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.graph import Graph
30
from bzrlib.index import (
35
GraphIndexPrefixAdapter,
37
from bzrlib.knit import KnitGraphIndex, _PackAccess, _KnitData
38
from bzrlib.osutils import rand_chars
39
from bzrlib.pack import ContainerWriter
40
from bzrlib.store import revision
41
from bzrlib import tsort
57
from bzrlib.decorators import needs_read_lock, needs_write_lock
58
from bzrlib.repofmt.knitrepo import KnitRepository
59
from bzrlib.repository import (
62
MetaDirRepositoryFormat,
65
import bzrlib.revision as _mod_revision
66
from bzrlib.store.revision.knit import KnitRevisionStore
67
from bzrlib.store.versioned import VersionedFileStore
68
from bzrlib.trace import mutter, note, warning
71
class PackCommitBuilder(CommitBuilder):
72
"""A subclass of CommitBuilder to add texts with pack semantics.
74
Specifically this uses one knit object rather than one knit object per
75
added text, reducing memory and object pressure.
78
def __init__(self, repository, parents, config, timestamp=None,
79
timezone=None, committer=None, revprops=None,
81
CommitBuilder.__init__(self, repository, parents, config,
82
timestamp=timestamp, timezone=timezone, committer=committer,
83
revprops=revprops, revision_id=revision_id)
84
self._file_graph = Graph(
85
repository._pack_collection.text_index.combined_index)
87
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
88
return self.repository._pack_collection._add_text_to_weave(file_id,
89
self._new_revision_id, new_lines, parents, nostore_sha,
92
def _heads(self, file_id, revision_ids):
93
keys = [(file_id, revision_id) for revision_id in revision_ids]
94
return set([key[1] for key in self._file_graph.heads(keys)])
97
class PackRootCommitBuilder(RootCommitBuilder):
98
"""A subclass of RootCommitBuilder to add texts with pack semantics.
100
Specifically this uses one knit object rather than one knit object per
101
added text, reducing memory and object pressure.
104
def __init__(self, repository, parents, config, timestamp=None,
105
timezone=None, committer=None, revprops=None,
107
CommitBuilder.__init__(self, repository, parents, config,
108
timestamp=timestamp, timezone=timezone, committer=committer,
109
revprops=revprops, revision_id=revision_id)
110
self._file_graph = Graph(
111
repository._pack_collection.text_index.combined_index)
113
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
114
return self.repository._pack_collection._add_text_to_weave(file_id,
115
self._new_revision_id, new_lines, parents, nostore_sha,
118
def _heads(self, file_id, revision_ids):
119
keys = [(file_id, revision_id) for revision_id in revision_ids]
120
return set([key[1] for key in self._file_graph.heads(keys)])
124
"""An in memory proxy for a pack and its indices.
126
This is a base class that is not directly used, instead the classes
127
ExistingPack and NewPack are used.
130
def __init__(self, revision_index, inventory_index, text_index,
132
"""Create a pack instance.
134
:param revision_index: A GraphIndex for determining what revisions are
135
present in the Pack and accessing the locations of their texts.
136
:param inventory_index: A GraphIndex for determining what inventories are
137
present in the Pack and accessing the locations of their
139
:param text_index: A GraphIndex for determining what file texts
140
are present in the pack and accessing the locations of their
141
texts/deltas (via (fileid, revisionid) tuples).
142
:param revision_index: A GraphIndex for determining what signatures are
143
present in the Pack and accessing the locations of their texts.
145
self.revision_index = revision_index
146
self.inventory_index = inventory_index
147
self.text_index = text_index
148
self.signature_index = signature_index
150
def access_tuple(self):
151
"""Return a tuple (transport, name) for the pack content."""
152
return self.pack_transport, self.file_name()
155
"""Get the file name for the pack on disk."""
156
return self.name + '.pack'
158
def get_revision_count(self):
159
return self.revision_index.key_count()
161
def inventory_index_name(self, name):
162
"""The inv index is the name + .iix."""
163
return self.index_name('inventory', name)
165
def revision_index_name(self, name):
166
"""The revision index is the name + .rix."""
167
return self.index_name('revision', name)
169
def signature_index_name(self, name):
170
"""The signature index is the name + .six."""
171
return self.index_name('signature', name)
173
def text_index_name(self, name):
174
"""The text index is the name + .tix."""
175
return self.index_name('text', name)
177
def _external_compression_parents_of_texts(self):
180
for node in self.text_index.iter_all_entries():
182
refs.update(node[3][1])
186
class ExistingPack(Pack):
187
"""An in memory proxy for an existing .pack and its disk indices."""
189
def __init__(self, pack_transport, name, revision_index, inventory_index,
190
text_index, signature_index):
191
"""Create an ExistingPack object.
193
:param pack_transport: The transport where the pack file resides.
194
:param name: The name of the pack on disk in the pack_transport.
196
Pack.__init__(self, revision_index, inventory_index, text_index,
199
self.pack_transport = pack_transport
200
assert None not in (revision_index, inventory_index, text_index,
201
signature_index, name, pack_transport)
203
def __eq__(self, other):
204
return self.__dict__ == other.__dict__
206
def __ne__(self, other):
207
return not self.__eq__(other)
210
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
211
id(self), self.transport, self.name)
215
"""An in memory proxy for a pack which is being created."""
217
# A map of index 'type' to the file extension and position in the
219
index_definitions = {
220
'revision': ('.rix', 0),
221
'inventory': ('.iix', 1),
223
'signature': ('.six', 3),
226
def __init__(self, upload_transport, index_transport, pack_transport,
227
upload_suffix='', file_mode=None):
228
"""Create a NewPack instance.
230
:param upload_transport: A writable transport for the pack to be
231
incrementally uploaded to.
232
:param index_transport: A writable transport for the pack's indices to
233
be written to when the pack is finished.
234
:param pack_transport: A writable transport for the pack to be renamed
235
to when the upload is complete. This *must* be the same as
236
upload_transport.clone('../packs').
237
:param upload_suffix: An optional suffix to be given to any temporary
238
files created during the pack creation. e.g '.autopack'
239
:param file_mode: An optional file mode to create the new files with.
241
# The relative locations of the packs are constrained, but all are
242
# passed in because the caller has them, so as to avoid object churn.
244
# Revisions: parents list, no text compression.
245
InMemoryGraphIndex(reference_lists=1),
246
# Inventory: We want to map compression only, but currently the
247
# knit code hasn't been updated enough to understand that, so we
248
# have a regular 2-list index giving parents and compression
250
InMemoryGraphIndex(reference_lists=2),
251
# Texts: compression and per file graph, for all fileids - so two
252
# reference lists and two elements in the key tuple.
253
InMemoryGraphIndex(reference_lists=2, key_elements=2),
254
# Signatures: Just blobs to store, no compression, no parents
256
InMemoryGraphIndex(reference_lists=0),
258
# where should the new pack be opened
259
self.upload_transport = upload_transport
260
# where are indices written out to
261
self.index_transport = index_transport
262
# where is the pack renamed to when it is finished?
263
self.pack_transport = pack_transport
264
# What file mode to upload the pack and indices with.
265
self._file_mode = file_mode
266
# tracks the content written to the .pack file.
267
self._hash = md5.new()
268
# a four-tuple with the length in bytes of the indices, once the pack
269
# is finalised. (rev, inv, text, sigs)
270
self.index_sizes = None
271
# How much data to cache when writing packs. Note that this is not
272
# synchronised with reads, because it's not in the transport layer, so
273
# is not safe unless the client knows it won't be reading from the pack
275
self._cache_limit = 0
276
# the temporary pack file name.
277
self.random_name = rand_chars(20) + upload_suffix
278
# when was this pack started ?
279
self.start_time = time.time()
280
# open an output stream for the data added to the pack.
281
self.write_stream = self.upload_transport.open_write_stream(
282
self.random_name, mode=self._file_mode)
283
if 'pack' in debug.debug_flags:
284
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
285
time.ctime(), self.upload_transport.base, self.random_name,
286
time.time() - self.start_time)
287
# A list of byte sequences to be written to the new pack, and the
288
# aggregate size of them. Stored as a list rather than separate
289
# variables so that the _write_data closure below can update them.
290
self._buffer = [[], 0]
291
# create a callable for adding data
293
# robertc says- this is a closure rather than a method on the object
294
# so that the variables are locals, and faster than accessing object
296
def _write_data(bytes, flush=False, _buffer=self._buffer,
297
_write=self.write_stream.write, _update=self._hash.update):
298
_buffer[0].append(bytes)
299
_buffer[1] += len(bytes)
301
if _buffer[1] > self._cache_limit or flush:
302
bytes = ''.join(_buffer[0])
306
# expose this on self, for the occasion when clients want to add data.
307
self._write_data = _write_data
308
# a pack writer object to serialise pack records.
309
self._writer = pack.ContainerWriter(self._write_data)
311
# what state is the pack in? (open, finished, aborted)
315
"""Cancel creating this pack."""
316
self._state = 'aborted'
317
self.write_stream.close()
318
# Remove the temporary pack file.
319
self.upload_transport.delete(self.random_name)
320
# The indices have no state on disk.
322
def access_tuple(self):
323
"""Return a tuple (transport, name) for the pack content."""
