
233
# is not safe unless the client knows it won't be reading from the pack
235
self._cache_limit = 0
236
# the temporary pack file name.
237
self.random_name = rand_chars(20) + upload_suffix
238
# when was this pack started ?
239
self.start_time = time.time()
240
# open an output stream for the data added to the pack.
241
self.write_stream = self.upload_transport.open_write_stream(
243
if 'pack' in debug.debug_flags:
244
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
245
time.ctime(), self.upload_transport.base, self.random_name,
246
time.time() - self.start_time)
247
# A list of byte sequences to be written to the new pack, and the
248
# aggregate size of them. Stored as a list rather than separate
249
# variables so that the _write_data closure below can update them.
250
self._buffer = [[], 0]
251
# create a callable for adding data
253
# robertc says- this is a closure rather than a method on the object
254
# so that the variables are locals, and faster than accessing object
256
def _write_data(bytes, flush=False, _buffer=self._buffer,
257
_write=self.write_stream.write, _update=self._hash.update):
258
_buffer[0].append(bytes)
259
_buffer[1] += len(bytes)
261
if _buffer[1] > self._cache_limit or flush:
262
bytes = ''.join(_buffer[0])
266
# expose this on self, for the occasion when clients want to add data.
267
self._write_data = _write_data
268
# a pack writer object to serialise pack records.
269
self._writer = pack.ContainerWriter(self._write_data)
271
# what state is the pack in? (open, finished, aborted)
275
"""Cancel creating this pack."""
276
self._state = 'aborted'
277
# Remove the temporary pack file.
278
self.upload_transport.delete(self.random_name)
279
# The indices have no state on disk.
281
def access_tuple(self):
282
"""Return a tuple (transport, name) for the pack content."""
283
assert self._state in ('open', 'finished')
284
if self._state == 'finished':
285
return Pack.access_tuple(self)
287
return self.upload_transport, self.random_name
289
def data_inserted(self):
290
"""True if data has been added to this pack."""
291
return bool(self.get_revision_count() or
292
self.inventory_index.key_count() or
293
self.text_index.key_count() or
294
self.signature_index.key_count())
297
"""Finish the new pack.
300
- finalises the content
301
- assigns a name (the md5 of the content, currently)
302
- writes out the associated indices
303
- renames the pack into place.
304
- stores the index size tuple for the pack in the index_sizes
309
self._write_data('', flush=True)
310
self.name = self._hash.hexdigest()
312
# XXX: It'd be better to write them all to temporary names, then
313
# rename them all into place, so that the window when only some are
314
# visible is smaller. On the other hand none will be seen until
315
# they're in the names list.
316
self.index_sizes = [None, None, None, None]
317
self._write_index('revision', self.revision_index, 'revision')
318
self._write_index('inventory', self.inventory_index, 'inventory')
319
self._write_index('text', self.text_index, 'file texts')
320
self._write_index('signature', self.signature_index,
321
'revision signatures')
322
self.write_stream.close()
323
# Note that this will clobber an existing pack with the same name,
324
# without checking for hash collisions. While this is undesirable this
325
# is something that can be rectified in a subsequent release. One way
326
# to rectify it may be to leave the pack at the original name, writing
327
# its pack-names entry as something like 'HASH: index-sizes
328
# temporary-name'. Allocate that and check for collisions, if it is
329
# collision free then rename it into place. If clients know this scheme
330
# they can handle missing-file errors by:
331
# - try for HASH.pack
332
# - try for temporary-name
333
# - refresh the pack-list to see if the pack is now absent
334
self.upload_transport.rename(self.random_name,
335
'../packs/' + self.name + '.pack')
336
self._state = 'finished'
337
if 'pack' in debug.debug_flags:
338
# XXX: size might be interesting?
339
mutter('%s: create_pack: pack renamed into place: %s%s->%s%s t+%6.3fs',
340
time.ctime(), self.upload_transport.base, self.random_name,
341
self.pack_transport, self.name,
342
time.time() - self.start_time)
344
def index_name(self, index_type, name):
345
"""Get the disk name of an index type for pack name 'name'."""
346
return name + NewPack.index_definitions[index_type][0]
348
def index_offset(self, index_type):
349
"""Get the position in a index_size array for a given index type."""
350
return NewPack.index_definitions[index_type][1]
352
def _replace_index_with_readonly(self, index_type):
353
setattr(self, index_type + '_index',
354
GraphIndex(self.index_transport,
355
self.index_name(index_type, self.name),
356
self.index_sizes[self.index_offset(index_type)]))
358
def set_write_cache_size(self, size):
359
self._cache_limit = size
361
def _write_index(self, index_type, index, label):
362
"""Write out an index.
364
:param index_type: The type of index to write - e.g. 'revision'.
365
:param index: The index object to serialise.
366
:param label: What label to give the index e.g. 'revision'.
368
index_name = self.index_name(index_type, self.name)
369
self.index_sizes[self.index_offset(index_type)] = \
370
self.index_transport.put_file(index_name, index.finish())
371
if 'pack' in debug.debug_flags:
372
# XXX: size might be interesting?
373
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
374
time.ctime(), label, self.upload_transport.base,
375
self.random_name, time.time() - self.start_time)
376
# Replace the writable index on this object with a readonly,
377
# presently unloaded index. We should alter
378
# the index layer to make its finish() error if add_node is
379
# subsequently used. RBC
380
self._replace_index_with_readonly(index_type)
383
class AggregateIndex(object):
384
"""An aggregated index for the RepositoryPackCollection.
386
AggregateIndex is reponsible for managing the PackAccess object,
387
Index-To-Pack mapping, and all indices list for a specific type of index
388
such as 'revision index'.
390
A CombinedIndex provides an index on a single key space built up
391
from several on-disk indices. The AggregateIndex builds on this
392
to provide a knit access layer, and allows having up to one writable
393
index within the collection.
395
# XXX: Probably 'can be written to' could/should be separated from 'acts
396
# like a knit index' -- mbp 20071024
399
"""Create an AggregateIndex."""
400
self.index_to_pack = {}
401
self.combined_index = CombinedGraphIndex([])
402
self.knit_access = _PackAccess(self.index_to_pack)
404
def replace_indices(self, index_to_pack, indices):
405
"""Replace the current mappings with fresh ones.
407
This should probably not be used eventually, rather incremental add and
408
removal of indices. It has been added during refactoring of existing
411
:param index_to_pack: A mapping from index objects to
412
(transport, name) tuples for the pack file data.
