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., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 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, flush_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
flush_func=flush_func)
548
self.add_callback = None
550
def replace_indices(self, index_to_pack, indices):
551
"""Replace the current mappings with fresh ones.
553
This should probably not be used eventually, rather incremental add and
554
removal of indices. It has been added during refactoring of existing
557
:param index_to_pack: A mapping from index objects to
558
(transport, name) tuples for the pack file data.
559
:param indices: A list of indices.
561
# refresh the revision pack map dict without replacing the instance.
562
self.index_to_pack.clear()
563
self.index_to_pack.update(index_to_pack)
564
# XXX: API break - clearly a 'replace' method would be good?
565
self.combined_index._indices[:] = indices
566
# the current add nodes callback for the current writable index if
568
self.add_callback = None
570
def add_index(self, index, pack):
571
"""Add index to the aggregate, which is an index for Pack pack.
573
Future searches on the aggregate index will seach this new index
574
before all previously inserted indices.
576
:param index: An Index for the pack.
577
:param pack: A Pack instance.
579
# expose it to the index map
580
self.index_to_pack[index] = pack.access_tuple()
581
# put it at the front of the linear index list
582
self.combined_index.insert_index(0, index)
584
def add_writable_index(self, index, pack):
585
"""Add an index which is able to have data added to it.
587
There can be at most one writable index at any time. Any
588
modifications made to the knit are put into this index.
590
:param index: An index from the pack parameter.
591
:param pack: A Pack instance.
593
if self.add_callback is not None:
594
raise AssertionError(
595
"%s already has a writable index through %s" % \
596
(self, self.add_callback))
597
# allow writing: queue writes to a new index
598
self.add_index(index, pack)
599
# Updates the index to packs mapping as a side effect,
600
self.data_access.set_writer(pack._writer, index, pack.access_tuple())
601
self.add_callback = index.add_nodes
604
"""Reset all the aggregate data to nothing."""
605
self.data_access.set_writer(None, None, (None, None))
606
self.index_to_pack.clear()
607
del self.combined_index._indices[:]
608
self.add_callback = None
610
def remove_index(self, index, pack):
611
"""Remove index from the indices used to answer queries.
613
:param index: An index from the pack parameter.
614
:param pack: A Pack instance.
616
del self.index_to_pack[index]
617
self.combined_index._indices.remove(index)
618
if (self.add_callback is not None and
619
getattr(index, 'add_nodes', None) == self.add_callback):
620
self.add_callback = None
621
self.data_access.set_writer(None, None, (None, None))
624
class Packer(object):
625
"""Create a pack from packs."""
627
def __init__(self, pack_collection, packs, suffix, revision_ids=None,
631
:param pack_collection: A RepositoryPackCollection object where the
632
new pack is being written to.
633
:param packs: The packs to combine.
634
:param suffix: The suffix to use on the temporary files for the pack.
635
:param revision_ids: Revision ids to limit the pack to.
636
:param reload_func: A function to call if a pack file/index goes
637
missing. The side effect of calling this function should be to
638
update self.packs. See also AggregateIndex
642
self.revision_ids = revision_ids
643
# The pack object we are creating.
645
self._pack_collection = pack_collection
646
self._reload_func = reload_func
647
# The index layer keys for the revisions being copied. None for 'all
649
self._revision_keys = None
650
# What text keys to copy. None for 'all texts'. This is set by
651
# _copy_inventory_texts
652
self._text_filter = None
655
def _extra_init(self):
656
"""A template hook to allow extending the constructor trivially."""
658
def _pack_map_and_index_list(self, index_attribute):
659
"""Convert a list of packs to an index pack map and index list.
661
:param index_attribute: The attribute that the desired index is found
663
:return: A tuple (map, list) where map contains the dict from
664
index:pack_tuple, and list contains the indices in the preferred
669
for pack_obj in self.packs:
670
index = getattr(pack_obj, index_attribute)
671
indices.append(index)
672
pack_map[index] = pack_obj
673
return pack_map, indices
675
def _index_contents(self, indices, key_filter=None):
676
"""Get an iterable of the index contents from a pack_map.
678
:param indices: The list of indices to query
679
:param key_filter: An optional filter to limit the keys returned.
681
all_index = CombinedGraphIndex(indices)
682
if key_filter is None:
683
return all_index.iter_all_entries()
685
return all_index.iter_entries(key_filter)
687
def pack(self, pb=None):
688
"""Create a new pack by reading data from other packs.
690
This does little more than a bulk copy of data. One key difference
691
is that data with the same item key across multiple packs is elided
692
from the output. The new pack is written into the current pack store
693
along with its indices, and the name added to the pack names. The
694
source packs are not altered and are not required to be in the current
697
:param pb: An optional progress bar to use. A nested bar is created if
699
:return: A Pack object, or None if nothing was copied.
701
# open a pack - using the same name as the last temporary file
702
# - which has already been flushed, so its safe.
703
# XXX: - duplicate code warning with start_write_group; fix before
704
# considering 'done'.
705
if self._pack_collection._new_pack is not None:
706
raise errors.BzrError('call to %s.pack() while another pack is'
708
% (self.__class__.__name__,))
709
if self.revision_ids is not None:
710
if len(self.revision_ids) == 0:
711
# silly fetch request.
714
self.revision_ids = frozenset(self.revision_ids)
715
self.revision_keys = frozenset((revid,) for revid in
718
self.pb = ui.ui_factory.nested_progress_bar()
722
return self._create_pack_from_packs()
728
"""Open a pack for the pack we are creating."""
729
new_pack = NewPack(self._pack_collection, upload_suffix=self.suffix,
730
file_mode=self._pack_collection.repo.bzrdir._get_file_mode())
731
# We know that we will process all nodes in order, and don't need to
732
# query, so don't combine any indices spilled to disk until we are done
733
new_pack.revision_index.set_optimize(combine_backing_indices=False)
734
new_pack.inventory_index.set_optimize(combine_backing_indices=False)
735
new_pack.text_index.set_optimize(combine_backing_indices=False)
736
new_pack.signature_index.set_optimize(combine_backing_indices=False)
739
def _update_pack_order(self, entries, index_to_pack_map):
740
"""Determine how we want our packs to be ordered.
742
This changes the sort order of the self.packs list so that packs unused
743
by 'entries' will be at the end of the list, so that future requests
744
can avoid probing them. Used packs will be at the front of the
745
self.packs list, in the order of their first use in 'entries'.
747
:param entries: A list of (index, ...) tuples
748
:param index_to_pack_map: A mapping from index objects to pack objects.
752
for entry in entries:
754
if index not in seen_indexes:
755
packs.append(index_to_pack_map[index])
756
seen_indexes.add(index)
757
if len(packs) == len(self.packs):
758
if 'pack' in debug.debug_flags:
759
mutter('Not changing pack list, all packs used.')
761
seen_packs = set(packs)
762
for pack in self.packs:
763
if pack not in seen_packs:
766
if 'pack' in debug.debug_flags:
767
old_names = [p.access_tuple()[1] for p in self.packs]
768
new_names = [p.access_tuple()[1] for p in packs]
769
mutter('Reordering packs\nfrom: %s\n to: %s',
770
old_names, new_names)
773
def _copy_revision_texts(self):
774
"""Copy revision data to the new pack."""
776
if self.revision_ids:
777
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
780
# select revision keys
781
revision_index_map, revision_indices = self._pack_map_and_index_list(
783
revision_nodes = self._index_contents(revision_indices, revision_keys)
784
revision_nodes = list(revision_nodes)
785
self._update_pack_order(revision_nodes, revision_index_map)
786
# copy revision keys and adjust values
787
self.pb.update("Copying revision texts", 1)
788
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
789
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
790
self.new_pack.revision_index, readv_group_iter, total_items))
791
if 'pack' in debug.debug_flags:
792
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
793
time.ctime(), self._pack_collection._upload_transport.base,
794
self.new_pack.random_name,
795
self.new_pack.revision_index.key_count(),
796
time.time() - self.new_pack.start_time)
797
self._revision_keys = revision_keys
799
def _copy_inventory_texts(self):
800
"""Copy the inventory texts to the new pack.
802
self._revision_keys is used to determine what inventories to copy.
804
Sets self._text_filter appropriately.