324
assert self._state in ('open', 'finished')
325
if self._state == 'finished':
326
return Pack.access_tuple(self)
328
return self.upload_transport, self.random_name
330
def data_inserted(self):
331
"""True if data has been added to this pack."""
332
return bool(self.get_revision_count() or
333
self.inventory_index.key_count() or
334
self.text_index.key_count() or
335
self.signature_index.key_count())
338
"""Finish the new pack.
341
- finalises the content
342
- assigns a name (the md5 of the content, currently)
343
- writes out the associated indices
344
- renames the pack into place.
345
- stores the index size tuple for the pack in the index_sizes
350
self._write_data('', flush=True)
351
self.name = self._hash.hexdigest()
353
# XXX: It'd be better to write them all to temporary names, then
354
# rename them all into place, so that the window when only some are
355
# visible is smaller. On the other hand none will be seen until
356
# they're in the names list.
357
self.index_sizes = [None, None, None, None]
358
self._write_index('revision', self.revision_index, 'revision')
359
self._write_index('inventory', self.inventory_index, 'inventory')
360
self._write_index('text', self.text_index, 'file texts')
361
self._write_index('signature', self.signature_index,
362
'revision signatures')
363
self.write_stream.close()
364
# Note that this will clobber an existing pack with the same name,
365
# without checking for hash collisions. While this is undesirable this
366
# is something that can be rectified in a subsequent release. One way
367
# to rectify it may be to leave the pack at the original name, writing
368
# its pack-names entry as something like 'HASH: index-sizes
369
# temporary-name'. Allocate that and check for collisions, if it is
370
# collision free then rename it into place. If clients know this scheme
371
# they can handle missing-file errors by:
372
# - try for HASH.pack
373
# - try for temporary-name
374
# - refresh the pack-list to see if the pack is now absent
375
self.upload_transport.rename(self.random_name,
376
'../packs/' + self.name + '.pack')
377
self._state = 'finished'
378
if 'pack' in debug.debug_flags:
379
# XXX: size might be interesting?
380
mutter('%s: create_pack: pack renamed into place: %s%s->%s%s t+%6.3fs',
381
time.ctime(), self.upload_transport.base, self.random_name,
382
self.pack_transport, self.name,
383
time.time() - self.start_time)
386
"""Flush any current data."""
388
bytes = ''.join(self._buffer[0])
389
self.write_stream.write(bytes)
390
self._hash.update(bytes)
391
self._buffer[:] = [[], 0]
393
def index_name(self, index_type, name):
394
"""Get the disk name of an index type for pack name 'name'."""
395
return name + NewPack.index_definitions[index_type][0]
397
def index_offset(self, index_type):
398
"""Get the position in a index_size array for a given index type."""
399
return NewPack.index_definitions[index_type][1]
401
def _replace_index_with_readonly(self, index_type):
402
setattr(self, index_type + '_index',
403
GraphIndex(self.index_transport,
404
self.index_name(index_type, self.name),
405
self.index_sizes[self.index_offset(index_type)]))
407
def set_write_cache_size(self, size):
408
self._cache_limit = size
410
def _write_index(self, index_type, index, label):
411
"""Write out an index.
413
:param index_type: The type of index to write - e.g. 'revision'.
414
:param index: The index object to serialise.
415
:param label: What label to give the index e.g. 'revision'.
417
index_name = self.index_name(index_type, self.name)
418
self.index_sizes[self.index_offset(index_type)] = \
419
self.index_transport.put_file(index_name, index.finish(),
420
mode=self._file_mode)
421
if 'pack' in debug.debug_flags:
422
# XXX: size might be interesting?
423
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
424
time.ctime(), label, self.upload_transport.base,
425
self.random_name, time.time() - self.start_time)
426
# Replace the writable index on this object with a readonly,
427
# presently unloaded index. We should alter
428
# the index layer to make its finish() error if add_node is
429
# subsequently used. RBC
430
self._replace_index_with_readonly(index_type)
433
class AggregateIndex(object):
434
"""An aggregated index for the RepositoryPackCollection.
436
AggregateIndex is reponsible for managing the PackAccess object,
437
Index-To-Pack mapping, and all indices list for a specific type of index
438
such as 'revision index'.
440
A CombinedIndex provides an index on a single key space built up
441
from several on-disk indices. The AggregateIndex builds on this
442
to provide a knit access layer, and allows having up to one writable
443
index within the collection.
445
# XXX: Probably 'can be written to' could/should be separated from 'acts
446
# like a knit index' -- mbp 20071024
449
"""Create an AggregateIndex."""
450
self.index_to_pack = {}
451
self.combined_index = CombinedGraphIndex([])
452
self.knit_access = _PackAccess(self.index_to_pack)
454
def replace_indices(self, index_to_pack, indices):
455
"""Replace the current mappings with fresh ones.
457
This should probably not be used eventually, rather incremental add and
458
removal of indices. It has been added during refactoring of existing
461
:param index_to_pack: A mapping from index objects to
462
(transport, name) tuples for the pack file data.
463
:param indices: A list of indices.
465
# refresh the revision pack map dict without replacing the instance.
466
self.index_to_pack.clear()
467
self.index_to_pack.update(index_to_pack)
468
# XXX: API break - clearly a 'replace' method would be good?
469
self.combined_index._indices[:] = indices
470
# the current add nodes callback for the current writable index if
472
self.add_callback = None
474
def add_index(self, index, pack):
475
"""Add index to the aggregate, which is an index for Pack pack.
477
Future searches on the aggregate index will seach this new index
478
before all previously inserted indices.
480
:param index: An Index for the pack.
481
:param pack: A Pack instance.
483
# expose it to the index map
484
self.index_to_pack[index] = pack.access_tuple()
485
# put it at the front of the linear index list
486
self.combined_index.insert_index(0, index)
488
def add_writable_index(self, index, pack):
489
"""Add an index which is able to have data added to it.
491
There can be at most one writable index at any time. Any
492
modifications made to the knit are put into this index.
494
:param index: An index from the pack parameter.
495
:param pack: A Pack instance.
497
assert self.add_callback is None, \
498
"%s already has a writable index through %s" % \
499
(self, self.add_callback)
500
# allow writing: queue writes to a new index
501
self.add_index(index, pack)
502
# Updates the index to packs mapping as a side effect,
503
self.knit_access.set_writer(pack._writer, index, pack.access_tuple())
504
self.add_callback = index.add_nodes
507
"""Reset all the aggregate data to nothing."""
508
self.knit_access.set_writer(None, None, (None, None))
509
self.index_to_pack.clear()
510
del self.combined_index._indices[:]
511
self.add_callback = None
513
def remove_index(self, index, pack):
514
"""Remove index from the indices used to answer queries.
516
:param index: An index from the pack parameter.
517
:param pack: A Pack instance.
519
del self.index_to_pack[index]
520
self.combined_index._indices.remove(index)
521
if (self.add_callback is not None and
522
getattr(index, 'add_nodes', None) == self.add_callback):
523
self.add_callback = None
524
self.knit_access.set_writer(None, None, (None, None))
527
class Packer(object):
528
"""Create a pack from packs."""
530
def __init__(self, pack_collection, packs, suffix, revision_ids=None):
533
:param pack_collection: A RepositoryPackCollection object where the
534
new pack is being written to.
535
:param packs: The packs to combine.
536
:param suffix: The suffix to use on the temporary files for the pack.
537
:param revision_ids: Revision ids to limit the pack to.
541
self.revision_ids = revision_ids
542
# The pack object we are creating.
544
self._pack_collection = pack_collection
545
# The index layer keys for the revisions being copied. None for 'all
547
self._revision_keys = None
548
# What text keys to copy. None for 'all texts'. This is set by
549
# _copy_inventory_texts
550
self._text_filter = None
553
def _extra_init(self):
554
"""A template hook to allow extending the constructor trivially."""
556
def pack(self, pb=None):
557
"""Create a new pack by reading data from other packs.
559
This does little more than a bulk copy of data. One key difference
560
is that data with the same item key across multiple packs is elided
561
from the output. The new pack is written into the current pack store
562
along with its indices, and the name added to the pack names. The
563
source packs are not altered and are not required to be in the current
566
:param pb: An optional progress bar to use. A nested bar is created if
568
:return: A Pack object, or None if nothing was copied.
570
# open a pack - using the same name as the last temporary file
571
# - which has already been flushed, so its safe.
572
# XXX: - duplicate code warning with start_write_group; fix before
573
# considering 'done'.
574
if self._pack_collection._new_pack is not None:
575
raise errors.BzrError('call to create_pack_from_packs while '
576
'another pack is being written.')
577
if self.revision_ids is not None:
578
if len(self.revision_ids) == 0:
579
# silly fetch request.
582
self.revision_ids = frozenset(self.revision_ids)
584
self.pb = ui.ui_factory.nested_progress_bar()
588
return self._create_pack_from_packs()
594
"""Open a pack for the pack we are creating."""
595
return NewPack(self._pack_collection._upload_transport,
596
self._pack_collection._index_transport,
597
self._pack_collection._pack_transport, upload_suffix=self.suffix,
598
file_mode=self._pack_collection.repo.control_files._file_mode)
600
def _copy_revision_texts(self):
601
"""Copy revision data to the new pack."""