413
:param indices: A list of indices.
415
# refresh the revision pack map dict without replacing the instance.
416
self.index_to_pack.clear()
417
self.index_to_pack.update(index_to_pack)
418
# XXX: API break - clearly a 'replace' method would be good?
419
self.combined_index._indices[:] = indices
420
# the current add nodes callback for the current writable index if
422
self.add_callback = None
424
def add_index(self, index, pack):
425
"""Add index to the aggregate, which is an index for Pack pack.
427
Future searches on the aggregate index will seach this new index
428
before all previously inserted indices.
430
:param index: An Index for the pack.
431
:param pack: A Pack instance.
433
# expose it to the index map
434
self.index_to_pack[index] = pack.access_tuple()
435
# put it at the front of the linear index list
436
self.combined_index.insert_index(0, index)
438
def add_writable_index(self, index, pack):
439
"""Add an index which is able to have data added to it.
441
There can be at most one writable index at any time. Any
442
modifications made to the knit are put into this index.
444
:param index: An index from the pack parameter.
445
:param pack: A Pack instance.
447
assert self.add_callback is None, \
448
"%s already has a writable index through %s" % \
449
(self, self.add_callback)
450
# allow writing: queue writes to a new index
451
self.add_index(index, pack)
452
# Updates the index to packs mapping as a side effect,
453
self.knit_access.set_writer(pack._writer, index, pack.access_tuple())
454
self.add_callback = index.add_nodes
457
"""Reset all the aggregate data to nothing."""
458
self.knit_access.set_writer(None, None, (None, None))
459
self.index_to_pack.clear()
460
del self.combined_index._indices[:]
461
self.add_callback = None
463
def remove_index(self, index, pack):
464
"""Remove index from the indices used to answer queries.
466
:param index: An index from the pack parameter.
467
:param pack: A Pack instance.
469
del self.index_to_pack[index]
470
self.combined_index._indices.remove(index)
471
if (self.add_callback is not None and
472
getattr(index, 'add_nodes', None) == self.add_callback):
473
self.add_callback = None
474
self.knit_access.set_writer(None, None, (None, None))
477
class RepositoryPackCollection(object):
478
"""Management of packs within a repository."""
480
def __init__(self, repo, transport, index_transport, upload_transport,
482
"""Create a new RepositoryPackCollection.
484
:param transport: Addresses the repository base directory
485
(typically .bzr/repository/).
486
:param index_transport: Addresses the directory containing indices.
487
:param upload_transport: Addresses the directory into which packs are written
488
while they're being created.
489
:param pack_transport: Addresses the directory of existing complete packs.
492
self.transport = transport
493
self._index_transport = index_transport
494
self._upload_transport = upload_transport
495
self._pack_transport = pack_transport
496
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
499
self._packs_by_name = {}
500
# the previous pack-names content
501
self._packs_at_load = None
502
# when a pack is being created by this object, the state of that pack.
503
self._new_pack = None
504
# aggregated revision index data
505
self.revision_index = AggregateIndex()
506
self.inventory_index = AggregateIndex()
507
self.text_index = AggregateIndex()
508
self.signature_index = AggregateIndex()
510
def add_pack_to_memory(self, pack):
511
"""Make a Pack object available to the repository to satisfy queries.
513
:param pack: A Pack object.
515
assert pack.name not in self._packs_by_name
516
self.packs.append(pack)
517
self._packs_by_name[pack.name] = pack
518
self.revision_index.add_index(pack.revision_index, pack)
519
self.inventory_index.add_index(pack.inventory_index, pack)
520
self.text_index.add_index(pack.text_index, pack)
521
self.signature_index.add_index(pack.signature_index, pack)
523
def _add_text_to_weave(self, file_id, revision_id, new_lines, parents,
524
nostore_sha, random_revid):
525
file_id_index = GraphIndexPrefixAdapter(
526
self.text_index.combined_index,
528
add_nodes_callback=self.text_index.add_callback)
529
self.repo._text_knit._index._graph_index = file_id_index
530
self.repo._text_knit._index._add_callback = file_id_index.add_nodes
531
return self.repo._text_knit.add_lines_with_ghosts(
532
revision_id, parents, new_lines, nostore_sha=nostore_sha,
533
random_id=random_revid, check_content=False)[0:2]
536
"""Return a list of all the Pack objects this repository has.
538
Note that an in-progress pack being created is not returned.
540
:return: A list of Pack objects for all the packs in the repository.
543
for name in self.names():
544
result.append(self.get_pack_by_name(name))
548
"""Pack the pack collection incrementally.
550
This will not attempt global reorganisation or recompression,
551
rather it will just ensure that the total number of packs does
552
not grow without bound. It uses the _max_pack_count method to
553
determine if autopacking is needed, and the pack_distribution
554
method to determine the number of revisions in each pack.
556
If autopacking takes place then the packs name collection will have
557
been flushed to disk - packing requires updating the name collection
558
in synchronisation with certain steps. Otherwise the names collection
561
:return: True if packing took place.
563
# XXX: Should not be needed when the management of indices is sane.
564
total_revisions = self.revision_index.combined_index.key_count()
565
total_packs = len(self._names)
566
if self._max_pack_count(total_revisions) >= total_packs:
568
# XXX: the following may want to be a class, to pack with a given
570
mutter('Auto-packing repository %s, which has %d pack files, '
571
'containing %d revisions into %d packs.', self, total_packs,
572
total_revisions, self._max_pack_count(total_revisions))
573
# determine which packs need changing
574
pack_distribution = self.pack_distribution(total_revisions)
576
for pack in self.all_packs():
577
revision_count = pack.get_revision_count()
578
if revision_count == 0:
579
# revision less packs are not generated by normal operation,
580
# only by operations like sign-my-commits, and thus will not
581
# tend to grow rapdily or without bound like commit containing
582
# packs do - leave them alone as packing them really should
583
# group their data with the relevant commit, and that may
584
# involve rewriting ancient history - which autopack tries to
585
# avoid. Alternatively we could not group the data but treat
586
# each of these as having a single revision, and thus add
587
# one revision for each to the total revision count, to get
588
# a matching distribution.
590
existing_packs.append((revision_count, pack))
591
pack_operations = self.plan_autopack_combinations(
592
existing_packs, pack_distribution)
593
self._execute_pack_operations(pack_operations)
596
def create_pack_from_packs(self, packs, suffix, revision_ids=None):
597
"""Create a new pack by reading data from other packs.
