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
20
from bzrlib.lazy_import import lazy_import
21
lazy_import(globals(), """
22
from itertools import izip
36
from bzrlib.index import (
40
GraphIndexPrefixAdapter,
43
from bzrlib.knit import (
49
from bzrlib import tsort
59
from bzrlib.decorators import needs_write_lock
60
from bzrlib.btree_index import (
64
from bzrlib.index import (
68
from bzrlib.repofmt.knitrepo import KnitRepository
69
from bzrlib.repository import (
71
MetaDirRepositoryFormat,
75
import bzrlib.revision as _mod_revision
76
from bzrlib.trace import (
82
class PackCommitBuilder(CommitBuilder):
83
"""A subclass of CommitBuilder to add texts with pack semantics.
85
Specifically this uses one knit object rather than one knit object per
86
added text, reducing memory and object pressure.
89
def __init__(self, repository, parents, config, timestamp=None,
90
timezone=None, committer=None, revprops=None,
92
CommitBuilder.__init__(self, repository, parents, config,
93
timestamp=timestamp, timezone=timezone, committer=committer,
94
revprops=revprops, revision_id=revision_id)
95
self._file_graph = graph.Graph(
96
repository._pack_collection.text_index.combined_index)
98
def _heads(self, file_id, revision_ids):
99
keys = [(file_id, revision_id) for revision_id in revision_ids]
100
return set([key[1] for key in self._file_graph.heads(keys)])
103
class PackRootCommitBuilder(RootCommitBuilder):
104
"""A subclass of RootCommitBuilder to add texts with pack semantics.
106
Specifically this uses one knit object rather than one knit object per
107
added text, reducing memory and object pressure.
110
def __init__(self, repository, parents, config, timestamp=None,
111
timezone=None, committer=None, revprops=None,
113
CommitBuilder.__init__(self, repository, parents, config,
114
timestamp=timestamp, timezone=timezone, committer=committer,
115
revprops=revprops, revision_id=revision_id)
116
self._file_graph = graph.Graph(
117
repository._pack_collection.text_index.combined_index)
119
def _heads(self, file_id, revision_ids):
120
keys = [(file_id, revision_id) for revision_id in revision_ids]
121
return set([key[1] for key in self._file_graph.heads(keys)])
125
"""An in memory proxy for a pack and its indices.
127
This is a base class that is not directly used, instead the classes
128
ExistingPack and NewPack are used.
131
# A map of index 'type' to the file extension and position in the
133
index_definitions = {
134
'revision': ('.rix', 0),
135
'inventory': ('.iix', 1),
137
'signature': ('.six', 3),
140
def __init__(self, revision_index, inventory_index, text_index,
142
"""Create a pack instance.
144
:param revision_index: A GraphIndex for determining what revisions are
145
present in the Pack and accessing the locations of their texts.
146
:param inventory_index: A GraphIndex for determining what inventories are
147
present in the Pack and accessing the locations of their
149
:param text_index: A GraphIndex for determining what file texts
150
are present in the pack and accessing the locations of their
151
texts/deltas (via (fileid, revisionid) tuples).
152
:param signature_index: A GraphIndex for determining what signatures are
153
present in the Pack and accessing the locations of their texts.
155
self.revision_index = revision_index
156
self.inventory_index = inventory_index
157
self.text_index = text_index
158
self.signature_index = signature_index
160
def access_tuple(self):
161
"""Return a tuple (transport, name) for the pack content."""
162
return self.pack_transport, self.file_name()
164
def _check_references(self):
165
"""Make sure our external references are present.
167
Packs are allowed to have deltas whose base is not in the pack, but it
168
must be present somewhere in this collection. It is not allowed to
169
have deltas based on a fallback repository.
170
(See <https://bugs.launchpad.net/bzr/+bug/288751>)
173
for (index_name, external_refs, index) in [
175
self._get_external_refs(self.text_index),
176
self._pack_collection.text_index.combined_index),
178
self._get_external_refs(self.inventory_index),
179
self._pack_collection.inventory_index.combined_index),
181
missing = external_refs.difference(
182
k for (idx, k, v, r) in
183
index.iter_entries(external_refs))
185
missing_items[index_name] = sorted(list(missing))
187
from pprint import pformat
188
raise errors.BzrCheckError(
189
"Newly created pack file %r has delta references to "
190
"items not in its repository:\n%s"
191
% (self, pformat(missing_items)))
194
"""Get the file name for the pack on disk."""
195
return self.name + '.pack'
197
def get_revision_count(self):
198
return self.revision_index.key_count()
200
def index_name(self, index_type, name):
201
"""Get the disk name of an index type for pack name 'name'."""
202
return name + Pack.index_definitions[index_type][0]
204
def index_offset(self, index_type):
205
"""Get the position in a index_size array for a given index type."""
206
return Pack.index_definitions[index_type][1]
208
def inventory_index_name(self, name):
209
"""The inv index is the name + .iix."""
210
return self.index_name('inventory', name)
212
def revision_index_name(self, name):
213
"""The revision index is the name + .rix."""
214
return self.index_name('revision', name)
216
def signature_index_name(self, name):
217
"""The signature index is the name + .six."""
218
return self.index_name('signature', name)
220
def text_index_name(self, name):
221
"""The text index is the name + .tix."""
222
return self.index_name('text', name)
224
def _replace_index_with_readonly(self, index_type):
225
setattr(self, index_type + '_index',
226
self.index_class(self.index_transport,
227
self.index_name(index_type, self.name),
228
self.index_sizes[self.index_offset(index_type)]))
231
class ExistingPack(Pack):
232
"""An in memory proxy for an existing .pack and its disk indices."""
234
def __init__(self, pack_transport, name, revision_index, inventory_index,
235
text_index, signature_index):
236
"""Create an ExistingPack object.
238
:param pack_transport: The transport where the pack file resides.
239
:param name: The name of the pack on disk in the pack_transport.
241
Pack.__init__(self, revision_index, inventory_index, text_index,
244
self.pack_transport = pack_transport
245
if None in (revision_index, inventory_index, text_index,
246
signature_index, name, pack_transport):
247
raise AssertionError()
249
def __eq__(self, other):
250
return self.__dict__ == other.__dict__
252
def __ne__(self, other):
253
return not self.__eq__(other)
256
return "<%s.%s object at 0x%x, %s, %s" % (
257
self.__class__.__module__, self.__class__.__name__, id(self),
258
self.pack_transport, self.name)
261
class ResumedPack(ExistingPack):
263
def __init__(self, name, revision_index, inventory_index, text_index,
264
signature_index, upload_transport, pack_transport, index_transport,
266
"""Create a ResumedPack object."""
267
ExistingPack.__init__(self, pack_transport, name, revision_index,
268
inventory_index, text_index, signature_index)
269
self.upload_transport = upload_transport
270
self.index_transport = index_transport
271
self.index_sizes = [None, None, None, None]
273
('revision', revision_index),
274
('inventory', inventory_index),
275
('text', text_index),
276
('signature', signature_index),
278
for index_type, index in indices:
279
offset = self.index_offset(index_type)
280
self.index_sizes[offset] = index._size
281
self.index_class = pack_collection._index_class
282
self._pack_collection = pack_collection
283
self._state = 'resumed'
284
# XXX: perhaps check that the .pack file exists?
286
def access_tuple(self):
287
if self._state == 'finished':
288
return Pack.access_tuple(self)
289
elif self._state == 'resumed':
290
return self.upload_transport, self.file_name()
292
raise AssertionError(self._state)
295
self.upload_transport.delete(self.file_name())
296
indices = [self.revision_index, self.inventory_index, self.text_index,
297
self.signature_index]
298
for index in indices:
299
index._transport.delete(index._name)
302
self._check_references()
303
new_name = '../packs/' + self.file_name()
304
self.upload_transport.rename(self.file_name(), new_name)
305
for index_type in ['revision', 'inventory', 'text', 'signature']:
306
old_name = self.index_name(index_type, self.name)
307
new_name = '../indices/' + old_name
308
self.upload_transport.rename(old_name, new_name)
309
self._replace_index_with_readonly(index_type)
310
self._state = 'finished'
312
def _get_external_refs(self, index):
313
return index.external_references(1)
317
"""An in memory proxy for a pack which is being created."""
319
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
320
"""Create a NewPack instance.
322
:param pack_collection: A PackCollection into which this is being inserted.
323
:param upload_suffix: An optional suffix to be given to any temporary
324
files created during the pack creation. e.g '.autopack'
325
:param file_mode: Unix permissions for newly created file.
327
# The relative locations of the packs are constrained, but all are
328
# passed in because the caller has them, so as to avoid object churn.
329
index_builder_class = pack_collection._index_builder_class
331
# Revisions: parents list, no text compression.
332
index_builder_class(reference_lists=1),
333
# Inventory: We want to map compression only, but currently the
334
# knit code hasn't been updated enough to understand that, so we
335
# have a regular 2-list index giving parents and compression
337
index_builder_class(reference_lists=2),
338
# Texts: compression and per file graph, for all fileids - so two
339
# reference lists and two elements in the key tuple.
340
index_builder_class(reference_lists=2, key_elements=2),
341
# Signatures: Just blobs to store, no compression, no parents
343
index_builder_class(reference_lists=0),
345
self._pack_collection = pack_collection
346
# When we make readonly indices, we need this.
347
self.index_class = pack_collection._index_class
348
# where should the new pack be opened
349
self.upload_transport = pack_collection._upload_transport
350
# where are indices written out to
351
self.index_transport = pack_collection._index_transport
352
# where is the pack renamed to when it is finished?
353
self.pack_transport = pack_collection._pack_transport
354
# What file mode to upload the pack and indices with.
355
self._file_mode = file_mode
356
# tracks the content written to the .pack file.
357
self._hash = osutils.md5()
358
# a four-tuple with the length in bytes of the indices, once the pack
359
# is finalised. (rev, inv, text, sigs)
360
self.index_sizes = None
361
# How much data to cache when writing packs. Note that this is not
362
# synchronised with reads, because it's not in the transport layer, so
363
# is not safe unless the client knows it won't be reading from the pack
365
self._cache_limit = 0
366
# the temporary pack file name.
367
self.random_name = osutils.rand_chars(20) + upload_suffix
368
# when was this pack started ?
369
self.start_time = time.time()
370
# open an output stream for the data added to the pack.
371
self.write_stream = self.upload_transport.open_write_stream(
372
self.random_name, mode=self._file_mode)
373
if 'pack' in debug.debug_flags:
374
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
375
time.ctime(), self.upload_transport.base, self.random_name,
376
time.time() - self.start_time)
377
# A list of byte sequences to be written to the new pack, and the
378
# aggregate size of them. Stored as a list rather than separate
379
# variables so that the _write_data closure below can update them.
380
self._buffer = [[], 0]
381
# create a callable for adding data
383
# robertc says- this is a closure rather than a method on the object
384
# so that the variables are locals, and faster than accessing object
386
def _write_data(bytes, flush=False, _buffer=self._buffer,
387
_write=self.write_stream.write, _update=self._hash.update):
388
_buffer[0].append(bytes)
389
_buffer[1] += len(bytes)
391
if _buffer[1] > self._cache_limit or flush:
392
bytes = ''.join(_buffer[0])
396
# expose this on self, for the occasion when clients want to add data.
397
self._write_data = _write_data
398
# a pack writer object to serialise pack records.
399
self._writer = pack.ContainerWriter(self._write_data)
401
# what state is the pack in? (open, finished, aborted)
405
"""Cancel creating this pack."""