603
if self.revision_ids:
604
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
607
# select revision keys
608
revision_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
609
self.packs, 'revision_index')[0]
610
revision_nodes = self._pack_collection._index_contents(revision_index_map, revision_keys)
611
# copy revision keys and adjust values
612
self.pb.update("Copying revision texts", 1)
613
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
614
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
615
self.new_pack.revision_index, readv_group_iter, total_items))
616
if 'pack' in debug.debug_flags:
617
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
618
time.ctime(), self._pack_collection._upload_transport.base,
619
self.new_pack.random_name,
620
self.new_pack.revision_index.key_count(),
621
time.time() - self.new_pack.start_time)
622
self._revision_keys = revision_keys
624
def _copy_inventory_texts(self):
625
"""Copy the inventory texts to the new pack.
627
self._revision_keys is used to determine what inventories to copy.
629
Sets self._text_filter appropriately.
631
# select inventory keys
632
inv_keys = self._revision_keys # currently the same keyspace, and note that
633
# querying for keys here could introduce a bug where an inventory item
634
# is missed, so do not change it to query separately without cross
635
# checking like the text key check below.
636
inventory_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
637
self.packs, 'inventory_index')[0]
638
inv_nodes = self._pack_collection._index_contents(inventory_index_map, inv_keys)
639
# copy inventory keys and adjust values
640
# XXX: Should be a helper function to allow different inv representation
642
self.pb.update("Copying inventory texts", 2)
643
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
644
inv_lines = self._copy_nodes_graph(inventory_index_map,
645
self.new_pack._writer, self.new_pack.inventory_index,
646
readv_group_iter, total_items, output_lines=True)
647
if self.revision_ids:
648
self._process_inventory_lines(inv_lines)
650
# eat the iterator to cause it to execute.
652
self._text_filter = None
653
if 'pack' in debug.debug_flags:
654
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
655
time.ctime(), self._pack_collection._upload_transport.base,
656
self.new_pack.random_name,
657
self.new_pack.inventory_index.key_count(),
658
time.time() - new_pack.start_time)
660
def _copy_text_texts(self):
662
text_index_map, text_nodes = self._get_text_nodes()
663
if self._text_filter is not None:
664
# We could return the keys copied as part of the return value from
665
# _copy_nodes_graph but this doesn't work all that well with the
666
# need to get line output too, so we check separately, and as we're
667
# going to buffer everything anyway, we check beforehand, which
668
# saves reading knit data over the wire when we know there are
670
text_nodes = set(text_nodes)
671
present_text_keys = set(_node[1] for _node in text_nodes)
672
missing_text_keys = set(self._text_filter) - present_text_keys
673
if missing_text_keys:
674
# TODO: raise a specific error that can handle many missing
676
a_missing_key = missing_text_keys.pop()
677
raise errors.RevisionNotPresent(a_missing_key[1],
679
# copy text keys and adjust values
680
self.pb.update("Copying content texts", 3)
681
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
682
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
683
self.new_pack.text_index, readv_group_iter, total_items))
684
self._log_copied_texts()
686
def _check_references(self):
687
"""Make sure our external refereneces are present."""
688
external_refs = self.new_pack._external_compression_parents_of_texts()
690
index = self._pack_collection.text_index.combined_index
691
found_items = list(index.iter_entries(external_refs))
692
if len(found_items) != len(external_refs):
693
found_keys = set(k for idx, k, refs, value in found_items)
694
missing_items = external_refs - found_keys
695
missing_file_id, missing_revision_id = missing_items.pop()
696
raise errors.RevisionNotPresent(missing_revision_id,
699
def _create_pack_from_packs(self):
700
self.pb.update("Opening pack", 0, 5)
701
self.new_pack = self.open_pack()
702
new_pack = self.new_pack
703
# buffer data - we won't be reading-back during the pack creation and
704
# this makes a significant difference on sftp pushes.
705
new_pack.set_write_cache_size(1024*1024)
706
if 'pack' in debug.debug_flags:
707
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
708
for a_pack in self.packs]
709
if self.revision_ids is not None:
710
rev_count = len(self.revision_ids)
713
mutter('%s: create_pack: creating pack from source packs: '
714
'%s%s %s revisions wanted %s t=0',
715
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
716
plain_pack_list, rev_count)
717
self._copy_revision_texts()
718
self._copy_inventory_texts()
719
self._copy_text_texts()
720
# select signature keys
721
signature_filter = self._revision_keys # same keyspace
722
signature_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
723
self.packs, 'signature_index')[0]
724
signature_nodes = self._pack_collection._index_contents(signature_index_map,
726
# copy signature keys and adjust values
727
self.pb.update("Copying signature texts", 4)
728
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
729
new_pack.signature_index)
730
if 'pack' in debug.debug_flags:
731
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
732
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
733
new_pack.signature_index.key_count(),
734
time.time() - new_pack.start_time)
735
self._check_references()
736
if not self._use_pack(new_pack):
739
self.pb.update("Finishing pack", 5)
741
self._pack_collection.allocate(new_pack)
744
def _copy_nodes(self, nodes, index_map, writer, write_index):
745
"""Copy knit nodes between packs with no graph references."""
746
pb = ui.ui_factory.nested_progress_bar()
748
return self._do_copy_nodes(nodes, index_map, writer,
753
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb):
754
# for record verification
755
knit_data = _KnitData(None)
756
# plan a readv on each source pack:
758
nodes = sorted(nodes)
759
# how to map this into knit.py - or knit.py into this?
760
# we don't want the typical knit logic, we want grouping by pack
761
# at this point - perhaps a helper library for the following code
762
# duplication points?
764
for index, key, value in nodes:
765
if index not in request_groups:
766
request_groups[index] = []
767
request_groups[index].append((key, value))
769
pb.update("Copied record", record_index, len(nodes))
770
for index, items in request_groups.iteritems():
771
pack_readv_requests = []
772
for key, value in items:
773
# ---- KnitGraphIndex.get_position
774
bits = value[1:].split(' ')
775
offset, length = int(bits[0]), int(bits[1])
776
pack_readv_requests.append((offset, length, (key, value[0])))
777
# linear scan up the pack
778
pack_readv_requests.sort()
780
transport, path = index_map[index]
781
reader = pack.make_readv_reader(transport, path,
782
[offset[0:2] for offset in pack_readv_requests])
783
for (names, read_func), (_1, _2, (key, eol_flag)) in \
784
izip(reader.iter_records(), pack_readv_requests):
785
raw_data = read_func(None)
786
# check the header only
787
df, _ = knit_data._parse_record_header(key[-1], raw_data)
789
pos, size = writer.add_bytes_record(raw_data, names)
790
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
791
pb.update("Copied record", record_index)
794
def _copy_nodes_graph(self, index_map, writer, write_index,
795
readv_group_iter, total_items, output_lines=False):
796
"""Copy knit nodes between packs.
798
:param output_lines: Return lines present in the copied data as
799
an iterator of line,version_id.
801
pb = ui.ui_factory.nested_progress_bar()
803
for result in self._do_copy_nodes_graph(index_map, writer,
804
write_index, output_lines, pb, readv_group_iter, total_items):
807
# Python 2.4 does not permit try:finally: in a generator.
813
def _do_copy_nodes_graph(self, index_map, writer, write_index,
814
output_lines, pb, readv_group_iter, total_items):
815
# for record verification
816
knit_data = _KnitData(None)
817
# for line extraction when requested (inventories only)
819
factory = knit.KnitPlainFactory()
821
pb.update("Copied record", record_index, total_items)
822
for index, readv_vector, node_vector in readv_group_iter:
824
transport, path = index_map[index]
825
reader = pack.make_readv_reader(transport, path, readv_vector)
826
for (names, read_func), (key, eol_flag, references) in \
827
izip(reader.iter_records(), node_vector):
828
raw_data = read_func(None)
831
# read the entire thing
832
content, _ = knit_data._parse_record(version_id, raw_data)
833
if len(references[-1]) == 0:
834
line_iterator = factory.get_fulltext_content(content)
836
line_iterator = factory.get_linedelta_content(content)
837
for line in line_iterator:
838
yield line, version_id
840
# check the header only
841
df, _ = knit_data._parse_record_header(version_id, raw_data)
843
pos, size = writer.add_bytes_record(raw_data, names)
844
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
845
pb.update("Copied record", record_index)
848
def _get_text_nodes(self):
849
text_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
850
self.packs, 'text_index')[0]
851
return text_index_map, self._pack_collection._index_contents(text_index_map,
854
def _least_readv_node_readv(self, nodes):
855
"""Generate request groups for nodes using the least readv's.
857
:param nodes: An iterable of graph index nodes.
858
:return: Total node count and an iterator of the data needed to perform
859
readvs to obtain the data for nodes. Each item yielded by the
860
iterator is a tuple with:
861
index, readv_vector, node_vector. readv_vector is a list ready to
862
hand to the transport readv method, and node_vector is a list of
863
(key, eol_flag, references) for the the node retrieved by the
864
matching readv_vector.
