1
# Copyright (C) 2005, 2006, 2007, 2008 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
from bzrlib.lazy_import import lazy_import
18
lazy_import(globals(), """
19
from itertools import izip
33
from bzrlib.index import (
37
GraphIndexPrefixAdapter,
40
from bzrlib.knit import (
46
from bzrlib import tsort
56
from bzrlib.decorators import needs_write_lock
57
from bzrlib.btree_index import (
61
from bzrlib.index import (
65
from bzrlib.repofmt.knitrepo import KnitRepository
66
from bzrlib.repository import (
68
MetaDirRepositoryFormat,
72
import bzrlib.revision as _mod_revision
73
from bzrlib.trace import (
79
class PackCommitBuilder(CommitBuilder):
80
"""A subclass of CommitBuilder to add texts with pack semantics.
82
Specifically this uses one knit object rather than one knit object per
83
added text, reducing memory and object pressure.
86
def __init__(self, repository, parents, config, timestamp=None,
87
timezone=None, committer=None, revprops=None,
89
CommitBuilder.__init__(self, repository, parents, config,
90
timestamp=timestamp, timezone=timezone, committer=committer,
91
revprops=revprops, revision_id=revision_id)
92
self._file_graph = graph.Graph(
93
repository._pack_collection.text_index.combined_index)
95
def _heads(self, file_id, revision_ids):
96
keys = [(file_id, revision_id) for revision_id in revision_ids]
97
return set([key[1] for key in self._file_graph.heads(keys)])
100
class PackRootCommitBuilder(RootCommitBuilder):
101
"""A subclass of RootCommitBuilder to add texts with pack semantics.
103
Specifically this uses one knit object rather than one knit object per
104
added text, reducing memory and object pressure.
107
def __init__(self, repository, parents, config, timestamp=None,
108
timezone=None, committer=None, revprops=None,
110
CommitBuilder.__init__(self, repository, parents, config,
111
timestamp=timestamp, timezone=timezone, committer=committer,
112
revprops=revprops, revision_id=revision_id)
113
self._file_graph = graph.Graph(
114
repository._pack_collection.text_index.combined_index)
116
def _heads(self, file_id, revision_ids):
117
keys = [(file_id, revision_id) for revision_id in revision_ids]
118
return set([key[1] for key in self._file_graph.heads(keys)])
122
"""An in memory proxy for a pack and its indices.
124
This is a base class that is not directly used, instead the classes
125
ExistingPack and NewPack are used.
128
def __init__(self, revision_index, inventory_index, text_index,
130
"""Create a pack instance.
132
:param revision_index: A GraphIndex for determining what revisions are
133
present in the Pack and accessing the locations of their texts.
134
:param inventory_index: A GraphIndex for determining what inventories are
135
present in the Pack and accessing the locations of their
137
:param text_index: A GraphIndex for determining what file texts
138
are present in the pack and accessing the locations of their
139
texts/deltas (via (fileid, revisionid) tuples).
140
:param signature_index: A GraphIndex for determining what signatures are
141
present in the Pack and accessing the locations of their texts.
143
self.revision_index = revision_index
144
self.inventory_index = inventory_index
145
self.text_index = text_index
146
self.signature_index = signature_index
148
def access_tuple(self):
149
"""Return a tuple (transport, name) for the pack content."""
150
return self.pack_transport, self.file_name()
153
"""Get the file name for the pack on disk."""
154
return self.name + '.pack'
156
def get_revision_count(self):
157
return self.revision_index.key_count()
159
def inventory_index_name(self, name):
160
"""The inv index is the name + .iix."""
161
return self.index_name('inventory', name)
163
def revision_index_name(self, name):
164
"""The revision index is the name + .rix."""
165
return self.index_name('revision', name)
167
def signature_index_name(self, name):
168
"""The signature index is the name + .six."""
169
return self.index_name('signature', name)
171
def text_index_name(self, name):
172
"""The text index is the name + .tix."""
173
return self.index_name('text', name)
176
class ExistingPack(Pack):
177
"""An in memory proxy for an existing .pack and its disk indices."""
179
def __init__(self, pack_transport, name, revision_index, inventory_index,
180
text_index, signature_index):
181
"""Create an ExistingPack object.
183
:param pack_transport: The transport where the pack file resides.
184
:param name: The name of the pack on disk in the pack_transport.
186
Pack.__init__(self, revision_index, inventory_index, text_index,
189
self.pack_transport = pack_transport
190
if None in (revision_index, inventory_index, text_index,
191
signature_index, name, pack_transport):
192
raise AssertionError()
194
def __eq__(self, other):
195
return self.__dict__ == other.__dict__
197
def __ne__(self, other):
198
return not self.__eq__(other)
201
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
202
id(self), self.pack_transport, self.name)
206
"""An in memory proxy for a pack which is being created."""
208
# A map of index 'type' to the file extension and position in the
210
index_definitions = {
211
'revision': ('.rix', 0),
212
'inventory': ('.iix', 1),
214
'signature': ('.six', 3),
217
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
218
"""Create a NewPack instance.
220
:param pack_collection: A PackCollection into which this is being inserted.
221
:param upload_suffix: An optional suffix to be given to any temporary
222
files created during the pack creation. e.g '.autopack'
223
:param file_mode: Unix permissions for newly created file.
225
# The relative locations of the packs are constrained, but all are
226
# passed in because the caller has them, so as to avoid object churn.
227
index_builder_class = pack_collection._index_builder_class
229
# Revisions: parents list, no text compression.
230
index_builder_class(reference_lists=1),
231
# Inventory: We want to map compression only, but currently the
232
# knit code hasn't been updated enough to understand that, so we
233
# have a regular 2-list index giving parents and compression
235
index_builder_class(reference_lists=2),
236
# Texts: compression and per file graph, for all fileids - so two
237
# reference lists and two elements in the key tuple.
238
index_builder_class(reference_lists=2, key_elements=2),
239
# Signatures: Just blobs to store, no compression, no parents
241
index_builder_class(reference_lists=0),
243
self._pack_collection = pack_collection
244
# When we make readonly indices, we need this.
245
self.index_class = pack_collection._index_class
246
# where should the new pack be opened
247
self.upload_transport = pack_collection._upload_transport
248
# where are indices written out to
249
self.index_transport = pack_collection._index_transport
250
# where is the pack renamed to when it is finished?
251
self.pack_transport = pack_collection._pack_transport
252
# What file mode to upload the pack and indices with.
253
self._file_mode = file_mode
254
# tracks the content written to the .pack file.
255
self._hash = osutils.md5()
256
# a four-tuple with the length in bytes of the indices, once the pack
257
# is finalised. (rev, inv, text, sigs)
258
self.index_sizes = None
259
# How much data to cache when writing packs. Note that this is not
260
# synchronised with reads, because it's not in the transport layer, so
261
# is not safe unless the client knows it won't be reading from the pack
263
self._cache_limit = 0
264
# the temporary pack file name.
265
self.random_name = osutils.rand_chars(20) + upload_suffix
266
# when was this pack started ?
267
self.start_time = time.time()
268
# open an output stream for the data added to the pack.
269
self.write_stream = self.upload_transport.open_write_stream(
270
self.random_name, mode=self._file_mode)
271
if 'pack' in debug.debug_flags:
272
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
273
time.ctime(), self.upload_transport.base, self.random_name,
274
time.time() - self.start_time)
275
# A list of byte sequences to be written to the new pack, and the
276
# aggregate size of them. Stored as a list rather than separate
277
# variables so that the _write_data closure below can update them.
278
self._buffer = [[], 0]
279
# create a callable for adding data
281
# robertc says- this is a closure rather than a method on the object
282
# so that the variables are locals, and faster than accessing object
284
def _write_data(bytes, flush=False, _buffer=self._buffer,
285
_write=self.write_stream.write, _update=self._hash.update):
286
_buffer[0].append(bytes)
287
_buffer[1] += len(bytes)
289
if _buffer[1] > self._cache_limit or flush:
290
bytes = ''.join(_buffer[0])
294
# expose this on self, for the occasion when clients want to add data.
295
self._write_data = _write_data
296
# a pack writer object to serialise pack records.
297
self._writer = pack.ContainerWriter(self._write_data)
299
# what state is the pack in? (open, finished, aborted)
303
"""Cancel creating this pack."""
304
self._state = 'aborted'
305
self.write_stream.close()
306
# Remove the temporary pack file.
307
self.upload_transport.delete(self.random_name)
308
# The indices have no state on disk.
310
def access_tuple(self):
311
"""Return a tuple (transport, name) for the pack content."""
312
if self._state == 'finished':
313
return Pack.access_tuple(self)
314
elif self._state == 'open':
315
return self.upload_transport, self.random_name
317
raise AssertionError(self._state)
319
def _check_references(self):
320
"""Make sure our external references are present.
322
Packs are allowed to have deltas whose base is not in the pack, but it
323
must be present somewhere in this collection. It is not allowed to
324
have deltas based on a fallback repository.
325
(See <https://bugs.launchpad.net/bzr/+bug/288751>)
328
for (index_name, external_refs, index) in [
330
self.text_index._external_references(),
331
self._pack_collection.text_index.combined_index),
333
self.inventory_index._external_references(),
334
self._pack_collection.inventory_index.combined_index),
336
missing = external_refs.difference(
337
k for (idx, k, v, r) in
338
index.iter_entries(external_refs))
340
missing_items[index_name] = sorted(list(missing))
342
from pprint import pformat
343
raise errors.BzrCheckError(
344
"Newly created pack file %r has delta references to "
345
"items not in its repository:\n%s"
346
% (self, pformat(missing_items)))
348
def data_inserted(self):
349
"""True if data has been added to this pack."""
350
return bool(self.get_revision_count() or
351
self.inventory_index.key_count() or
352
self.text_index.key_count() or
353
self.signature_index.key_count())
356
"""Finish the new pack.