599
This does little more than a bulk copy of data. One key difference
600
is that data with the same item key across multiple packs is elided
601
from the output. The new pack is written into the current pack store
602
along with its indices, and the name added to the pack names. The
603
source packs are not altered and are not required to be in the current
606
:param packs: An iterable of Packs to combine.
607
:param revision_ids: Either None, to copy all data, or a list
608
of revision_ids to limit the copied data to the data they
610
:return: A Pack object, or None if nothing was copied.
612
# open a pack - using the same name as the last temporary file
613
# - which has already been flushed, so its safe.
614
# XXX: - duplicate code warning with start_write_group; fix before
615
# considering 'done'.
616
if self._new_pack is not None:
617
raise errors.BzrError('call to create_pack_from_packs while '
618
'another pack is being written.')
619
if revision_ids is not None and len(revision_ids) == 0:
620
# silly fetch request.
622
new_pack = NewPack(self._upload_transport, self._index_transport,
623
self._pack_transport, upload_suffix=suffix)
624
# buffer data - we won't be reading-back during the pack creation and
625
# this makes a significant difference on sftp pushes.
626
new_pack.set_write_cache_size(1024*1024)
627
if 'pack' in debug.debug_flags:
628
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
630
if revision_ids is not None:
631
rev_count = len(revision_ids)
634
mutter('%s: create_pack: creating pack from source packs: '
635
'%s%s %s revisions wanted %s t=0',
636
time.ctime(), self._upload_transport.base, new_pack.random_name,
637
plain_pack_list, rev_count)
640
revision_keys = [(revision_id,) for revision_id in revision_ids]
644
# select revision keys
645
revision_index_map = self._packs_list_to_pack_map_and_index_list(
646
packs, 'revision_index')[0]
647
revision_nodes = self._index_contents(revision_index_map, revision_keys)
648
# copy revision keys and adjust values
649
list(self._copy_nodes_graph(revision_nodes, revision_index_map,
650
new_pack._writer, new_pack.revision_index))
651
if 'pack' in debug.debug_flags:
652
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
653
time.ctime(), self._upload_transport.base, new_pack.random_name,
654
new_pack.revision_index.key_count(),
655
time.time() - new_pack.start_time)
656
# select inventory keys
657
inv_keys = revision_keys # currently the same keyspace, and note that
658
# querying for keys here could introduce a bug where an inventory item
659
# is missed, so do not change it to query separately without cross
660
# checking like the text key check below.
661
inventory_index_map = self._packs_list_to_pack_map_and_index_list(
662
packs, 'inventory_index')[0]
663
inv_nodes = self._index_contents(inventory_index_map, inv_keys)
664
# copy inventory keys and adjust values
665
# XXX: Should be a helper function to allow different inv representation
667
inv_lines = self._copy_nodes_graph(inv_nodes, inventory_index_map,
668
new_pack._writer, new_pack.inventory_index, output_lines=True)
670
fileid_revisions = self.repo._find_file_ids_from_xml_inventory_lines(
671
inv_lines, revision_ids)
673
for fileid, file_revids in fileid_revisions.iteritems():
675
[(fileid, file_revid) for file_revid in file_revids])
677
# eat the iterator to cause it to execute.
680
if 'pack' in debug.debug_flags:
681
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
682
time.ctime(), self._upload_transport.base, new_pack.random_name,
683
new_pack.inventory_index.key_count(),
684
time.time() - new_pack.start_time)
686
text_index_map = self._packs_list_to_pack_map_and_index_list(
687
packs, 'text_index')[0]
688
text_nodes = self._index_contents(text_index_map, text_filter)
689
if text_filter is not None:
690
# We could return the keys copied as part of the return value from
691
# _copy_nodes_graph but this doesn't work all that well with the
692
# need to get line output too, so we check separately, and as we're
693
# going to buffer everything anyway, we check beforehand, which
694
# saves reading knit data over the wire when we know there are
696
text_nodes = set(text_nodes)
697
present_text_keys = set(_node[1] for _node in text_nodes)
698
missing_text_keys = set(text_filter) - present_text_keys
699
if missing_text_keys:
700
# TODO: raise a specific error that can handle many missing
702
a_missing_key = missing_text_keys.pop()
703
raise errors.RevisionNotPresent(a_missing_key[1],
705
# copy text keys and adjust values
706
list(self._copy_nodes_graph(text_nodes, text_index_map,
707
new_pack._writer, new_pack.text_index))
708
if 'pack' in debug.debug_flags:
709
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
710
time.ctime(), self._upload_transport.base, new_pack.random_name,
711
new_pack.text_index.key_count(),
712
time.time() - new_pack.start_time)
713
# select signature keys
714
signature_filter = revision_keys # same keyspace
715
signature_index_map = self._packs_list_to_pack_map_and_index_list(
716
packs, 'signature_index')[0]
717
signature_nodes = self._index_contents(signature_index_map,
719
# copy signature keys and adjust values
720
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
721
new_pack.signature_index)
722
if 'pack' in debug.debug_flags:
723
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
724
time.ctime(), self._upload_transport.base, new_pack.random_name,
725
new_pack.signature_index.key_count(),
726
time.time() - new_pack.start_time)
727
if not new_pack.data_inserted():
731
self.allocate(new_pack)
734
def _execute_pack_operations(self, pack_operations):
735
"""Execute a series of pack operations.
737
:param pack_operations: A list of [revision_count, packs_to_combine].
740
for revision_count, packs in pack_operations:
741
# we may have no-ops from the setup logic
744
# have a progress bar?
745
self.create_pack_from_packs(packs, '.autopack')
747
self._remove_pack_from_memory(pack)
748
# record the newly available packs and stop advertising the old
750
self._save_pack_names()
751
# Move the old packs out of the way now they are no longer referenced.
752
for revision_count, packs in pack_operations:
753
self._obsolete_packs(packs)
755
def lock_names(self):
756
"""Acquire the mutex around the pack-names index.
758
This cannot be used in the middle of a read-only transaction on the
761
self.repo.control_files.lock_write()
764
"""Pack the pack collection totally."""
766
total_packs = len(self._names)
769
total_revisions = self.revision_index.combined_index.key_count()
770
# XXX: the following may want to be a class, to pack with a given
772
mutter('Packing repository %s, which has %d pack files, '
773
'containing %d revisions into 1 packs.', self, total_packs,
775
# determine which packs need changing
776
pack_distribution = [1]
777
pack_operations = [[0, []]]
778
for pack in self.all_packs():
779
revision_count = pack.get_revision_count()
780
pack_operations[-1][0] += revision_count
781
pack_operations[-1][1].append(pack)
782
self._execute_pack_operations(pack_operations)
784
def plan_autopack_combinations(self, existing_packs, pack_distribution):
785
"""Plan a pack operation.