406
self._state = 'aborted'
407
self.write_stream.close()
408
# Remove the temporary pack file.
409
self.upload_transport.delete(self.random_name)
410
# The indices have no state on disk.
412
def access_tuple(self):
413
"""Return a tuple (transport, name) for the pack content."""
414
if self._state == 'finished':
415
return Pack.access_tuple(self)
416
elif self._state == 'open':
417
return self.upload_transport, self.random_name
419
raise AssertionError(self._state)
421
def data_inserted(self):
422
"""True if data has been added to this pack."""
423
return bool(self.get_revision_count() or
424
self.inventory_index.key_count() or
425
self.text_index.key_count() or
426
self.signature_index.key_count())
428
def finish(self, suspend=False):
429
"""Finish the new pack.
432
- finalises the content
433
- assigns a name (the md5 of the content, currently)
434
- writes out the associated indices
435
- renames the pack into place.
436
- stores the index size tuple for the pack in the index_sizes
441
self._write_data('', flush=True)
442
self.name = self._hash.hexdigest()
444
self._check_references()
446
# XXX: It'd be better to write them all to temporary names, then
447
# rename them all into place, so that the window when only some are
448
# visible is smaller. On the other hand none will be seen until
449
# they're in the names list.
450
self.index_sizes = [None, None, None, None]
451
self._write_index('revision', self.revision_index, 'revision', suspend)
452
self._write_index('inventory', self.inventory_index, 'inventory',
454
self._write_index('text', self.text_index, 'file texts', suspend)
455
self._write_index('signature', self.signature_index,
456
'revision signatures', suspend)
457
self.write_stream.close()
458
# Note that this will clobber an existing pack with the same name,
459
# without checking for hash collisions. While this is undesirable this
460
# is something that can be rectified in a subsequent release. One way
461
# to rectify it may be to leave the pack at the original name, writing
462
# its pack-names entry as something like 'HASH: index-sizes
463
# temporary-name'. Allocate that and check for collisions, if it is
464
# collision free then rename it into place. If clients know this scheme
465
# they can handle missing-file errors by:
466
# - try for HASH.pack
467
# - try for temporary-name
468
# - refresh the pack-list to see if the pack is now absent
469
new_name = self.name + '.pack'
471
new_name = '../packs/' + new_name
472
self.upload_transport.rename(self.random_name, new_name)
473
self._state = 'finished'
474
if 'pack' in debug.debug_flags:
475
# XXX: size might be interesting?
476
mutter('%s: create_pack: pack finished: %s%s->%s t+%6.3fs',
477
time.ctime(), self.upload_transport.base, self.random_name,
478
new_name, time.time() - self.start_time)
481
"""Flush any current data."""
483
bytes = ''.join(self._buffer[0])
484
self.write_stream.write(bytes)
485
self._hash.update(bytes)
486
self._buffer[:] = [[], 0]
488
def _get_external_refs(self, index):
489
return index._external_references()
491
def set_write_cache_size(self, size):
492
self._cache_limit = size
494
def _write_index(self, index_type, index, label, suspend=False):
495
"""Write out an index.
497
:param index_type: The type of index to write - e.g. 'revision'.
498
:param index: The index object to serialise.
499
:param label: What label to give the index e.g. 'revision'.
501
index_name = self.index_name(index_type, self.name)
503
transport = self.upload_transport
505
transport = self.index_transport
506
self.index_sizes[self.index_offset(index_type)] = transport.put_file(
507
index_name, index.finish(), mode=self._file_mode)
508
if 'pack' in debug.debug_flags:
509
# XXX: size might be interesting?
510
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
511
time.ctime(), label, self.upload_transport.base,
512
self.random_name, time.time() - self.start_time)
513
# Replace the writable index on this object with a readonly,
514
# presently unloaded index. We should alter
515
# the index layer to make its finish() error if add_node is
516
# subsequently used. RBC
517
self._replace_index_with_readonly(index_type)
520
class AggregateIndex(object):
521
"""An aggregated index for the RepositoryPackCollection.
523
AggregateIndex is reponsible for managing the PackAccess object,
524
Index-To-Pack mapping, and all indices list for a specific type of index
525
such as 'revision index'.
527
A CombinedIndex provides an index on a single key space built up
528
from several on-disk indices. The AggregateIndex builds on this
529
to provide a knit access layer, and allows having up to one writable
530
index within the collection.
532
# XXX: Probably 'can be written to' could/should be separated from 'acts
533
# like a knit index' -- mbp 20071024
535
def __init__(self, reload_func=None):
536
"""Create an AggregateIndex.
538
:param reload_func: A function to call if we find we are missing an
539
index. Should have the form reload_func() => True if the list of
540
active pack files has changed.
542
self._reload_func = reload_func
543
self.index_to_pack = {}
544
self.combined_index = CombinedGraphIndex([], reload_func=reload_func)
545
self.data_access = _DirectPackAccess(self.index_to_pack,
546
reload_func=reload_func)
547
self.add_callback = None
549
def replace_indices(self, index_to_pack, indices):
550
"""Replace the current mappings with fresh ones.
552
This should probably not be used eventually, rather incremental add and
553
removal of indices. It has been added during refactoring of existing
556
:param index_to_pack: A mapping from index objects to
557
(transport, name) tuples for the pack file data.
558
:param indices: A list of indices.
560
# refresh the revision pack map dict without replacing the instance.
561
self.index_to_pack.clear()
562
self.index_to_pack.update(index_to_pack)
563
# XXX: API break - clearly a 'replace' method would be good?
564
self.combined_index._indices[:] = indices
565
# the current add nodes callback for the current writable index if
567
self.add_callback = None
569
def add_index(self, index, pack):
570
"""Add index to the aggregate, which is an index for Pack pack.
572
Future searches on the aggregate index will seach this new index
573
before all previously inserted indices.
575
:param index: An Index for the pack.
576
:param pack: A Pack instance.
578
# expose it to the index map
579
self.index_to_pack[index] = pack.access_tuple()
580
# put it at the front of the linear index list
581
self.combined_index.insert_index(0, index)
583
def add_writable_index(self, index, pack):
584
"""Add an index which is able to have data added to it.
586
There can be at most one writable index at any time. Any
587
modifications made to the knit are put into this index.
589
:param index: An index from the pack parameter.
590
:param pack: A Pack instance.
592
if self.add_callback is not None:
593
raise AssertionError(
594
"%s already has a writable index through %s" % \
595
(self, self.add_callback))
596
# allow writing: queue writes to a new index
597
self.add_index(index, pack)
598
# Updates the index to packs mapping as a side effect,
599
self.data_access.set_writer(pack._writer, index, pack.access_tuple())
600
self.add_callback = index.add_nodes
603
"""Reset all the aggregate data to nothing."""
604
self.data_access.set_writer(None, None, (None, None))
605
self.index_to_pack.clear()
606
del self.combined_index._indices[:]
607
self.add_callback = None
609
def remove_index(self, index, pack):
610
"""Remove index from the indices used to answer queries.
612
:param index: An index from the pack parameter.
613
:param pack: A Pack instance.
615
del self.index_to_pack[index]
616
self.combined_index._indices.remove(index)
617
if (self.add_callback is not None and
618
getattr(index, 'add_nodes', None) == self.add_callback):
619
self.add_callback = None
620
self.data_access.set_writer(None, None, (None, None))
623
class Packer(object):
624
"""Create a pack from packs."""
626
def __init__(self, pack_collection, packs, suffix, revision_ids=None,
630
:param pack_collection: A RepositoryPackCollection object where the
631
new pack is being written to.
632
:param packs: The packs to combine.
633
:param suffix: The suffix to use on the temporary files for the pack.
634
:param revision_ids: Revision ids to limit the pack to.
635
:param reload_func: A function to call if a pack file/index goes
636
missing. The side effect of calling this function should be to
637
update self.packs. See also AggregateIndex
641
self.revision_ids = revision_ids
642
# The pack object we are creating.
644
self._pack_collection = pack_collection
645
self._reload_func = reload_func
646
# The index layer keys for the revisions being copied. None for 'all
648
self._revision_keys = None
649
# What text keys to copy. None for 'all texts'. This is set by
650
# _copy_inventory_texts
651
self._text_filter = None
654
def _extra_init(self):
655
"""A template hook to allow extending the constructor trivially."""
657
def _pack_map_and_index_list(self, index_attribute):
658
"""Convert a list of packs to an index pack map and index list.
660
:param index_attribute: The attribute that the desired index is found
662
:return: A tuple (map, list) where map contains the dict from
663
index:pack_tuple, and list contains the indices in the preferred
668
for pack_obj in self.packs:
669
index = getattr(pack_obj, index_attribute)
670
indices.append(index)
671
pack_map[index] = pack_obj
672
return pack_map, indices
674
def _index_contents(self, indices, key_filter=None):
675
"""Get an iterable of the index contents from a pack_map.
677
:param indices: The list of indices to query
678
:param key_filter: An optional filter to limit the keys returned.
680
all_index = CombinedGraphIndex(indices)
681
if key_filter is None:
682
return all_index.iter_all_entries()
684
return all_index.iter_entries(key_filter)
686
def pack(self, pb=None):
687
"""Create a new pack by reading data from other packs.
689
This does little more than a bulk copy of data. One key difference
690
is that data with the same item key across multiple packs is elided
691
from the output. The new pack is written into the current pack store
692
along with its indices, and the name added to the pack names. The
693
source packs are not altered and are not required to be in the current
696
:param pb: An optional progress bar to use. A nested bar is created if
698
:return: A Pack object, or None if nothing was copied.
700
# open a pack - using the same name as the last temporary file
701
# - which has already been flushed, so its safe.
702
# XXX: - duplicate code warning with start_write_group; fix before
703
# considering 'done'.
704
if self._pack_collection._new_pack is not None:
705
raise errors.BzrError('call to %s.pack() while another pack is'
707
% (self.__class__.__name__,))
708
if self.revision_ids is not None:
709
if len(self.revision_ids) == 0:
710
# silly fetch request.
713
self.revision_ids = frozenset(self.revision_ids)
714
self.revision_keys = frozenset((revid,) for revid in
717
self.pb = ui.ui_factory.nested_progress_bar()
721
return self._create_pack_from_packs()
727
"""Open a pack for the pack we are creating."""