359
- finalises the content
360
- assigns a name (the md5 of the content, currently)
361
- writes out the associated indices
362
- renames the pack into place.
363
- stores the index size tuple for the pack in the index_sizes
368
self._write_data('', flush=True)
369
self.name = self._hash.hexdigest()
370
self._check_references()
372
# XXX: It'd be better to write them all to temporary names, then
373
# rename them all into place, so that the window when only some are
374
# visible is smaller. On the other hand none will be seen until
375
# they're in the names list.
376
self.index_sizes = [None, None, None, None]
377
self._write_index('revision', self.revision_index, 'revision')
378
self._write_index('inventory', self.inventory_index, 'inventory')
379
self._write_index('text', self.text_index, 'file texts')
380
self._write_index('signature', self.signature_index,
381
'revision signatures')
382
self.write_stream.close()
383
# Note that this will clobber an existing pack with the same name,
384
# without checking for hash collisions. While this is undesirable this
385
# is something that can be rectified in a subsequent release. One way
386
# to rectify it may be to leave the pack at the original name, writing
387
# its pack-names entry as something like 'HASH: index-sizes
388
# temporary-name'. Allocate that and check for collisions, if it is
389
# collision free then rename it into place. If clients know this scheme
390
# they can handle missing-file errors by:
391
# - try for HASH.pack
392
# - try for temporary-name
393
# - refresh the pack-list to see if the pack is now absent
394
self.upload_transport.rename(self.random_name,
395
'../packs/' + self.name + '.pack')
396
self._state = 'finished'
397
if 'pack' in debug.debug_flags:
398
# XXX: size might be interesting?
399
mutter('%s: create_pack: pack renamed into place: %s%s->%s%s t+%6.3fs',
400
time.ctime(), self.upload_transport.base, self.random_name,
401
self.pack_transport, self.name,
402
time.time() - self.start_time)
405
"""Flush any current data."""
407
bytes = ''.join(self._buffer[0])
408
self.write_stream.write(bytes)
409
self._hash.update(bytes)
410
self._buffer[:] = [[], 0]
412
def index_name(self, index_type, name):
413
"""Get the disk name of an index type for pack name 'name'."""
414
return name + NewPack.index_definitions[index_type][0]
416
def index_offset(self, index_type):
417
"""Get the position in a index_size array for a given index type."""
418
return NewPack.index_definitions[index_type][1]
420
def _replace_index_with_readonly(self, index_type):
421
setattr(self, index_type + '_index',
422
self.index_class(self.index_transport,
423
self.index_name(index_type, self.name),
424
self.index_sizes[self.index_offset(index_type)]))
426
def set_write_cache_size(self, size):
427
self._cache_limit = size
429
def _write_index(self, index_type, index, label):
430
"""Write out an index.
432
:param index_type: The type of index to write - e.g. 'revision'.
433
:param index: The index object to serialise.
434
:param label: What label to give the index e.g. 'revision'.
436
index_name = self.index_name(index_type, self.name)
437
self.index_sizes[self.index_offset(index_type)] = \
438
self.index_transport.put_file(index_name, index.finish(),
439
mode=self._file_mode)
440
if 'pack' in debug.debug_flags:
441
# XXX: size might be interesting?
442
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
443
time.ctime(), label, self.upload_transport.base,
444
self.random_name, time.time() - self.start_time)
445
# Replace the writable index on this object with a readonly,
446
# presently unloaded index. We should alter
447
# the index layer to make its finish() error if add_node is
448
# subsequently used. RBC
449
self._replace_index_with_readonly(index_type)
452
class AggregateIndex(object):
453
"""An aggregated index for the RepositoryPackCollection.
455
AggregateIndex is reponsible for managing the PackAccess object,
456
Index-To-Pack mapping, and all indices list for a specific type of index
457
such as 'revision index'.
459
A CombinedIndex provides an index on a single key space built up
460
from several on-disk indices. The AggregateIndex builds on this
461
to provide a knit access layer, and allows having up to one writable
462
index within the collection.
464
# XXX: Probably 'can be written to' could/should be separated from 'acts
465
# like a knit index' -- mbp 20071024
467
def __init__(self, reload_func=None):
468
"""Create an AggregateIndex.
470
:param reload_func: A function to call if we find we are missing an
471
index. Should have the form reload_func() => True if the list of
472
active pack files has changed.
474
self._reload_func = reload_func
475
self.index_to_pack = {}
476
self.combined_index = CombinedGraphIndex([], reload_func=reload_func)
477
self.data_access = _DirectPackAccess(self.index_to_pack)
478
self.add_callback = None
480
def replace_indices(self, index_to_pack, indices):
481
"""Replace the current mappings with fresh ones.
483
This should probably not be used eventually, rather incremental add and
484
removal of indices. It has been added during refactoring of existing
487
:param index_to_pack: A mapping from index objects to
488
(transport, name) tuples for the pack file data.
489
:param indices: A list of indices.
491
# refresh the revision pack map dict without replacing the instance.
492
self.index_to_pack.clear()
493
self.index_to_pack.update(index_to_pack)
494
# XXX: API break - clearly a 'replace' method would be good?
495
self.combined_index._indices[:] = indices
496
# the current add nodes callback for the current writable index if
498
self.add_callback = None
500
def add_index(self, index, pack):
501
"""Add index to the aggregate, which is an index for Pack pack.
503
Future searches on the aggregate index will seach this new index
504
before all previously inserted indices.
506
:param index: An Index for the pack.
507
:param pack: A Pack instance.
509
# expose it to the index map
510
self.index_to_pack[index] = pack.access_tuple()
511
# put it at the front of the linear index list
512
self.combined_index.insert_index(0, index)
514
def add_writable_index(self, index, pack):
515
"""Add an index which is able to have data added to it.
517
There can be at most one writable index at any time. Any
518
modifications made to the knit are put into this index.
520
:param index: An index from the pack parameter.
521
:param pack: A Pack instance.
523
if self.add_callback is not None:
524
raise AssertionError(
525
"%s already has a writable index through %s" % \
526
(self, self.add_callback))
527
# allow writing: queue writes to a new index
528
self.add_index(index, pack)
529
# Updates the index to packs mapping as a side effect,
530
self.data_access.set_writer(pack._writer, index, pack.access_tuple())
531
self.add_callback = index.add_nodes
534
"""Reset all the aggregate data to nothing."""
535
self.data_access.set_writer(None, None, (None, None))
536
self.index_to_pack.clear()
537
del self.combined_index._indices[:]
538
self.add_callback = None
540
def remove_index(self, index, pack):
541
"""Remove index from the indices used to answer queries.
543
:param index: An index from the pack parameter.
544
:param pack: A Pack instance.
546
del self.index_to_pack[index]
547
self.combined_index._indices.remove(index)
548
if (self.add_callback is not None and
549
getattr(index, 'add_nodes', None) == self.add_callback):
550
self.add_callback = None
551
self.data_access.set_writer(None, None, (None, None))
554
class Packer(object):
555
"""Create a pack from packs."""
557
def __init__(self, pack_collection, packs, suffix, revision_ids=None):
560
:param pack_collection: A RepositoryPackCollection object where the
561
new pack is being written to.
562
:param packs: The packs to combine.
563
:param suffix: The suffix to use on the temporary files for the pack.
564
:param revision_ids: Revision ids to limit the pack to.
568
self.revision_ids = revision_ids
569
# The pack object we are creating.
571
self._pack_collection = pack_collection
572
# The index layer keys for the revisions being copied. None for 'all
574
self._revision_keys = None
575
# What text keys to copy. None for 'all texts'. This is set by
576
# _copy_inventory_texts
577
self._text_filter = None
580
def _extra_init(self):
581
"""A template hook to allow extending the constructor trivially."""
583
def pack(self, pb=None):
584
"""Create a new pack by reading data from other packs.
586
This does little more than a bulk copy of data. One key difference
587
is that data with the same item key across multiple packs is elided
588
from the output. The new pack is written into the current pack store
589
along with its indices, and the name added to the pack names. The
590
source packs are not altered and are not required to be in the current
593
:param pb: An optional progress bar to use. A nested bar is created if
595
:return: A Pack object, or None if nothing was copied.
597
# open a pack - using the same name as the last temporary file
598
# - which has already been flushed, so its safe.
599
# XXX: - duplicate code warning with start_write_group; fix before
600
# considering 'done'.
601
if self._pack_collection._new_pack is not None:
602
raise errors.BzrError('call to create_pack_from_packs while '
603
'another pack is being written.')
604
if self.revision_ids is not None:
605
if len(self.revision_ids) == 0:
606
# silly fetch request.
609
self.revision_ids = frozenset(self.revision_ids)
610
self.revision_keys = frozenset((revid,) for revid in
613
self.pb = ui.ui_factory.nested_progress_bar()
617
return self._create_pack_from_packs()
623
"""Open a pack for the pack we are creating."""
624
return NewPack(self._pack_collection, upload_suffix=self.suffix,
625
file_mode=self._pack_collection.repo.bzrdir._get_file_mode())
627
def _copy_revision_texts(self):
628
"""Copy revision data to the new pack."""
630
if self.revision_ids:
631
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
634
# select revision keys
635
revision_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
636
self.packs, 'revision_index')[0]
637
revision_nodes = self._pack_collection._index_contents(revision_index_map, revision_keys)
638
# copy revision keys and adjust values
639
self.pb.update("Copying revision texts", 1)
640
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
641
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
642
self.new_pack.revision_index, readv_group_iter, total_items))
643
if 'pack' in debug.debug_flags:
644
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
645
time.ctime(), self._pack_collection._upload_transport.base,
646
self.new_pack.random_name,
647
self.new_pack.revision_index.key_count(),
648
time.time() - self.new_pack.start_time)
649
self._revision_keys = revision_keys
651
def _copy_inventory_texts(self):
652
"""Copy the inventory texts to the new pack.