806
# select inventory keys
807
inv_keys = self._revision_keys # currently the same keyspace, and note that
808
# querying for keys here could introduce a bug where an inventory item
809
# is missed, so do not change it to query separately without cross
810
# checking like the text key check below.
811
inventory_index_map, inventory_indices = self._pack_map_and_index_list(
813
inv_nodes = self._index_contents(inventory_indices, inv_keys)
814
# copy inventory keys and adjust values
815
# XXX: Should be a helper function to allow different inv representation
817
self.pb.update("Copying inventory texts", 2)
818
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
819
# Only grab the output lines if we will be processing them
820
output_lines = bool(self.revision_ids)
821
inv_lines = self._copy_nodes_graph(inventory_index_map,
822
self.new_pack._writer, self.new_pack.inventory_index,
823
readv_group_iter, total_items, output_lines=output_lines)
824
if self.revision_ids:
825
self._process_inventory_lines(inv_lines)
827
# eat the iterator to cause it to execute.
829
self._text_filter = None
830
if 'pack' in debug.debug_flags:
831
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
832
time.ctime(), self._pack_collection._upload_transport.base,
833
self.new_pack.random_name,
834
self.new_pack.inventory_index.key_count(),
835
time.time() - self.new_pack.start_time)
837
def _copy_text_texts(self):
839
text_index_map, text_nodes = self._get_text_nodes()
840
if self._text_filter is not None:
841
# We could return the keys copied as part of the return value from
842
# _copy_nodes_graph but this doesn't work all that well with the
843
# need to get line output too, so we check separately, and as we're
844
# going to buffer everything anyway, we check beforehand, which
845
# saves reading knit data over the wire when we know there are
847
text_nodes = set(text_nodes)
848
present_text_keys = set(_node[1] for _node in text_nodes)
849
missing_text_keys = set(self._text_filter) - present_text_keys
850
if missing_text_keys:
851
# TODO: raise a specific error that can handle many missing
853
mutter("missing keys during fetch: %r", missing_text_keys)
854
a_missing_key = missing_text_keys.pop()
855
raise errors.RevisionNotPresent(a_missing_key[1],
857
# copy text keys and adjust values
858
self.pb.update("Copying content texts", 3)
859
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
860
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
861
self.new_pack.text_index, readv_group_iter, total_items))
862
self._log_copied_texts()
864
def _create_pack_from_packs(self):
865
self.pb.update("Opening pack", 0, 5)
866
self.new_pack = self.open_pack()
867
new_pack = self.new_pack
868
# buffer data - we won't be reading-back during the pack creation and
869
# this makes a significant difference on sftp pushes.
870
new_pack.set_write_cache_size(1024*1024)
871
if 'pack' in debug.debug_flags:
872
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
873
for a_pack in self.packs]
874
if self.revision_ids is not None:
875
rev_count = len(self.revision_ids)
878
mutter('%s: create_pack: creating pack from source packs: '
879
'%s%s %s revisions wanted %s t=0',
880
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
881
plain_pack_list, rev_count)
882
self._copy_revision_texts()
883
self._copy_inventory_texts()
884
self._copy_text_texts()
885
# select signature keys
886
signature_filter = self._revision_keys # same keyspace
887
signature_index_map, signature_indices = self._pack_map_and_index_list(
889
signature_nodes = self._index_contents(signature_indices,
891
# copy signature keys and adjust values
892
self.pb.update("Copying signature texts", 4)
893
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
894
new_pack.signature_index)
895
if 'pack' in debug.debug_flags:
896
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
897
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
898
new_pack.signature_index.key_count(),
899
time.time() - new_pack.start_time)
900
new_pack._check_references()
901
if not self._use_pack(new_pack):
904
self.pb.update("Finishing pack", 5)
906
self._pack_collection.allocate(new_pack)
909
def _copy_nodes(self, nodes, index_map, writer, write_index):
910
"""Copy knit nodes between packs with no graph references."""
911
pb = ui.ui_factory.nested_progress_bar()
913
return self._do_copy_nodes(nodes, index_map, writer,
918
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb):
919
# for record verification
920
knit = KnitVersionedFiles(None, None)
921
# plan a readv on each source pack:
923
nodes = sorted(nodes)
924
# how to map this into knit.py - or knit.py into this?
925
# we don't want the typical knit logic, we want grouping by pack
926
# at this point - perhaps a helper library for the following code
927
# duplication points?
929
for index, key, value in nodes:
930
if index not in request_groups:
931
request_groups[index] = []
932
request_groups[index].append((key, value))
934
pb.update("Copied record", record_index, len(nodes))
935
for index, items in request_groups.iteritems():
936
pack_readv_requests = []
937
for key, value in items:
938
# ---- KnitGraphIndex.get_position
939
bits = value[1:].split(' ')
940
offset, length = int(bits[0]), int(bits[1])
941
pack_readv_requests.append((offset, length, (key, value[0])))
942
# linear scan up the pack
943
pack_readv_requests.sort()
945
pack_obj = index_map[index]
946
transport, path = pack_obj.access_tuple()
948
reader = pack.make_readv_reader(transport, path,
949
[offset[0:2] for offset in pack_readv_requests])
950
except errors.NoSuchFile:
951
if self._reload_func is not None:
954
for (names, read_func), (_1, _2, (key, eol_flag)) in \
955
izip(reader.iter_records(), pack_readv_requests):
956
raw_data = read_func(None)
957
# check the header only
958
df, _ = knit._parse_record_header(key, raw_data)
960
pos, size = writer.add_bytes_record(raw_data, names)
961
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
962
pb.update("Copied record", record_index)
965
def _copy_nodes_graph(self, index_map, writer, write_index,
966
readv_group_iter, total_items, output_lines=False):
967
"""Copy knit nodes between packs.
969
:param output_lines: Return lines present in the copied data as
970
an iterator of line,version_id.
972
pb = ui.ui_factory.nested_progress_bar()
974
for result in self._do_copy_nodes_graph(index_map, writer,
975
write_index, output_lines, pb, readv_group_iter, total_items):
978
# Python 2.4 does not permit try:finally: in a generator.
984
def _do_copy_nodes_graph(self, index_map, writer, write_index,
985
output_lines, pb, readv_group_iter, total_items):
986
# for record verification
987
knit = KnitVersionedFiles(None, None)
988
# for line extraction when requested (inventories only)
990
factory = KnitPlainFactory()
992
pb.update("Copied record", record_index, total_items)
993
for index, readv_vector, node_vector in readv_group_iter:
995
pack_obj = index_map[index]
996
transport, path = pack_obj.access_tuple()
998
reader = pack.make_readv_reader(transport, path, readv_vector)
999
except errors.NoSuchFile:
1000
if self._reload_func is not None:
1003
for (names, read_func), (key, eol_flag, references) in \
1004
izip(reader.iter_records(), node_vector):
1005
raw_data = read_func(None)
1007
# read the entire thing
1008
content, _ = knit._parse_record(key[-1], raw_data)
1009
if len(references[-1]) == 0:
1010
line_iterator = factory.get_fulltext_content(content)
1012
line_iterator = factory.get_linedelta_content(content)
1013
for line in line_iterator:
1016
# check the header only
1017
df, _ = knit._parse_record_header(key, raw_data)
1019
pos, size = writer.add_bytes_record(raw_data, names)
1020
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1021
pb.update("Copied record", record_index)
1024
def _get_text_nodes(self):
1025
text_index_map, text_indices = self._pack_map_and_index_list(
1027
return text_index_map, self._index_contents(text_indices,
1030
def _least_readv_node_readv(self, nodes):
1031
"""Generate request groups for nodes using the least readv's.
1033
:param nodes: An iterable of graph index nodes.
1034
:return: Total node count and an iterator of the data needed to perform
1035
readvs to obtain the data for nodes. Each item yielded by the
1036
iterator is a tuple with:
1037
index, readv_vector, node_vector. readv_vector is a list ready to
1038
hand to the transport readv method, and node_vector is a list of
1039
(key, eol_flag, references) for the the node retrieved by the
1040
matching readv_vector.