728
return NewPack(self._pack_collection, upload_suffix=self.suffix,
729
file_mode=self._pack_collection.repo.bzrdir._get_file_mode())
731
def _update_pack_order(self, entries, index_to_pack_map):
732
"""Determine how we want our packs to be ordered.
734
This changes the sort order of the self.packs list so that packs unused
735
by 'entries' will be at the end of the list, so that future requests
736
can avoid probing them. Used packs will be at the front of the
737
self.packs list, in the order of their first use in 'entries'.
739
:param entries: A list of (index, ...) tuples
740
:param index_to_pack_map: A mapping from index objects to pack objects.
744
for entry in entries:
746
if index not in seen_indexes:
747
packs.append(index_to_pack_map[index])
748
seen_indexes.add(index)
749
if len(packs) == len(self.packs):
750
if 'pack' in debug.debug_flags:
751
mutter('Not changing pack list, all packs used.')
753
seen_packs = set(packs)
754
for pack in self.packs:
755
if pack not in seen_packs:
758
if 'pack' in debug.debug_flags:
759
old_names = [p.access_tuple()[1] for p in self.packs]
760
new_names = [p.access_tuple()[1] for p in packs]
761
mutter('Reordering packs\nfrom: %s\n to: %s',
762
old_names, new_names)
765
def _copy_revision_texts(self):
766
"""Copy revision data to the new pack."""
768
if self.revision_ids:
769
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
772
# select revision keys
773
revision_index_map, revision_indices = self._pack_map_and_index_list(
775
revision_nodes = self._index_contents(revision_indices, revision_keys)
776
revision_nodes = list(revision_nodes)
777
self._update_pack_order(revision_nodes, revision_index_map)
778
# copy revision keys and adjust values
779
self.pb.update("Copying revision texts", 1)
780
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
781
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
782
self.new_pack.revision_index, readv_group_iter, total_items))
783
if 'pack' in debug.debug_flags:
784
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
785
time.ctime(), self._pack_collection._upload_transport.base,
786
self.new_pack.random_name,
787
self.new_pack.revision_index.key_count(),
788
time.time() - self.new_pack.start_time)
789
self._revision_keys = revision_keys
791
def _copy_inventory_texts(self):
792
"""Copy the inventory texts to the new pack.
794
self._revision_keys is used to determine what inventories to copy.
796
Sets self._text_filter appropriately.
798
# select inventory keys
799
inv_keys = self._revision_keys # currently the same keyspace, and note that
800
# querying for keys here could introduce a bug where an inventory item
801
# is missed, so do not change it to query separately without cross
802
# checking like the text key check below.
803
inventory_index_map, inventory_indices = self._pack_map_and_index_list(
805
inv_nodes = self._index_contents(inventory_indices, inv_keys)
806
# copy inventory keys and adjust values
807
# XXX: Should be a helper function to allow different inv representation
809
self.pb.update("Copying inventory texts", 2)
810
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
811
# Only grab the output lines if we will be processing them
812
output_lines = bool(self.revision_ids)
813
inv_lines = self._copy_nodes_graph(inventory_index_map,
814
self.new_pack._writer, self.new_pack.inventory_index,
815
readv_group_iter, total_items, output_lines=output_lines)
816
if self.revision_ids:
817
self._process_inventory_lines(inv_lines)
819
# eat the iterator to cause it to execute.
821
self._text_filter = None
822
if 'pack' in debug.debug_flags:
823
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
824
time.ctime(), self._pack_collection._upload_transport.base,
825
self.new_pack.random_name,
826
self.new_pack.inventory_index.key_count(),
827
time.time() - self.new_pack.start_time)
829
def _copy_text_texts(self):
831
text_index_map, text_nodes = self._get_text_nodes()
832
if self._text_filter is not None:
833
# We could return the keys copied as part of the return value from
834
# _copy_nodes_graph but this doesn't work all that well with the
835
# need to get line output too, so we check separately, and as we're
836
# going to buffer everything anyway, we check beforehand, which
837
# saves reading knit data over the wire when we know there are
839
text_nodes = set(text_nodes)
840
present_text_keys = set(_node[1] for _node in text_nodes)
841
missing_text_keys = set(self._text_filter) - present_text_keys
842
if missing_text_keys:
843
# TODO: raise a specific error that can handle many missing
845
a_missing_key = missing_text_keys.pop()
846
raise errors.RevisionNotPresent(a_missing_key[1],
848
# copy text keys and adjust values
849
self.pb.update("Copying content texts", 3)
850
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
851
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
852
self.new_pack.text_index, readv_group_iter, total_items))
853
self._log_copied_texts()
855
def _create_pack_from_packs(self):
856
self.pb.update("Opening pack", 0, 5)
857
self.new_pack = self.open_pack()
858
new_pack = self.new_pack
859
# buffer data - we won't be reading-back during the pack creation and
860
# this makes a significant difference on sftp pushes.
861
new_pack.set_write_cache_size(1024*1024)
862
if 'pack' in debug.debug_flags:
863
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
864
for a_pack in self.packs]
865
if self.revision_ids is not None:
866
rev_count = len(self.revision_ids)
869
mutter('%s: create_pack: creating pack from source packs: '
870
'%s%s %s revisions wanted %s t=0',
871
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
872
plain_pack_list, rev_count)
873
self._copy_revision_texts()
874
self._copy_inventory_texts()
875
self._copy_text_texts()
876
# select signature keys
877
signature_filter = self._revision_keys # same keyspace
878
signature_index_map, signature_indices = self._pack_map_and_index_list(
880
signature_nodes = self._index_contents(signature_indices,
882
# copy signature keys and adjust values
883
self.pb.update("Copying signature texts", 4)
884
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
885
new_pack.signature_index)
886
if 'pack' in debug.debug_flags:
887
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
888
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
889
new_pack.signature_index.key_count(),
890
time.time() - new_pack.start_time)
891
new_pack._check_references()
892
if not self._use_pack(new_pack):
895
self.pb.update("Finishing pack", 5)
897
self._pack_collection.allocate(new_pack)
900
def _copy_nodes(self, nodes, index_map, writer, write_index):
901
"""Copy knit nodes between packs with no graph references."""
902
pb = ui.ui_factory.nested_progress_bar()
904
return self._do_copy_nodes(nodes, index_map, writer,
909
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb):
910
# for record verification
911
knit = KnitVersionedFiles(None, None)
912
# plan a readv on each source pack:
914
nodes = sorted(nodes)
915
# how to map this into knit.py - or knit.py into this?
916
# we don't want the typical knit logic, we want grouping by pack
917
# at this point - perhaps a helper library for the following code
918
# duplication points?
920
for index, key, value in nodes:
921
if index not in request_groups:
922
request_groups[index] = []
923
request_groups[index].append((key, value))
925
pb.update("Copied record", record_index, len(nodes))
926
for index, items in request_groups.iteritems():
927
pack_readv_requests = []
928
for key, value in items:
929
# ---- KnitGraphIndex.get_position
930
bits = value[1:].split(' ')
931
offset, length = int(bits[0]), int(bits[1])
932
pack_readv_requests.append((offset, length, (key, value[0])))
933
# linear scan up the pack
934
pack_readv_requests.sort()
936
pack_obj = index_map[index]
937
transport, path = pack_obj.access_tuple()
939
reader = pack.make_readv_reader(transport, path,
940
[offset[0:2] for offset in pack_readv_requests])
941
except errors.NoSuchFile:
942
if self._reload_func is not None:
945
for (names, read_func), (_1, _2, (key, eol_flag)) in \
946
izip(reader.iter_records(), pack_readv_requests):
947
raw_data = read_func(None)
948
# check the header only
949
df, _ = knit._parse_record_header(key, raw_data)
951
pos, size = writer.add_bytes_record(raw_data, names)
952
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
953
pb.update("Copied record", record_index)
956
def _copy_nodes_graph(self, index_map, writer, write_index,
957
readv_group_iter, total_items, output_lines=False):
958
"""Copy knit nodes between packs.
960
:param output_lines: Return lines present in the copied data as
961
an iterator of line,version_id.
963
pb = ui.ui_factory.nested_progress_bar()
965
for result in self._do_copy_nodes_graph(index_map, writer,
966
write_index, output_lines, pb, readv_group_iter, total_items):
969
# Python 2.4 does not permit try:finally: in a generator.
975
def _do_copy_nodes_graph(self, index_map, writer, write_index,
976
output_lines, pb, readv_group_iter, total_items):
977
# for record verification
978
knit = KnitVersionedFiles(None, None)
979
# for line extraction when requested (inventories only)
981
factory = KnitPlainFactory()
983
pb.update("Copied record", record_index, total_items)
984
for index, readv_vector, node_vector in readv_group_iter:
986
pack_obj = index_map[index]
987
transport, path = pack_obj.access_tuple()
989
reader = pack.make_readv_reader(transport, path, readv_vector)
990
except errors.NoSuchFile:
991
if self._reload_func is not None:
994
for (names, read_func), (key, eol_flag, references) in \
995
izip(reader.iter_records(), node_vector):
996
raw_data = read_func(None)
998
# read the entire thing
999
content, _ = knit._parse_record(key[-1], raw_data)
1000
if len(references[-1]) == 0:
1001
line_iterator = factory.get_fulltext_content(content)
1003
line_iterator = factory.get_linedelta_content(content)
1004
for line in line_iterator:
1007
# check the header only
1008
df, _ = knit._parse_record_header(key, raw_data)
1010
pos, size = writer.add_bytes_record(raw_data, names)
1011
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1012
pb.update("Copied record", record_index)
1015
def _get_text_nodes(self):
1016
text_index_map, text_indices = self._pack_map_and_index_list(
1018
return text_index_map, self._index_contents(text_indices,
1021
def _least_readv_node_readv(self, nodes):
1022
"""Generate request groups for nodes using the least readv's.
1024
:param nodes: An iterable of graph index nodes.
1025
:return: Total node count and an iterator of the data needed to perform
1026
readvs to obtain the data for nodes. Each item yielded by the
1027
iterator is a tuple with:
1028
index, readv_vector, node_vector. readv_vector is a list ready to
1029
hand to the transport readv method, and node_vector is a list of
1030
(key, eol_flag, references) for the the node retrieved by the
1031
matching readv_vector.