654
self._revision_keys is used to determine what inventories to copy.
656
Sets self._text_filter appropriately.
658
# select inventory keys
659
inv_keys = self._revision_keys # currently the same keyspace, and note that
660
# querying for keys here could introduce a bug where an inventory item
661
# is missed, so do not change it to query separately without cross
662
# checking like the text key check below.
663
inventory_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
664
self.packs, 'inventory_index')[0]
665
inv_nodes = self._pack_collection._index_contents(inventory_index_map, inv_keys)
666
# copy inventory keys and adjust values
667
# XXX: Should be a helper function to allow different inv representation
669
self.pb.update("Copying inventory texts", 2)
670
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
671
# Only grab the output lines if we will be processing them
672
output_lines = bool(self.revision_ids)
673
inv_lines = self._copy_nodes_graph(inventory_index_map,
674
self.new_pack._writer, self.new_pack.inventory_index,
675
readv_group_iter, total_items, output_lines=output_lines)
676
if self.revision_ids:
677
self._process_inventory_lines(inv_lines)
679
# eat the iterator to cause it to execute.
681
self._text_filter = None
682
if 'pack' in debug.debug_flags:
683
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
684
time.ctime(), self._pack_collection._upload_transport.base,
685
self.new_pack.random_name,
686
self.new_pack.inventory_index.key_count(),
687
time.time() - self.new_pack.start_time)
689
def _copy_text_texts(self):
691
text_index_map, text_nodes = self._get_text_nodes()
692
if self._text_filter is not None:
693
# We could return the keys copied as part of the return value from
694
# _copy_nodes_graph but this doesn't work all that well with the
695
# need to get line output too, so we check separately, and as we're
696
# going to buffer everything anyway, we check beforehand, which
697
# saves reading knit data over the wire when we know there are
699
text_nodes = set(text_nodes)
700
present_text_keys = set(_node[1] for _node in text_nodes)
701
missing_text_keys = set(self._text_filter) - present_text_keys
702
if missing_text_keys:
703
# TODO: raise a specific error that can handle many missing
705
a_missing_key = missing_text_keys.pop()
706
raise errors.RevisionNotPresent(a_missing_key[1],
708
# copy text keys and adjust values
709
self.pb.update("Copying content texts", 3)
710
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
711
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
712
self.new_pack.text_index, readv_group_iter, total_items))
713
self._log_copied_texts()
715
def _create_pack_from_packs(self):
716
self.pb.update("Opening pack", 0, 5)
717
self.new_pack = self.open_pack()
718
new_pack = self.new_pack
719
# buffer data - we won't be reading-back during the pack creation and
720
# this makes a significant difference on sftp pushes.
721
new_pack.set_write_cache_size(1024*1024)
722
if 'pack' in debug.debug_flags:
723
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
724
for a_pack in self.packs]
725
if self.revision_ids is not None:
726
rev_count = len(self.revision_ids)
729
mutter('%s: create_pack: creating pack from source packs: '
730
'%s%s %s revisions wanted %s t=0',
731
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
732
plain_pack_list, rev_count)
733
self._copy_revision_texts()
734
self._copy_inventory_texts()
735
self._copy_text_texts()
736
# select signature keys
737
signature_filter = self._revision_keys # same keyspace
738
signature_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
739
self.packs, 'signature_index')[0]
740
signature_nodes = self._pack_collection._index_contents(signature_index_map,
742
# copy signature keys and adjust values
743
self.pb.update("Copying signature texts", 4)
744
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
745
new_pack.signature_index)
746
if 'pack' in debug.debug_flags:
747
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
748
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
749
new_pack.signature_index.key_count(),
750
time.time() - new_pack.start_time)
751
new_pack._check_references()
752
if not self._use_pack(new_pack):
755
self.pb.update("Finishing pack", 5)
757
self._pack_collection.allocate(new_pack)
760
def _copy_nodes(self, nodes, index_map, writer, write_index):
761
"""Copy knit nodes between packs with no graph references."""
762
pb = ui.ui_factory.nested_progress_bar()
764
return self._do_copy_nodes(nodes, index_map, writer,
769
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb):
770
# for record verification
771
knit = KnitVersionedFiles(None, None)
772
# plan a readv on each source pack:
774
nodes = sorted(nodes)
775
# how to map this into knit.py - or knit.py into this?
776
# we don't want the typical knit logic, we want grouping by pack
777
# at this point - perhaps a helper library for the following code
778
# duplication points?
780
for index, key, value in nodes:
781
if index not in request_groups:
782
request_groups[index] = []
783
request_groups[index].append((key, value))
785
pb.update("Copied record", record_index, len(nodes))
786
for index, items in request_groups.iteritems():
787
pack_readv_requests = []
788
for key, value in items:
789
# ---- KnitGraphIndex.get_position
790
bits = value[1:].split(' ')
791
offset, length = int(bits[0]), int(bits[1])
792
pack_readv_requests.append((offset, length, (key, value[0])))
793
# linear scan up the pack
794
pack_readv_requests.sort()
796
transport, path = index_map[index]
797
reader = pack.make_readv_reader(transport, path,
798
[offset[0:2] for offset in pack_readv_requests])
799
for (names, read_func), (_1, _2, (key, eol_flag)) in \
800
izip(reader.iter_records(), pack_readv_requests):
801
raw_data = read_func(None)
802
# check the header only
803
df, _ = knit._parse_record_header(key, raw_data)
805
pos, size = writer.add_bytes_record(raw_data, names)
806
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
807
pb.update("Copied record", record_index)
810
def _copy_nodes_graph(self, index_map, writer, write_index,
811
readv_group_iter, total_items, output_lines=False):
812
"""Copy knit nodes between packs.
814
:param output_lines: Return lines present in the copied data as
815
an iterator of line,version_id.
817
pb = ui.ui_factory.nested_progress_bar()
819
for result in self._do_copy_nodes_graph(index_map, writer,
820
write_index, output_lines, pb, readv_group_iter, total_items):
823
# Python 2.4 does not permit try:finally: in a generator.
829
def _do_copy_nodes_graph(self, index_map, writer, write_index,
830
output_lines, pb, readv_group_iter, total_items):
831
# for record verification
832
knit = KnitVersionedFiles(None, None)
833
# for line extraction when requested (inventories only)
835
factory = KnitPlainFactory()
837
pb.update("Copied record", record_index, total_items)
838
for index, readv_vector, node_vector in readv_group_iter:
840
transport, path = index_map[index]
841
reader = pack.make_readv_reader(transport, path, readv_vector)
842
for (names, read_func), (key, eol_flag, references) in \
843
izip(reader.iter_records(), node_vector):
844
raw_data = read_func(None)
846
# read the entire thing
847
content, _ = knit._parse_record(key[-1], raw_data)
848
if len(references[-1]) == 0:
849
line_iterator = factory.get_fulltext_content(content)
851
line_iterator = factory.get_linedelta_content(content)
852
for line in line_iterator:
855
# check the header only
856
df, _ = knit._parse_record_header(key, raw_data)
858
pos, size = writer.add_bytes_record(raw_data, names)
859
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
860
pb.update("Copied record", record_index)
863
def _get_text_nodes(self):
864
text_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
865
self.packs, 'text_index')[0]
866
return text_index_map, self._pack_collection._index_contents(text_index_map,
869
def _least_readv_node_readv(self, nodes):
870
"""Generate request groups for nodes using the least readv's.
872
:param nodes: An iterable of graph index nodes.
873
:return: Total node count and an iterator of the data needed to perform
874
readvs to obtain the data for nodes. Each item yielded by the
875
iterator is a tuple with:
876
index, readv_vector, node_vector. readv_vector is a list ready to
877
hand to the transport readv method, and node_vector is a list of
878
(key, eol_flag, references) for the the node retrieved by the
879
matching readv_vector.
881
# group by pack so we do one readv per pack
882
nodes = sorted(nodes)
885
for index, key, value, references in nodes:
886
if index not in request_groups:
887
request_groups[index] = []
888
request_groups[index].append((key, value, references))
890
for index, items in request_groups.iteritems():
891
pack_readv_requests = []
892
for key, value, references in items:
893
# ---- KnitGraphIndex.get_position
894
bits = value[1:].split(' ')
895
offset, length = int(bits[0]), int(bits[1])
896
pack_readv_requests.append(
897
((offset, length), (key, value[0], references)))
898
# linear scan up the pack to maximum range combining.
899
pack_readv_requests.sort()
900
# split out the readv and the node data.
901
pack_readv = [readv for readv, node in pack_readv_requests]
902
node_vector = [node for readv, node in pack_readv_requests]
903
result.append((index, pack_readv, node_vector))
906
def _log_copied_texts(self):
907
if 'pack' in debug.debug_flags:
908
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
909
time.ctime(), self._pack_collection._upload_transport.base,
910
self.new_pack.random_name,
911
self.new_pack.text_index.key_count(),
912
time.time() - self.new_pack.start_time)
914
def _process_inventory_lines(self, inv_lines):
915
"""Use up the inv_lines generator and setup a text key filter."""
916
repo = self._pack_collection.repo
917
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
918
inv_lines, self.revision_keys)
920
for fileid, file_revids in fileid_revisions.iteritems():
921
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
922
self._text_filter = text_filter
924
def _revision_node_readv(self, revision_nodes):
925
"""Return the total revisions and the readv's to issue.
927
:param revision_nodes: The revision index contents for the packs being
928
incorporated into the new pack.
929
:return: As per _least_readv_node_readv.
931
return self._least_readv_node_readv(revision_nodes)
933
def _use_pack(self, new_pack):
934
"""Return True if new_pack should be used.
936
:param new_pack: The pack that has just been created.
937
:return: True if the pack should be used.
939
return new_pack.data_inserted()
942
class OptimisingPacker(Packer):
943
"""A packer which spends more time to create better disk layouts."""