1042
# group by pack so we do one readv per pack
1043
nodes = sorted(nodes)
1046
for index, key, value, references in nodes:
1047
if index not in request_groups:
1048
request_groups[index] = []
1049
request_groups[index].append((key, value, references))
1051
for index, items in request_groups.iteritems():
1052
pack_readv_requests = []
1053
for key, value, references in items:
1054
# ---- KnitGraphIndex.get_position
1055
bits = value[1:].split(' ')
1056
offset, length = int(bits[0]), int(bits[1])
1057
pack_readv_requests.append(
1058
((offset, length), (key, value[0], references)))
1059
# linear scan up the pack to maximum range combining.
1060
pack_readv_requests.sort()
1061
# split out the readv and the node data.
1062
pack_readv = [readv for readv, node in pack_readv_requests]
1063
node_vector = [node for readv, node in pack_readv_requests]
1064
result.append((index, pack_readv, node_vector))
1065
return total, result
1067
def _log_copied_texts(self):
1068
if 'pack' in debug.debug_flags:
1069
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
1070
time.ctime(), self._pack_collection._upload_transport.base,
1071
self.new_pack.random_name,
1072
self.new_pack.text_index.key_count(),
1073
time.time() - self.new_pack.start_time)
1075
def _process_inventory_lines(self, inv_lines):
1076
"""Use up the inv_lines generator and setup a text key filter."""
1077
repo = self._pack_collection.repo
1078
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
1079
inv_lines, self.revision_keys)
1081
for fileid, file_revids in fileid_revisions.iteritems():
1082
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
1083
self._text_filter = text_filter
1085
def _revision_node_readv(self, revision_nodes):
1086
"""Return the total revisions and the readv's to issue.
1088
:param revision_nodes: The revision index contents for the packs being
1089
incorporated into the new pack.
1090
:return: As per _least_readv_node_readv.
1092
return self._least_readv_node_readv(revision_nodes)
1094
def _use_pack(self, new_pack):
1095
"""Return True if new_pack should be used.
1097
:param new_pack: The pack that has just been created.
1098
:return: True if the pack should be used.
1100
return new_pack.data_inserted()
1103
class OptimisingPacker(Packer):
1104
"""A packer which spends more time to create better disk layouts."""
1106
def _revision_node_readv(self, revision_nodes):
1107
"""Return the total revisions and the readv's to issue.
1109
This sort places revisions in topological order with the ancestors
1112
:param revision_nodes: The revision index contents for the packs being
1113
incorporated into the new pack.
1114
:return: As per _least_readv_node_readv.
1116
# build an ancestors dict
1119
for index, key, value, references in revision_nodes:
1120
ancestors[key] = references[0]
1121
by_key[key] = (index, value, references)
1122
order = tsort.topo_sort(ancestors)
1124
# Single IO is pathological, but it will work as a starting point.
1126
for key in reversed(order):
1127
index, value, references = by_key[key]
1128
# ---- KnitGraphIndex.get_position
1129
bits = value[1:].split(' ')
1130
offset, length = int(bits[0]), int(bits[1])
1132
(index, [(offset, length)], [(key, value[0], references)]))
1133
# TODO: combine requests in the same index that are in ascending order.
1134
return total, requests
1136
def open_pack(self):
1137
"""Open a pack for the pack we are creating."""
1138
new_pack = super(OptimisingPacker, self).open_pack()
1139
# Turn on the optimization flags for all the index builders.
1140
new_pack.revision_index.set_optimize(for_size=True)
1141
new_pack.inventory_index.set_optimize(for_size=True)
1142
new_pack.text_index.set_optimize(for_size=True)
1143
new_pack.signature_index.set_optimize(for_size=True)
1147
class ReconcilePacker(Packer):
1148
"""A packer which regenerates indices etc as it copies.
1150
This is used by ``bzr reconcile`` to cause parent text pointers to be
1154
def _extra_init(self):
1155
self._data_changed = False
1157
def _process_inventory_lines(self, inv_lines):
1158
"""Generate a text key reference map rather for reconciling with."""
1159
repo = self._pack_collection.repo
1160
refs = repo._find_text_key_references_from_xml_inventory_lines(
1162
self._text_refs = refs
1163
# during reconcile we:
1164
# - convert unreferenced texts to full texts
1165
# - correct texts which reference a text not copied to be full texts
1166
# - copy all others as-is but with corrected parents.
1167
# - so at this point we don't know enough to decide what becomes a full
1169
self._text_filter = None
1171
def _copy_text_texts(self):
1172
"""generate what texts we should have and then copy."""
1173
self.pb.update("Copying content texts", 3)
1174
# we have three major tasks here:
1175
# 1) generate the ideal index
1176
repo = self._pack_collection.repo
1177
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1178
_1, key, _2, refs in
1179
self.new_pack.revision_index.iter_all_entries()])
1180
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1181
# 2) generate a text_nodes list that contains all the deltas that can
1182
# be used as-is, with corrected parents.
1185
discarded_nodes = []
1186
NULL_REVISION = _mod_revision.NULL_REVISION
1187
text_index_map, text_nodes = self._get_text_nodes()
1188
for node in text_nodes:
1194
ideal_parents = tuple(ideal_index[node[1]])
1196
discarded_nodes.append(node)
1197
self._data_changed = True
1199
if ideal_parents == (NULL_REVISION,):
1201
if ideal_parents == node[3][0]:
1203
ok_nodes.append(node)
1204
elif ideal_parents[0:1] == node[3][0][0:1]:
1205
# the left most parent is the same, or there are no parents
1206
# today. Either way, we can preserve the representation as
1207
# long as we change the refs to be inserted.
1208
self._data_changed = True
1209
ok_nodes.append((node[0], node[1], node[2],
1210
(ideal_parents, node[3][1])))
1211
self._data_changed = True
1213
# Reinsert this text completely
1214
bad_texts.append((node[1], ideal_parents))
1215
self._data_changed = True
1216
# we're finished with some data.
1219
# 3) bulk copy the ok data
1220
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1221
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1222
self.new_pack.text_index, readv_group_iter, total_items))
1223
# 4) adhoc copy all the other texts.
1224
# We have to topologically insert all texts otherwise we can fail to
1225
# reconcile when parts of a single delta chain are preserved intact,
1226
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1227
# reinserted, and if d3 has incorrect parents it will also be
1228
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1229
# copied), so we will try to delta, but d2 is not currently able to be
1230
# extracted because it's basis d1 is not present. Topologically sorting
1231
# addresses this. The following generates a sort for all the texts that
1232
# are being inserted without having to reference the entire text key
1233
# space (we only topo sort the revisions, which is smaller).
1234
topo_order = tsort.topo_sort(ancestors)
1235
rev_order = dict(zip(topo_order, range(len(topo_order))))
1236
bad_texts.sort(key=lambda key:rev_order[key[0][1]])
1237
transaction = repo.get_transaction()
1238
file_id_index = GraphIndexPrefixAdapter(
1239
self.new_pack.text_index,
1241
add_nodes_callback=self.new_pack.text_index.add_nodes)
1242
data_access = _DirectPackAccess(
1243
{self.new_pack.text_index:self.new_pack.access_tuple()})
1244
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1245
self.new_pack.access_tuple())
1246
output_texts = KnitVersionedFiles(
1247
_KnitGraphIndex(self.new_pack.text_index,
1248
add_callback=self.new_pack.text_index.add_nodes,
1249
deltas=True, parents=True, is_locked=repo.is_locked),
1250
data_access=data_access, max_delta_chain=200)
1251
for key, parent_keys in bad_texts:
1252
# We refer to the new pack to delta data being output.
1253
# A possible improvement would be to catch errors on short reads
1254
# and only flush then.
1255
self.new_pack.flush()
1257
for parent_key in parent_keys:
1258
if parent_key[0] != key[0]:
1259
# Graph parents must match the fileid
1260
raise errors.BzrError('Mismatched key parent %r:%r' %
1262
parents.append(parent_key[1])
1263
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1264
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1265
output_texts.add_lines(key, parent_keys, text_lines,
1266
random_id=True, check_content=False)
1267
# 5) check that nothing inserted has a reference outside the keyspace.
1268
missing_text_keys = self.new_pack.text_index._external_references()
1269
if missing_text_keys:
1270
raise errors.BzrCheckError('Reference to missing compression parents %r'
1271
% (missing_text_keys,))
1272
self._log_copied_texts()
1274
def _use_pack(self, new_pack):
1275
"""Override _use_pack to check for reconcile having changed content."""
1276
# XXX: we might be better checking this at the copy time.
1277
original_inventory_keys = set()
1278
inv_index = self._pack_collection.inventory_index.combined_index
1279
for entry in inv_index.iter_all_entries():
1280
original_inventory_keys.add(entry[1])
1281
new_inventory_keys = set()
1282
for entry in new_pack.inventory_index.iter_all_entries():
1283
new_inventory_keys.add(entry[1])
1284
if new_inventory_keys != original_inventory_keys:
1285
self._data_changed = True
1286
return new_pack.data_inserted() and self._data_changed
1289
class RepositoryPackCollection(object):
1290
"""Management of packs within a repository.