1033
# group by pack so we do one readv per pack
1034
nodes = sorted(nodes)
1037
for index, key, value, references in nodes:
1038
if index not in request_groups:
1039
request_groups[index] = []
1040
request_groups[index].append((key, value, references))
1042
for index, items in request_groups.iteritems():
1043
pack_readv_requests = []
1044
for key, value, references in items:
1045
# ---- KnitGraphIndex.get_position
1046
bits = value[1:].split(' ')
1047
offset, length = int(bits[0]), int(bits[1])
1048
pack_readv_requests.append(
1049
((offset, length), (key, value[0], references)))
1050
# linear scan up the pack to maximum range combining.
1051
pack_readv_requests.sort()
1052
# split out the readv and the node data.
1053
pack_readv = [readv for readv, node in pack_readv_requests]
1054
node_vector = [node for readv, node in pack_readv_requests]
1055
result.append((index, pack_readv, node_vector))
1056
return total, result
1058
def _log_copied_texts(self):
1059
if 'pack' in debug.debug_flags:
1060
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
1061
time.ctime(), self._pack_collection._upload_transport.base,
1062
self.new_pack.random_name,
1063
self.new_pack.text_index.key_count(),
1064
time.time() - self.new_pack.start_time)
1066
def _process_inventory_lines(self, inv_lines):
1067
"""Use up the inv_lines generator and setup a text key filter."""
1068
repo = self._pack_collection.repo
1069
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
1070
inv_lines, self.revision_keys)
1072
for fileid, file_revids in fileid_revisions.iteritems():
1073
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
1074
self._text_filter = text_filter
1076
def _revision_node_readv(self, revision_nodes):
1077
"""Return the total revisions and the readv's to issue.
1079
:param revision_nodes: The revision index contents for the packs being
1080
incorporated into the new pack.
1081
:return: As per _least_readv_node_readv.
1083
return self._least_readv_node_readv(revision_nodes)
1085
def _use_pack(self, new_pack):
1086
"""Return True if new_pack should be used.
1088
:param new_pack: The pack that has just been created.
1089
:return: True if the pack should be used.
1091
return new_pack.data_inserted()
1094
class OptimisingPacker(Packer):
1095
"""A packer which spends more time to create better disk layouts."""
1097
def _revision_node_readv(self, revision_nodes):
1098
"""Return the total revisions and the readv's to issue.
1100
This sort places revisions in topological order with the ancestors
1103
:param revision_nodes: The revision index contents for the packs being
1104
incorporated into the new pack.
1105
:return: As per _least_readv_node_readv.
1107
# build an ancestors dict
1110
for index, key, value, references in revision_nodes:
1111
ancestors[key] = references[0]
1112
by_key[key] = (index, value, references)
1113
order = tsort.topo_sort(ancestors)
1115
# Single IO is pathological, but it will work as a starting point.
1117
for key in reversed(order):
1118
index, value, references = by_key[key]
1119
# ---- KnitGraphIndex.get_position
1120
bits = value[1:].split(' ')
1121
offset, length = int(bits[0]), int(bits[1])
1123
(index, [(offset, length)], [(key, value[0], references)]))
1124
# TODO: combine requests in the same index that are in ascending order.
1125
return total, requests
1127
def open_pack(self):
1128
"""Open a pack for the pack we are creating."""
1129
new_pack = super(OptimisingPacker, self).open_pack()
1130
# Turn on the optimization flags for all the index builders.
1131
new_pack.revision_index.set_optimize(for_size=True)
1132
new_pack.inventory_index.set_optimize(for_size=True)
1133
new_pack.text_index.set_optimize(for_size=True)
1134
new_pack.signature_index.set_optimize(for_size=True)
1138
class ReconcilePacker(Packer):
1139
"""A packer which regenerates indices etc as it copies.
1141
This is used by ``bzr reconcile`` to cause parent text pointers to be
1145
def _extra_init(self):
1146
self._data_changed = False
1148
def _process_inventory_lines(self, inv_lines):
1149
"""Generate a text key reference map rather for reconciling with."""
1150
repo = self._pack_collection.repo
1151
refs = repo._find_text_key_references_from_xml_inventory_lines(
1153
self._text_refs = refs
1154
# during reconcile we:
1155
# - convert unreferenced texts to full texts
1156
# - correct texts which reference a text not copied to be full texts
1157
# - copy all others as-is but with corrected parents.
1158
# - so at this point we don't know enough to decide what becomes a full
1160
self._text_filter = None
1162
def _copy_text_texts(self):
1163
"""generate what texts we should have and then copy."""
1164
self.pb.update("Copying content texts", 3)
1165
# we have three major tasks here:
1166
# 1) generate the ideal index
1167
repo = self._pack_collection.repo
1168
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1169
_1, key, _2, refs in
1170
self.new_pack.revision_index.iter_all_entries()])
1171
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1172
# 2) generate a text_nodes list that contains all the deltas that can
1173
# be used as-is, with corrected parents.
1176
discarded_nodes = []
1177
NULL_REVISION = _mod_revision.NULL_REVISION
1178
text_index_map, text_nodes = self._get_text_nodes()
1179
for node in text_nodes:
1185
ideal_parents = tuple(ideal_index[node[1]])
1187
discarded_nodes.append(node)
1188
self._data_changed = True
1190
if ideal_parents == (NULL_REVISION,):
1192
if ideal_parents == node[3][0]:
1194
ok_nodes.append(node)
1195
elif ideal_parents[0:1] == node[3][0][0:1]:
1196
# the left most parent is the same, or there are no parents
1197
# today. Either way, we can preserve the representation as
1198
# long as we change the refs to be inserted.
1199
self._data_changed = True
1200
ok_nodes.append((node[0], node[1], node[2],
1201
(ideal_parents, node[3][1])))
1202
self._data_changed = True
1204
# Reinsert this text completely
1205
bad_texts.append((node[1], ideal_parents))
1206
self._data_changed = True
1207
# we're finished with some data.
1210
# 3) bulk copy the ok data
1211
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1212
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1213
self.new_pack.text_index, readv_group_iter, total_items))
1214
# 4) adhoc copy all the other texts.
1215
# We have to topologically insert all texts otherwise we can fail to
1216
# reconcile when parts of a single delta chain are preserved intact,
1217
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1218
# reinserted, and if d3 has incorrect parents it will also be
1219
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1220
# copied), so we will try to delta, but d2 is not currently able to be
1221
# extracted because it's basis d1 is not present. Topologically sorting
1222
# addresses this. The following generates a sort for all the texts that
1223
# are being inserted without having to reference the entire text key
1224
# space (we only topo sort the revisions, which is smaller).
1225
topo_order = tsort.topo_sort(ancestors)
1226
rev_order = dict(zip(topo_order, range(len(topo_order))))
1227
bad_texts.sort(key=lambda key:rev_order[key[0][1]])
1228
transaction = repo.get_transaction()
1229
file_id_index = GraphIndexPrefixAdapter(
1230
self.new_pack.text_index,
1232
add_nodes_callback=self.new_pack.text_index.add_nodes)
1233
data_access = _DirectPackAccess(
1234
{self.new_pack.text_index:self.new_pack.access_tuple()})
1235
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1236
self.new_pack.access_tuple())
1237
output_texts = KnitVersionedFiles(
1238
_KnitGraphIndex(self.new_pack.text_index,
1239
add_callback=self.new_pack.text_index.add_nodes,
1240
deltas=True, parents=True, is_locked=repo.is_locked),
1241
data_access=data_access, max_delta_chain=200)
1242
for key, parent_keys in bad_texts:
1243
# We refer to the new pack to delta data being output.
1244
# A possible improvement would be to catch errors on short reads
1245
# and only flush then.
1246
self.new_pack.flush()
1248
for parent_key in parent_keys:
1249
if parent_key[0] != key[0]:
1250
# Graph parents must match the fileid
1251
raise errors.BzrError('Mismatched key parent %r:%r' %
1253
parents.append(parent_key[1])
1254
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1255
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1256
output_texts.add_lines(key, parent_keys, text_lines,
1257
random_id=True, check_content=False)
1258
# 5) check that nothing inserted has a reference outside the keyspace.
1259
missing_text_keys = self.new_pack.text_index._external_references()
1260
if missing_text_keys:
1261
raise errors.BzrCheckError('Reference to missing compression parents %r'
1262
% (missing_text_keys,))
1263
self._log_copied_texts()
1265
def _use_pack(self, new_pack):
1266
"""Override _use_pack to check for reconcile having changed content."""
1267
# XXX: we might be better checking this at the copy time.
1268
original_inventory_keys = set()
1269
inv_index = self._pack_collection.inventory_index.combined_index
1270
for entry in inv_index.iter_all_entries():
1271
original_inventory_keys.add(entry[1])
1272
new_inventory_keys = set()
1273
for entry in new_pack.inventory_index.iter_all_entries():
1274
new_inventory_keys.add(entry[1])
1275
if new_inventory_keys != original_inventory_keys:
1276
self._data_changed = True
1277
return new_pack.data_inserted() and self._data_changed
1280
class RepositoryPackCollection(object):
1281
"""Management of packs within a repository.
1283
:ivar _names: map of {pack_name: (index_size,)}
1286
def __init__(self, repo, transport, index_transport, upload_transport,
1287
pack_transport, index_builder_class, index_class):
1288
"""Create a new RepositoryPackCollection.
1290
:param transport: Addresses the repository base directory
1291
(typically .bzr/repository/).
1292
:param index_transport: Addresses the directory containing indices.
1293
:param upload_transport: Addresses the directory into which packs are written
1294
while they're being created.
1295
:param pack_transport: Addresses the directory of existing complete packs.
1296
:param index_builder_class: The index builder class to use.
1297
:param index_class: The index class to use.
1300
self.transport = transport
1301
self._index_transport = index_transport
1302
self._upload_transport = upload_transport
1303
self._pack_transport = pack_transport
1304
self._index_builder_class = index_builder_class
1305
self._index_class = index_class
1306
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
1309
self._packs_by_name = {}
1310
# the previous pack-names content
1311
self._packs_at_load = None
1312
# when a pack is being created by this object, the state of that pack.
1313
self._new_pack = None
1314
# aggregated revision index data
1315
self.revision_index = AggregateIndex(self.reload_pack_names)
1316
self.inventory_index = AggregateIndex(self.reload_pack_names)
1317
self.text_index = AggregateIndex(self.reload_pack_names)
1318
self.signature_index = AggregateIndex(self.reload_pack_names)
1320
self._resumed_packs = []
1322
def add_pack_to_memory(self, pack):
1323
"""Make a Pack object available to the repository to satisfy queries.
1325
:param pack: A Pack object.
1327
if pack.name in self._packs_by_name:
1328
raise AssertionError(
1329
'pack %s already in _packs_by_name' % (pack.name,))
1330
self.packs.append(pack)
1331
self._packs_by_name[pack.name] = pack
1332
self.revision_index.add_index(pack.revision_index, pack)
1333
self.inventory_index.add_index(pack.inventory_index, pack)
1334
self.text_index.add_index(pack.text_index, pack)
1335
self.signature_index.add_index(pack.signature_index, pack)
1337
def all_packs(self):
1338
"""Return a list of all the Pack objects this repository has.
1340
Note that an in-progress pack being created is not returned.
1342
:return: A list of Pack objects for all the packs in the repository.