945
def _revision_node_readv(self, revision_nodes):
946
"""Return the total revisions and the readv's to issue.
948
This sort places revisions in topological order with the ancestors
951
:param revision_nodes: The revision index contents for the packs being
952
incorporated into the new pack.
953
:return: As per _least_readv_node_readv.
955
# build an ancestors dict
958
for index, key, value, references in revision_nodes:
959
ancestors[key] = references[0]
960
by_key[key] = (index, value, references)
961
order = tsort.topo_sort(ancestors)
963
# Single IO is pathological, but it will work as a starting point.
965
for key in reversed(order):
966
index, value, references = by_key[key]
967
# ---- KnitGraphIndex.get_position
968
bits = value[1:].split(' ')
969
offset, length = int(bits[0]), int(bits[1])
971
(index, [(offset, length)], [(key, value[0], references)]))
972
# TODO: combine requests in the same index that are in ascending order.
973
return total, requests
976
"""Open a pack for the pack we are creating."""
977
new_pack = super(OptimisingPacker, self).open_pack()
978
# Turn on the optimization flags for all the index builders.
979
new_pack.revision_index.set_optimize(for_size=True)
980
new_pack.inventory_index.set_optimize(for_size=True)
981
new_pack.text_index.set_optimize(for_size=True)
982
new_pack.signature_index.set_optimize(for_size=True)
986
class ReconcilePacker(Packer):
987
"""A packer which regenerates indices etc as it copies.
989
This is used by ``bzr reconcile`` to cause parent text pointers to be
993
def _extra_init(self):
994
self._data_changed = False
996
def _process_inventory_lines(self, inv_lines):
997
"""Generate a text key reference map rather for reconciling with."""
998
repo = self._pack_collection.repo
999
refs = repo._find_text_key_references_from_xml_inventory_lines(
1001
self._text_refs = refs
1002
# during reconcile we:
1003
# - convert unreferenced texts to full texts
1004
# - correct texts which reference a text not copied to be full texts
1005
# - copy all others as-is but with corrected parents.
1006
# - so at this point we don't know enough to decide what becomes a full
1008
self._text_filter = None
1010
def _copy_text_texts(self):
1011
"""generate what texts we should have and then copy."""
1012
self.pb.update("Copying content texts", 3)
1013
# we have three major tasks here:
1014
# 1) generate the ideal index
1015
repo = self._pack_collection.repo
1016
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1017
_1, key, _2, refs in
1018
self.new_pack.revision_index.iter_all_entries()])
1019
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1020
# 2) generate a text_nodes list that contains all the deltas that can
1021
# be used as-is, with corrected parents.
1024
discarded_nodes = []
1025
NULL_REVISION = _mod_revision.NULL_REVISION
1026
text_index_map, text_nodes = self._get_text_nodes()
1027
for node in text_nodes:
1033
ideal_parents = tuple(ideal_index[node[1]])
1035
discarded_nodes.append(node)
1036
self._data_changed = True
1038
if ideal_parents == (NULL_REVISION,):
1040
if ideal_parents == node[3][0]:
1042
ok_nodes.append(node)
1043
elif ideal_parents[0:1] == node[3][0][0:1]:
1044
# the left most parent is the same, or there are no parents
1045
# today. Either way, we can preserve the representation as
1046
# long as we change the refs to be inserted.
1047
self._data_changed = True
1048
ok_nodes.append((node[0], node[1], node[2],
1049
(ideal_parents, node[3][1])))
1050
self._data_changed = True
1052
# Reinsert this text completely
1053
bad_texts.append((node[1], ideal_parents))
1054
self._data_changed = True
1055
# we're finished with some data.
1058
# 3) bulk copy the ok data
1059
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1060
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1061
self.new_pack.text_index, readv_group_iter, total_items))
1062
# 4) adhoc copy all the other texts.
1063
# We have to topologically insert all texts otherwise we can fail to
1064
# reconcile when parts of a single delta chain are preserved intact,
1065
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1066
# reinserted, and if d3 has incorrect parents it will also be
1067
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1068
# copied), so we will try to delta, but d2 is not currently able to be
1069
# extracted because it's basis d1 is not present. Topologically sorting
1070
# addresses this. The following generates a sort for all the texts that
1071
# are being inserted without having to reference the entire text key
1072
# space (we only topo sort the revisions, which is smaller).
1073
topo_order = tsort.topo_sort(ancestors)
1074
rev_order = dict(zip(topo_order, range(len(topo_order))))
1075
bad_texts.sort(key=lambda key:rev_order[key[0][1]])
1076
transaction = repo.get_transaction()
1077
file_id_index = GraphIndexPrefixAdapter(
1078
self.new_pack.text_index,
1080
add_nodes_callback=self.new_pack.text_index.add_nodes)
1081
data_access = _DirectPackAccess(
1082
{self.new_pack.text_index:self.new_pack.access_tuple()})
1083
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1084
self.new_pack.access_tuple())
1085
output_texts = KnitVersionedFiles(
1086
_KnitGraphIndex(self.new_pack.text_index,
1087
add_callback=self.new_pack.text_index.add_nodes,
1088
deltas=True, parents=True, is_locked=repo.is_locked),
1089
data_access=data_access, max_delta_chain=200)
1090
for key, parent_keys in bad_texts:
1091
# We refer to the new pack to delta data being output.
1092
# A possible improvement would be to catch errors on short reads
1093
# and only flush then.
1094
self.new_pack.flush()
1096
for parent_key in parent_keys:
1097
if parent_key[0] != key[0]:
1098
# Graph parents must match the fileid
1099
raise errors.BzrError('Mismatched key parent %r:%r' %
1101
parents.append(parent_key[1])
1102
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1103
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1104
output_texts.add_lines(key, parent_keys, text_lines,
1105
random_id=True, check_content=False)
1106
# 5) check that nothing inserted has a reference outside the keyspace.
1107
missing_text_keys = self.new_pack.text_index._external_references()
1108
if missing_text_keys:
1109
raise errors.BzrCheckError('Reference to missing compression parents %r'
1110
% (missing_text_keys,))
1111
self._log_copied_texts()
1113
def _use_pack(self, new_pack):
1114
"""Override _use_pack to check for reconcile having changed content."""
1115
# XXX: we might be better checking this at the copy time.
1116
original_inventory_keys = set()
1117
inv_index = self._pack_collection.inventory_index.combined_index
1118
for entry in inv_index.iter_all_entries():
1119
original_inventory_keys.add(entry[1])
1120
new_inventory_keys = set()
1121
for entry in new_pack.inventory_index.iter_all_entries():
1122
new_inventory_keys.add(entry[1])
1123
if new_inventory_keys != original_inventory_keys:
1124
self._data_changed = True
1125
return new_pack.data_inserted() and self._data_changed
1128
class RepositoryPackCollection(object):
1129
"""Management of packs within a repository.
1131
:ivar _names: map of {pack_name: (index_size,)}
1134
def __init__(self, repo, transport, index_transport, upload_transport,
1135
pack_transport, index_builder_class, index_class):
1136
"""Create a new RepositoryPackCollection.
1138
:param transport: Addresses the repository base directory
1139
(typically .bzr/repository/).
1140
:param index_transport: Addresses the directory containing indices.
1141
:param upload_transport: Addresses the directory into which packs are written
1142
while they're being created.
1143
:param pack_transport: Addresses the directory of existing complete packs.
1144
:param index_builder_class: The index builder class to use.
1145
:param index_class: The index class to use.
1148
self.transport = transport
1149
self._index_transport = index_transport
1150
self._upload_transport = upload_transport
1151
self._pack_transport = pack_transport
1152
self._index_builder_class = index_builder_class
1153
self._index_class = index_class
1154
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
1157
self._packs_by_name = {}
1158
# the previous pack-names content
1159
self._packs_at_load = None
1160
# when a pack is being created by this object, the state of that pack.
1161
self._new_pack = None
1162
# aggregated revision index data
1163
self.revision_index = AggregateIndex(self.reload_pack_names)
1164
self.inventory_index = AggregateIndex(self.reload_pack_names)
1165
self.text_index = AggregateIndex(self.reload_pack_names)
1166
self.signature_index = AggregateIndex(self.reload_pack_names)
1168
def add_pack_to_memory(self, pack):
1169
"""Make a Pack object available to the repository to satisfy queries.
1171
:param pack: A Pack object.
1173
if pack.name in self._packs_by_name:
1174
raise AssertionError()
1175
self.packs.append(pack)
1176
self._packs_by_name[pack.name] = pack
1177
self.revision_index.add_index(pack.revision_index, pack)
1178
self.inventory_index.add_index(pack.inventory_index, pack)
1179
self.text_index.add_index(pack.text_index, pack)
1180
self.signature_index.add_index(pack.signature_index, pack)
1182
def all_packs(self):
1183
"""Return a list of all the Pack objects this repository has.
1185
Note that an in-progress pack being created is not returned.
1187
:return: A list of Pack objects for all the packs in the repository.
1190
for name in self.names():
1191
result.append(self.get_pack_by_name(name))
1195
"""Pack the pack collection incrementally.
1197
This will not attempt global reorganisation or recompression,
1198
rather it will just ensure that the total number of packs does
1199
not grow without bound. It uses the _max_pack_count method to
1200
determine if autopacking is needed, and the pack_distribution
1201
method to determine the number of revisions in each pack.
1203
If autopacking takes place then the packs name collection will have
1204
been flushed to disk - packing requires updating the name collection
1205
in synchronisation with certain steps. Otherwise the names collection
1208
:return: True if packing took place.