866
# group by pack so we do one readv per pack
867
nodes = sorted(nodes)
870
for index, key, value, references in nodes:
871
if index not in request_groups:
872
request_groups[index] = []
873
request_groups[index].append((key, value, references))
875
for index, items in request_groups.iteritems():
876
pack_readv_requests = []
877
for key, value, references in items:
878
# ---- KnitGraphIndex.get_position
879
bits = value[1:].split(' ')
880
offset, length = int(bits[0]), int(bits[1])
881
pack_readv_requests.append(
882
((offset, length), (key, value[0], references)))
883
# linear scan up the pack to maximum range combining.
884
pack_readv_requests.sort()
885
# split out the readv and the node data.
886
pack_readv = [readv for readv, node in pack_readv_requests]
887
node_vector = [node for readv, node in pack_readv_requests]
888
result.append((index, pack_readv, node_vector))
891
def _log_copied_texts(self):
892
if 'pack' in debug.debug_flags:
893
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
894
time.ctime(), self._pack_collection._upload_transport.base,
895
self.new_pack.random_name,
896
self.new_pack.text_index.key_count(),
897
time.time() - self.new_pack.start_time)
899
def _process_inventory_lines(self, inv_lines):
900
"""Use up the inv_lines generator and setup a text key filter."""
901
repo = self._pack_collection.repo
902
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
903
inv_lines, self.revision_ids)
905
for fileid, file_revids in fileid_revisions.iteritems():
906
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
907
self._text_filter = text_filter
909
def _revision_node_readv(self, revision_nodes):
910
"""Return the total revisions and the readv's to issue.
912
:param revision_nodes: The revision index contents for the packs being
913
incorporated into the new pack.
914
:return: As per _least_readv_node_readv.
916
return self._least_readv_node_readv(revision_nodes)
918
def _use_pack(self, new_pack):
919
"""Return True if new_pack should be used.
921
:param new_pack: The pack that has just been created.
922
:return: True if the pack should be used.
924
return new_pack.data_inserted()
927
class OptimisingPacker(Packer):
928
"""A packer which spends more time to create better disk layouts."""
930
def _revision_node_readv(self, revision_nodes):
931
"""Return the total revisions and the readv's to issue.
933
This sort places revisions in topological order with the ancestors
936
:param revision_nodes: The revision index contents for the packs being
937
incorporated into the new pack.
938
:return: As per _least_readv_node_readv.
940
# build an ancestors dict
943
for index, key, value, references in revision_nodes:
944
ancestors[key] = references[0]
945
by_key[key] = (index, value, references)
946
order = tsort.topo_sort(ancestors)
948
# Single IO is pathological, but it will work as a starting point.
950
for key in reversed(order):
951
index, value, references = by_key[key]
952
# ---- KnitGraphIndex.get_position
953
bits = value[1:].split(' ')
954
offset, length = int(bits[0]), int(bits[1])
956
(index, [(offset, length)], [(key, value[0], references)]))
957
# TODO: combine requests in the same index that are in ascending order.
958
return total, requests
961
class ReconcilePacker(Packer):
962
"""A packer which regenerates indices etc as it copies.
964
This is used by ``bzr reconcile`` to cause parent text pointers to be
968
def _extra_init(self):
969
self._data_changed = False
971
def _process_inventory_lines(self, inv_lines):
972
"""Generate a text key reference map rather for reconciling with."""
973
repo = self._pack_collection.repo
974
refs = repo._find_text_key_references_from_xml_inventory_lines(
976
self._text_refs = refs
977
# during reconcile we:
978
# - convert unreferenced texts to full texts
979
# - correct texts which reference a text not copied to be full texts
980
# - copy all others as-is but with corrected parents.
981
# - so at this point we don't know enough to decide what becomes a full
983
self._text_filter = None
985
def _copy_text_texts(self):
986
"""generate what texts we should have and then copy."""
987
self.pb.update("Copying content texts", 3)
988
# we have three major tasks here:
989
# 1) generate the ideal index
990
repo = self._pack_collection.repo
991
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
993
self.new_pack.revision_index.iter_all_entries()])
994
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
995
# 2) generate a text_nodes list that contains all the deltas that can
996
# be used as-is, with corrected parents.
1000
NULL_REVISION = _mod_revision.NULL_REVISION
1001
text_index_map, text_nodes = self._get_text_nodes()
1002
for node in text_nodes:
1008
ideal_parents = tuple(ideal_index[node[1]])
1010
discarded_nodes.append(node)
1011
self._data_changed = True
1013
if ideal_parents == (NULL_REVISION,):
1015
if ideal_parents == node[3][0]:
1017
ok_nodes.append(node)
1018
elif ideal_parents[0:1] == node[3][0][0:1]:
1019
# the left most parent is the same, or there are no parents
1020
# today. Either way, we can preserve the representation as
1021
# long as we change the refs to be inserted.
1022
self._data_changed = True
1023
ok_nodes.append((node[0], node[1], node[2],
1024
(ideal_parents, node[3][1])))
1025
self._data_changed = True
1027
# Reinsert this text completely
1028
bad_texts.append((node[1], ideal_parents))
1029
self._data_changed = True
1030
# we're finished with some data.
1033
# 3) bulk copy the ok data
1034
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1035
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1036
self.new_pack.text_index, readv_group_iter, total_items))
1037
# 4) adhoc copy all the other texts.
1038
# We have to topologically insert all texts otherwise we can fail to
1039
# reconcile when parts of a single delta chain are preserved intact,
1040
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1041
# reinserted, and if d3 has incorrect parents it will also be
1042
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1043
# copied), so we will try to delta, but d2 is not currently able to be
1044
# extracted because it's basis d1 is not present. Topologically sorting
1045
# addresses this. The following generates a sort for all the texts that
1046
# are being inserted without having to reference the entire text key
1047
# space (we only topo sort the revisions, which is smaller).
1048
topo_order = tsort.topo_sort(ancestors)
1049
rev_order = dict(zip(topo_order, range(len(topo_order))))
1050
bad_texts.sort(key=lambda key:rev_order[key[0][1]])
1051
transaction = repo.get_transaction()
1052
file_id_index = GraphIndexPrefixAdapter(
1053
self.new_pack.text_index,
1055
add_nodes_callback=self.new_pack.text_index.add_nodes)
1056
knit_index = KnitGraphIndex(file_id_index,
1057
add_callback=file_id_index.add_nodes,
1058
deltas=True, parents=True)
1059
output_knit = knit.KnitVersionedFile('reconcile-texts',
1060
self._pack_collection.transport,
1063
access_method=_PackAccess(
1064
{self.new_pack.text_index:self.new_pack.access_tuple()},
1065
(self.new_pack._writer, self.new_pack.text_index)),
1066
factory=knit.KnitPlainFactory())
1067
for key, parent_keys in bad_texts:
1068
# We refer to the new pack to delta data being output.
1069
# A possible improvement would be to catch errors on short reads
1070
# and only flush then.
1071
self.new_pack.flush()
1073
for parent_key in parent_keys:
1074
if parent_key[0] != key[0]:
1075
# Graph parents must match the fileid
1076
raise errors.BzrError('Mismatched key parent %r:%r' %
1078
parents.append(parent_key[1])
1079
source_weave = repo.weave_store.get_weave(key[0], transaction)
1080
text_lines = source_weave.get_lines(key[1])
1081
# adapt the 'knit' to the current file_id.
1082
file_id_index = GraphIndexPrefixAdapter(
1083
self.new_pack.text_index,
1085
add_nodes_callback=self.new_pack.text_index.add_nodes)
1086
knit_index._graph_index = file_id_index
1087
knit_index._add_callback = file_id_index.add_nodes
1088
output_knit.add_lines_with_ghosts(
1089
key[1], parents, text_lines, random_id=True, check_content=False)
1090
# 5) check that nothing inserted has a reference outside the keyspace.
1091
missing_text_keys = self.new_pack._external_compression_parents_of_texts()
1092
if missing_text_keys:
1093
raise errors.BzrError('Reference to missing compression parents %r'
1095
self._log_copied_texts()
1097
def _use_pack(self, new_pack):
1098
"""Override _use_pack to check for reconcile having changed content."""
1099
# XXX: we might be better checking this at the copy time.
1100
original_inventory_keys = set()
1101
inv_index = self._pack_collection.inventory_index.combined_index
1102
for entry in inv_index.iter_all_entries():
1103
original_inventory_keys.add(entry[1])
1104
new_inventory_keys = set()
1105
for entry in new_pack.inventory_index.iter_all_entries():
1106
new_inventory_keys.add(entry[1])
1107
if new_inventory_keys != original_inventory_keys:
1108
self._data_changed = True
1109
return new_pack.data_inserted() and self._data_changed
1112
class RepositoryPackCollection(object):
1113
"""Management of packs within a repository."""
1115
def __init__(self, repo, transport, index_transport, upload_transport,
1117
"""Create a new RepositoryPackCollection.
1119
:param transport: Addresses the repository base directory
1120
(typically .bzr/repository/).
1121
:param index_transport: Addresses the directory containing indices.
1122
:param upload_transport: Addresses the directory into which packs are written
1123
while they're being created.
1124
:param pack_transport: Addresses the directory of existing complete packs.
1127
self.transport = transport
1128
self._index_transport = index_transport
1129
self._upload_transport = upload_transport
1130
self._pack_transport = pack_transport
1131
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
1134
self._packs_by_name = {}
1135
# the previous pack-names content
1136
self._packs_at_load = None
1137
# when a pack is being created by this object, the state of that pack.
1138
self._new_pack = None
1139
# aggregated revision index data
1140
self.revision_index = AggregateIndex()
1141
self.inventory_index = AggregateIndex()
1142
self.text_index = AggregateIndex()
1143
self.signature_index = AggregateIndex()
1145
def add_pack_to_memory(self, pack):
1146
"""Make a Pack object available to the repository to satisfy queries.