1292
:ivar _names: map of {pack_name: (index_size,)}
1295
def __init__(self, repo, transport, index_transport, upload_transport,
1296
pack_transport, index_builder_class, index_class):
1297
"""Create a new RepositoryPackCollection.
1299
:param transport: Addresses the repository base directory
1300
(typically .bzr/repository/).
1301
:param index_transport: Addresses the directory containing indices.
1302
:param upload_transport: Addresses the directory into which packs are written
1303
while they're being created.
1304
:param pack_transport: Addresses the directory of existing complete packs.
1305
:param index_builder_class: The index builder class to use.
1306
:param index_class: The index class to use.
1308
# XXX: This should call self.reset()
1310
self.transport = transport
1311
self._index_transport = index_transport
1312
self._upload_transport = upload_transport
1313
self._pack_transport = pack_transport
1314
self._index_builder_class = index_builder_class
1315
self._index_class = index_class
1316
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
1320
self._packs_by_name = {}
1321
# the previous pack-names content
1322
self._packs_at_load = None
1323
# when a pack is being created by this object, the state of that pack.
1324
self._new_pack = None
1325
# aggregated revision index data
1326
flush = self._flush_new_pack
1327
self.revision_index = AggregateIndex(self.reload_pack_names, flush)
1328
self.inventory_index = AggregateIndex(self.reload_pack_names, flush)
1329
self.text_index = AggregateIndex(self.reload_pack_names, flush)
1330
self.signature_index = AggregateIndex(self.reload_pack_names, flush)
1332
self._resumed_packs = []
1334
def add_pack_to_memory(self, pack):
1335
"""Make a Pack object available to the repository to satisfy queries.
1337
:param pack: A Pack object.
1339
if pack.name in self._packs_by_name:
1340
raise AssertionError(
1341
'pack %s already in _packs_by_name' % (pack.name,))
1342
self.packs.append(pack)
1343
self._packs_by_name[pack.name] = pack
1344
self.revision_index.add_index(pack.revision_index, pack)
1345
self.inventory_index.add_index(pack.inventory_index, pack)
1346
self.text_index.add_index(pack.text_index, pack)
1347
self.signature_index.add_index(pack.signature_index, pack)
1349
def all_packs(self):
1350
"""Return a list of all the Pack objects this repository has.
1352
Note that an in-progress pack being created is not returned.
1354
:return: A list of Pack objects for all the packs in the repository.
1357
for name in self.names():
1358
result.append(self.get_pack_by_name(name))
1362
"""Pack the pack collection incrementally.
1364
This will not attempt global reorganisation or recompression,
1365
rather it will just ensure that the total number of packs does
1366
not grow without bound. It uses the _max_pack_count method to
1367
determine if autopacking is needed, and the pack_distribution
1368
method to determine the number of revisions in each pack.
1370
If autopacking takes place then the packs name collection will have
1371
been flushed to disk - packing requires updating the name collection
1372
in synchronisation with certain steps. Otherwise the names collection
1375
:return: True if packing took place.
1379
return self._do_autopack()
1380
except errors.RetryAutopack, e:
1381
# If we get a RetryAutopack exception, we should abort the
1382
# current action, and retry.
1385
def _do_autopack(self):
1386
# XXX: Should not be needed when the management of indices is sane.
1387
total_revisions = self.revision_index.combined_index.key_count()
1388
total_packs = len(self._names)
1389
if self._max_pack_count(total_revisions) >= total_packs:
1391
# XXX: the following may want to be a class, to pack with a given
1393
# determine which packs need changing
1394
pack_distribution = self.pack_distribution(total_revisions)
1396
for pack in self.all_packs():
1397
revision_count = pack.get_revision_count()
1398
if revision_count == 0:
1399
# revision less packs are not generated by normal operation,
1400
# only by operations like sign-my-commits, and thus will not
1401
# tend to grow rapdily or without bound like commit containing
1402
# packs do - leave them alone as packing them really should
1403
# group their data with the relevant commit, and that may
1404
# involve rewriting ancient history - which autopack tries to
1405
# avoid. Alternatively we could not group the data but treat
1406
# each of these as having a single revision, and thus add
1407
# one revision for each to the total revision count, to get
1408
# a matching distribution.
1410
existing_packs.append((revision_count, pack))
1411
pack_operations = self.plan_autopack_combinations(
1412
existing_packs, pack_distribution)
1413
num_new_packs = len(pack_operations)
1414
num_old_packs = sum([len(po[1]) for po in pack_operations])
1415
num_revs_affected = sum([po[0] for po in pack_operations])
1416
mutter('Auto-packing repository %s, which has %d pack files, '
1417
'containing %d revisions. Packing %d files into %d affecting %d'
1418
' revisions', self, total_packs, total_revisions, num_old_packs,
1419
num_new_packs, num_revs_affected)
1420
self._execute_pack_operations(pack_operations,
1421
reload_func=self._restart_autopack)
1424
def _execute_pack_operations(self, pack_operations, _packer_class=Packer,
1426
"""Execute a series of pack operations.
1428
:param pack_operations: A list of [revision_count, packs_to_combine].
1429
:param _packer_class: The class of packer to use (default: Packer).
1432
for revision_count, packs in pack_operations:
1433
# we may have no-ops from the setup logic
1436
packer = _packer_class(self, packs, '.autopack',
1437
reload_func=reload_func)
1440
except errors.RetryWithNewPacks:
1441
# An exception is propagating out of this context, make sure
1442
# this packer has cleaned up. Packer() doesn't set its new_pack
1443
# state into the RepositoryPackCollection object, so we only
1444
# have access to it directly here.
1445
if packer.new_pack is not None:
1446
packer.new_pack.abort()
1449
self._remove_pack_from_memory(pack)
1450
# record the newly available packs and stop advertising the old
1452
self._save_pack_names(clear_obsolete_packs=True)
1453
# Move the old packs out of the way now they are no longer referenced.
1454
for revision_count, packs in pack_operations:
1455
self._obsolete_packs(packs)
1457
def _flush_new_pack(self):
1458
if self._new_pack is not None:
1459
self._new_pack.flush()
1461
def lock_names(self):
1462
"""Acquire the mutex around the pack-names index.
1464
This cannot be used in the middle of a read-only transaction on the
1467
self.repo.control_files.lock_write()
1470
"""Pack the pack collection totally."""
1471
self.ensure_loaded()
1472
total_packs = len(self._names)
1474
# This is arguably wrong because we might not be optimal, but for
1475
# now lets leave it in. (e.g. reconcile -> one pack. But not
1478
total_revisions = self.revision_index.combined_index.key_count()
1479
# XXX: the following may want to be a class, to pack with a given
1481
mutter('Packing repository %s, which has %d pack files, '
1482
'containing %d revisions into 1 packs.', self, total_packs,
1484
# determine which packs need changing
1485
pack_distribution = [1]
1486
pack_operations = [[0, []]]
1487
for pack in self.all_packs():
1488
pack_operations[-1][0] += pack.get_revision_count()
1489
pack_operations[-1][1].append(pack)
1490
self._execute_pack_operations(pack_operations, OptimisingPacker)
1492
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1493
"""Plan a pack operation.
1495
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1497
:param pack_distribution: A list with the number of revisions desired
1500
if len(existing_packs) <= len(pack_distribution):
1502
existing_packs.sort(reverse=True)
1503
pack_operations = [[0, []]]
1504
# plan out what packs to keep, and what to reorganise
1505
while len(existing_packs):
1506
# take the largest pack, and if its less than the head of the
1507
# distribution chart we will include its contents in the new pack
1508
# for that position. If its larger, we remove its size from the
1509
# distribution chart
1510
next_pack_rev_count, next_pack = existing_packs.pop(0)
1511
if next_pack_rev_count >= pack_distribution[0]:
1512
# this is already packed 'better' than this, so we can
1513
# not waste time packing it.
1514
while next_pack_rev_count > 0:
1515
next_pack_rev_count -= pack_distribution[0]
1516
if next_pack_rev_count >= 0:
1518
del pack_distribution[0]
1520
# didn't use that entire bucket up
1521
pack_distribution[0] = -next_pack_rev_count
1523
# add the revisions we're going to add to the next output pack
1524
pack_operations[-1][0] += next_pack_rev_count
1525
# allocate this pack to the next pack sub operation
1526
pack_operations[-1][1].append(next_pack)
1527
if pack_operations[-1][0] >= pack_distribution[0]:
1528
# this pack is used up, shift left.
1529
del pack_distribution[0]
1530
pack_operations.append([0, []])
1531
# Now that we know which pack files we want to move, shove them all
1532
# into a single pack file.
1534
final_pack_list = []
1535
for num_revs, pack_files in pack_operations:
1536
final_rev_count += num_revs
1537
final_pack_list.extend(pack_files)
1538
if len(final_pack_list) == 1:
1539
raise AssertionError('We somehow generated an autopack with a'
1540
' single pack file being moved.')