787
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
789
:param pack_distribution: A list with the number of revisions desired
792
if len(existing_packs) <= len(pack_distribution):
794
existing_packs.sort(reverse=True)
795
pack_operations = [[0, []]]
796
# plan out what packs to keep, and what to reorganise
797
while len(existing_packs):
798
# take the largest pack, and if its less than the head of the
799
# distribution chart we will include its contents in the new pack for
800
# that position. If its larger, we remove its size from the
802
next_pack_rev_count, next_pack = existing_packs.pop(0)
803
if next_pack_rev_count >= pack_distribution[0]:
804
# this is already packed 'better' than this, so we can
805
# not waste time packing it.
806
while next_pack_rev_count > 0:
807
next_pack_rev_count -= pack_distribution[0]
808
if next_pack_rev_count >= 0:
810
del pack_distribution[0]
812
# didn't use that entire bucket up
813
pack_distribution[0] = -next_pack_rev_count
815
# add the revisions we're going to add to the next output pack
816
pack_operations[-1][0] += next_pack_rev_count
817
# allocate this pack to the next pack sub operation
818
pack_operations[-1][1].append(next_pack)
819
if pack_operations[-1][0] >= pack_distribution[0]:
820
# this pack is used up, shift left.
821
del pack_distribution[0]
822
pack_operations.append([0, []])
824
return pack_operations
826
def _copy_nodes(self, nodes, index_map, writer, write_index):
827
# plan a readv on each source pack:
829
nodes = sorted(nodes)
830
# how to map this into knit.py - or knit.py into this?
831
# we don't want the typical knit logic, we want grouping by pack
832
# at this point - perhaps a helper library for the following code
833
# duplication points?
835
for index, key, value in nodes:
836
if index not in request_groups:
837
request_groups[index] = []
838
request_groups[index].append((key, value))
839
for index, items in request_groups.iteritems():
840
pack_readv_requests = []
841
for key, value in items:
842
# ---- KnitGraphIndex.get_position
843
bits = value[1:].split(' ')
844
offset, length = int(bits[0]), int(bits[1])
845
pack_readv_requests.append((offset, length, (key, value[0])))
846
# linear scan up the pack
847
pack_readv_requests.sort()
849
transport, path = index_map[index]
850
reader = pack.make_readv_reader(transport, path,
851
[offset[0:2] for offset in pack_readv_requests])
852
for (names, read_func), (_1, _2, (key, eol_flag)) in \
853
izip(reader.iter_records(), pack_readv_requests):
854
raw_data = read_func(None)
855
pos, size = writer.add_bytes_record(raw_data, names)
856
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
858
def _copy_nodes_graph(self, nodes, index_map, writer, write_index,
860
"""Copy knit nodes between packs.
862
:param output_lines: Return lines present in the copied data as
865
# for record verification
866
knit_data = _KnitData(None)
867
# for line extraction when requested (inventories only)
869
factory = knit.KnitPlainFactory()
870
# plan a readv on each source pack:
872
nodes = sorted(nodes)
873
# how to map this into knit.py - or knit.py into this?
874
# we don't want the typical knit logic, we want grouping by pack
875
# at this point - perhaps a helper library for the following code
876
# duplication points?
878
for index, key, value, references in nodes:
879
if index not in request_groups:
880
request_groups[index] = []
881
request_groups[index].append((key, value, references))
882
for index, items in request_groups.iteritems():
883
pack_readv_requests = []
884
for key, value, references in items:
885
# ---- KnitGraphIndex.get_position
886
bits = value[1:].split(' ')
887
offset, length = int(bits[0]), int(bits[1])
888
pack_readv_requests.append((offset, length, (key, value[0], references)))
889
# linear scan up the pack
890
pack_readv_requests.sort()
892
transport, path = index_map[index]
893
reader = pack.make_readv_reader(transport, path,
894
[offset[0:2] for offset in pack_readv_requests])
895
for (names, read_func), (_1, _2, (key, eol_flag, references)) in \
896
izip(reader.iter_records(), pack_readv_requests):
897
raw_data = read_func(None)
899
# read the entire thing
900
content, _ = knit_data._parse_record(key[-1], raw_data)
901
if len(references[-1]) == 0:
902
line_iterator = factory.get_fulltext_content(content)
904
line_iterator = factory.get_linedelta_content(content)
905
for line in line_iterator:
908
# check the header only
909
df, _ = knit_data._parse_record_header(key[-1], raw_data)
911
pos, size = writer.add_bytes_record(raw_data, names)
912
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
914
def ensure_loaded(self):
915
# NB: if you see an assertion error here, its probably access against
916
# an unlocked repo. Naughty.
917
assert self.repo.is_locked()
918
if self._names is None:
920
self._packs_at_load = set()
921
for index, key, value in self._iter_disk_pack_index():
923
self._names[name] = self._parse_index_sizes(value)
924
self._packs_at_load.add((key, value))
925
# populate all the metadata.
928
def _parse_index_sizes(self, value):
929
"""Parse a string of index sizes."""
930
return tuple([int(digits) for digits in value.split(' ')])
932
def get_pack_by_name(self, name):
933
"""Get a Pack object by name.
935
:param name: The name of the pack - e.g. '123456'
936
:return: A Pack object.
939
return self._packs_by_name[name]
941
rev_index = self._make_index(name, '.rix')
942
inv_index = self._make_index(name, '.iix')
943
txt_index = self._make_index(name, '.tix')
944
sig_index = self._make_index(name, '.six')
945
result = ExistingPack(self._pack_transport, name, rev_index,
946
inv_index, txt_index, sig_index)
947
self.add_pack_to_memory(result)
950
def allocate(self, a_new_pack):
951
"""Allocate name in the list of packs.
953
:param a_new_pack: A NewPack instance to be added to the collection of
954
packs for this repository.
957
if a_new_pack.name in self._names:
958
# a collision with the packs we know about (not the only possible
959
# collision, see NewPack.finish() for some discussion). Remove our
960
# prior reference to it.
961
self._remove_pack_from_memory(a_new_pack)
962
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
963
self.add_pack_to_memory(a_new_pack)
965
def _iter_disk_pack_index(self):
966
"""Iterate over the contents of the pack-names index.