1148
:param pack: A Pack object.
1150
assert pack.name not in self._packs_by_name
1151
self.packs.append(pack)
1152
self._packs_by_name[pack.name] = pack
1153
self.revision_index.add_index(pack.revision_index, pack)
1154
self.inventory_index.add_index(pack.inventory_index, pack)
1155
self.text_index.add_index(pack.text_index, pack)
1156
self.signature_index.add_index(pack.signature_index, pack)
1158
def _add_text_to_weave(self, file_id, revision_id, new_lines, parents,
1159
nostore_sha, random_revid):
1160
file_id_index = GraphIndexPrefixAdapter(
1161
self.text_index.combined_index,
1163
add_nodes_callback=self.text_index.add_callback)
1164
self.repo._text_knit._index._graph_index = file_id_index
1165
self.repo._text_knit._index._add_callback = file_id_index.add_nodes
1166
return self.repo._text_knit.add_lines_with_ghosts(
1167
revision_id, parents, new_lines, nostore_sha=nostore_sha,
1168
random_id=random_revid, check_content=False)[0:2]
1170
def all_packs(self):
1171
"""Return a list of all the Pack objects this repository has.
1173
Note that an in-progress pack being created is not returned.
1175
:return: A list of Pack objects for all the packs in the repository.
1178
for name in self.names():
1179
result.append(self.get_pack_by_name(name))
1183
"""Pack the pack collection incrementally.
1185
This will not attempt global reorganisation or recompression,
1186
rather it will just ensure that the total number of packs does
1187
not grow without bound. It uses the _max_pack_count method to
1188
determine if autopacking is needed, and the pack_distribution
1189
method to determine the number of revisions in each pack.
1191
If autopacking takes place then the packs name collection will have
1192
been flushed to disk - packing requires updating the name collection
1193
in synchronisation with certain steps. Otherwise the names collection
1196
:return: True if packing took place.
1198
# XXX: Should not be needed when the management of indices is sane.
1199
total_revisions = self.revision_index.combined_index.key_count()
1200
total_packs = len(self._names)
1201
if self._max_pack_count(total_revisions) >= total_packs:
1203
# XXX: the following may want to be a class, to pack with a given
1205
mutter('Auto-packing repository %s, which has %d pack files, '
1206
'containing %d revisions into %d packs.', self, total_packs,
1207
total_revisions, self._max_pack_count(total_revisions))
1208
# determine which packs need changing
1209
pack_distribution = self.pack_distribution(total_revisions)
1211
for pack in self.all_packs():
1212
revision_count = pack.get_revision_count()
1213
if revision_count == 0:
1214
# revision less packs are not generated by normal operation,
1215
# only by operations like sign-my-commits, and thus will not
1216
# tend to grow rapdily or without bound like commit containing
1217
# packs do - leave them alone as packing them really should
1218
# group their data with the relevant commit, and that may
1219
# involve rewriting ancient history - which autopack tries to
1220
# avoid. Alternatively we could not group the data but treat
1221
# each of these as having a single revision, and thus add
1222
# one revision for each to the total revision count, to get
1223
# a matching distribution.
1225
existing_packs.append((revision_count, pack))
1226
pack_operations = self.plan_autopack_combinations(
1227
existing_packs, pack_distribution)
1228
self._execute_pack_operations(pack_operations)
1231
def _execute_pack_operations(self, pack_operations, _packer_class=Packer):
1232
"""Execute a series of pack operations.
1234
:param pack_operations: A list of [revision_count, packs_to_combine].
1235
:param _packer_class: The class of packer to use (default: Packer).
1238
for revision_count, packs in pack_operations:
1239
# we may have no-ops from the setup logic
1242
_packer_class(self, packs, '.autopack').pack()
1244
self._remove_pack_from_memory(pack)
1245
# record the newly available packs and stop advertising the old
1247
self._save_pack_names(clear_obsolete_packs=True)
1248
# Move the old packs out of the way now they are no longer referenced.
1249
for revision_count, packs in pack_operations:
1250
self._obsolete_packs(packs)
1252
def lock_names(self):
1253
"""Acquire the mutex around the pack-names index.
1255
This cannot be used in the middle of a read-only transaction on the
1258
self.repo.control_files.lock_write()
1261
"""Pack the pack collection totally."""
1262
self.ensure_loaded()
1263
total_packs = len(self._names)
1265
# This is arguably wrong because we might not be optimal, but for
1266
# now lets leave it in. (e.g. reconcile -> one pack. But not
1269
total_revisions = self.revision_index.combined_index.key_count()
1270
# XXX: the following may want to be a class, to pack with a given
1272
mutter('Packing repository %s, which has %d pack files, '
1273
'containing %d revisions into 1 packs.', self, total_packs,
1275
# determine which packs need changing
1276
pack_distribution = [1]
1277
pack_operations = [[0, []]]
1278
for pack in self.all_packs():
1279
pack_operations[-1][0] += pack.get_revision_count()
1280
pack_operations[-1][1].append(pack)
1281
self._execute_pack_operations(pack_operations, OptimisingPacker)
1283
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1284
"""Plan a pack operation.
1286
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1288
:param pack_distribution: A list with the number of revisions desired
1291
if len(existing_packs) <= len(pack_distribution):
1293
existing_packs.sort(reverse=True)
1294
pack_operations = [[0, []]]
1295
# plan out what packs to keep, and what to reorganise
1296
while len(existing_packs):
1297
# take the largest pack, and if its less than the head of the
1298
# distribution chart we will include its contents in the new pack for
1299
# that position. If its larger, we remove its size from the
1300
# distribution chart
1301
next_pack_rev_count, next_pack = existing_packs.pop(0)
1302
if next_pack_rev_count >= pack_distribution[0]:
1303
# this is already packed 'better' than this, so we can
1304
# not waste time packing it.
1305
while next_pack_rev_count > 0:
1306
next_pack_rev_count -= pack_distribution[0]
1307
if next_pack_rev_count >= 0:
1309
del pack_distribution[0]
1311
# didn't use that entire bucket up
1312
pack_distribution[0] = -next_pack_rev_count
1314
# add the revisions we're going to add to the next output pack
1315
pack_operations[-1][0] += next_pack_rev_count
1316
# allocate this pack to the next pack sub operation
1317
pack_operations[-1][1].append(next_pack)
1318
if pack_operations[-1][0] >= pack_distribution[0]:
1319
# this pack is used up, shift left.
1320
del pack_distribution[0]
1321
pack_operations.append([0, []])
1323
return pack_operations
1325
def ensure_loaded(self):
1326
# NB: if you see an assertion error here, its probably access against
1327
# an unlocked repo. Naughty.
1328
if not self.repo.is_locked():
1329
raise errors.ObjectNotLocked(self.repo)
1330
if self._names is None:
1332
self._packs_at_load = set()
1333
for index, key, value in self._iter_disk_pack_index():
1335
self._names[name] = self._parse_index_sizes(value)
1336
self._packs_at_load.add((key, value))
1337
# populate all the metadata.
1340
def _parse_index_sizes(self, value):
1341
"""Parse a string of index sizes."""
1342
return tuple([int(digits) for digits in value.split(' ')])
1344
def get_pack_by_name(self, name):
1345
"""Get a Pack object by name.
1347
:param name: The name of the pack - e.g. '123456'
1348
:return: A Pack object.
1351
return self._packs_by_name[name]
1353
rev_index = self._make_index(name, '.rix')
1354
inv_index = self._make_index(name, '.iix')
1355
txt_index = self._make_index(name, '.tix')
1356
sig_index = self._make_index(name, '.six')
1357
result = ExistingPack(self._pack_transport, name, rev_index,
1358
inv_index, txt_index, sig_index)
1359
self.add_pack_to_memory(result)
1362
def allocate(self, a_new_pack):
1363
"""Allocate name in the list of packs.
1365
:param a_new_pack: A NewPack instance to be added to the collection of
1366
packs for this repository.
1368
self.ensure_loaded()
1369
if a_new_pack.name in self._names:
1370
raise errors.BzrError(
1371
'Pack %r already exists in %s' % (a_new_pack.name, self))
1372
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1373
self.add_pack_to_memory(a_new_pack)
1375
def _iter_disk_pack_index(self):
1376
"""Iterate over the contents of the pack-names index.
1378
This is used when loading the list from disk, and before writing to
1379
detect updates from others during our write operation.
1380
:return: An iterator of the index contents.
1382
return GraphIndex(self.transport, 'pack-names', None
1383
).iter_all_entries()
1385
def _make_index(self, name, suffix):
1386
size_offset = self._suffix_offsets[suffix]
1387
index_name = name + suffix
1388
index_size = self._names[name][size_offset]
1390
self._index_transport, index_name, index_size)
1392
def _max_pack_count(self, total_revisions):
1393
"""Return the maximum number of packs to use for total revisions.
1395
:param total_revisions: The total number of revisions in the
1398
if not total_revisions:
1400
digits = str(total_revisions)
1402
for digit in digits:
1403
result += int(digit)
1407
"""Provide an order to the underlying names."""