1542
return [[final_rev_count, final_pack_list]]
1544
def ensure_loaded(self):
1545
"""Ensure we have read names from disk.
1547
:return: True if the disk names had not been previously read.
1549
# NB: if you see an assertion error here, its probably access against
1550
# an unlocked repo. Naughty.
1551
if not self.repo.is_locked():
1552
raise errors.ObjectNotLocked(self.repo)
1553
if self._names is None:
1555
self._packs_at_load = set()
1556
for index, key, value in self._iter_disk_pack_index():
1558
self._names[name] = self._parse_index_sizes(value)
1559
self._packs_at_load.add((key, value))
1563
# populate all the metadata.
1567
def _parse_index_sizes(self, value):
1568
"""Parse a string of index sizes."""
1569
return tuple([int(digits) for digits in value.split(' ')])
1571
def get_pack_by_name(self, name):
1572
"""Get a Pack object by name.
1574
:param name: The name of the pack - e.g. '123456'
1575
:return: A Pack object.
1578
return self._packs_by_name[name]
1580
rev_index = self._make_index(name, '.rix')
1581
inv_index = self._make_index(name, '.iix')
1582
txt_index = self._make_index(name, '.tix')
1583
sig_index = self._make_index(name, '.six')
1584
result = ExistingPack(self._pack_transport, name, rev_index,
1585
inv_index, txt_index, sig_index)
1586
self.add_pack_to_memory(result)
1589
def _resume_pack(self, name):
1590
"""Get a suspended Pack object by name.
1592
:param name: The name of the pack - e.g. '123456'
1593
:return: A Pack object.
1595
if not re.match('[a-f0-9]{32}', name):
1596
# Tokens should be md5sums of the suspended pack file, i.e. 32 hex
1598
raise errors.UnresumableWriteGroup(
1599
self.repo, [name], 'Malformed write group token')
1601
rev_index = self._make_index(name, '.rix', resume=True)
1602
inv_index = self._make_index(name, '.iix', resume=True)
1603
txt_index = self._make_index(name, '.tix', resume=True)
1604
sig_index = self._make_index(name, '.six', resume=True)
1605
result = ResumedPack(name, rev_index, inv_index, txt_index,
1606
sig_index, self._upload_transport, self._pack_transport,
1607
self._index_transport, self)
1608
except errors.NoSuchFile, e:
1609
raise errors.UnresumableWriteGroup(self.repo, [name], str(e))
1610
self.add_pack_to_memory(result)
1611
self._resumed_packs.append(result)
1614
def allocate(self, a_new_pack):
1615
"""Allocate name in the list of packs.
1617
:param a_new_pack: A NewPack instance to be added to the collection of
1618
packs for this repository.
1620
self.ensure_loaded()
1621
if a_new_pack.name in self._names:
1622
raise errors.BzrError(
1623
'Pack %r already exists in %s' % (a_new_pack.name, self))
1624
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1625
self.add_pack_to_memory(a_new_pack)
1627
def _iter_disk_pack_index(self):
1628
"""Iterate over the contents of the pack-names index.
1630
This is used when loading the list from disk, and before writing to
1631
detect updates from others during our write operation.
1632
:return: An iterator of the index contents.
1634
return self._index_class(self.transport, 'pack-names', None
1635
).iter_all_entries()
1637
def _make_index(self, name, suffix, resume=False):
1638
size_offset = self._suffix_offsets[suffix]
1639
index_name = name + suffix
1641
transport = self._upload_transport
1642
index_size = transport.stat(index_name).st_size
1644
transport = self._index_transport
1645
index_size = self._names[name][size_offset]
1646
return self._index_class(transport, index_name, index_size)
1648
def _max_pack_count(self, total_revisions):
1649
"""Return the maximum number of packs to use for total revisions.
1651
:param total_revisions: The total number of revisions in the
1654
if not total_revisions:
1656
digits = str(total_revisions)
1658
for digit in digits:
1659
result += int(digit)
1663
"""Provide an order to the underlying names."""
1664
return sorted(self._names.keys())
1666
def _obsolete_packs(self, packs):
1667
"""Move a number of packs which have been obsoleted out of the way.
1669
Each pack and its associated indices are moved out of the way.
1671
Note: for correctness this function should only be called after a new
1672
pack names index has been written without these pack names, and with
1673
the names of packs that contain the data previously available via these
1676
:param packs: The packs to obsolete.
1677
:param return: None.
1680
pack.pack_transport.rename(pack.file_name(),
1681
'../obsolete_packs/' + pack.file_name())
1682
# TODO: Probably needs to know all possible indices for this pack
1683
# - or maybe list the directory and move all indices matching this
1684
# name whether we recognize it or not?
1685
for suffix in ('.iix', '.six', '.tix', '.rix'):
1686
self._index_transport.rename(pack.name + suffix,
1687
'../obsolete_packs/' + pack.name + suffix)
1689
def pack_distribution(self, total_revisions):
1690
"""Generate a list of the number of revisions to put in each pack.
1692
:param total_revisions: The total number of revisions in the
1695
if total_revisions == 0:
1697
digits = reversed(str(total_revisions))
1699
for exponent, count in enumerate(digits):
1700
size = 10 ** exponent
1701
for pos in range(int(count)):
1703
return list(reversed(result))
1705
def _pack_tuple(self, name):
1706
"""Return a tuple with the transport and file name for a pack name."""
1707
return self._pack_transport, name + '.pack'
1709
def _remove_pack_from_memory(self, pack):
1710
"""Remove pack from the packs accessed by this repository.
1712
Only affects memory state, until self._save_pack_names() is invoked.
1714
self._names.pop(pack.name)
1715
self._packs_by_name.pop(pack.name)
1716
self._remove_pack_indices(pack)
1717
self.packs.remove(pack)
1719
def _remove_pack_indices(self, pack):
1720
"""Remove the indices for pack from the aggregated indices."""
1721
self.revision_index.remove_index(pack.revision_index, pack)
1722
self.inventory_index.remove_index(pack.inventory_index, pack)
1723
self.text_index.remove_index(pack.text_index, pack)
1724
self.signature_index.remove_index(pack.signature_index, pack)
1727
"""Clear all cached data."""
1728
# cached revision data
1729
self.repo._revision_knit = None
1730
self.revision_index.clear()
1731
# cached signature data
1732
self.repo._signature_knit = None
1733
self.signature_index.clear()
1734
# cached file text data
1735
self.text_index.clear()
1736
self.repo._text_knit = None
1737
# cached inventory data
1738
self.inventory_index.clear()
1739
# remove the open pack
1740
self._new_pack = None
1741
# information about packs.
1744
self._packs_by_name = {}
1745
self._packs_at_load = None
1747
def _unlock_names(self):
1748
"""Release the mutex around the pack-names index."""
1749
self.repo.control_files.unlock()
1751
def _diff_pack_names(self):
1752
"""Read the pack names from disk, and compare it to the one in memory.
1754
:return: (disk_nodes, deleted_nodes, new_nodes)
1755
disk_nodes The final set of nodes that should be referenced
1756
deleted_nodes Nodes which have been removed from when we started
1757
new_nodes Nodes that are newly introduced
1759
# load the disk nodes across
1761
for index, key, value in self._iter_disk_pack_index():
1762
disk_nodes.add((key, value))
1764
# do a two-way diff against our original content
1765
current_nodes = set()
1766
for name, sizes in self._names.iteritems():
1768
((name, ), ' '.join(str(size) for size in sizes)))
1770
# Packs no longer present in the repository, which were present when we
1771
# locked the repository
1772
deleted_nodes = self._packs_at_load - current_nodes
1773
# Packs which this process is adding
1774
new_nodes = current_nodes - self._packs_at_load
1776
# Update the disk_nodes set to include the ones we are adding, and
1777
# remove the ones which were removed by someone else
1778
disk_nodes.difference_update(deleted_nodes)
1779
disk_nodes.update(new_nodes)
1781
return disk_nodes, deleted_nodes, new_nodes
1783
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1784
"""Given the correct set of pack files, update our saved info.