1345
for name in self.names():
1346
result.append(self.get_pack_by_name(name))
1350
"""Pack the pack collection incrementally.
1352
This will not attempt global reorganisation or recompression,
1353
rather it will just ensure that the total number of packs does
1354
not grow without bound. It uses the _max_pack_count method to
1355
determine if autopacking is needed, and the pack_distribution
1356
method to determine the number of revisions in each pack.
1358
If autopacking takes place then the packs name collection will have
1359
been flushed to disk - packing requires updating the name collection
1360
in synchronisation with certain steps. Otherwise the names collection
1363
:return: True if packing took place.
1367
return self._do_autopack()
1368
except errors.RetryAutopack, e:
1369
# If we get a RetryAutopack exception, we should abort the
1370
# current action, and retry.
1373
def _do_autopack(self):
1374
# XXX: Should not be needed when the management of indices is sane.
1375
total_revisions = self.revision_index.combined_index.key_count()
1376
total_packs = len(self._names)
1377
if self._max_pack_count(total_revisions) >= total_packs:
1379
# XXX: the following may want to be a class, to pack with a given
1381
# determine which packs need changing
1382
pack_distribution = self.pack_distribution(total_revisions)
1384
for pack in self.all_packs():
1385
revision_count = pack.get_revision_count()
1386
if revision_count == 0:
1387
# revision less packs are not generated by normal operation,
1388
# only by operations like sign-my-commits, and thus will not
1389
# tend to grow rapdily or without bound like commit containing
1390
# packs do - leave them alone as packing them really should
1391
# group their data with the relevant commit, and that may
1392
# involve rewriting ancient history - which autopack tries to
1393
# avoid. Alternatively we could not group the data but treat
1394
# each of these as having a single revision, and thus add
1395
# one revision for each to the total revision count, to get
1396
# a matching distribution.
1398
existing_packs.append((revision_count, pack))
1399
pack_operations = self.plan_autopack_combinations(
1400
existing_packs, pack_distribution)
1401
num_new_packs = len(pack_operations)
1402
num_old_packs = sum([len(po[1]) for po in pack_operations])
1403
num_revs_affected = sum([po[0] for po in pack_operations])
1404
mutter('Auto-packing repository %s, which has %d pack files, '
1405
'containing %d revisions. Packing %d files into %d affecting %d'
1406
' revisions', self, total_packs, total_revisions, num_old_packs,
1407
num_new_packs, num_revs_affected)
1408
self._execute_pack_operations(pack_operations,
1409
reload_func=self._restart_autopack)
1412
def _execute_pack_operations(self, pack_operations, _packer_class=Packer,
1414
"""Execute a series of pack operations.
1416
:param pack_operations: A list of [revision_count, packs_to_combine].
1417
:param _packer_class: The class of packer to use (default: Packer).
1420
for revision_count, packs in pack_operations:
1421
# we may have no-ops from the setup logic
1424
packer = _packer_class(self, packs, '.autopack',
1425
reload_func=reload_func)
1428
except errors.RetryWithNewPacks:
1429
# An exception is propagating out of this context, make sure
1430
# this packer has cleaned up. Packer() doesn't set its new_pack
1431
# state into the RepositoryPackCollection object, so we only
1432
# have access to it directly here.
1433
if packer.new_pack is not None:
1434
packer.new_pack.abort()
1437
self._remove_pack_from_memory(pack)
1438
# record the newly available packs and stop advertising the old
1440
self._save_pack_names(clear_obsolete_packs=True)
1441
# Move the old packs out of the way now they are no longer referenced.
1442
for revision_count, packs in pack_operations:
1443
self._obsolete_packs(packs)
1445
def lock_names(self):
1446
"""Acquire the mutex around the pack-names index.
1448
This cannot be used in the middle of a read-only transaction on the
1451
self.repo.control_files.lock_write()
1454
"""Pack the pack collection totally."""
1455
self.ensure_loaded()
1456
total_packs = len(self._names)
1458
# This is arguably wrong because we might not be optimal, but for
1459
# now lets leave it in. (e.g. reconcile -> one pack. But not
1462
total_revisions = self.revision_index.combined_index.key_count()
1463
# XXX: the following may want to be a class, to pack with a given
1465
mutter('Packing repository %s, which has %d pack files, '
1466
'containing %d revisions into 1 packs.', self, total_packs,
1468
# determine which packs need changing
1469
pack_distribution = [1]
1470
pack_operations = [[0, []]]
1471
for pack in self.all_packs():
1472
pack_operations[-1][0] += pack.get_revision_count()
1473
pack_operations[-1][1].append(pack)
1474
self._execute_pack_operations(pack_operations, OptimisingPacker)
1476
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1477
"""Plan a pack operation.
1479
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1481
:param pack_distribution: A list with the number of revisions desired
1484
if len(existing_packs) <= len(pack_distribution):
1486
existing_packs.sort(reverse=True)
1487
pack_operations = [[0, []]]
1488
# plan out what packs to keep, and what to reorganise
1489
while len(existing_packs):
1490
# take the largest pack, and if its less than the head of the
1491
# distribution chart we will include its contents in the new pack
1492
# for that position. If its larger, we remove its size from the
1493
# distribution chart
1494
next_pack_rev_count, next_pack = existing_packs.pop(0)
1495
if next_pack_rev_count >= pack_distribution[0]:
1496
# this is already packed 'better' than this, so we can
1497
# not waste time packing it.
1498
while next_pack_rev_count > 0:
1499
next_pack_rev_count -= pack_distribution[0]
1500
if next_pack_rev_count >= 0:
1502
del pack_distribution[0]
1504
# didn't use that entire bucket up
1505
pack_distribution[0] = -next_pack_rev_count
1507
# add the revisions we're going to add to the next output pack
1508
pack_operations[-1][0] += next_pack_rev_count
1509
# allocate this pack to the next pack sub operation
1510
pack_operations[-1][1].append(next_pack)
1511
if pack_operations[-1][0] >= pack_distribution[0]:
1512
# this pack is used up, shift left.
1513
del pack_distribution[0]
1514
pack_operations.append([0, []])
1515
# Now that we know which pack files we want to move, shove them all
1516
# into a single pack file.
1518
final_pack_list = []
1519
for num_revs, pack_files in pack_operations:
1520
final_rev_count += num_revs
1521
final_pack_list.extend(pack_files)
1522
if len(final_pack_list) == 1:
1523
raise AssertionError('We somehow generated an autopack with a'
1524
' single pack file being moved.')
1526
return [[final_rev_count, final_pack_list]]
1528
def ensure_loaded(self):
1529
# NB: if you see an assertion error here, its probably access against
1530
# an unlocked repo. Naughty.
1531
if not self.repo.is_locked():
1532
raise errors.ObjectNotLocked(self.repo)
1533
if self._names is None:
1535
self._packs_at_load = set()
1536
for index, key, value in self._iter_disk_pack_index():
1538
self._names[name] = self._parse_index_sizes(value)
1539
self._packs_at_load.add((key, value))
1540
# populate all the metadata.
1543
def _parse_index_sizes(self, value):
1544
"""Parse a string of index sizes."""
1545
return tuple([int(digits) for digits in value.split(' ')])
1547
def get_pack_by_name(self, name):
1548
"""Get a Pack object by name.
1550
:param name: The name of the pack - e.g. '123456'
1551
:return: A Pack object.
1554
return self._packs_by_name[name]
1556
rev_index = self._make_index(name, '.rix')
1557
inv_index = self._make_index(name, '.iix')
1558
txt_index = self._make_index(name, '.tix')
1559
sig_index = self._make_index(name, '.six')
1560
result = ExistingPack(self._pack_transport, name, rev_index,
1561
inv_index, txt_index, sig_index)
1562
self.add_pack_to_memory(result)
1565
def _resume_pack(self, name):
1566
"""Get a suspended Pack object by name.
1568
:param name: The name of the pack - e.g. '123456'
1569
:return: A Pack object.
1571
if not re.match('[a-f0-9]{32}', name):
1572
# Tokens should be md5sums of the suspended pack file, i.e. 32 hex
1574
raise errors.UnresumableWriteGroup(
1575
self.repo, [name], 'Malformed write group token')
1577
rev_index = self._make_index(name, '.rix', resume=True)
1578
inv_index = self._make_index(name, '.iix', resume=True)
1579
txt_index = self._make_index(name, '.tix', resume=True)
1580
sig_index = self._make_index(name, '.six', resume=True)
1581
result = ResumedPack(name, rev_index, inv_index, txt_index,
1582
sig_index, self._upload_transport, self._pack_transport,
1583
self._index_transport, self)
1584
except errors.NoSuchFile, e:
1585
raise errors.UnresumableWriteGroup(self.repo, [name], str(e))
1586
self.add_pack_to_memory(result)
1587
self._resumed_packs.append(result)
1590
def allocate(self, a_new_pack):
1591
"""Allocate name in the list of packs.
1593
:param a_new_pack: A NewPack instance to be added to the collection of
1594
packs for this repository.
1596
self.ensure_loaded()
1597
if a_new_pack.name in self._names:
1598
raise errors.BzrError(
1599
'Pack %r already exists in %s' % (a_new_pack.name, self))
1600
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1601
self.add_pack_to_memory(a_new_pack)
1603
def _iter_disk_pack_index(self):
1604
"""Iterate over the contents of the pack-names index.
1606
This is used when loading the list from disk, and before writing to
1607
detect updates from others during our write operation.
1608
:return: An iterator of the index contents.
1610
return self._index_class(self.transport, 'pack-names', None
1611
).iter_all_entries()
1613
def _make_index(self, name, suffix, resume=False):
1614
size_offset = self._suffix_offsets[suffix]
1615
index_name = name + suffix
1617
transport = self._upload_transport
1618
index_size = transport.stat(index_name).st_size
1620
transport = self._index_transport
1621
index_size = self._names[name][size_offset]
1622
return self._index_class(transport, index_name, index_size)
1624
def _max_pack_count(self, total_revisions):
1625
"""Return the maximum number of packs to use for total revisions.
1627
:param total_revisions: The total number of revisions in the
1630
if not total_revisions:
1632
digits = str(total_revisions)
1634
for digit in digits:
1635
result += int(digit)
1639
"""Provide an order to the underlying names."""
1640
return sorted(self._names.keys())
1642
def _obsolete_packs(self, packs):
1643
"""Move a number of packs which have been obsoleted out of the way.
1645
Each pack and its associated indices are moved out of the way.
1647
Note: for correctness this function should only be called after a new
1648
pack names index has been written without these pack names, and with
1649
the names of packs that contain the data previously available via these
1652
:param packs: The packs to obsolete.
1653
:param return: None.