968
This is used when loading the list from disk, and before writing to
969
detect updates from others during our write operation.
970
:return: An iterator of the index contents.
972
return GraphIndex(self.transport, 'pack-names', None
975
def _make_index(self, name, suffix):
976
size_offset = self._suffix_offsets[suffix]
977
index_name = name + suffix
978
index_size = self._names[name][size_offset]
980
self._index_transport, index_name, index_size)
982
def _max_pack_count(self, total_revisions):
983
"""Return the maximum number of packs to use for total revisions.
985
:param total_revisions: The total number of revisions in the
988
if not total_revisions:
990
digits = str(total_revisions)
997
"""Provide an order to the underlying names."""
998
return sorted(self._names.keys())
1000
def _obsolete_packs(self, packs):
1001
"""Move a number of packs which have been obsoleted out of the way.
1003
Each pack and its associated indices are moved out of the way.
1005
Note: for correctness this function should only be called after a new
1006
pack names index has been written without these pack names, and with
1007
the names of packs that contain the data previously available via these
1010
:param packs: The packs to obsolete.
1011
:param return: None.
1014
pack.pack_transport.rename(pack.file_name(),
1015
'../obsolete_packs/' + pack.file_name())
1016
# TODO: Probably needs to know all possible indices for this pack
1017
# - or maybe list the directory and move all indices matching this
1018
# name whether we recognize it or not?
1019
for suffix in ('.iix', '.six', '.tix', '.rix'):
1020
self._index_transport.rename(pack.name + suffix,
1021
'../obsolete_packs/' + pack.name + suffix)
1023
def pack_distribution(self, total_revisions):
1024
"""Generate a list of the number of revisions to put in each pack.
1026
:param total_revisions: The total number of revisions in the
1029
if total_revisions == 0:
1031
digits = reversed(str(total_revisions))
1033
for exponent, count in enumerate(digits):
1034
size = 10 ** exponent
1035
for pos in range(int(count)):
1037
return list(reversed(result))
1039
def _pack_tuple(self, name):
1040
"""Return a tuple with the transport and file name for a pack name."""
1041
return self._pack_transport, name + '.pack'
1043
def _remove_pack_from_memory(self, pack):
1044
"""Remove pack from the packs accessed by this repository.
1046
Only affects memory state, until self._save_pack_names() is invoked.
1048
self._names.pop(pack.name)
1049
self._packs_by_name.pop(pack.name)
1050
self._remove_pack_indices(pack)
1052
def _remove_pack_indices(self, pack):
1053
"""Remove the indices for pack from the aggregated indices."""
1054
self.revision_index.remove_index(pack.revision_index, pack)
1055
self.inventory_index.remove_index(pack.inventory_index, pack)
1056
self.text_index.remove_index(pack.text_index, pack)
1057
self.signature_index.remove_index(pack.signature_index, pack)
1060
"""Clear all cached data."""
1061
# cached revision data
1062
self.repo._revision_knit = None
1063
self.revision_index.clear()
1064
# cached signature data
1065
self.repo._signature_knit = None
1066
self.signature_index.clear()
1067
# cached file text data
1068
self.text_index.clear()
1069
self.repo._text_knit = None
1070
# cached inventory data
1071
self.inventory_index.clear()
1072
# remove the open pack
1073
self._new_pack = None
1074
# information about packs.
1077
self._packs_by_name = {}
1078
self._packs_at_load = None
1080
def _make_index_map(self, index_suffix):
1081
"""Return information on existing indices.
1083
:param suffix: Index suffix added to pack name.
1085
:returns: (pack_map, indices) where indices is a list of GraphIndex
1086
objects, and pack_map is a mapping from those objects to the
1087
pack tuple they describe.
1089
# TODO: stop using this; it creates new indices unnecessarily.
1090
self.ensure_loaded()
1091
suffix_map = {'.rix': 'revision_index',
1092
'.six': 'signature_index',
1093
'.iix': 'inventory_index',
1094
'.tix': 'text_index',
1096
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1097
suffix_map[index_suffix])
1099
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1100
"""Convert a list of packs to an index pack map and index list.
1102
:param packs: The packs list to process.
1103
:param index_attribute: The attribute that the desired index is found
1105
:return: A tuple (map, list) where map contains the dict from
1106
index:pack_tuple, and lsit contains the indices in the same order
1112
index = getattr(pack, index_attribute)
1113
indices.append(index)
1114
pack_map[index] = (pack.pack_transport, pack.file_name())
1115
return pack_map, indices
1117
def _index_contents(self, pack_map, key_filter=None):
1118
"""Get an iterable of the index contents from a pack_map.
1120
:param pack_map: A map from indices to pack details.
1121
:param key_filter: An optional filter to limit the
1124
indices = [index for index in pack_map.iterkeys()]
1125
all_index = CombinedGraphIndex(indices)
1126
if key_filter is None:
1127
return all_index.iter_all_entries()
1129
return all_index.iter_entries(key_filter)
1131
def _unlock_names(self):
1132
"""Release the mutex around the pack-names index."""
1133
self.repo.control_files.unlock()
1135
def _save_pack_names(self):
1136
"""Save the list of packs.
1138
This will take out the mutex around the pack names list for the
1139
duration of the method call. If concurrent updates have been made, a
1140
three-way merge between the current list and the current in memory list
1145
builder = GraphIndexBuilder()
1146
# load the disk nodes across
1148
for index, key, value in self._iter_disk_pack_index():
1149
disk_nodes.add((key, value))
1150
# do a two-way diff against our original content
1151
current_nodes = set()
1152
for name, sizes in self._names.iteritems():
1154
((name, ), ' '.join(str(size) for size in sizes)))
1155
deleted_nodes = self._packs_at_load - current_nodes
1156
new_nodes = current_nodes - self._packs_at_load
1157
disk_nodes.difference_update(deleted_nodes)
1158
disk_nodes.update(new_nodes)
1159
# TODO: handle same-name, index-size-changes here -
1160
# e.g. use the value from disk, not ours, *unless* we're the one
1162
for key, value in disk_nodes:
1163
builder.add_node(key, value)
1164
self.transport.put_file('pack-names', builder.finish())
1165
# move the baseline forward
1166
self._packs_at_load = disk_nodes
1168
self._unlock_names()
1169
# synchronise the memory packs list with what we just wrote:
1170
new_names = dict(disk_nodes)
1171
# drop no longer present nodes
1172
for pack in self.all_packs():
1173
if (pack.name,) not in new_names:
1174
self._remove_pack_from_memory(pack)
1175
# add new nodes/refresh existing ones
1176
for key, value in disk_nodes:
1178
sizes = self._parse_index_sizes(value)
1179
if name in self._names:
1181
if sizes != self._names[name]:
1182
# the pack for name has had its indices replaced - rare but
1183
# important to handle. XXX: probably can never happen today
1184
# because the three-way merge code above does not handle it
1185
# - you may end up adding the same key twice to the new
1186
# disk index because the set values are the same, unless
1187
# the only index shows up as deleted by the set difference
1188
# - which it may. Until there is a specific test for this,
1189
# assume its broken. RBC 20071017.
1190
self._remove_pack_from_memory(self.get_pack_by_name(name))
1191
self._names[name] = sizes
1192
self.get_pack_by_name(name)
1195
self._names[name] = sizes
1196
self.get_pack_by_name(name)
1198
def _start_write_group(self):
1199
# Do not permit preparation for writing if we're not in a 'write lock'.
1200
if not self.repo.is_write_locked():
1201
raise errors.NotWriteLocked(self)
1202
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1203
self._pack_transport, upload_suffix='.pack')
1204
# allow writing: queue writes to a new index
1205
self.revision_index.add_writable_index(self._new_pack.revision_index,
1207
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
1209
self.text_index.add_writable_index(self._new_pack.text_index,
1211
self.signature_index.add_writable_index(self._new_pack.signature_index,
1214
# reused revision and signature knits may need updating
1216
# "Hysterical raisins. client code in bzrlib grabs those knits outside
1217
# of write groups and then mutates it inside the write group."
1218
if self.repo._revision_knit is not None:
1219
self.repo._revision_knit._index._add_callback = \
1220
self.revision_index.add_callback
1221
if self.repo._signature_knit is not None:
1222
self.repo._signature_knit._index._add_callback = \
1223
self.signature_index.add_callback
1224
# create a reused knit object for text addition in commit.
1225
self.repo._text_knit = self.repo.weave_store.get_weave_or_empty(
1228
def _abort_write_group(self):
1229
# FIXME: just drop the transient index.
1230
# forget what names there are
1231
self._new_pack.abort()
1232
self._remove_pack_indices(self._new_pack)
1233
self._new_pack = None
1234
self.repo._text_knit = None
1236
def _commit_write_group(self):
1237
self._remove_pack_indices(self._new_pack)
1238
if self._new_pack.data_inserted():
1239
# get all the data to disk and read to use
1240
self._new_pack.finish()
1241
self.allocate(self._new_pack)
1242
self._new_pack = None
1243
if not self.autopack():
1244
# when autopack takes no steps, the names list is still
1246
self._save_pack_names()
1248
self._new_pack.abort()
1249
self.repo._text_knit = None
1252
class KnitPackRevisionStore(KnitRevisionStore):
1253
"""An object to adapt access from RevisionStore's to use KnitPacks.
1255
This class works by replacing the original RevisionStore.
1256
We need to do this because the KnitPackRevisionStore is less
1257
isolated in its layering - it uses services from the repo.
1260
def __init__(self, repo, transport, revisionstore):
1261
"""Create a KnitPackRevisionStore on repo with revisionstore.
1263
This will store its state in the Repository, use the
1264
indices to provide a KnitGraphIndex,
1265
and at the end of transactions write new indices.
1267
KnitRevisionStore.__init__(self, revisionstore.versioned_file_store)
1269
self._serializer = revisionstore._serializer
1270
self.transport = transport
1272
def get_revision_file(self, transaction):
1273
"""Get the revision versioned file object."""
1274
if getattr(self.repo, '_revision_knit', None) is not None:
1275
return self.repo._revision_knit
1276
self.repo._pack_collection.ensure_loaded()
1277
add_callback = self.repo._pack_collection.revision_index.add_callback
1278
# setup knit specific objects
1279
knit_index = KnitGraphIndex(
1280
self.repo._pack_collection.revision_index.combined_index,
1281
add_callback=add_callback)
1282
self.repo._revision_knit = knit.KnitVersionedFile(
1283
'revisions', self.transport.clone('..'),
1284
self.repo.control_files._file_mode,
1285
create=False, access_mode=self.repo._access_mode(),
1286
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1287
access_method=self.repo._pack_collection.revision_index.knit_access)
1288
return self.repo._revision_knit
1290
def get_signature_file(self, transaction):
1291
"""Get the signature versioned file object."""
1292
if getattr(self.repo, '_signature_knit', None) is not None:
1293
return self.repo._signature_knit
1294
self.repo._pack_collection.ensure_loaded()
1295
add_callback = self.repo._pack_collection.signature_index.add_callback
1296
# setup knit specific objects
1297
knit_index = KnitGraphIndex(
1298
self.repo._pack_collection.signature_index.combined_index,
1299
add_callback=add_callback, parents=False)
1300
self.repo._signature_knit = knit.KnitVersionedFile(
1301
'signatures', self.transport.clone('..'),
1302
self.repo.control_files._file_mode,
1303
create=False, access_mode=self.repo._access_mode(),
1304
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1305
access_method=self.repo._pack_collection.signature_index.knit_access)
1306
return self.repo._signature_knit
1309
class KnitPackTextStore(VersionedFileStore):
1310
"""Presents a TextStore abstraction on top of packs.
1312
This class works by replacing the original VersionedFileStore.
1313
We need to do this because the KnitPackRevisionStore is less
1314
isolated in its layering - it uses services from the repo and shares them
1315
with all the data written in a single write group.
1318
def __init__(self, repo, transport, weavestore):
1319
"""Create a KnitPackTextStore on repo with weavestore.
1321
This will store its state in the Repository, use the
1322
indices FileNames to provide a KnitGraphIndex,
1323
and at the end of transactions write new indices.
1325
# don't call base class constructor - it's not suitable.
1326
# no transient data stored in the transaction
1328
self._precious = False
1330
self.transport = transport
1331
self.weavestore = weavestore
1332
# XXX for check() which isn't updated yet
1333
self._transport = weavestore._transport
1335
def get_weave_or_empty(self, file_id, transaction):
1336
"""Get a 'Knit' backed by the .tix indices.
1338
The transaction parameter is ignored.