1786
:return: (removed, added, modified)
1787
removed pack names removed from self._names
1788
added pack names added to self._names
1789
modified pack names that had changed value
1794
## self._packs_at_load = disk_nodes
1795
new_names = dict(disk_nodes)
1796
# drop no longer present nodes
1797
for pack in self.all_packs():
1798
if (pack.name,) not in new_names:
1799
removed.append(pack.name)
1800
self._remove_pack_from_memory(pack)
1801
# add new nodes/refresh existing ones
1802
for key, value in disk_nodes:
1804
sizes = self._parse_index_sizes(value)
1805
if name in self._names:
1807
if sizes != self._names[name]:
1808
# the pack for name has had its indices replaced - rare but
1809
# important to handle. XXX: probably can never happen today
1810
# because the three-way merge code above does not handle it
1811
# - you may end up adding the same key twice to the new
1812
# disk index because the set values are the same, unless
1813
# the only index shows up as deleted by the set difference
1814
# - which it may. Until there is a specific test for this,
1815
# assume its broken. RBC 20071017.
1816
self._remove_pack_from_memory(self.get_pack_by_name(name))
1817
self._names[name] = sizes
1818
self.get_pack_by_name(name)
1819
modified.append(name)
1822
self._names[name] = sizes
1823
self.get_pack_by_name(name)
1825
return removed, added, modified
1827
def _save_pack_names(self, clear_obsolete_packs=False):
1828
"""Save the list of packs.
1830
This will take out the mutex around the pack names list for the
1831
duration of the method call. If concurrent updates have been made, a
1832
three-way merge between the current list and the current in memory list
1835
:param clear_obsolete_packs: If True, clear out the contents of the
1836
obsolete_packs directory.
1840
builder = self._index_builder_class()
1841
disk_nodes, deleted_nodes, new_nodes = self._diff_pack_names()
1842
# TODO: handle same-name, index-size-changes here -
1843
# e.g. use the value from disk, not ours, *unless* we're the one
1845
for key, value in disk_nodes:
1846
builder.add_node(key, value)
1847
self.transport.put_file('pack-names', builder.finish(),
1848
mode=self.repo.bzrdir._get_file_mode())
1849
# move the baseline forward
1850
self._packs_at_load = disk_nodes
1851
if clear_obsolete_packs:
1852
self._clear_obsolete_packs()
1854
self._unlock_names()
1855
# synchronise the memory packs list with what we just wrote:
1856
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1858
def reload_pack_names(self):
1859
"""Sync our pack listing with what is present in the repository.
1861
This should be called when we find out that something we thought was
1862
present is now missing. This happens when another process re-packs the
1865
:return: True if the in-memory list of packs has been altered at all.
1867
# The ensure_loaded call is to handle the case where the first call
1868
# made involving the collection was to reload_pack_names, where we
1869
# don't have a view of disk contents. Its a bit of a bandaid, and
1870
# causes two reads of pack-names, but its a rare corner case not struck
1871
# with regular push/pull etc.
1872
first_read = self.ensure_loaded()
1875
# out the new value.
1876
disk_nodes, _, _ = self._diff_pack_names()
1877
self._packs_at_load = disk_nodes
1879
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1880
if removed or added or modified:
1884
def _restart_autopack(self):
1885
"""Reload the pack names list, and restart the autopack code."""
1886
if not self.reload_pack_names():
1887
# Re-raise the original exception, because something went missing
1888
# and a restart didn't find it
1890
raise errors.RetryAutopack(self.repo, False, sys.exc_info())
1892
def _clear_obsolete_packs(self):
1893
"""Delete everything from the obsolete-packs directory.
1895
obsolete_pack_transport = self.transport.clone('obsolete_packs')
1896
for filename in obsolete_pack_transport.list_dir('.'):
1898
obsolete_pack_transport.delete(filename)
1899
except (errors.PathError, errors.TransportError), e:
1900
warning("couldn't delete obsolete pack, skipping it:\n%s" % (e,))
1902
def _start_write_group(self):
1903
# Do not permit preparation for writing if we're not in a 'write lock'.
1904
if not self.repo.is_write_locked():
1905
raise errors.NotWriteLocked(self)
1906
self._new_pack = NewPack(self, upload_suffix='.pack',
1907
file_mode=self.repo.bzrdir._get_file_mode())
1908
# allow writing: queue writes to a new index
1909
self.revision_index.add_writable_index(self._new_pack.revision_index,
1911
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
1913
self.text_index.add_writable_index(self._new_pack.text_index,
1915
self.signature_index.add_writable_index(self._new_pack.signature_index,
1918
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
1919
self.repo.revisions._index._add_callback = self.revision_index.add_callback
1920
self.repo.signatures._index._add_callback = self.signature_index.add_callback
1921
self.repo.texts._index._add_callback = self.text_index.add_callback
1923
def _abort_write_group(self):
1924
# FIXME: just drop the transient index.
1925
# forget what names there are
1926
if self._new_pack is not None:
1928
self._new_pack.abort()
1930
# XXX: If we aborted while in the middle of finishing the write
1931
# group, _remove_pack_indices can fail because the indexes are
1932
# already gone. If they're not there we shouldn't fail in this
1933
# case. -- mbp 20081113
1934
self._remove_pack_indices(self._new_pack)
1935
self._new_pack = None
1936
for resumed_pack in self._resumed_packs:
1938
resumed_pack.abort()
1940
# See comment in previous finally block.
1942
self._remove_pack_indices(resumed_pack)
1945
del self._resumed_packs[:]
1946
self.repo._text_knit = None
1948
def _remove_resumed_pack_indices(self):
1949
for resumed_pack in self._resumed_packs:
1950
self._remove_pack_indices(resumed_pack)
1951
del self._resumed_packs[:]
1953
def _commit_write_group(self):
1955
for prefix, versioned_file in (
1956
('revisions', self.repo.revisions),
1957
('inventories', self.repo.inventories),
1958
('texts', self.repo.texts),
1959
('signatures', self.repo.signatures),
1961
missing = versioned_file.get_missing_compression_parent_keys()
1962
all_missing.update([(prefix,) + key for key in missing])
1964
raise errors.BzrCheckError(
1965
"Repository %s has missing compression parent(s) %r "
1966
% (self.repo, sorted(all_missing)))
1967
self._remove_pack_indices(self._new_pack)
1968
should_autopack = False
1969
if self._new_pack.data_inserted():
1970
# get all the data to disk and read to use
1971
self._new_pack.finish()
1972
self.allocate(self._new_pack)
1973
self._new_pack = None
1974
should_autopack = True
1976
self._new_pack.abort()
1977
self._new_pack = None
1978
for resumed_pack in self._resumed_packs:
1979
# XXX: this is a pretty ugly way to turn the resumed pack into a
1980
# properly committed pack.
1981
self._names[resumed_pack.name] = None
1982
self._remove_pack_from_memory(resumed_pack)
1983
resumed_pack.finish()
1984
self.allocate(resumed_pack)
1985
should_autopack = True
1986
del self._resumed_packs[:]
1988
if not self.autopack():
1989
# when autopack takes no steps, the names list is still
1991
self._save_pack_names()
1992
self.repo._text_knit = None
1994
def _suspend_write_group(self):
1995
tokens = [pack.name for pack in self._resumed_packs]
1996
self._remove_pack_indices(self._new_pack)
1997
if self._new_pack.data_inserted():
1998
# get all the data to disk and read to use
1999
self._new_pack.finish(suspend=True)
2000
tokens.append(self._new_pack.name)
2001
self._new_pack = None
2003
self._new_pack.abort()
2004
self._new_pack = None
2005
self._remove_resumed_pack_indices()
2006
self.repo._text_knit = None
2009
def _resume_write_group(self, tokens):
2010
for token in tokens:
2011
self._resume_pack(token)
2014
class KnitPackRepository(KnitRepository):
2015
"""Repository with knit objects stored inside pack containers.
2017
The layering for a KnitPackRepository is:
2019
Graph | HPSS | Repository public layer |
2020
===================================================
2021
Tuple based apis below, string based, and key based apis above
2022
---------------------------------------------------
2024
Provides .texts, .revisions etc
2025
This adapts the N-tuple keys to physical knit records which only have a
2026
single string identifier (for historical reasons), which in older formats
2027
was always the revision_id, and in the mapped code for packs is always
2028
the last element of key tuples.