1408
return sorted(self._names.keys())
1410
def _obsolete_packs(self, packs):
1411
"""Move a number of packs which have been obsoleted out of the way.
1413
Each pack and its associated indices are moved out of the way.
1415
Note: for correctness this function should only be called after a new
1416
pack names index has been written without these pack names, and with
1417
the names of packs that contain the data previously available via these
1420
:param packs: The packs to obsolete.
1421
:param return: None.
1424
pack.pack_transport.rename(pack.file_name(),
1425
'../obsolete_packs/' + pack.file_name())
1426
# TODO: Probably needs to know all possible indices for this pack
1427
# - or maybe list the directory and move all indices matching this
1428
# name whether we recognize it or not?
1429
for suffix in ('.iix', '.six', '.tix', '.rix'):
1430
self._index_transport.rename(pack.name + suffix,
1431
'../obsolete_packs/' + pack.name + suffix)
1433
def pack_distribution(self, total_revisions):
1434
"""Generate a list of the number of revisions to put in each pack.
1436
:param total_revisions: The total number of revisions in the
1439
if total_revisions == 0:
1441
digits = reversed(str(total_revisions))
1443
for exponent, count in enumerate(digits):
1444
size = 10 ** exponent
1445
for pos in range(int(count)):
1447
return list(reversed(result))
1449
def _pack_tuple(self, name):
1450
"""Return a tuple with the transport and file name for a pack name."""
1451
return self._pack_transport, name + '.pack'
1453
def _remove_pack_from_memory(self, pack):
1454
"""Remove pack from the packs accessed by this repository.
1456
Only affects memory state, until self._save_pack_names() is invoked.
1458
self._names.pop(pack.name)
1459
self._packs_by_name.pop(pack.name)
1460
self._remove_pack_indices(pack)
1462
def _remove_pack_indices(self, pack):
1463
"""Remove the indices for pack from the aggregated indices."""
1464
self.revision_index.remove_index(pack.revision_index, pack)
1465
self.inventory_index.remove_index(pack.inventory_index, pack)
1466
self.text_index.remove_index(pack.text_index, pack)
1467
self.signature_index.remove_index(pack.signature_index, pack)
1470
"""Clear all cached data."""
1471
# cached revision data
1472
self.repo._revision_knit = None
1473
self.revision_index.clear()
1474
# cached signature data
1475
self.repo._signature_knit = None
1476
self.signature_index.clear()
1477
# cached file text data
1478
self.text_index.clear()
1479
self.repo._text_knit = None
1480
# cached inventory data
1481
self.inventory_index.clear()
1482
# remove the open pack
1483
self._new_pack = None
1484
# information about packs.
1487
self._packs_by_name = {}
1488
self._packs_at_load = None
1490
def _make_index_map(self, index_suffix):
1491
"""Return information on existing indices.
1493
:param suffix: Index suffix added to pack name.
1495
:returns: (pack_map, indices) where indices is a list of GraphIndex
1496
objects, and pack_map is a mapping from those objects to the
1497
pack tuple they describe.
1499
# TODO: stop using this; it creates new indices unnecessarily.
1500
self.ensure_loaded()
1501
suffix_map = {'.rix': 'revision_index',
1502
'.six': 'signature_index',
1503
'.iix': 'inventory_index',
1504
'.tix': 'text_index',
1506
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1507
suffix_map[index_suffix])
1509
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1510
"""Convert a list of packs to an index pack map and index list.
1512
:param packs: The packs list to process.
1513
:param index_attribute: The attribute that the desired index is found
1515
:return: A tuple (map, list) where map contains the dict from
1516
index:pack_tuple, and lsit contains the indices in the same order
1522
index = getattr(pack, index_attribute)
1523
indices.append(index)
1524
pack_map[index] = (pack.pack_transport, pack.file_name())
1525
return pack_map, indices
1527
def _index_contents(self, pack_map, key_filter=None):
1528
"""Get an iterable of the index contents from a pack_map.
1530
:param pack_map: A map from indices to pack details.
1531
:param key_filter: An optional filter to limit the
1534
indices = [index for index in pack_map.iterkeys()]
1535
all_index = CombinedGraphIndex(indices)
1536
if key_filter is None:
1537
return all_index.iter_all_entries()
1539
return all_index.iter_entries(key_filter)
1541
def _unlock_names(self):
1542
"""Release the mutex around the pack-names index."""
1543
self.repo.control_files.unlock()
1545
def _save_pack_names(self, clear_obsolete_packs=False):
1546
"""Save the list of packs.
1548
This will take out the mutex around the pack names list for the
1549
duration of the method call. If concurrent updates have been made, a
1550
three-way merge between the current list and the current in memory list
1553
:param clear_obsolete_packs: If True, clear out the contents of the
1554
obsolete_packs directory.
1558
builder = GraphIndexBuilder()
1559
# load the disk nodes across
1561
for index, key, value in self._iter_disk_pack_index():
1562
disk_nodes.add((key, value))
1563
# do a two-way diff against our original content
1564
current_nodes = set()
1565
for name, sizes in self._names.iteritems():
1567
((name, ), ' '.join(str(size) for size in sizes)))
1568
deleted_nodes = self._packs_at_load - current_nodes
1569
new_nodes = current_nodes - self._packs_at_load
1570
disk_nodes.difference_update(deleted_nodes)
1571
disk_nodes.update(new_nodes)
1572
# TODO: handle same-name, index-size-changes here -
1573
# e.g. use the value from disk, not ours, *unless* we're the one
1575
for key, value in disk_nodes:
1576
builder.add_node(key, value)
1577
self.transport.put_file('pack-names', builder.finish(),
1578
mode=self.repo.control_files._file_mode)
1579
# move the baseline forward
1580
self._packs_at_load = disk_nodes
1581
# now clear out the obsolete packs directory
1582
if clear_obsolete_packs:
1583
self.transport.clone('obsolete_packs').delete_multi(
1584
self.transport.list_dir('obsolete_packs'))
1586
self._unlock_names()
1587
# synchronise the memory packs list with what we just wrote:
1588
new_names = dict(disk_nodes)
1589
# drop no longer present nodes
1590
for pack in self.all_packs():
1591
if (pack.name,) not in new_names:
1592
self._remove_pack_from_memory(pack)
1593
# add new nodes/refresh existing ones
1594
for key, value in disk_nodes:
1596
sizes = self._parse_index_sizes(value)
1597
if name in self._names:
1599
if sizes != self._names[name]:
1600
# the pack for name has had its indices replaced - rare but
1601
# important to handle. XXX: probably can never happen today
1602
# because the three-way merge code above does not handle it
1603
# - you may end up adding the same key twice to the new
1604
# disk index because the set values are the same, unless
1605
# the only index shows up as deleted by the set difference
1606
# - which it may. Until there is a specific test for this,
1607
# assume its broken. RBC 20071017.
1608
self._remove_pack_from_memory(self.get_pack_by_name(name))
1609
self._names[name] = sizes
1610
self.get_pack_by_name(name)
1613
self._names[name] = sizes
1614
self.get_pack_by_name(name)
1616
def _start_write_group(self):
1617
# Do not permit preparation for writing if we're not in a 'write lock'.
1618
if not self.repo.is_write_locked():
1619
raise errors.NotWriteLocked(self)
1620
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1621
self._pack_transport, upload_suffix='.pack',
1622
file_mode=self.repo.control_files._file_mode)
1623
# allow writing: queue writes to a new index
1624
self.revision_index.add_writable_index(self._new_pack.revision_index,
1626
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
1628
self.text_index.add_writable_index(self._new_pack.text_index,
1630
self.signature_index.add_writable_index(self._new_pack.signature_index,
1633
# reused revision and signature knits may need updating
1635
# "Hysterical raisins. client code in bzrlib grabs those knits outside
1636
# of write groups and then mutates it inside the write group."
1637
if self.repo._revision_knit is not None:
1638
self.repo._revision_knit._index._add_callback = \
1639
self.revision_index.add_callback
1640
if self.repo._signature_knit is not None:
1641
self.repo._signature_knit._index._add_callback = \
1642
self.signature_index.add_callback
1643
# create a reused knit object for text addition in commit.
1644
self.repo._text_knit = self.repo.weave_store.get_weave_or_empty(
1647
def _abort_write_group(self):
1648
# FIXME: just drop the transient index.
1649
# forget what names there are
1650
self._new_pack.abort()
1651
self._remove_pack_indices(self._new_pack)
1652
self._new_pack = None
1653
self.repo._text_knit = None
1655
def _commit_write_group(self):
1656
self._remove_pack_indices(self._new_pack)
1657
if self._new_pack.data_inserted():
1658
# get all the data to disk and read to use
1659
self._new_pack.finish()
1660
self.allocate(self._new_pack)
1661
self._new_pack = None
1662
if not self.autopack():
1663
# when autopack takes no steps, the names list is still
1665
self._save_pack_names()
1667
self._new_pack.abort()
1668
self._new_pack = None
1669
self.repo._text_knit = None
1672
class KnitPackRevisionStore(KnitRevisionStore):
1673
"""An object to adapt access from RevisionStore's to use KnitPacks.
1675
This class works by replacing the original RevisionStore.
1676
We need to do this because the KnitPackRevisionStore is less
1677
isolated in its layering - it uses services from the repo.