1210
# XXX: Should not be needed when the management of indices is sane.
1211
total_revisions = self.revision_index.combined_index.key_count()
1212
total_packs = len(self._names)
1213
if self._max_pack_count(total_revisions) >= total_packs:
1215
# XXX: the following may want to be a class, to pack with a given
1217
mutter('Auto-packing repository %s, which has %d pack files, '
1218
'containing %d revisions into %d packs.', self, total_packs,
1219
total_revisions, self._max_pack_count(total_revisions))
1220
# determine which packs need changing
1221
pack_distribution = self.pack_distribution(total_revisions)
1223
for pack in self.all_packs():
1224
revision_count = pack.get_revision_count()
1225
if revision_count == 0:
1226
# revision less packs are not generated by normal operation,
1227
# only by operations like sign-my-commits, and thus will not
1228
# tend to grow rapdily or without bound like commit containing
1229
# packs do - leave them alone as packing them really should
1230
# group their data with the relevant commit, and that may
1231
# involve rewriting ancient history - which autopack tries to
1232
# avoid. Alternatively we could not group the data but treat
1233
# each of these as having a single revision, and thus add
1234
# one revision for each to the total revision count, to get
1235
# a matching distribution.
1237
existing_packs.append((revision_count, pack))
1238
pack_operations = self.plan_autopack_combinations(
1239
existing_packs, pack_distribution)
1240
self._execute_pack_operations(pack_operations)
1243
def _execute_pack_operations(self, pack_operations, _packer_class=Packer):
1244
"""Execute a series of pack operations.
1246
:param pack_operations: A list of [revision_count, packs_to_combine].
1247
:param _packer_class: The class of packer to use (default: Packer).
1250
for revision_count, packs in pack_operations:
1251
# we may have no-ops from the setup logic
1254
_packer_class(self, packs, '.autopack').pack()
1256
self._remove_pack_from_memory(pack)
1257
# record the newly available packs and stop advertising the old
1259
self._save_pack_names(clear_obsolete_packs=True)
1260
# Move the old packs out of the way now they are no longer referenced.
1261
for revision_count, packs in pack_operations:
1262
self._obsolete_packs(packs)
1264
def lock_names(self):
1265
"""Acquire the mutex around the pack-names index.
1267
This cannot be used in the middle of a read-only transaction on the
1270
self.repo.control_files.lock_write()
1273
"""Pack the pack collection totally."""
1274
self.ensure_loaded()
1275
total_packs = len(self._names)
1277
# This is arguably wrong because we might not be optimal, but for
1278
# now lets leave it in. (e.g. reconcile -> one pack. But not
1281
total_revisions = self.revision_index.combined_index.key_count()
1282
# XXX: the following may want to be a class, to pack with a given
1284
mutter('Packing repository %s, which has %d pack files, '
1285
'containing %d revisions into 1 packs.', self, total_packs,
1287
# determine which packs need changing
1288
pack_distribution = [1]
1289
pack_operations = [[0, []]]
1290
for pack in self.all_packs():
1291
pack_operations[-1][0] += pack.get_revision_count()
1292
pack_operations[-1][1].append(pack)
1293
self._execute_pack_operations(pack_operations, OptimisingPacker)
1295
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1296
"""Plan a pack operation.
1298
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1300
:param pack_distribution: A list with the number of revisions desired
1303
if len(existing_packs) <= len(pack_distribution):
1305
existing_packs.sort(reverse=True)
1306
pack_operations = [[0, []]]
1307
# plan out what packs to keep, and what to reorganise
1308
while len(existing_packs):
1309
# take the largest pack, and if its less than the head of the
1310
# distribution chart we will include its contents in the new pack
1311
# for that position. If its larger, we remove its size from the
1312
# distribution chart
1313
next_pack_rev_count, next_pack = existing_packs.pop(0)
1314
if next_pack_rev_count >= pack_distribution[0]:
1315
# this is already packed 'better' than this, so we can
1316
# not waste time packing it.
1317
while next_pack_rev_count > 0:
1318
next_pack_rev_count -= pack_distribution[0]
1319
if next_pack_rev_count >= 0:
1321
del pack_distribution[0]
1323
# didn't use that entire bucket up
1324
pack_distribution[0] = -next_pack_rev_count
1326
# add the revisions we're going to add to the next output pack
1327
pack_operations[-1][0] += next_pack_rev_count
1328
# allocate this pack to the next pack sub operation
1329
pack_operations[-1][1].append(next_pack)
1330
if pack_operations[-1][0] >= pack_distribution[0]:
1331
# this pack is used up, shift left.
1332
del pack_distribution[0]
1333
pack_operations.append([0, []])
1334
# Now that we know which pack files we want to move, shove them all
1335
# into a single pack file.
1337
final_pack_list = []
1338
for num_revs, pack_files in pack_operations:
1339
final_rev_count += num_revs
1340
final_pack_list.extend(pack_files)
1341
if len(final_pack_list) == 1:
1342
raise AssertionError('We somehow generated an autopack with a'
1343
' single pack file being moved.')
1345
return [[final_rev_count, final_pack_list]]
1347
def ensure_loaded(self):
1348
# NB: if you see an assertion error here, its probably access against
1349
# an unlocked repo. Naughty.
1350
if not self.repo.is_locked():
1351
raise errors.ObjectNotLocked(self.repo)
1352
if self._names is None:
1354
self._packs_at_load = set()
1355
for index, key, value in self._iter_disk_pack_index():
1357
self._names[name] = self._parse_index_sizes(value)
1358
self._packs_at_load.add((key, value))
1359
# populate all the metadata.
1362
def _parse_index_sizes(self, value):
1363
"""Parse a string of index sizes."""
1364
return tuple([int(digits) for digits in value.split(' ')])
1366
def get_pack_by_name(self, name):
1367
"""Get a Pack object by name.
1369
:param name: The name of the pack - e.g. '123456'
1370
:return: A Pack object.
1373
return self._packs_by_name[name]
1375
rev_index = self._make_index(name, '.rix')
1376
inv_index = self._make_index(name, '.iix')
1377
txt_index = self._make_index(name, '.tix')
1378
sig_index = self._make_index(name, '.six')
1379
result = ExistingPack(self._pack_transport, name, rev_index,
1380
inv_index, txt_index, sig_index)
1381
self.add_pack_to_memory(result)
1384
def allocate(self, a_new_pack):
1385
"""Allocate name in the list of packs.
1387
:param a_new_pack: A NewPack instance to be added to the collection of
1388
packs for this repository.
1390
self.ensure_loaded()
1391
if a_new_pack.name in self._names:
1392
raise errors.BzrError(
1393
'Pack %r already exists in %s' % (a_new_pack.name, self))
1394
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1395
self.add_pack_to_memory(a_new_pack)
1397
def _iter_disk_pack_index(self):
1398
"""Iterate over the contents of the pack-names index.
1400
This is used when loading the list from disk, and before writing to
1401
detect updates from others during our write operation.
1402
:return: An iterator of the index contents.
1404
return self._index_class(self.transport, 'pack-names', None
1405
).iter_all_entries()
1407
def _make_index(self, name, suffix):
1408
size_offset = self._suffix_offsets[suffix]
1409
index_name = name + suffix
1410
index_size = self._names[name][size_offset]
1411
return self._index_class(
1412
self._index_transport, index_name, index_size)
1414
def _max_pack_count(self, total_revisions):
1415
"""Return the maximum number of packs to use for total revisions.
1417
:param total_revisions: The total number of revisions in the
1420
if not total_revisions:
1422
digits = str(total_revisions)
1424
for digit in digits:
1425
result += int(digit)
1429
"""Provide an order to the underlying names."""
1430
return sorted(self._names.keys())
1432
def _obsolete_packs(self, packs):
1433
"""Move a number of packs which have been obsoleted out of the way.
1435
Each pack and its associated indices are moved out of the way.
1437
Note: for correctness this function should only be called after a new
1438
pack names index has been written without these pack names, and with
1439
the names of packs that contain the data previously available via these
1442
:param packs: The packs to obsolete.
1443
:param return: None.
1446
pack.pack_transport.rename(pack.file_name(),
1447
'../obsolete_packs/' + pack.file_name())
1448
# TODO: Probably needs to know all possible indices for this pack
1449
# - or maybe list the directory and move all indices matching this
1450
# name whether we recognize it or not?
1451
for suffix in ('.iix', '.six', '.tix', '.rix'):
1452
self._index_transport.rename(pack.name + suffix,
1453
'../obsolete_packs/' + pack.name + suffix)
1455
def pack_distribution(self, total_revisions):
1456
"""Generate a list of the number of revisions to put in each pack.
1458
:param total_revisions: The total number of revisions in the
1461
if total_revisions == 0:
1463
digits = reversed(str(total_revisions))
1465
for exponent, count in enumerate(digits):
1466
size = 10 ** exponent
1467
for pos in range(int(count)):
1469
return list(reversed(result))
1471
def _pack_tuple(self, name):
1472
"""Return a tuple with the transport and file name for a pack name."""
1473
return self._pack_transport, name + '.pack'
1475
def _remove_pack_from_memory(self, pack):
1476
"""Remove pack from the packs accessed by this repository.
1478
Only affects memory state, until self._save_pack_names() is invoked.
1480
self._names.pop(pack.name)
1481
self._packs_by_name.pop(pack.name)
1482
self._remove_pack_indices(pack)
1483
self.packs.remove(pack)
1485
def _remove_pack_indices(self, pack):
1486
"""Remove the indices for pack from the aggregated indices."""
1487
self.revision_index.remove_index(pack.revision_index, pack)
1488
self.inventory_index.remove_index(pack.inventory_index, pack)
1489
self.text_index.remove_index(pack.text_index, pack)
1490
self.signature_index.remove_index(pack.signature_index, pack)
1493
"""Clear all cached data."""
1494
# cached revision data
1495
self.repo._revision_knit = None
1496
self.revision_index.clear()
1497
# cached signature data
1498
self.repo._signature_knit = None
1499
self.signature_index.clear()
1500
# cached file text data
1501
self.text_index.clear()
1502
self.repo._text_knit = None
1503
# cached inventory data
1504
self.inventory_index.clear()
1505
# remove the open pack
1506
self._new_pack = None
1507
# information about packs.
1510
self._packs_by_name = {}
1511
self._packs_at_load = None
1513
def _make_index_map(self, index_suffix):
1514
"""Return information on existing indices.