1340
self.repo._pack_collection.ensure_loaded()
1341
add_callback = self.repo._pack_collection.text_index.add_callback
1342
# setup knit specific objects
1343
file_id_index = GraphIndexPrefixAdapter(
1344
self.repo._pack_collection.text_index.combined_index,
1345
(file_id, ), 1, add_nodes_callback=add_callback)
1346
knit_index = KnitGraphIndex(file_id_index,
1347
add_callback=file_id_index.add_nodes,
1348
deltas=True, parents=True)
1349
return knit.KnitVersionedFile('text:' + file_id,
1350
self.transport.clone('..'),
1353
access_method=self.repo._pack_collection.text_index.knit_access,
1354
factory=knit.KnitPlainFactory())
1356
get_weave = get_weave_or_empty
1359
"""Generate a list of the fileids inserted, for use by check."""
1360
self.repo._pack_collection.ensure_loaded()
1362
for index, key, value, refs in \
1363
self.repo._pack_collection.text_index.combined_index.iter_all_entries():
1368
class InventoryKnitThunk(object):
1369
"""An object to manage thunking get_inventory_weave to pack based knits."""
1371
def __init__(self, repo, transport):
1372
"""Create an InventoryKnitThunk for repo at transport.
1374
This will store its state in the Repository, use the
1375
indices FileNames to provide a KnitGraphIndex,
1376
and at the end of transactions write a new index..
1379
self.transport = transport
1381
def get_weave(self):
1382
"""Get a 'Knit' that contains inventory data."""
1383
self.repo._pack_collection.ensure_loaded()
1384
add_callback = self.repo._pack_collection.inventory_index.add_callback
1385
# setup knit specific objects
1386
knit_index = KnitGraphIndex(
1387
self.repo._pack_collection.inventory_index.combined_index,
1388
add_callback=add_callback, deltas=True, parents=True)
1389
return knit.KnitVersionedFile(
1390
'inventory', self.transport.clone('..'),
1391
self.repo.control_files._file_mode,
1392
create=False, access_mode=self.repo._access_mode(),
1393
index=knit_index, delta=True, factory=knit.KnitPlainFactory(),
1394
access_method=self.repo._pack_collection.inventory_index.knit_access)
1397
class KnitPackRepository(KnitRepository):
1398
"""Experimental graph-knit using repository."""
1400
def __init__(self, _format, a_bzrdir, control_files, _revision_store,
1401
control_store, text_store, _commit_builder_class, _serializer):
1402
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
1403
_revision_store, control_store, text_store, _commit_builder_class,
1405
index_transport = control_files._transport.clone('indices')
1406
self._pack_collection = RepositoryPackCollection(self, control_files._transport,
1408
control_files._transport.clone('upload'),
1409
control_files._transport.clone('packs'))
1410
self._revision_store = KnitPackRevisionStore(self, index_transport, self._revision_store)
1411
self.weave_store = KnitPackTextStore(self, index_transport, self.weave_store)
1412
self._inv_thunk = InventoryKnitThunk(self, index_transport)
1413
# True when the repository object is 'write locked' (as opposed to the
1414
# physical lock only taken out around changes to the pack-names list.)
1415
# Another way to represent this would be a decorator around the control
1416
# files object that presents logical locks as physical ones - if this
1417
# gets ugly consider that alternative design. RBC 20071011
1418
self._write_lock_count = 0
1419
self._transaction = None
1421
self._reconcile_does_inventory_gc = False
1422
self._reconcile_fixes_text_parents = False
1424
def _abort_write_group(self):
1425
self._pack_collection._abort_write_group()
1427
def _access_mode(self):
1428
"""Return 'w' or 'r' for depending on whether a write lock is active.
1430
This method is a helper for the Knit-thunking support objects.
1432
if self.is_write_locked():
1436
def get_parents(self, revision_ids):
1437
"""See StackedParentsProvider.get_parents.
1439
This implementation accesses the combined revision index to provide
1442
index = self._pack_collection.revision_index.combined_index
1444
for revision_id in revision_ids:
1445
if revision_id != _mod_revision.NULL_REVISION:
1446
search_keys.add((revision_id,))
1447
found_parents = {_mod_revision.NULL_REVISION:[]}
1448
for index, key, value, refs in index.iter_entries(search_keys):
1451
parents = (_mod_revision.NULL_REVISION,)
1453
parents = tuple(parent[0] for parent in parents)
1454
found_parents[key[0]] = parents
1456
for revision_id in revision_ids:
1458
result.append(found_parents[revision_id])
1463
def _make_parents_provider(self):
1466
def _refresh_data(self):
1467
if self._write_lock_count == 1 or self.control_files._lock_count == 1:
1468
# forget what names there are
1469
self._pack_collection.reset()
1470
# XXX: Better to do an in-memory merge when acquiring a new lock -
1471
# factor out code from _save_pack_names.
1473
def _start_write_group(self):
1474
self._pack_collection._start_write_group()
1476
def _commit_write_group(self):
1477
return self._pack_collection._commit_write_group()
1479
def get_inventory_weave(self):
1480
return self._inv_thunk.get_weave()
1482
def get_transaction(self):
1483
if self._write_lock_count:
1484
return self._transaction
1486
return self.control_files.get_transaction()
1488
def is_locked(self):
1489
return self._write_lock_count or self.control_files.is_locked()
1491
def is_write_locked(self):
1492
return self._write_lock_count
1494
def lock_write(self, token=None):
1495
if not self._write_lock_count and self.is_locked():
1496
raise errors.ReadOnlyError(self)
1497
self._write_lock_count += 1
1498
if self._write_lock_count == 1:
1499
from bzrlib import transactions
1500
self._transaction = transactions.WriteTransaction()
1501
self._refresh_data()
1503
def lock_read(self):
1504
if self._write_lock_count:
1505
self._write_lock_count += 1
1507
self.control_files.lock_read()
1508
self._refresh_data()
1510
def leave_lock_in_place(self):
1511
# not supported - raise an error
1512
raise NotImplementedError(self.leave_lock_in_place)
1514
def dont_leave_lock_in_place(self):
1515
# not supported - raise an error
1516
raise NotImplementedError(self.dont_leave_lock_in_place)
1520
"""Compress the data within the repository.
1522
This will pack all the data to a single pack. In future it may
1523
recompress deltas or do other such expensive operations.
1525
self._pack_collection.pack()
1528
def reconcile(self, other=None, thorough=False):
1529
"""Reconcile this repository."""