2029
---------------------------------------------------
2031
A separate GraphIndex is used for each of the
2032
texts/inventories/revisions/signatures contained within each individual
2033
pack file. The GraphIndex layer works in N-tuples and is unaware of any
2035
===================================================
2039
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
2041
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2042
_commit_builder_class, _serializer)
2043
index_transport = self._transport.clone('indices')
2044
self._pack_collection = RepositoryPackCollection(self, self._transport,
2046
self._transport.clone('upload'),
2047
self._transport.clone('packs'),
2048
_format.index_builder_class,
2049
_format.index_class)
2050
self.inventories = KnitVersionedFiles(
2051
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2052
add_callback=self._pack_collection.inventory_index.add_callback,
2053
deltas=True, parents=True, is_locked=self.is_locked),
2054
data_access=self._pack_collection.inventory_index.data_access,
2055
max_delta_chain=200)
2056
self.revisions = KnitVersionedFiles(
2057
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2058
add_callback=self._pack_collection.revision_index.add_callback,
2059
deltas=False, parents=True, is_locked=self.is_locked),
2060
data_access=self._pack_collection.revision_index.data_access,
2062
self.signatures = KnitVersionedFiles(
2063
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2064
add_callback=self._pack_collection.signature_index.add_callback,
2065
deltas=False, parents=False, is_locked=self.is_locked),
2066
data_access=self._pack_collection.signature_index.data_access,
2068
self.texts = KnitVersionedFiles(
2069
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2070
add_callback=self._pack_collection.text_index.add_callback,
2071
deltas=True, parents=True, is_locked=self.is_locked),
2072
data_access=self._pack_collection.text_index.data_access,
2073
max_delta_chain=200)
2074
self.chk_bytes = None
2075
# True when the repository object is 'write locked' (as opposed to the
2076
# physical lock only taken out around changes to the pack-names list.)
2077
# Another way to represent this would be a decorator around the control
2078
# files object that presents logical locks as physical ones - if this
2079
# gets ugly consider that alternative design. RBC 20071011
2080
self._write_lock_count = 0
2081
self._transaction = None
2083
self._reconcile_does_inventory_gc = True
2084
self._reconcile_fixes_text_parents = True
2085
self._reconcile_backsup_inventory = False
2087
def _warn_if_deprecated(self):
2088
# This class isn't deprecated, but one sub-format is
2089
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2090
from bzrlib import repository
2091
if repository._deprecation_warning_done:
2093
repository._deprecation_warning_done = True
2094
warning("Format %s for %s is deprecated - please use"
2095
" 'bzr upgrade --1.6.1-rich-root'"
2096
% (self._format, self.bzrdir.transport.base))
2098
def _abort_write_group(self):
2099
self._pack_collection._abort_write_group()
2101
def _find_inconsistent_revision_parents(self):
2102
"""Find revisions with incorrectly cached parents.
2104
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
2105
parents-in-revision).
2107
if not self.is_locked():
2108
raise errors.ObjectNotLocked(self)
2109
pb = ui.ui_factory.nested_progress_bar()
2112
revision_nodes = self._pack_collection.revision_index \
2113
.combined_index.iter_all_entries()
2114
index_positions = []
2115
# Get the cached index values for all revisions, and also the location
2116
# in each index of the revision text so we can perform linear IO.
2117
for index, key, value, refs in revision_nodes:
2118
pos, length = value[1:].split(' ')
2119
index_positions.append((index, int(pos), key[0],
2120
tuple(parent[0] for parent in refs[0])))
2121
pb.update("Reading revision index", 0, 0)
2122
index_positions.sort()
2123
batch_count = len(index_positions) / 1000 + 1
2124
pb.update("Checking cached revision graph", 0, batch_count)
2125
for offset in xrange(batch_count):
2126
pb.update("Checking cached revision graph", offset)
2127
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
2130
rev_ids = [item[2] for item in to_query]
2131
revs = self.get_revisions(rev_ids)
2132
for revision, item in zip(revs, to_query):
2133
index_parents = item[3]
2134
rev_parents = tuple(revision.parent_ids)
2135
if index_parents != rev_parents:
2136
result.append((revision.revision_id, index_parents, rev_parents))
2141
def _make_parents_provider(self):
2142
return graph.CachingParentsProvider(self)
2144
def _refresh_data(self):
2145
if not self.is_locked():
2147
self._pack_collection.reload_pack_names()
2149
def _start_write_group(self):
2150
self._pack_collection._start_write_group()
2152
def _commit_write_group(self):
2153
return self._pack_collection._commit_write_group()
2155
def suspend_write_group(self):
2156
# XXX check self._write_group is self.get_transaction()?
2157
tokens = self._pack_collection._suspend_write_group()
2158
self._write_group = None
2161
def _resume_write_group(self, tokens):
2162
self._start_write_group()
2163
self._pack_collection._resume_write_group(tokens)
2165
def get_transaction(self):
2166
if self._write_lock_count:
2167
return self._transaction
2169
return self.control_files.get_transaction()
2171
def is_locked(self):
2172
return self._write_lock_count or self.control_files.is_locked()
2174
def is_write_locked(self):
2175
return self._write_lock_count
2177
def lock_write(self, token=None):
2178
locked = self.is_locked()
2179
if not self._write_lock_count and locked:
2180
raise errors.ReadOnlyError(self)
2181
self._write_lock_count += 1
2182
if self._write_lock_count == 1:
2183
self._transaction = transactions.WriteTransaction()
2184
for repo in self._fallback_repositories:
2185
# Writes don't affect fallback repos
2188
self._refresh_data()
2190
def lock_read(self):
2191
locked = self.is_locked()
2192
if self._write_lock_count:
2193
self._write_lock_count += 1
2195
self.control_files.lock_read()
2196
for repo in self._fallback_repositories:
2197
# Writes don't affect fallback repos
2200
self._refresh_data()
2202
def leave_lock_in_place(self):
2203
# not supported - raise an error
2204
raise NotImplementedError(self.leave_lock_in_place)
2206
def dont_leave_lock_in_place(self):
2207
# not supported - raise an error
2208
raise NotImplementedError(self.dont_leave_lock_in_place)
2212
"""Compress the data within the repository.
2214
This will pack all the data to a single pack. In future it may
2215
recompress deltas or do other such expensive operations.
2217
self._pack_collection.pack()
2220
def reconcile(self, other=None, thorough=False):
2221
"""Reconcile this repository."""
2222
from bzrlib.reconcile import PackReconciler
2223
reconciler = PackReconciler(self, thorough=thorough)
2224
reconciler.reconcile()
2227
def _reconcile_pack(self, collection, packs, extension, revs, pb):
2228
packer = ReconcilePacker(collection, packs, extension, revs)
2229
return packer.pack(pb)
2232
if self._write_lock_count == 1 and self._write_group is not None:
2233
self.abort_write_group()
2234
self._transaction = None
2235
self._write_lock_count = 0
2236
raise errors.BzrError(
2237
'Must end write group before releasing write lock on %s'
2239
if self._write_lock_count:
2240
self._write_lock_count -= 1
2241
if not self._write_lock_count:
2242
transaction = self._transaction
2243
self._transaction = None
2244
transaction.finish()
2245
for repo in self._fallback_repositories:
2248
self.control_files.unlock()
2249
for repo in self._fallback_repositories:
2253
class RepositoryFormatPack(MetaDirRepositoryFormat):
2254
"""Format logic for pack structured repositories.
2256
This repository format has:
2257
- a list of packs in pack-names
2258
- packs in packs/NAME.pack
2259
- indices in indices/NAME.{iix,six,tix,rix}
2260
- knit deltas in the packs, knit indices mapped to the indices.
2261
- thunk objects to support the knits programming API.
2262
- a format marker of its own
2263
- an optional 'shared-storage' flag
2264
- an optional 'no-working-trees' flag
2268
# Set this attribute in derived classes to control the repository class
2269
# created by open and initialize.
2270
repository_class = None
2271
# Set this attribute in derived classes to control the
2272
# _commit_builder_class that the repository objects will have passed to
2273
# their constructor.
2274
_commit_builder_class = None
2275
# Set this attribute in derived clases to control the _serializer that the
2276
# repository objects will have passed to their constructor.
2278
# Packs are not confused by ghosts.
2279
supports_ghosts = True
2280
# External references are not supported in pack repositories yet.
2281
supports_external_lookups = False
2282
# Most pack formats do not use chk lookups.
2283
supports_chks = False
2284
# What index classes to use
2285
index_builder_class = None
2287
_fetch_uses_deltas = True
2290
def initialize(self, a_bzrdir, shared=False):
2291
"""Create a pack based repository.