1516
:param suffix: Index suffix added to pack name.
1518
:returns: (pack_map, indices) where indices is a list of GraphIndex
1519
objects, and pack_map is a mapping from those objects to the
1520
pack tuple they describe.
1522
# TODO: stop using this; it creates new indices unnecessarily.
1523
self.ensure_loaded()
1524
suffix_map = {'.rix': 'revision_index',
1525
'.six': 'signature_index',
1526
'.iix': 'inventory_index',
1527
'.tix': 'text_index',
1529
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1530
suffix_map[index_suffix])
1532
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1533
"""Convert a list of packs to an index pack map and index list.
1535
:param packs: The packs list to process.
1536
:param index_attribute: The attribute that the desired index is found
1538
:return: A tuple (map, list) where map contains the dict from
1539
index:pack_tuple, and lsit contains the indices in the same order
1545
index = getattr(pack, index_attribute)
1546
indices.append(index)
1547
pack_map[index] = (pack.pack_transport, pack.file_name())
1548
return pack_map, indices
1550
def _index_contents(self, pack_map, key_filter=None):
1551
"""Get an iterable of the index contents from a pack_map.
1553
:param pack_map: A map from indices to pack details.
1554
:param key_filter: An optional filter to limit the
1557
indices = [index for index in pack_map.iterkeys()]
1558
all_index = CombinedGraphIndex(indices)
1559
if key_filter is None:
1560
return all_index.iter_all_entries()
1562
return all_index.iter_entries(key_filter)
1564
def _unlock_names(self):
1565
"""Release the mutex around the pack-names index."""
1566
self.repo.control_files.unlock()
1568
def _diff_pack_names(self):
1569
"""Read the pack names from disk, and compare it to the one in memory.
1571
:return: (disk_nodes, deleted_nodes, new_nodes)
1572
disk_nodes The final set of nodes that should be referenced
1573
deleted_nodes Nodes which have been removed from when we started
1574
new_nodes Nodes that are newly introduced
1576
# load the disk nodes across
1578
for index, key, value in self._iter_disk_pack_index():
1579
disk_nodes.add((key, value))
1581
# do a two-way diff against our original content
1582
current_nodes = set()
1583
for name, sizes in self._names.iteritems():
1585
((name, ), ' '.join(str(size) for size in sizes)))
1587
# Packs no longer present in the repository, which were present when we
1588
# locked the repository
1589
deleted_nodes = self._packs_at_load - current_nodes
1590
# Packs which this process is adding
1591
new_nodes = current_nodes - self._packs_at_load
1593
# Update the disk_nodes set to include the ones we are adding, and
1594
# remove the ones which were removed by someone else
1595
disk_nodes.difference_update(deleted_nodes)
1596
disk_nodes.update(new_nodes)
1598
return disk_nodes, deleted_nodes, new_nodes
1600
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1601
"""Given the correct set of pack files, update our saved info.
1603
:return: (removed, added, modified)
1604
removed pack names removed from self._names
1605
added pack names added to self._names
1606
modified pack names that had changed value
1611
## self._packs_at_load = disk_nodes
1612
new_names = dict(disk_nodes)
1613
# drop no longer present nodes
1614
for pack in self.all_packs():
1615
if (pack.name,) not in new_names:
1616
removed.append(pack.name)
1617
self._remove_pack_from_memory(pack)
1618
# add new nodes/refresh existing ones
1619
for key, value in disk_nodes:
1621
sizes = self._parse_index_sizes(value)
1622
if name in self._names:
1624
if sizes != self._names[name]:
1625
# the pack for name has had its indices replaced - rare but
1626
# important to handle. XXX: probably can never happen today
1627
# because the three-way merge code above does not handle it
1628
# - you may end up adding the same key twice to the new
1629
# disk index because the set values are the same, unless
1630
# the only index shows up as deleted by the set difference
1631
# - which it may. Until there is a specific test for this,
1632
# assume its broken. RBC 20071017.
1633
self._remove_pack_from_memory(self.get_pack_by_name(name))
1634
self._names[name] = sizes
1635
self.get_pack_by_name(name)
1636
modified.append(name)
1639
self._names[name] = sizes
1640
self.get_pack_by_name(name)
1642
return removed, added, modified
1644
def _save_pack_names(self, clear_obsolete_packs=False):
1645
"""Save the list of packs.
1647
This will take out the mutex around the pack names list for the
1648
duration of the method call. If concurrent updates have been made, a
1649
three-way merge between the current list and the current in memory list
1652
:param clear_obsolete_packs: If True, clear out the contents of the
1653
obsolete_packs directory.
1657
builder = self._index_builder_class()
1658
disk_nodes, deleted_nodes, new_nodes = self._diff_pack_names()
1659
# TODO: handle same-name, index-size-changes here -
1660
# e.g. use the value from disk, not ours, *unless* we're the one
1662
for key, value in disk_nodes:
1663
builder.add_node(key, value)
1664
self.transport.put_file('pack-names', builder.finish(),
1665
mode=self.repo.bzrdir._get_file_mode())
1666
# move the baseline forward
1667
self._packs_at_load = disk_nodes
1668
if clear_obsolete_packs:
1669
self._clear_obsolete_packs()
1671
self._unlock_names()
1672
# synchronise the memory packs list with what we just wrote:
1673
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1675
def reload_pack_names(self):
1676
"""Sync our pack listing with what is present in the repository.
1678
This should be called when we find out that something we thought was
1679
present is now missing. This happens when another process re-packs the
1682
# This is functionally similar to _save_pack_names, but we don't write
1683
# out the new value.
1684
disk_nodes, _, _ = self._diff_pack_names()
1685
self._packs_at_load = disk_nodes
1687
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1688
if removed or added or modified:
1692
def _clear_obsolete_packs(self):
1693
"""Delete everything from the obsolete-packs directory.
1695
obsolete_pack_transport = self.transport.clone('obsolete_packs')
1696
for filename in obsolete_pack_transport.list_dir('.'):
1698
obsolete_pack_transport.delete(filename)
1699
except (errors.PathError, errors.TransportError), e:
1700
warning("couldn't delete obsolete pack, skipping it:\n%s" % (e,))
1702
def _start_write_group(self):
1703
# Do not permit preparation for writing if we're not in a 'write lock'.
1704
if not self.repo.is_write_locked():
1705
raise errors.NotWriteLocked(self)
1706
self._new_pack = NewPack(self, upload_suffix='.pack',
1707
file_mode=self.repo.bzrdir._get_file_mode())
1708
# allow writing: queue writes to a new index
1709
self.revision_index.add_writable_index(self._new_pack.revision_index,
1711
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
1713
self.text_index.add_writable_index(self._new_pack.text_index,
1715
self.signature_index.add_writable_index(self._new_pack.signature_index,
1718
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
1719
self.repo.revisions._index._add_callback = self.revision_index.add_callback
1720
self.repo.signatures._index._add_callback = self.signature_index.add_callback
1721
self.repo.texts._index._add_callback = self.text_index.add_callback
1723
def _abort_write_group(self):
1724
# FIXME: just drop the transient index.
1725
# forget what names there are
1726
if self._new_pack is not None:
1728
self._new_pack.abort()
1730
# XXX: If we aborted while in the middle of finishing the write
1731
# group, _remove_pack_indices can fail because the indexes are
1732
# already gone. If they're not there we shouldn't fail in this
1733
# case. -- mbp 20081113
1734
self._remove_pack_indices(self._new_pack)
1735
self._new_pack = None
1736
self.repo._text_knit = None
1738
def _commit_write_group(self):
1739
self._remove_pack_indices(self._new_pack)
1740
if self._new_pack.data_inserted():
1741
# get all the data to disk and read to use
1742
self._new_pack.finish()
1743
self.allocate(self._new_pack)
1744
self._new_pack = None
1745
if not self.autopack():
1746
# when autopack takes no steps, the names list is still
1748
self._save_pack_names()
1750
self._new_pack.abort()
1751
self._new_pack = None
1752
self.repo._text_knit = None
1755
class KnitPackRepository(KnitRepository):
1756
"""Repository with knit objects stored inside pack containers.
1758
The layering for a KnitPackRepository is:
1760
Graph | HPSS | Repository public layer |
1761
===================================================
1762
Tuple based apis below, string based, and key based apis above
1763
---------------------------------------------------
1765
Provides .texts, .revisions etc
1766
This adapts the N-tuple keys to physical knit records which only have a
1767
single string identifier (for historical reasons), which in older formats
1768
was always the revision_id, and in the mapped code for packs is always
1769
the last element of key tuples.