1656
pack.pack_transport.rename(pack.file_name(),
1657
'../obsolete_packs/' + pack.file_name())
1658
# TODO: Probably needs to know all possible indices for this pack
1659
# - or maybe list the directory and move all indices matching this
1660
# name whether we recognize it or not?
1661
for suffix in ('.iix', '.six', '.tix', '.rix'):
1662
self._index_transport.rename(pack.name + suffix,
1663
'../obsolete_packs/' + pack.name + suffix)
1665
def pack_distribution(self, total_revisions):
1666
"""Generate a list of the number of revisions to put in each pack.
1668
:param total_revisions: The total number of revisions in the
1671
if total_revisions == 0:
1673
digits = reversed(str(total_revisions))
1675
for exponent, count in enumerate(digits):
1676
size = 10 ** exponent
1677
for pos in range(int(count)):
1679
return list(reversed(result))
1681
def _pack_tuple(self, name):
1682
"""Return a tuple with the transport and file name for a pack name."""
1683
return self._pack_transport, name + '.pack'
1685
def _remove_pack_from_memory(self, pack):
1686
"""Remove pack from the packs accessed by this repository.
1688
Only affects memory state, until self._save_pack_names() is invoked.
1690
self._names.pop(pack.name)
1691
self._packs_by_name.pop(pack.name)
1692
self._remove_pack_indices(pack)
1693
self.packs.remove(pack)
1695
def _remove_pack_indices(self, pack):
1696
"""Remove the indices for pack from the aggregated indices."""
1697
self.revision_index.remove_index(pack.revision_index, pack)
1698
self.inventory_index.remove_index(pack.inventory_index, pack)
1699
self.text_index.remove_index(pack.text_index, pack)
1700
self.signature_index.remove_index(pack.signature_index, pack)
1703
"""Clear all cached data."""
1704
# cached revision data
1705
self.repo._revision_knit = None
1706
self.revision_index.clear()
1707
# cached signature data
1708
self.repo._signature_knit = None
1709
self.signature_index.clear()
1710
# cached file text data
1711
self.text_index.clear()
1712
self.repo._text_knit = None
1713
# cached inventory data
1714
self.inventory_index.clear()
1715
# remove the open pack
1716
self._new_pack = None
1717
# information about packs.
1720
self._packs_by_name = {}
1721
self._packs_at_load = None
1723
def _unlock_names(self):
1724
"""Release the mutex around the pack-names index."""
1725
self.repo.control_files.unlock()
1727
def _diff_pack_names(self):
1728
"""Read the pack names from disk, and compare it to the one in memory.
1730
:return: (disk_nodes, deleted_nodes, new_nodes)
1731
disk_nodes The final set of nodes that should be referenced
1732
deleted_nodes Nodes which have been removed from when we started
1733
new_nodes Nodes that are newly introduced
1735
# load the disk nodes across
1737
for index, key, value in self._iter_disk_pack_index():
1738
disk_nodes.add((key, value))
1740
# do a two-way diff against our original content
1741
current_nodes = set()
1742
for name, sizes in self._names.iteritems():
1744
((name, ), ' '.join(str(size) for size in sizes)))
1746
# Packs no longer present in the repository, which were present when we
1747
# locked the repository
1748
deleted_nodes = self._packs_at_load - current_nodes
1749
# Packs which this process is adding
1750
new_nodes = current_nodes - self._packs_at_load
1752
# Update the disk_nodes set to include the ones we are adding, and
1753
# remove the ones which were removed by someone else
1754
disk_nodes.difference_update(deleted_nodes)
1755
disk_nodes.update(new_nodes)
1757
return disk_nodes, deleted_nodes, new_nodes
1759
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1760
"""Given the correct set of pack files, update our saved info.
1762
:return: (removed, added, modified)
1763
removed pack names removed from self._names
1764
added pack names added to self._names
1765
modified pack names that had changed value
1770
## self._packs_at_load = disk_nodes
1771
new_names = dict(disk_nodes)
1772
# drop no longer present nodes
1773
for pack in self.all_packs():
1774
if (pack.name,) not in new_names:
1775
removed.append(pack.name)
1776
self._remove_pack_from_memory(pack)
1777
# add new nodes/refresh existing ones
1778
for key, value in disk_nodes:
1780
sizes = self._parse_index_sizes(value)
1781
if name in self._names:
1783
if sizes != self._names[name]:
1784
# the pack for name has had its indices replaced - rare but
1785
# important to handle. XXX: probably can never happen today
1786
# because the three-way merge code above does not handle it
1787
# - you may end up adding the same key twice to the new
1788
# disk index because the set values are the same, unless
1789
# the only index shows up as deleted by the set difference
1790
# - which it may. Until there is a specific test for this,
1791
# assume its broken. RBC 20071017.
1792
self._remove_pack_from_memory(self.get_pack_by_name(name))
1793
self._names[name] = sizes
1794
self.get_pack_by_name(name)
1795
modified.append(name)
1798
self._names[name] = sizes
1799
self.get_pack_by_name(name)
1801
return removed, added, modified
1803
def _save_pack_names(self, clear_obsolete_packs=False):
1804
"""Save the list of packs.
1806
This will take out the mutex around the pack names list for the
1807
duration of the method call. If concurrent updates have been made, a
1808
three-way merge between the current list and the current in memory list
1811
:param clear_obsolete_packs: If True, clear out the contents of the
1812
obsolete_packs directory.
1816
builder = self._index_builder_class()
1817
disk_nodes, deleted_nodes, new_nodes = self._diff_pack_names()
1818
# TODO: handle same-name, index-size-changes here -
1819
# e.g. use the value from disk, not ours, *unless* we're the one
1821
for key, value in disk_nodes:
1822
builder.add_node(key, value)
1823
self.transport.put_file('pack-names', builder.finish(),
1824
mode=self.repo.bzrdir._get_file_mode())
1825
# move the baseline forward
1826
self._packs_at_load = disk_nodes
1827
if clear_obsolete_packs:
1828
self._clear_obsolete_packs()
1830
self._unlock_names()
1831
# synchronise the memory packs list with what we just wrote:
1832
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1834
def reload_pack_names(self):
1835
"""Sync our pack listing with what is present in the repository.
1837
This should be called when we find out that something we thought was
1838
present is now missing. This happens when another process re-packs the
1841
# This is functionally similar to _save_pack_names, but we don't write
1842
# out the new value.
1843
disk_nodes, _, _ = self._diff_pack_names()
1844
self._packs_at_load = disk_nodes
1846
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1847
if removed or added or modified:
1851
def _restart_autopack(self):
1852
"""Reload the pack names list, and restart the autopack code."""
1853
if not self.reload_pack_names():
1854
# Re-raise the original exception, because something went missing
1855
# and a restart didn't find it
1857
raise errors.RetryAutopack(self.repo, False, sys.exc_info())
1859
def _clear_obsolete_packs(self):
1860
"""Delete everything from the obsolete-packs directory.
1862
obsolete_pack_transport = self.transport.clone('obsolete_packs')
1863
for filename in obsolete_pack_transport.list_dir('.'):
1865
obsolete_pack_transport.delete(filename)
1866
except (errors.PathError, errors.TransportError), e:
1867
warning("couldn't delete obsolete pack, skipping it:\n%s" % (e,))
1869
def _start_write_group(self):
1870
# Do not permit preparation for writing if we're not in a 'write lock'.
1871
if not self.repo.is_write_locked():
1872
raise errors.NotWriteLocked(self)
1873
self._new_pack = NewPack(self, upload_suffix='.pack',
1874
file_mode=self.repo.bzrdir._get_file_mode())
1875
# allow writing: queue writes to a new index
1876
self.revision_index.add_writable_index(self._new_pack.revision_index,
1878
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
1880
self.text_index.add_writable_index(self._new_pack.text_index,
1882
self.signature_index.add_writable_index(self._new_pack.signature_index,
1885
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
1886
self.repo.revisions._index._add_callback = self.revision_index.add_callback
1887
self.repo.signatures._index._add_callback = self.signature_index.add_callback
1888
self.repo.texts._index._add_callback = self.text_index.add_callback
1890
def _abort_write_group(self):
1891
# FIXME: just drop the transient index.
1892
# forget what names there are
1893
if self._new_pack is not None:
1895
self._new_pack.abort()
1897
# XXX: If we aborted while in the middle of finishing the write
1898
# group, _remove_pack_indices can fail because the indexes are
1899
# already gone. If they're not there we shouldn't fail in this
1900
# case. -- mbp 20081113
1901
self._remove_pack_indices(self._new_pack)
1902
self._new_pack = None
1903
for resumed_pack in self._resumed_packs:
1905
resumed_pack.abort()
1907
# See comment in previous finally block.
1909
self._remove_pack_indices(resumed_pack)
1912
del self._resumed_packs[:]
1913
self.repo._text_knit = None
1915
def _remove_resumed_pack_indices(self):
1916
for resumed_pack in self._resumed_packs:
1917
self._remove_pack_indices(resumed_pack)
1918
del self._resumed_packs[:]
1920
def _commit_write_group(self):
1922
for prefix, versioned_file in (
1923
('revisions', self.repo.revisions),
1924
('inventories', self.repo.inventories),
1925
('texts', self.repo.texts),
1926
('signatures', self.repo.signatures),
1928
missing = versioned_file.get_missing_compression_parent_keys()
1929
all_missing.update([(prefix,) + key for key in missing])
1931
raise errors.BzrCheckError(
1932
"Repository %s has missing compression parent(s) %r "
1933
% (self.repo, sorted(all_missing)))
1934
self._remove_pack_indices(self._new_pack)
1935
should_autopack = False
1936
if self._new_pack.data_inserted():
1937
# get all the data to disk and read to use
1938
self._new_pack.finish()
1939
self.allocate(self._new_pack)
1940
self._new_pack = None
1941
should_autopack = True
1943
self._new_pack.abort()
1944
self._new_pack = None
1945
for resumed_pack in self._resumed_packs:
1946
# XXX: this is a pretty ugly way to turn the resumed pack into a
1947
# properly committed pack.
1948
self._names[resumed_pack.name] = None
1949
self._remove_pack_from_memory(resumed_pack)
1950
resumed_pack.finish()
1951
self.allocate(resumed_pack)
1952
should_autopack = True
1953
del self._resumed_packs[:]
1955
if not self.autopack():
1956
# when autopack takes no steps, the names list is still
1958
self._save_pack_names()
1959
self.repo._text_knit = None
1961
def _suspend_write_group(self):
1962
tokens = [pack.name for pack in self._resumed_packs]
1963
self._remove_pack_indices(self._new_pack)
1964
if self._new_pack.data_inserted():
1965
# get all the data to disk and read to use
1966
self._new_pack.finish(suspend=True)
1967
tokens.append(self._new_pack.name)
1968
self._new_pack = None
1970
self._new_pack.abort()
1971
self._new_pack = None
1972
self._remove_resumed_pack_indices()
1973
self.repo._text_knit = None
1976
def _resume_write_group(self, tokens):
1977
for token in tokens:
1978
self._resume_pack(token)
1981
class KnitPackRepository(KnitRepository):
1982
"""Repository with knit objects stored inside pack containers.