1680
def __init__(self, repo, transport, revisionstore):
1681
"""Create a KnitPackRevisionStore on repo with revisionstore.
1683
This will store its state in the Repository, use the
1684
indices to provide a KnitGraphIndex,
1685
and at the end of transactions write new indices.
1687
KnitRevisionStore.__init__(self, revisionstore.versioned_file_store)
1689
self._serializer = revisionstore._serializer
1690
self.transport = transport
1692
def get_revision_file(self, transaction):
1693
"""Get the revision versioned file object."""
1694
if getattr(self.repo, '_revision_knit', None) is not None:
1695
return self.repo._revision_knit
1696
self.repo._pack_collection.ensure_loaded()
1697
add_callback = self.repo._pack_collection.revision_index.add_callback
1698
# setup knit specific objects
1699
knit_index = KnitGraphIndex(
1700
self.repo._pack_collection.revision_index.combined_index,
1701
add_callback=add_callback)
1702
self.repo._revision_knit = knit.KnitVersionedFile(
1703
'revisions', self.transport.clone('..'),
1704
self.repo.control_files._file_mode,
1705
create=False, access_mode=self.repo._access_mode(),
1706
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1707
access_method=self.repo._pack_collection.revision_index.knit_access)
1708
return self.repo._revision_knit
1710
def get_signature_file(self, transaction):
1711
"""Get the signature versioned file object."""
1712
if getattr(self.repo, '_signature_knit', None) is not None:
1713
return self.repo._signature_knit
1714
self.repo._pack_collection.ensure_loaded()
1715
add_callback = self.repo._pack_collection.signature_index.add_callback
1716
# setup knit specific objects
1717
knit_index = KnitGraphIndex(
1718
self.repo._pack_collection.signature_index.combined_index,
1719
add_callback=add_callback, parents=False)
1720
self.repo._signature_knit = knit.KnitVersionedFile(
1721
'signatures', self.transport.clone('..'),
1722
self.repo.control_files._file_mode,
1723
create=False, access_mode=self.repo._access_mode(),
1724
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1725
access_method=self.repo._pack_collection.signature_index.knit_access)
1726
return self.repo._signature_knit
1729
class KnitPackTextStore(VersionedFileStore):
1730
"""Presents a TextStore abstraction on top of packs.
1732
This class works by replacing the original VersionedFileStore.
1733
We need to do this because the KnitPackRevisionStore is less
1734
isolated in its layering - it uses services from the repo and shares them
1735
with all the data written in a single write group.
1738
def __init__(self, repo, transport, weavestore):
1739
"""Create a KnitPackTextStore on repo with weavestore.
1741
This will store its state in the Repository, use the
1742
indices FileNames to provide a KnitGraphIndex,
1743
and at the end of transactions write new indices.
1745
# don't call base class constructor - it's not suitable.
1746
# no transient data stored in the transaction
1748
self._precious = False
1750
self.transport = transport
1751
self.weavestore = weavestore
1752
# XXX for check() which isn't updated yet
1753
self._transport = weavestore._transport
1755
def get_weave_or_empty(self, file_id, transaction):
1756
"""Get a 'Knit' backed by the .tix indices.
1758
The transaction parameter is ignored.
1760
self.repo._pack_collection.ensure_loaded()
1761
add_callback = self.repo._pack_collection.text_index.add_callback
1762
# setup knit specific objects
1763
file_id_index = GraphIndexPrefixAdapter(
1764
self.repo._pack_collection.text_index.combined_index,
1765
(file_id, ), 1, add_nodes_callback=add_callback)
1766
knit_index = KnitGraphIndex(file_id_index,
1767
add_callback=file_id_index.add_nodes,
1768
deltas=True, parents=True)
1769
return knit.KnitVersionedFile('text:' + file_id,
1770
self.transport.clone('..'),
1773
access_method=self.repo._pack_collection.text_index.knit_access,
1774
factory=knit.KnitPlainFactory())
1776
get_weave = get_weave_or_empty
1779
"""Generate a list of the fileids inserted, for use by check."""
1780
self.repo._pack_collection.ensure_loaded()
1782
for index, key, value, refs in \
1783
self.repo._pack_collection.text_index.combined_index.iter_all_entries():
1788
class InventoryKnitThunk(object):
1789
"""An object to manage thunking get_inventory_weave to pack based knits."""
1791
def __init__(self, repo, transport):
1792
"""Create an InventoryKnitThunk for repo at transport.
1794
This will store its state in the Repository, use the
1795
indices FileNames to provide a KnitGraphIndex,
1796
and at the end of transactions write a new index..
1799
self.transport = transport
1801
def get_weave(self):
1802
"""Get a 'Knit' that contains inventory data."""
1803
self.repo._pack_collection.ensure_loaded()
1804
add_callback = self.repo._pack_collection.inventory_index.add_callback
1805
# setup knit specific objects
1806
knit_index = KnitGraphIndex(
1807
self.repo._pack_collection.inventory_index.combined_index,
1808
add_callback=add_callback, deltas=True, parents=True)
1809
return knit.KnitVersionedFile(
1810
'inventory', self.transport.clone('..'),
1811
self.repo.control_files._file_mode,
1812
create=False, access_mode=self.repo._access_mode(),
1813
index=knit_index, delta=True, factory=knit.KnitPlainFactory(),
1814
access_method=self.repo._pack_collection.inventory_index.knit_access)
1817
class KnitPackRepository(KnitRepository):
1818
"""Experimental graph-knit using repository."""
1820
def __init__(self, _format, a_bzrdir, control_files, _revision_store,
1821
control_store, text_store, _commit_builder_class, _serializer):
1822
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
1823
_revision_store, control_store, text_store, _commit_builder_class,
1825
index_transport = control_files._transport.clone('indices')
1826
self._pack_collection = RepositoryPackCollection(self, control_files._transport,
1828
control_files._transport.clone('upload'),
1829
control_files._transport.clone('packs'))
1830
self._revision_store = KnitPackRevisionStore(self, index_transport, self._revision_store)
1831
self.weave_store = KnitPackTextStore(self, index_transport, self.weave_store)
1832
self._inv_thunk = InventoryKnitThunk(self, index_transport)
1833
# True when the repository object is 'write locked' (as opposed to the
1834
# physical lock only taken out around changes to the pack-names list.)
1835
# Another way to represent this would be a decorator around the control
1836
# files object that presents logical locks as physical ones - if this
1837
# gets ugly consider that alternative design. RBC 20071011
1838
self._write_lock_count = 0
1839
self._transaction = None
1841
self._reconcile_does_inventory_gc = True
1842
self._reconcile_fixes_text_parents = True
1843
self._reconcile_backsup_inventory = False
1845
def _abort_write_group(self):
1846
self._pack_collection._abort_write_group()
1848
def _access_mode(self):
1849
"""Return 'w' or 'r' for depending on whether a write lock is active.
1851
This method is a helper for the Knit-thunking support objects.
1853
if self.is_write_locked():
1857
def _find_inconsistent_revision_parents(self):
1858
"""Find revisions with incorrectly cached parents.
1860
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1861
parents-in-revision).
1863
if not self.is_locked():
1864
raise errors.ObjectNotLocked(self)
1865
pb = ui.ui_factory.nested_progress_bar()
1868
revision_nodes = self._pack_collection.revision_index \
1869
.combined_index.iter_all_entries()
1870
index_positions = []
1871
# Get the cached index values for all revisions, and also the location
1872
# in each index of the revision text so we can perform linear IO.
1873
for index, key, value, refs in revision_nodes:
1874
pos, length = value[1:].split(' ')
1875
index_positions.append((index, int(pos), key[0],
1876
tuple(parent[0] for parent in refs[0])))
1877
pb.update("Reading revision index.", 0, 0)
1878
index_positions.sort()
1879
batch_count = len(index_positions) / 1000 + 1
1880
pb.update("Checking cached revision graph.", 0, batch_count)
1881
for offset in xrange(batch_count):
1882
pb.update("Checking cached revision graph.", offset)
1883
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1886
rev_ids = [item[2] for item in to_query]
1887
revs = self.get_revisions(rev_ids)
1888
for revision, item in zip(revs, to_query):
1889
index_parents = item[3]
1890
rev_parents = tuple(revision.parent_ids)
1891
if index_parents != rev_parents:
1892
result.append((revision.revision_id, index_parents, rev_parents))
1897
def get_parents(self, revision_ids):
1898
"""See StackedParentsProvider.get_parents.