1770
---------------------------------------------------
1772
A separate GraphIndex is used for each of the
1773
texts/inventories/revisions/signatures contained within each individual
1774
pack file. The GraphIndex layer works in N-tuples and is unaware of any
1776
===================================================
1780
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
1782
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
1783
_commit_builder_class, _serializer)
1784
index_transport = self._transport.clone('indices')
1785
self._pack_collection = RepositoryPackCollection(self, self._transport,
1787
self._transport.clone('upload'),
1788
self._transport.clone('packs'),
1789
_format.index_builder_class,
1790
_format.index_class)
1791
self.inventories = KnitVersionedFiles(
1792
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
1793
add_callback=self._pack_collection.inventory_index.add_callback,
1794
deltas=True, parents=True, is_locked=self.is_locked),
1795
data_access=self._pack_collection.inventory_index.data_access,
1796
max_delta_chain=200)
1797
self.revisions = KnitVersionedFiles(
1798
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
1799
add_callback=self._pack_collection.revision_index.add_callback,
1800
deltas=False, parents=True, is_locked=self.is_locked),
1801
data_access=self._pack_collection.revision_index.data_access,
1803
self.signatures = KnitVersionedFiles(
1804
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
1805
add_callback=self._pack_collection.signature_index.add_callback,
1806
deltas=False, parents=False, is_locked=self.is_locked),
1807
data_access=self._pack_collection.signature_index.data_access,
1809
self.texts = KnitVersionedFiles(
1810
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
1811
add_callback=self._pack_collection.text_index.add_callback,
1812
deltas=True, parents=True, is_locked=self.is_locked),
1813
data_access=self._pack_collection.text_index.data_access,
1814
max_delta_chain=200)
1815
# True when the repository object is 'write locked' (as opposed to the
1816
# physical lock only taken out around changes to the pack-names list.)
1817
# Another way to represent this would be a decorator around the control
1818
# files object that presents logical locks as physical ones - if this
1819
# gets ugly consider that alternative design. RBC 20071011
1820
self._write_lock_count = 0
1821
self._transaction = None
1823
self._reconcile_does_inventory_gc = True
1824
self._reconcile_fixes_text_parents = True
1825
self._reconcile_backsup_inventory = False
1826
self._fetch_order = 'unordered'
1828
def _warn_if_deprecated(self):
1829
# This class isn't deprecated, but one sub-format is
1830
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
1831
from bzrlib import repository
1832
if repository._deprecation_warning_done:
1834
repository._deprecation_warning_done = True
1835
warning("Format %s for %s is deprecated - please use"
1836
" 'bzr upgrade --1.6.1-rich-root'"
1837
% (self._format, self.bzrdir.transport.base))
1839
def _abort_write_group(self):
1840
self._pack_collection._abort_write_group()
1842
def _find_inconsistent_revision_parents(self):
1843
"""Find revisions with incorrectly cached parents.
1845
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1846
parents-in-revision).
1848
if not self.is_locked():
1849
raise errors.ObjectNotLocked(self)
1850
pb = ui.ui_factory.nested_progress_bar()
1853
revision_nodes = self._pack_collection.revision_index \
1854
.combined_index.iter_all_entries()
1855
index_positions = []
1856
# Get the cached index values for all revisions, and also the location
1857
# in each index of the revision text so we can perform linear IO.
1858
for index, key, value, refs in revision_nodes:
1859
pos, length = value[1:].split(' ')
1860
index_positions.append((index, int(pos), key[0],
1861
tuple(parent[0] for parent in refs[0])))
1862
pb.update("Reading revision index.", 0, 0)
1863
index_positions.sort()
1864
batch_count = len(index_positions) / 1000 + 1
1865
pb.update("Checking cached revision graph.", 0, batch_count)
1866
for offset in xrange(batch_count):
1867
pb.update("Checking cached revision graph.", offset)
1868
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1871
rev_ids = [item[2] for item in to_query]
1872
revs = self.get_revisions(rev_ids)
1873
for revision, item in zip(revs, to_query):
1874
index_parents = item[3]
1875
rev_parents = tuple(revision.parent_ids)
1876
if index_parents != rev_parents:
1877
result.append((revision.revision_id, index_parents, rev_parents))
1882
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
1883
def get_parents(self, revision_ids):
1884
"""See graph._StackedParentsProvider.get_parents."""
1885
parent_map = self.get_parent_map(revision_ids)
1886
return [parent_map.get(r, None) for r in revision_ids]
1888
def _make_parents_provider(self):
1889
return graph.CachingParentsProvider(self)
1891
def _refresh_data(self):
1892
if self._write_lock_count == 1 or (
1893
self.control_files._lock_count == 1 and
1894
self.control_files._lock_mode == 'r'):
1895
# forget what names there are
1896
self._pack_collection.reset()
1897
# XXX: Better to do an in-memory merge when acquiring a new lock -
1898
# factor out code from _save_pack_names.
1899
self._pack_collection.ensure_loaded()
1901
def _start_write_group(self):
1902
self._pack_collection._start_write_group()
1904
def _commit_write_group(self):
1905
return self._pack_collection._commit_write_group()
1907
def get_transaction(self):
1908
if self._write_lock_count:
1909
return self._transaction
1911
return self.control_files.get_transaction()
1913
def is_locked(self):
1914
return self._write_lock_count or self.control_files.is_locked()
1916
def is_write_locked(self):
1917
return self._write_lock_count
1919
def lock_write(self, token=None):
1920
if not self._write_lock_count and self.is_locked():
1921
raise errors.ReadOnlyError(self)
1922
self._write_lock_count += 1
1923
if self._write_lock_count == 1:
1924
self._transaction = transactions.WriteTransaction()
1925
for repo in self._fallback_repositories:
1926
# Writes don't affect fallback repos
1928
self._refresh_data()
1930
def lock_read(self):
1931
if self._write_lock_count:
1932
self._write_lock_count += 1
1934
self.control_files.lock_read()
1935
for repo in self._fallback_repositories:
1936
# Writes don't affect fallback repos
1938
self._refresh_data()
1940
def leave_lock_in_place(self):
1941
# not supported - raise an error
1942
raise NotImplementedError(self.leave_lock_in_place)
1944
def dont_leave_lock_in_place(self):
1945
# not supported - raise an error
1946
raise NotImplementedError(self.dont_leave_lock_in_place)
1950
"""Compress the data within the repository.
1952
This will pack all the data to a single pack. In future it may
1953
recompress deltas or do other such expensive operations.
1955
self._pack_collection.pack()
1958
def reconcile(self, other=None, thorough=False):
1959
"""Reconcile this repository."""
1960
from bzrlib.reconcile import PackReconciler
1961
reconciler = PackReconciler(self, thorough=thorough)
1962
reconciler.reconcile()
1966
if self._write_lock_count == 1 and self._write_group is not None:
1967
self.abort_write_group()
1968
self._transaction = None
1969
self._write_lock_count = 0
1970
raise errors.BzrError(
1971
'Must end write group before releasing write lock on %s'
1973
if self._write_lock_count:
1974
self._write_lock_count -= 1
1975
if not self._write_lock_count:
1976
transaction = self._transaction
1977
self._transaction = None
1978
transaction.finish()
1979
for repo in self._fallback_repositories:
1982
self.control_files.unlock()
1983
for repo in self._fallback_repositories:
1987
class RepositoryFormatPack(MetaDirRepositoryFormat):
1988
"""Format logic for pack structured repositories.
1990
This repository format has:
1991
- a list of packs in pack-names
1992
- packs in packs/NAME.pack
1993
- indices in indices/NAME.{iix,six,tix,rix}
1994
- knit deltas in the packs, knit indices mapped to the indices.
1995
- thunk objects to support the knits programming API.
1996
- a format marker of its own
1997
- an optional 'shared-storage' flag
1998
- an optional 'no-working-trees' flag
2002
# Set this attribute in derived classes to control the repository class
2003
# created by open and initialize.
2004
repository_class = None
2005
# Set this attribute in derived classes to control the
2006
# _commit_builder_class that the repository objects will have passed to
2007
# their constructor.
2008
_commit_builder_class = None
2009
# Set this attribute in derived clases to control the _serializer that the
2010
# repository objects will have passed to their constructor.
2012
# External references are not supported in pack repositories yet.
2013
supports_external_lookups = False
2014
# What index classes to use
2015
index_builder_class = None
2018
def initialize(self, a_bzrdir, shared=False):
2019
"""Create a pack based repository.
2021
:param a_bzrdir: bzrdir to contain the new repository; must already
2023
:param shared: If true the repository will be initialized as a shared
2026
mutter('creating repository in %s.', a_bzrdir.transport.base)
2027
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
2028
builder = self.index_builder_class()
2029
files = [('pack-names', builder.finish())]
2030
utf8_files = [('format', self.get_format_string())]
2032
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
2033
return self.open(a_bzrdir=a_bzrdir, _found=True)
2035
def open(self, a_bzrdir, _found=False, _override_transport=None):
2036
"""See RepositoryFormat.open().
2038
:param _override_transport: INTERNAL USE ONLY. Allows opening the
2039
repository at a slightly different url
2040
than normal. I.e. during 'upgrade'.
2043
format = RepositoryFormat.find_format(a_bzrdir)
2044
if _override_transport is not None:
2045
repo_transport = _override_transport
2047
repo_transport = a_bzrdir.get_repository_transport(None)
2048
control_files = lockable_files.LockableFiles(repo_transport,
2049
'lock', lockdir.LockDir)
2050
return self.repository_class(_format=self,
2052
control_files=control_files,
2053
_commit_builder_class=self._commit_builder_class,
2054
_serializer=self._serializer)
2057
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2058
"""A no-subtrees parameterized Pack repository.
2060
This format was introduced in 0.92.
2063
repository_class = KnitPackRepository
2064
_commit_builder_class = PackCommitBuilder
2066
def _serializer(self):
2067
return xml5.serializer_v5
2068
# What index classes to use
2069
index_builder_class = InMemoryGraphIndex
2070
index_class = GraphIndex
2072
def _get_matching_bzrdir(self):
2073
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2075
def _ignore_setting_bzrdir(self, format):
2078
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2080
def get_format_string(self):
2081
"""See RepositoryFormat.get_format_string()."""
2082
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2084
def get_format_description(self):
2085
"""See RepositoryFormat.get_format_description()."""
2086
return "Packs containing knits without subtree support"
2088
def check_conversion_target(self, target_format):
2092
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2093
"""A subtrees parameterized Pack repository.
2095
This repository format uses the xml7 serializer to get:
2096
- support for recording full info about the tree root
2097
- support for recording tree-references
2099
This format was introduced in 0.92.