1984
The layering for a KnitPackRepository is:
1986
Graph | HPSS | Repository public layer |
1987
===================================================
1988
Tuple based apis below, string based, and key based apis above
1989
---------------------------------------------------
1991
Provides .texts, .revisions etc
1992
This adapts the N-tuple keys to physical knit records which only have a
1993
single string identifier (for historical reasons), which in older formats
1994
was always the revision_id, and in the mapped code for packs is always
1995
the last element of key tuples.
1996
---------------------------------------------------
1998
A separate GraphIndex is used for each of the
1999
texts/inventories/revisions/signatures contained within each individual
2000
pack file. The GraphIndex layer works in N-tuples and is unaware of any
2002
===================================================
2006
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
2008
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2009
_commit_builder_class, _serializer)
2010
index_transport = self._transport.clone('indices')
2011
self._pack_collection = RepositoryPackCollection(self, self._transport,
2013
self._transport.clone('upload'),
2014
self._transport.clone('packs'),
2015
_format.index_builder_class,
2016
_format.index_class)
2017
self.inventories = KnitVersionedFiles(
2018
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2019
add_callback=self._pack_collection.inventory_index.add_callback,
2020
deltas=True, parents=True, is_locked=self.is_locked),
2021
data_access=self._pack_collection.inventory_index.data_access,
2022
max_delta_chain=200)
2023
self.revisions = KnitVersionedFiles(
2024
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2025
add_callback=self._pack_collection.revision_index.add_callback,
2026
deltas=False, parents=True, is_locked=self.is_locked),
2027
data_access=self._pack_collection.revision_index.data_access,
2029
self.signatures = KnitVersionedFiles(
2030
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2031
add_callback=self._pack_collection.signature_index.add_callback,
2032
deltas=False, parents=False, is_locked=self.is_locked),
2033
data_access=self._pack_collection.signature_index.data_access,
2035
self.texts = KnitVersionedFiles(
2036
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2037
add_callback=self._pack_collection.text_index.add_callback,
2038
deltas=True, parents=True, is_locked=self.is_locked),
2039
data_access=self._pack_collection.text_index.data_access,
2040
max_delta_chain=200)
2041
# True when the repository object is 'write locked' (as opposed to the
2042
# physical lock only taken out around changes to the pack-names list.)
2043
# Another way to represent this would be a decorator around the control
2044
# files object that presents logical locks as physical ones - if this
2045
# gets ugly consider that alternative design. RBC 20071011
2046
self._write_lock_count = 0
2047
self._transaction = None
2049
self._reconcile_does_inventory_gc = True
2050
self._reconcile_fixes_text_parents = True
2051
self._reconcile_backsup_inventory = False
2053
def _warn_if_deprecated(self):
2054
# This class isn't deprecated, but one sub-format is
2055
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2056
from bzrlib import repository
2057
if repository._deprecation_warning_done:
2059
repository._deprecation_warning_done = True
2060
warning("Format %s for %s is deprecated - please use"
2061
" 'bzr upgrade --1.6.1-rich-root'"
2062
% (self._format, self.bzrdir.transport.base))
2064
def _abort_write_group(self):
2065
self._pack_collection._abort_write_group()
2067
def _find_inconsistent_revision_parents(self):
2068
"""Find revisions with incorrectly cached parents.
2070
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
2071
parents-in-revision).
2073
if not self.is_locked():
2074
raise errors.ObjectNotLocked(self)
2075
pb = ui.ui_factory.nested_progress_bar()
2078
revision_nodes = self._pack_collection.revision_index \
2079
.combined_index.iter_all_entries()
2080
index_positions = []
2081
# Get the cached index values for all revisions, and also the location
2082
# in each index of the revision text so we can perform linear IO.
2083
for index, key, value, refs in revision_nodes:
2084
pos, length = value[1:].split(' ')
2085
index_positions.append((index, int(pos), key[0],
2086
tuple(parent[0] for parent in refs[0])))
2087
pb.update("Reading revision index.", 0, 0)
2088
index_positions.sort()
2089
batch_count = len(index_positions) / 1000 + 1
2090
pb.update("Checking cached revision graph.", 0, batch_count)
2091
for offset in xrange(batch_count):
2092
pb.update("Checking cached revision graph.", offset)
2093
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
2096
rev_ids = [item[2] for item in to_query]
2097
revs = self.get_revisions(rev_ids)
2098
for revision, item in zip(revs, to_query):
2099
index_parents = item[3]
2100
rev_parents = tuple(revision.parent_ids)
2101
if index_parents != rev_parents:
2102
result.append((revision.revision_id, index_parents, rev_parents))
2107
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
2108
def get_parents(self, revision_ids):
2109
"""See graph._StackedParentsProvider.get_parents."""
2110
parent_map = self.get_parent_map(revision_ids)
2111
return [parent_map.get(r, None) for r in revision_ids]
2113
def _make_parents_provider(self):
2114
return graph.CachingParentsProvider(self)
2116
def _refresh_data(self):
2117
if self._write_lock_count == 1 or (
2118
self.control_files._lock_count == 1 and
2119
self.control_files._lock_mode == 'r'):
2120
# forget what names there are
2121
self._pack_collection.reset()
2122
# XXX: Better to do an in-memory merge when acquiring a new lock -
2123
# factor out code from _save_pack_names.
2124
self._pack_collection.ensure_loaded()
2126
def _start_write_group(self):
2127
self._pack_collection._start_write_group()
2129
def _commit_write_group(self):
2130
return self._pack_collection._commit_write_group()
2132
def suspend_write_group(self):
2133
# XXX check self._write_group is self.get_transaction()?
2134
tokens = self._pack_collection._suspend_write_group()
2135
self._write_group = None
2138
def _resume_write_group(self, tokens):
2139
self._start_write_group()
2140
self._pack_collection._resume_write_group(tokens)
2142
def get_transaction(self):
2143
if self._write_lock_count:
2144
return self._transaction
2146
return self.control_files.get_transaction()
2148
def is_locked(self):
2149
return self._write_lock_count or self.control_files.is_locked()
2151
def is_write_locked(self):
2152
return self._write_lock_count
2154
def lock_write(self, token=None):
2155
if not self._write_lock_count and self.is_locked():
2156
raise errors.ReadOnlyError(self)
2157
self._write_lock_count += 1
2158
if self._write_lock_count == 1:
2159
self._transaction = transactions.WriteTransaction()
2160
for repo in self._fallback_repositories:
2161
# Writes don't affect fallback repos
2163
self._refresh_data()
2165
def lock_read(self):
2166
if self._write_lock_count:
2167
self._write_lock_count += 1
2169
self.control_files.lock_read()
2170
for repo in self._fallback_repositories:
2171
# Writes don't affect fallback repos
2173
self._refresh_data()
2175
def leave_lock_in_place(self):
2176
# not supported - raise an error
2177
raise NotImplementedError(self.leave_lock_in_place)
2179
def dont_leave_lock_in_place(self):
2180
# not supported - raise an error
2181
raise NotImplementedError(self.dont_leave_lock_in_place)
2185
"""Compress the data within the repository.
2187
This will pack all the data to a single pack. In future it may
2188
recompress deltas or do other such expensive operations.
2190
self._pack_collection.pack()
2193
def reconcile(self, other=None, thorough=False):
2194
"""Reconcile this repository."""
2195
from bzrlib.reconcile import PackReconciler
2196
reconciler = PackReconciler(self, thorough=thorough)
2197
reconciler.reconcile()
2201
if self._write_lock_count == 1 and self._write_group is not None:
2202
self.abort_write_group()
2203
self._transaction = None
2204
self._write_lock_count = 0
2205
raise errors.BzrError(
2206
'Must end write group before releasing write lock on %s'
2208
if self._write_lock_count:
2209
self._write_lock_count -= 1
2210
if not self._write_lock_count:
2211
transaction = self._transaction
2212
self._transaction = None
2213
transaction.finish()
2214
for repo in self._fallback_repositories:
2217
self.control_files.unlock()
2218
for repo in self._fallback_repositories:
2222
class RepositoryFormatPack(MetaDirRepositoryFormat):
2223
"""Format logic for pack structured repositories.
2225
This repository format has:
2226
- a list of packs in pack-names
2227
- packs in packs/NAME.pack
2228
- indices in indices/NAME.{iix,six,tix,rix}
2229
- knit deltas in the packs, knit indices mapped to the indices.
2230
- thunk objects to support the knits programming API.
2231
- a format marker of its own
2232
- an optional 'shared-storage' flag
2233
- an optional 'no-working-trees' flag
2237
# Set this attribute in derived classes to control the repository class
2238
# created by open and initialize.
2239
repository_class = None
2240
# Set this attribute in derived classes to control the
2241
# _commit_builder_class that the repository objects will have passed to
2242
# their constructor.
2243
_commit_builder_class = None
2244
# Set this attribute in derived clases to control the _serializer that the
2245
# repository objects will have passed to their constructor.
2247
# External references are not supported in pack repositories yet.
2248
supports_external_lookups = False
2249
# What index classes to use
2250
index_builder_class = None
2252
_fetch_uses_deltas = True
2254
def initialize(self, a_bzrdir, shared=False):
2255
"""Create a pack based repository.
2257
:param a_bzrdir: bzrdir to contain the new repository; must already
2259
:param shared: If true the repository will be initialized as a shared
2262
mutter('creating repository in %s.', a_bzrdir.transport.base)
2263
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
2264
builder = self.index_builder_class()
2265
files = [('pack-names', builder.finish())]
2266
utf8_files = [('format', self.get_format_string())]
2268
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
2269
return self.open(a_bzrdir=a_bzrdir, _found=True)
2271
def open(self, a_bzrdir, _found=False, _override_transport=None):
2272
"""See RepositoryFormat.open().
2274
:param _override_transport: INTERNAL USE ONLY. Allows opening the
2275
repository at a slightly different url
2276
than normal. I.e. during 'upgrade'.
2279
format = RepositoryFormat.find_format(a_bzrdir)
2280
if _override_transport is not None:
2281
repo_transport = _override_transport
2283
repo_transport = a_bzrdir.get_repository_transport(None)
2284
control_files = lockable_files.LockableFiles(repo_transport,
2285
'lock', lockdir.LockDir)
2286
return self.repository_class(_format=self,
2288
control_files=control_files,
2289
_commit_builder_class=self._commit_builder_class,
2290
_serializer=self._serializer)
2293
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2294
"""A no-subtrees parameterized Pack repository.
2296
This format was introduced in 0.92.
2299
repository_class = KnitPackRepository
2300
_commit_builder_class = PackCommitBuilder
2302
def _serializer(self):
2303
return xml5.serializer_v5
2304
# What index classes to use
2305
index_builder_class = InMemoryGraphIndex
2306
index_class = GraphIndex
2308
def _get_matching_bzrdir(self):
2309
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2311
def _ignore_setting_bzrdir(self, format):
2314
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2316
def get_format_string(self):
2317
"""See RepositoryFormat.get_format_string()."""