1530
from bzrlib.reconcile import PackReconciler
1531
reconciler = PackReconciler(self, thorough=thorough)
1532
reconciler.reconcile()
1536
if self._write_lock_count == 1 and self._write_group is not None:
1537
raise errors.BzrError(
1538
'Must end write groups before releasing write locks.')
1539
if self._write_lock_count:
1540
self._write_lock_count -= 1
1541
if not self._write_lock_count:
1542
transaction = self._transaction
1543
self._transaction = None
1544
transaction.finish()
1546
self.control_files.unlock()
1549
class RepositoryFormatPack(MetaDirRepositoryFormat):
1550
"""Format logic for pack structured repositories.
1552
This repository format has:
1553
- a list of packs in pack-names
1554
- packs in packs/NAME.pack
1555
- indices in indices/NAME.{iix,six,tix,rix}
1556
- knit deltas in the packs, knit indices mapped to the indices.
1557
- thunk objects to support the knits programming API.
1558
- a format marker of its own
1559
- an optional 'shared-storage' flag
1560
- an optional 'no-working-trees' flag
1564
# Set this attribute in derived classes to control the repository class
1565
# created by open and initialize.
1566
repository_class = None
1567
# Set this attribute in derived classes to control the
1568
# _commit_builder_class that the repository objects will have passed to
1569
# their constructor.
1570
_commit_builder_class = None
1571
# Set this attribute in derived clases to control the _serializer that the
1572
# repository objects will have passed to their constructor.
1575
def _get_control_store(self, repo_transport, control_files):
1576
"""Return the control store for this repository."""
1577
return VersionedFileStore(
1580
file_mode=control_files._file_mode,
1581
versionedfile_class=knit.KnitVersionedFile,
1582
versionedfile_kwargs={'factory': knit.KnitPlainFactory()},
1585
def _get_revision_store(self, repo_transport, control_files):
1586
"""See RepositoryFormat._get_revision_store()."""
1587
versioned_file_store = VersionedFileStore(
1589
file_mode=control_files._file_mode,
1592
versionedfile_class=knit.KnitVersionedFile,
1593
versionedfile_kwargs={'delta': False,
1594
'factory': knit.KnitPlainFactory(),
1598
return KnitRevisionStore(versioned_file_store)
1600
def _get_text_store(self, transport, control_files):
1601
"""See RepositoryFormat._get_text_store()."""
1602
return self._get_versioned_file_store('knits',
1605
versionedfile_class=knit.KnitVersionedFile,
1606
versionedfile_kwargs={
1607
'create_parent_dir': True,
1608
'delay_create': True,
1609
'dir_mode': control_files._dir_mode,
1613
def initialize(self, a_bzrdir, shared=False):
1614
"""Create a pack based repository.
1616
:param a_bzrdir: bzrdir to contain the new repository; must already
1618
:param shared: If true the repository will be initialized as a shared
1621
mutter('creating repository in %s.', a_bzrdir.transport.base)
1622
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
1623
builder = GraphIndexBuilder()
1624
files = [('pack-names', builder.finish())]
1625
utf8_files = [('format', self.get_format_string())]
1627
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
1628
return self.open(a_bzrdir=a_bzrdir, _found=True)
1630
def open(self, a_bzrdir, _found=False, _override_transport=None):
1631
"""See RepositoryFormat.open().
1633
:param _override_transport: INTERNAL USE ONLY. Allows opening the
1634
repository at a slightly different url
1635
than normal. I.e. during 'upgrade'.
1638
format = RepositoryFormat.find_format(a_bzrdir)
1639
assert format.__class__ == self.__class__
1640
if _override_transport is not None:
1641
repo_transport = _override_transport
1643
repo_transport = a_bzrdir.get_repository_transport(None)
1644
control_files = lockable_files.LockableFiles(repo_transport,
1645
'lock', lockdir.LockDir)
1646
text_store = self._get_text_store(repo_transport, control_files)
1647
control_store = self._get_control_store(repo_transport, control_files)
1648
_revision_store = self._get_revision_store(repo_transport, control_files)
1649
return self.repository_class(_format=self,
1651
control_files=control_files,
1652
_revision_store=_revision_store,
1653
control_store=control_store,
1654
text_store=text_store,
1655
_commit_builder_class=self._commit_builder_class,
1656
_serializer=self._serializer)
1659
class RepositoryFormatKnitPack1(RepositoryFormatPack):
1660
"""A no-subtrees parameterised Pack repository.
1662
This format was introduced in bzr.dev.
1665
repository_class = KnitPackRepository
1666
_commit_builder_class = PackCommitBuilder
1667
_serializer = xml5.serializer_v5
1669
def _get_matching_bzrdir(self):
1670
return bzrdir.format_registry.make_bzrdir('experimental')
1672
def _ignore_setting_bzrdir(self, format):
1675
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1677
def get_format_string(self):
1678
"""See RepositoryFormat.get_format_string()."""
1679
return "Bazaar Experimental no-subtrees\n"
1681
def get_format_description(self):
1682
"""See RepositoryFormat.get_format_description()."""
1683
return "Experimental no-subtrees"
1685
def check_conversion_target(self, target_format):
1689
class RepositoryFormatKnitPack3(RepositoryFormatPack):
1690
"""A subtrees parameterised Pack repository.
1692
This repository format uses the xml7 serializer to get:
1693
- support for recording full info about the tree root
1694
- support for recording tree-references
1696
This format was introduced in bzr.dev.
1699
repository_class = KnitPackRepository
1700
_commit_builder_class = PackRootCommitBuilder
1701
rich_root_data = True
1702
supports_tree_reference = True
1703
_serializer = xml7.serializer_v7
1705
def _get_matching_bzrdir(self):
1706
return bzrdir.format_registry.make_bzrdir('experimental-subtree')
1708
def _ignore_setting_bzrdir(self, format):
1711
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1713
def check_conversion_target(self, target_format):
1714
if not target_format.rich_root_data:
1715
raise errors.BadConversionTarget(
1716
'Does not support rich root data.', target_format)
1717
if not getattr(target_format, 'supports_tree_reference', False):
1718
raise errors.BadConversionTarget(
1719
'Does not support nested trees', target_format)
1721
def get_format_string(self):
1722
"""See RepositoryFormat.get_format_string()."""
1723
return "Bazaar Experimental subtrees\n"
1725
def get_format_description(self):
1726
"""See RepositoryFormat.get_format_description()."""
1727
return "Experimental subtrees\n"
added
removed
bzrlib/repofmt/pack_repo.py