2102
repository_class = KnitPackRepository
2103
_commit_builder_class = PackRootCommitBuilder
2104
rich_root_data = True
2105
supports_tree_reference = True
2107
def _serializer(self):
2108
return xml7.serializer_v7
2109
# What index classes to use
2110
index_builder_class = InMemoryGraphIndex
2111
index_class = GraphIndex
2113
def _get_matching_bzrdir(self):
2114
return bzrdir.format_registry.make_bzrdir(
2115
'pack-0.92-subtree')
2117
def _ignore_setting_bzrdir(self, format):
2120
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2122
def check_conversion_target(self, target_format):
2123
if not target_format.rich_root_data:
2124
raise errors.BadConversionTarget(
2125
'Does not support rich root data.', target_format)
2126
if not getattr(target_format, 'supports_tree_reference', False):
2127
raise errors.BadConversionTarget(
2128
'Does not support nested trees', target_format)
2130
def get_format_string(self):
2131
"""See RepositoryFormat.get_format_string()."""
2132
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2134
def get_format_description(self):
2135
"""See RepositoryFormat.get_format_description()."""
2136
return "Packs containing knits with subtree support\n"
2139
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2140
"""A rich-root, no subtrees parameterized Pack repository.
2142
This repository format uses the xml6 serializer to get:
2143
- support for recording full info about the tree root
2145
This format was introduced in 1.0.
2148
repository_class = KnitPackRepository
2149
_commit_builder_class = PackRootCommitBuilder
2150
rich_root_data = True
2151
supports_tree_reference = False
2153
def _serializer(self):
2154
return xml6.serializer_v6
2155
# What index classes to use
2156
index_builder_class = InMemoryGraphIndex
2157
index_class = GraphIndex
2159
def _get_matching_bzrdir(self):
2160
return bzrdir.format_registry.make_bzrdir(
2163
def _ignore_setting_bzrdir(self, format):
2166
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2168
def check_conversion_target(self, target_format):
2169
if not target_format.rich_root_data:
2170
raise errors.BadConversionTarget(
2171
'Does not support rich root data.', target_format)
2173
def get_format_string(self):
2174
"""See RepositoryFormat.get_format_string()."""
2175
return ("Bazaar pack repository format 1 with rich root"
2176
" (needs bzr 1.0)\n")
2178
def get_format_description(self):
2179
"""See RepositoryFormat.get_format_description()."""
2180
return "Packs containing knits with rich root support\n"
2183
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2184
"""Repository that supports external references to allow stacking.
2188
Supports external lookups, which results in non-truncated ghosts after
2189
reconcile compared to pack-0.92 formats.
2192
repository_class = KnitPackRepository
2193
_commit_builder_class = PackCommitBuilder
2194
supports_external_lookups = True
2195
# What index classes to use
2196
index_builder_class = InMemoryGraphIndex
2197
index_class = GraphIndex
2200
def _serializer(self):
2201
return xml5.serializer_v5
2203
def _get_matching_bzrdir(self):
2204
return bzrdir.format_registry.make_bzrdir('1.6')
2206
def _ignore_setting_bzrdir(self, format):
2209
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2211
def get_format_string(self):
2212
"""See RepositoryFormat.get_format_string()."""
2213
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2215
def get_format_description(self):
2216
"""See RepositoryFormat.get_format_description()."""
2217
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2219
def check_conversion_target(self, target_format):
2223
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2224
"""A repository with rich roots and stacking.
2226
New in release 1.6.1.
2228
Supports stacking on other repositories, allowing data to be accessed
2229
without being stored locally.
2232
repository_class = KnitPackRepository
2233
_commit_builder_class = PackRootCommitBuilder
2234
rich_root_data = True
2235
supports_tree_reference = False # no subtrees
2236
supports_external_lookups = True
2237
# What index classes to use
2238
index_builder_class = InMemoryGraphIndex
2239
index_class = GraphIndex
2242
def _serializer(self):
2243
return xml6.serializer_v6
2245
def _get_matching_bzrdir(self):
2246
return bzrdir.format_registry.make_bzrdir(
2249
def _ignore_setting_bzrdir(self, format):
2252
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2254
def check_conversion_target(self, target_format):
2255
if not target_format.rich_root_data:
2256
raise errors.BadConversionTarget(
2257
'Does not support rich root data.', target_format)
2259
def get_format_string(self):
2260
"""See RepositoryFormat.get_format_string()."""
2261
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2263
def get_format_description(self):
2264
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2267
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2268
"""A repository with rich roots and external references.
2272
Supports external lookups, which results in non-truncated ghosts after
2273
reconcile compared to pack-0.92 formats.
2275
This format was deprecated because the serializer it uses accidentally
2276
supported subtrees, when the format was not intended to. This meant that
2277
someone could accidentally fetch from an incorrect repository.
2280
repository_class = KnitPackRepository
2281
_commit_builder_class = PackRootCommitBuilder
2282
rich_root_data = True
2283
supports_tree_reference = False # no subtrees
2285
supports_external_lookups = True
2286
# What index classes to use
2287
index_builder_class = InMemoryGraphIndex
2288
index_class = GraphIndex
2291
def _serializer(self):
2292
return xml7.serializer_v7
2294
def _get_matching_bzrdir(self):
2295
return bzrdir.format_registry.make_bzrdir(
2298
def _ignore_setting_bzrdir(self, format):
2301
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2303
def check_conversion_target(self, target_format):
2304
if not target_format.rich_root_data:
2305
raise errors.BadConversionTarget(
2306
'Does not support rich root data.', target_format)
2308
def get_format_string(self):
2309
"""See RepositoryFormat.get_format_string()."""
2310
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2312
def get_format_description(self):
2313
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2317
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2318
"""A repository with stacking and btree indexes,
2319
without rich roots or subtrees.
2321
This is equivalent to pack-1.6 with B+Tree indices.
2324
repository_class = KnitPackRepository
2325
_commit_builder_class = PackCommitBuilder
2326
supports_external_lookups = True
2327
# What index classes to use
2328
index_builder_class = BTreeBuilder
2329
index_class = BTreeGraphIndex
2332
def _serializer(self):
2333
return xml5.serializer_v5
2335
def _get_matching_bzrdir(self):
2336
return bzrdir.format_registry.make_bzrdir('1.9')
2338
def _ignore_setting_bzrdir(self, format):
2341
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2343
def get_format_string(self):
2344
"""See RepositoryFormat.get_format_string()."""
2345
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2347
def get_format_description(self):
2348
"""See RepositoryFormat.get_format_description()."""
2349
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2351
def check_conversion_target(self, target_format):
2355
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2356
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2358
1.6-rich-root with B+Tree indices.
2361
repository_class = KnitPackRepository
2362
_commit_builder_class = PackRootCommitBuilder
2363
rich_root_data = True
2364
supports_tree_reference = False # no subtrees
2365
supports_external_lookups = True
2366
# What index classes to use
2367
index_builder_class = BTreeBuilder
2368
index_class = BTreeGraphIndex
2371
def _serializer(self):
2372
return xml6.serializer_v6
2374
def _get_matching_bzrdir(self):
2375
return bzrdir.format_registry.make_bzrdir(
2378
def _ignore_setting_bzrdir(self, format):
2381
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2383
def check_conversion_target(self, target_format):
2384
if not target_format.rich_root_data:
2385
raise errors.BadConversionTarget(
2386
'Does not support rich root data.', target_format)
2388
def get_format_string(self):
2389
"""See RepositoryFormat.get_format_string()."""
2390
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2392
def get_format_description(self):
2393
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2396
class RepositoryFormatPackDevelopment2(RepositoryFormatPack):
2397
"""A no-subtrees development repository.
2399
This format should be retained until the second release after bzr 1.7.
2401
This is pack-1.6.1 with B+Tree indices.
2404
repository_class = KnitPackRepository
2405
_commit_builder_class = PackCommitBuilder
2406
supports_external_lookups = True
2407
# What index classes to use
2408
index_builder_class = BTreeBuilder
2409
index_class = BTreeGraphIndex
2412
def _serializer(self):
2413
return xml5.serializer_v5
2415
def _get_matching_bzrdir(self):
2416
return bzrdir.format_registry.make_bzrdir('development2')
2418
def _ignore_setting_bzrdir(self, format):
2421
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2423
def get_format_string(self):
2424
"""See RepositoryFormat.get_format_string()."""
2425
return "Bazaar development format 2 (needs bzr.dev from before 1.8)\n"
2427
def get_format_description(self):
2428
"""See RepositoryFormat.get_format_description()."""
2429
return ("Development repository format, currently the same as "
2430
"1.6.1 with B+Trees.\n")
2432
def check_conversion_target(self, target_format):
2436
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2437
"""A subtrees development repository.
2439
This format should be retained until the second release after bzr 1.7.
2441
1.6.1-subtree[as it might have been] with B+Tree indices.
2444
repository_class = KnitPackRepository
2445
_commit_builder_class = PackRootCommitBuilder
2446
rich_root_data = True
2447
supports_tree_reference = True
2448
supports_external_lookups = True
2449
# What index classes to use
2450
index_builder_class = BTreeBuilder
2451
index_class = BTreeGraphIndex
2454
def _serializer(self):
2455
return xml7.serializer_v7
2457
def _get_matching_bzrdir(self):
2458
return bzrdir.format_registry.make_bzrdir(
2459
'development2-subtree')
2461
def _ignore_setting_bzrdir(self, format):
2464
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2466
def check_conversion_target(self, target_format):
2467
if not target_format.rich_root_data:
2468
raise errors.BadConversionTarget(
2469
'Does not support rich root data.', target_format)
2470
if not getattr(target_format, 'supports_tree_reference', False):
2471
raise errors.BadConversionTarget(
2472
'Does not support nested trees', target_format)
2474
def get_format_string(self):
2475
"""See RepositoryFormat.get_format_string()."""
2476
return ("Bazaar development format 2 with subtree support "
2477
"(needs bzr.dev from before 1.8)\n")
2479
def get_format_description(self):
2480
"""See RepositoryFormat.get_format_description()."""
2481
return ("Development repository format, currently the same as "
2482
"1.6.1-subtree with B+Tree indices.\n")
added
removed
bzrlib/repofmt/pack_repo.py