2318
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2320
def get_format_description(self):
2321
"""See RepositoryFormat.get_format_description()."""
2322
return "Packs containing knits without subtree support"
2324
def check_conversion_target(self, target_format):
2328
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2329
"""A subtrees parameterized Pack repository.
2331
This repository format uses the xml7 serializer to get:
2332
- support for recording full info about the tree root
2333
- support for recording tree-references
2335
This format was introduced in 0.92.
2338
repository_class = KnitPackRepository
2339
_commit_builder_class = PackRootCommitBuilder
2340
rich_root_data = True
2341
supports_tree_reference = True
2343
def _serializer(self):
2344
return xml7.serializer_v7
2345
# What index classes to use
2346
index_builder_class = InMemoryGraphIndex
2347
index_class = GraphIndex
2349
def _get_matching_bzrdir(self):
2350
return bzrdir.format_registry.make_bzrdir(
2351
'pack-0.92-subtree')
2353
def _ignore_setting_bzrdir(self, format):
2356
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2358
def check_conversion_target(self, target_format):
2359
if not target_format.rich_root_data:
2360
raise errors.BadConversionTarget(
2361
'Does not support rich root data.', target_format)
2362
if not getattr(target_format, 'supports_tree_reference', False):
2363
raise errors.BadConversionTarget(
2364
'Does not support nested trees', target_format)
2366
def get_format_string(self):
2367
"""See RepositoryFormat.get_format_string()."""
2368
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2370
def get_format_description(self):
2371
"""See RepositoryFormat.get_format_description()."""
2372
return "Packs containing knits with subtree support\n"
2375
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2376
"""A rich-root, no subtrees parameterized Pack repository.
2378
This repository format uses the xml6 serializer to get:
2379
- support for recording full info about the tree root
2381
This format was introduced in 1.0.
2384
repository_class = KnitPackRepository
2385
_commit_builder_class = PackRootCommitBuilder
2386
rich_root_data = True
2387
supports_tree_reference = False
2389
def _serializer(self):
2390
return xml6.serializer_v6
2391
# What index classes to use
2392
index_builder_class = InMemoryGraphIndex
2393
index_class = GraphIndex
2395
def _get_matching_bzrdir(self):
2396
return bzrdir.format_registry.make_bzrdir(
2399
def _ignore_setting_bzrdir(self, format):
2402
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2404
def check_conversion_target(self, target_format):
2405
if not target_format.rich_root_data:
2406
raise errors.BadConversionTarget(
2407
'Does not support rich root data.', target_format)
2409
def get_format_string(self):
2410
"""See RepositoryFormat.get_format_string()."""
2411
return ("Bazaar pack repository format 1 with rich root"
2412
" (needs bzr 1.0)\n")
2414
def get_format_description(self):
2415
"""See RepositoryFormat.get_format_description()."""
2416
return "Packs containing knits with rich root support\n"
2419
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2420
"""Repository that supports external references to allow stacking.
2424
Supports external lookups, which results in non-truncated ghosts after
2425
reconcile compared to pack-0.92 formats.
2428
repository_class = KnitPackRepository
2429
_commit_builder_class = PackCommitBuilder
2430
supports_external_lookups = True
2431
# What index classes to use
2432
index_builder_class = InMemoryGraphIndex
2433
index_class = GraphIndex
2436
def _serializer(self):
2437
return xml5.serializer_v5
2439
def _get_matching_bzrdir(self):
2440
return bzrdir.format_registry.make_bzrdir('1.6')
2442
def _ignore_setting_bzrdir(self, format):
2445
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2447
def get_format_string(self):
2448
"""See RepositoryFormat.get_format_string()."""
2449
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2451
def get_format_description(self):
2452
"""See RepositoryFormat.get_format_description()."""
2453
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2455
def check_conversion_target(self, target_format):
2459
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2460
"""A repository with rich roots and stacking.
2462
New in release 1.6.1.
2464
Supports stacking on other repositories, allowing data to be accessed
2465
without being stored locally.
2468
repository_class = KnitPackRepository
2469
_commit_builder_class = PackRootCommitBuilder
2470
rich_root_data = True
2471
supports_tree_reference = False # no subtrees
2472
supports_external_lookups = True
2473
# What index classes to use
2474
index_builder_class = InMemoryGraphIndex
2475
index_class = GraphIndex
2478
def _serializer(self):
2479
return xml6.serializer_v6
2481
def _get_matching_bzrdir(self):
2482
return bzrdir.format_registry.make_bzrdir(
2485
def _ignore_setting_bzrdir(self, format):
2488
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2490
def check_conversion_target(self, target_format):
2491
if not target_format.rich_root_data:
2492
raise errors.BadConversionTarget(
2493
'Does not support rich root data.', target_format)
2495
def get_format_string(self):
2496
"""See RepositoryFormat.get_format_string()."""
2497
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2499
def get_format_description(self):
2500
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2503
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2504
"""A repository with rich roots and external references.
2508
Supports external lookups, which results in non-truncated ghosts after
2509
reconcile compared to pack-0.92 formats.
2511
This format was deprecated because the serializer it uses accidentally
2512
supported subtrees, when the format was not intended to. This meant that
2513
someone could accidentally fetch from an incorrect repository.
2516
repository_class = KnitPackRepository
2517
_commit_builder_class = PackRootCommitBuilder
2518
rich_root_data = True
2519
supports_tree_reference = False # no subtrees
2521
supports_external_lookups = True
2522
# What index classes to use
2523
index_builder_class = InMemoryGraphIndex
2524
index_class = GraphIndex
2527
def _serializer(self):
2528
return xml7.serializer_v7
2530
def _get_matching_bzrdir(self):
2531
matching = bzrdir.format_registry.make_bzrdir(
2533
matching.repository_format = self
2536
def _ignore_setting_bzrdir(self, format):
2539
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2541
def check_conversion_target(self, target_format):
2542
if not target_format.rich_root_data:
2543
raise errors.BadConversionTarget(
2544
'Does not support rich root data.', target_format)
2546
def get_format_string(self):
2547
"""See RepositoryFormat.get_format_string()."""
2548
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2550
def get_format_description(self):
2551
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2555
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2556
"""A repository with stacking and btree indexes,
2557
without rich roots or subtrees.
2559
This is equivalent to pack-1.6 with B+Tree indices.
2562
repository_class = KnitPackRepository
2563
_commit_builder_class = PackCommitBuilder
2564
supports_external_lookups = True
2565
# What index classes to use
2566
index_builder_class = BTreeBuilder
2567
index_class = BTreeGraphIndex
2570
def _serializer(self):
2571
return xml5.serializer_v5
2573
def _get_matching_bzrdir(self):
2574
return bzrdir.format_registry.make_bzrdir('1.9')
2576
def _ignore_setting_bzrdir(self, format):
2579
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2581
def get_format_string(self):
2582
"""See RepositoryFormat.get_format_string()."""
2583
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2585
def get_format_description(self):
2586
"""See RepositoryFormat.get_format_description()."""
2587
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2589
def check_conversion_target(self, target_format):
2593
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2594
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2596
1.6-rich-root with B+Tree indices.
2599
repository_class = KnitPackRepository
2600
_commit_builder_class = PackRootCommitBuilder
2601
rich_root_data = True
2602
supports_tree_reference = False # no subtrees
2603
supports_external_lookups = True
2604
# What index classes to use
2605
index_builder_class = BTreeBuilder
2606
index_class = BTreeGraphIndex
2609
def _serializer(self):
2610
return xml6.serializer_v6
2612
def _get_matching_bzrdir(self):
2613
return bzrdir.format_registry.make_bzrdir(
2616
def _ignore_setting_bzrdir(self, format):
2619
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2621
def check_conversion_target(self, target_format):
2622
if not target_format.rich_root_data:
2623
raise errors.BadConversionTarget(
2624
'Does not support rich root data.', target_format)
2626
def get_format_string(self):
2627
"""See RepositoryFormat.get_format_string()."""
2628
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2630
def get_format_description(self):
2631
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2634
class RepositoryFormatPackDevelopment2(RepositoryFormatPack):
2635
"""A no-subtrees development repository.
2637
This format should be retained until the second release after bzr 1.7.
2639
This is pack-1.6.1 with B+Tree indices.
2642
repository_class = KnitPackRepository
2643
_commit_builder_class = PackCommitBuilder
2644
supports_external_lookups = True
2645
# What index classes to use
2646
index_builder_class = BTreeBuilder
2647
index_class = BTreeGraphIndex
2650
def _serializer(self):
2651
return xml5.serializer_v5
2653
def _get_matching_bzrdir(self):
2654
return bzrdir.format_registry.make_bzrdir('development2')
2656
def _ignore_setting_bzrdir(self, format):
2659
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2661
def get_format_string(self):
2662
"""See RepositoryFormat.get_format_string()."""
2663
return "Bazaar development format 2 (needs bzr.dev from before 1.8)\n"
2665
def get_format_description(self):
2666
"""See RepositoryFormat.get_format_description()."""
2667
return ("Development repository format, currently the same as "
2668
"1.6.1 with B+Trees.\n")
2670
def check_conversion_target(self, target_format):
2674
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2675
"""A subtrees development repository.
2677
This format should be retained until the second release after bzr 1.7.
2679
1.6.1-subtree[as it might have been] with B+Tree indices.
2682
repository_class = KnitPackRepository
2683
_commit_builder_class = PackRootCommitBuilder
2684
rich_root_data = True
2685
supports_tree_reference = True
2686
supports_external_lookups = True
2687
# What index classes to use
2688
index_builder_class = BTreeBuilder
2689
index_class = BTreeGraphIndex
2692
def _serializer(self):
2693
return xml7.serializer_v7
2695
def _get_matching_bzrdir(self):
2696
return bzrdir.format_registry.make_bzrdir(
2697
'development2-subtree')
2699
def _ignore_setting_bzrdir(self, format):
2702
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2704
def check_conversion_target(self, target_format):
2705
if not target_format.rich_root_data:
2706
raise errors.BadConversionTarget(
2707
'Does not support rich root data.', target_format)
2708
if not getattr(target_format, 'supports_tree_reference', False):
2709
raise errors.BadConversionTarget(
2710
'Does not support nested trees', target_format)
2712
def get_format_string(self):
2713
"""See RepositoryFormat.get_format_string()."""
2714
return ("Bazaar development format 2 with subtree support "
2715
"(needs bzr.dev from before 1.8)\n")
2717
def get_format_description(self):
2718
"""See RepositoryFormat.get_format_description()."""
2719
return ("Development repository format, currently the same as "
2720
"1.6.1-subtree with B+Tree indices.\n")
added
removed
bzrlib/repofmt/pack_repo.py