1900
This implementation accesses the combined revision index to provide
1903
self._pack_collection.ensure_loaded()
1904
index = self._pack_collection.revision_index.combined_index
1906
for revision_id in revision_ids:
1907
if revision_id != _mod_revision.NULL_REVISION:
1908
search_keys.add((revision_id,))
1909
found_parents = {_mod_revision.NULL_REVISION:[]}
1910
for index, key, value, refs in index.iter_entries(search_keys):
1913
parents = (_mod_revision.NULL_REVISION,)
1915
parents = tuple(parent[0] for parent in parents)
1916
found_parents[key[0]] = parents
1918
for revision_id in revision_ids:
1920
result.append(found_parents[revision_id])
1925
def _make_parents_provider(self):
1928
def _refresh_data(self):
1929
if self._write_lock_count == 1 or (
1930
self.control_files._lock_count == 1 and
1931
self.control_files._lock_mode == 'r'):
1932
# forget what names there are
1933
self._pack_collection.reset()
1934
# XXX: Better to do an in-memory merge when acquiring a new lock -
1935
# factor out code from _save_pack_names.
1936
self._pack_collection.ensure_loaded()
1938
def _start_write_group(self):
1939
self._pack_collection._start_write_group()
1941
def _commit_write_group(self):
1942
return self._pack_collection._commit_write_group()
1944
def get_inventory_weave(self):
1945
return self._inv_thunk.get_weave()
1947
def get_transaction(self):
1948
if self._write_lock_count:
1949
return self._transaction
1951
return self.control_files.get_transaction()
1953
def is_locked(self):
1954
return self._write_lock_count or self.control_files.is_locked()
1956
def is_write_locked(self):
1957
return self._write_lock_count
1959
def lock_write(self, token=None):
1960
if not self._write_lock_count and self.is_locked():
1961
raise errors.ReadOnlyError(self)
1962
self._write_lock_count += 1
1963
if self._write_lock_count == 1:
1964
from bzrlib import transactions
1965
self._transaction = transactions.WriteTransaction()
1966
self._refresh_data()
1968
def lock_read(self):
1969
if self._write_lock_count:
1970
self._write_lock_count += 1
1972
self.control_files.lock_read()
1973
self._refresh_data()
1975
def leave_lock_in_place(self):
1976
# not supported - raise an error
1977
raise NotImplementedError(self.leave_lock_in_place)
1979
def dont_leave_lock_in_place(self):
1980
# not supported - raise an error
1981
raise NotImplementedError(self.dont_leave_lock_in_place)
1985
"""Compress the data within the repository.
1987
This will pack all the data to a single pack. In future it may
1988
recompress deltas or do other such expensive operations.
1990
self._pack_collection.pack()
1993
def reconcile(self, other=None, thorough=False):
1994
"""Reconcile this repository."""
1995
from bzrlib.reconcile import PackReconciler
1996
reconciler = PackReconciler(self, thorough=thorough)
1997
reconciler.reconcile()
2001
if self._write_lock_count == 1 and self._write_group is not None:
2002
self.abort_write_group()
2003
self._transaction = None
2004
self._write_lock_count = 0
2005
raise errors.BzrError(
2006
'Must end write group before releasing write lock on %s'
2008
if self._write_lock_count:
2009
self._write_lock_count -= 1
2010
if not self._write_lock_count:
2011
transaction = self._transaction
2012
self._transaction = None
2013
transaction.finish()
2015
self.control_files.unlock()
2018
class RepositoryFormatPack(MetaDirRepositoryFormat):
2019
"""Format logic for pack structured repositories.
2021
This repository format has:
2022
- a list of packs in pack-names
2023
- packs in packs/NAME.pack
2024
- indices in indices/NAME.{iix,six,tix,rix}
2025
- knit deltas in the packs, knit indices mapped to the indices.
2026
- thunk objects to support the knits programming API.
2027
- a format marker of its own
2028
- an optional 'shared-storage' flag
2029
- an optional 'no-working-trees' flag
2033
# Set this attribute in derived classes to control the repository class
2034
# created by open and initialize.
2035
repository_class = None
2036
# Set this attribute in derived classes to control the
2037
# _commit_builder_class that the repository objects will have passed to
2038
# their constructor.
2039
_commit_builder_class = None
2040
# Set this attribute in derived clases to control the _serializer that the
2041
# repository objects will have passed to their constructor.
2044
def _get_control_store(self, repo_transport, control_files):
2045
"""Return the control store for this repository."""
2046
return VersionedFileStore(
2049
file_mode=control_files._file_mode,
2050
versionedfile_class=knit.KnitVersionedFile,
2051
versionedfile_kwargs={'factory': knit.KnitPlainFactory()},
2054
def _get_revision_store(self, repo_transport, control_files):
2055
"""See RepositoryFormat._get_revision_store()."""
2056
versioned_file_store = VersionedFileStore(
2058
file_mode=control_files._file_mode,
2061
versionedfile_class=knit.KnitVersionedFile,
2062
versionedfile_kwargs={'delta': False,
2063
'factory': knit.KnitPlainFactory(),
2067
return KnitRevisionStore(versioned_file_store)
2069
def _get_text_store(self, transport, control_files):
2070
"""See RepositoryFormat._get_text_store()."""
2071
return self._get_versioned_file_store('knits',
2074
versionedfile_class=knit.KnitVersionedFile,
2075
versionedfile_kwargs={
2076
'create_parent_dir': True,
2077
'delay_create': True,
2078
'dir_mode': control_files._dir_mode,
2082
def initialize(self, a_bzrdir, shared=False):
2083
"""Create a pack based repository.
2085
:param a_bzrdir: bzrdir to contain the new repository; must already
2087
:param shared: If true the repository will be initialized as a shared
2090
mutter('creating repository in %s.', a_bzrdir.transport.base)
2091
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
2092
builder = GraphIndexBuilder()
2093
files = [('pack-names', builder.finish())]
2094
utf8_files = [('format', self.get_format_string())]
2096
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
2097
return self.open(a_bzrdir=a_bzrdir, _found=True)
2099
def open(self, a_bzrdir, _found=False, _override_transport=None):
2100
"""See RepositoryFormat.open().
2102
:param _override_transport: INTERNAL USE ONLY. Allows opening the
2103
repository at a slightly different url
2104
than normal. I.e. during 'upgrade'.
2107
format = RepositoryFormat.find_format(a_bzrdir)
2108
assert format.__class__ == self.__class__
2109
if _override_transport is not None:
2110
repo_transport = _override_transport
2112
repo_transport = a_bzrdir.get_repository_transport(None)
2113
control_files = lockable_files.LockableFiles(repo_transport,
2114
'lock', lockdir.LockDir)
2115
text_store = self._get_text_store(repo_transport, control_files)
2116
control_store = self._get_control_store(repo_transport, control_files)
2117
_revision_store = self._get_revision_store(repo_transport, control_files)
2118
return self.repository_class(_format=self,
2120
control_files=control_files,
2121
_revision_store=_revision_store,
2122
control_store=control_store,
2123
text_store=text_store,
2124
_commit_builder_class=self._commit_builder_class,
2125
_serializer=self._serializer)
2128
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2129
"""A no-subtrees parameterised Pack repository.
2131
This format was introduced in 0.92.
2134
repository_class = KnitPackRepository
2135
_commit_builder_class = PackCommitBuilder
2136
_serializer = xml5.serializer_v5
2138
def _get_matching_bzrdir(self):
2139
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2141
def _ignore_setting_bzrdir(self, format):
2144
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2146
def get_format_string(self):
2147
"""See RepositoryFormat.get_format_string()."""
2148
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2150
def get_format_description(self):
2151
"""See RepositoryFormat.get_format_description()."""
2152
return "Packs containing knits without subtree support"
2154
def check_conversion_target(self, target_format):
2158
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2159
"""A subtrees parameterised Pack repository.
2161
This repository format uses the xml7 serializer to get:
2162
- support for recording full info about the tree root
2163
- support for recording tree-references
2165
This format was introduced in 0.92.
2168
repository_class = KnitPackRepository
2169
_commit_builder_class = PackRootCommitBuilder
2170
rich_root_data = True
2171
supports_tree_reference = True
2172
_serializer = xml7.serializer_v7
2174
def _get_matching_bzrdir(self):
2175
return bzrdir.format_registry.make_bzrdir(
2176
'pack-0.92-subtree')
2178
def _ignore_setting_bzrdir(self, format):
2181
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2183
def check_conversion_target(self, target_format):
2184
if not target_format.rich_root_data:
2185
raise errors.BadConversionTarget(
2186
'Does not support rich root data.', target_format)
2187
if not getattr(target_format, 'supports_tree_reference', False):
2188
raise errors.BadConversionTarget(
2189
'Does not support nested trees', target_format)
2191
def get_format_string(self):
2192
"""See RepositoryFormat.get_format_string()."""
2193
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2195
def get_format_description(self):
2196
"""See RepositoryFormat.get_format_description()."""
2197
return "Packs containing knits with subtree support\n"
2200
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2201
"""A rich-root, no subtrees parameterised Pack repository.
2203
This repository format uses the xml6 serializer to get:
2204
- support for recording full info about the tree root
2206
This format was introduced in 1.0.
2209
repository_class = KnitPackRepository
2210
_commit_builder_class = PackRootCommitBuilder
2211
rich_root_data = True
2212
supports_tree_reference = False
2213
_serializer = xml6.serializer_v6
2215
def _get_matching_bzrdir(self):
2216
return bzrdir.format_registry.make_bzrdir(
2219
def _ignore_setting_bzrdir(self, format):
2222
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2224
def check_conversion_target(self, target_format):
2225
if not target_format.rich_root_data:
2226
raise errors.BadConversionTarget(
2227
'Does not support rich root data.', target_format)
2229
def get_format_string(self):
2230
"""See RepositoryFormat.get_format_string()."""
2231
return ("Bazaar pack repository format 1 with rich root"
2232
" (needs bzr 1.0)\n")
2234
def get_format_description(self):
2235
"""See RepositoryFormat.get_format_description()."""
2236
return "Packs containing knits with rich root support\n"
added
removed
bzrlib/repofmt/pack_repo.py