2293
:param a_bzrdir: bzrdir to contain the new repository; must already
2295
:param shared: If true the repository will be initialized as a shared
2298
mutter('creating repository in %s.', a_bzrdir.transport.base)
2299
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
2300
builder = self.index_builder_class()
2301
files = [('pack-names', builder.finish())]
2302
utf8_files = [('format', self.get_format_string())]
2304
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
2305
return self.open(a_bzrdir=a_bzrdir, _found=True)
2307
def open(self, a_bzrdir, _found=False, _override_transport=None):
2308
"""See RepositoryFormat.open().
2310
:param _override_transport: INTERNAL USE ONLY. Allows opening the
2311
repository at a slightly different url
2312
than normal. I.e. during 'upgrade'.
2315
format = RepositoryFormat.find_format(a_bzrdir)
2316
if _override_transport is not None:
2317
repo_transport = _override_transport
2319
repo_transport = a_bzrdir.get_repository_transport(None)
2320
control_files = lockable_files.LockableFiles(repo_transport,
2321
'lock', lockdir.LockDir)
2322
return self.repository_class(_format=self,
2324
control_files=control_files,
2325
_commit_builder_class=self._commit_builder_class,
2326
_serializer=self._serializer)
2329
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2330
"""A no-subtrees parameterized Pack repository.
2332
This format was introduced in 0.92.
2335
repository_class = KnitPackRepository
2336
_commit_builder_class = PackCommitBuilder
2338
def _serializer(self):
2339
return xml5.serializer_v5
2340
# What index classes to use
2341
index_builder_class = InMemoryGraphIndex
2342
index_class = GraphIndex
2344
def _get_matching_bzrdir(self):
2345
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2347
def _ignore_setting_bzrdir(self, format):
2350
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2352
def get_format_string(self):
2353
"""See RepositoryFormat.get_format_string()."""
2354
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2356
def get_format_description(self):
2357
"""See RepositoryFormat.get_format_description()."""
2358
return "Packs containing knits without subtree support"
2360
def check_conversion_target(self, target_format):
2364
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2365
"""A subtrees parameterized Pack repository.
2367
This repository format uses the xml7 serializer to get:
2368
- support for recording full info about the tree root
2369
- support for recording tree-references
2371
This format was introduced in 0.92.
2374
repository_class = KnitPackRepository
2375
_commit_builder_class = PackRootCommitBuilder
2376
rich_root_data = True
2377
supports_tree_reference = True
2379
def _serializer(self):
2380
return xml7.serializer_v7
2381
# What index classes to use
2382
index_builder_class = InMemoryGraphIndex
2383
index_class = GraphIndex
2385
def _get_matching_bzrdir(self):
2386
return bzrdir.format_registry.make_bzrdir(
2387
'pack-0.92-subtree')
2389
def _ignore_setting_bzrdir(self, format):
2392
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2394
def check_conversion_target(self, target_format):
2395
if not target_format.rich_root_data:
2396
raise errors.BadConversionTarget(
2397
'Does not support rich root data.', target_format)
2398
if not getattr(target_format, 'supports_tree_reference', False):
2399
raise errors.BadConversionTarget(
2400
'Does not support nested trees', target_format)
2402
def get_format_string(self):
2403
"""See RepositoryFormat.get_format_string()."""
2404
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2406
def get_format_description(self):
2407
"""See RepositoryFormat.get_format_description()."""
2408
return "Packs containing knits with subtree support\n"
2411
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2412
"""A rich-root, no subtrees parameterized Pack repository.
2414
This repository format uses the xml6 serializer to get:
2415
- support for recording full info about the tree root
2417
This format was introduced in 1.0.
2420
repository_class = KnitPackRepository
2421
_commit_builder_class = PackRootCommitBuilder
2422
rich_root_data = True
2423
supports_tree_reference = False
2425
def _serializer(self):
2426
return xml6.serializer_v6
2427
# What index classes to use
2428
index_builder_class = InMemoryGraphIndex
2429
index_class = GraphIndex
2431
def _get_matching_bzrdir(self):
2432
return bzrdir.format_registry.make_bzrdir(
2435
def _ignore_setting_bzrdir(self, format):
2438
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2440
def check_conversion_target(self, target_format):
2441
if not target_format.rich_root_data:
2442
raise errors.BadConversionTarget(
2443
'Does not support rich root data.', target_format)
2445
def get_format_string(self):
2446
"""See RepositoryFormat.get_format_string()."""
2447
return ("Bazaar pack repository format 1 with rich root"
2448
" (needs bzr 1.0)\n")
2450
def get_format_description(self):
2451
"""See RepositoryFormat.get_format_description()."""
2452
return "Packs containing knits with rich root support\n"
2455
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2456
"""Repository that supports external references to allow stacking.
2460
Supports external lookups, which results in non-truncated ghosts after
2461
reconcile compared to pack-0.92 formats.
2464
repository_class = KnitPackRepository
2465
_commit_builder_class = PackCommitBuilder
2466
supports_external_lookups = True
2467
# What index classes to use
2468
index_builder_class = InMemoryGraphIndex
2469
index_class = GraphIndex
2472
def _serializer(self):
2473
return xml5.serializer_v5
2475
def _get_matching_bzrdir(self):
2476
return bzrdir.format_registry.make_bzrdir('1.6')
2478
def _ignore_setting_bzrdir(self, format):
2481
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2483
def get_format_string(self):
2484
"""See RepositoryFormat.get_format_string()."""
2485
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2487
def get_format_description(self):
2488
"""See RepositoryFormat.get_format_description()."""
2489
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2491
def check_conversion_target(self, target_format):
2495
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2496
"""A repository with rich roots and stacking.
2498
New in release 1.6.1.
2500
Supports stacking on other repositories, allowing data to be accessed
2501
without being stored locally.
2504
repository_class = KnitPackRepository
2505
_commit_builder_class = PackRootCommitBuilder
2506
rich_root_data = True
2507
supports_tree_reference = False # no subtrees
2508
supports_external_lookups = True
2509
# What index classes to use
2510
index_builder_class = InMemoryGraphIndex
2511
index_class = GraphIndex
2514
def _serializer(self):
2515
return xml6.serializer_v6
2517
def _get_matching_bzrdir(self):
2518
return bzrdir.format_registry.make_bzrdir(
2521
def _ignore_setting_bzrdir(self, format):
2524
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2526
def check_conversion_target(self, target_format):
2527
if not target_format.rich_root_data:
2528
raise errors.BadConversionTarget(
2529
'Does not support rich root data.', target_format)
2531
def get_format_string(self):
2532
"""See RepositoryFormat.get_format_string()."""
2533
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2535
def get_format_description(self):
2536
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2539
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2540
"""A repository with rich roots and external references.
2544
Supports external lookups, which results in non-truncated ghosts after
2545
reconcile compared to pack-0.92 formats.
2547
This format was deprecated because the serializer it uses accidentally
2548
supported subtrees, when the format was not intended to. This meant that
2549
someone could accidentally fetch from an incorrect repository.
2552
repository_class = KnitPackRepository
2553
_commit_builder_class = PackRootCommitBuilder
2554
rich_root_data = True
2555
supports_tree_reference = False # no subtrees
2557
supports_external_lookups = True
2558
# What index classes to use
2559
index_builder_class = InMemoryGraphIndex
2560
index_class = GraphIndex
2563
def _serializer(self):
2564
return xml7.serializer_v7
2566
def _get_matching_bzrdir(self):
2567
matching = bzrdir.format_registry.make_bzrdir(
2569
matching.repository_format = self
2572
def _ignore_setting_bzrdir(self, format):
2575
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2577
def check_conversion_target(self, target_format):
2578
if not target_format.rich_root_data:
2579
raise errors.BadConversionTarget(
2580
'Does not support rich root data.', target_format)
2582
def get_format_string(self):
2583
"""See RepositoryFormat.get_format_string()."""
2584
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2586
def get_format_description(self):
2587
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2591
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2592
"""A repository with stacking and btree indexes,
2593
without rich roots or subtrees.
2595
This is equivalent to pack-1.6 with B+Tree indices.
2598
repository_class = KnitPackRepository
2599
_commit_builder_class = PackCommitBuilder
2600
supports_external_lookups = True
2601
# What index classes to use
2602
index_builder_class = BTreeBuilder
2603
index_class = BTreeGraphIndex
2606
def _serializer(self):
2607
return xml5.serializer_v5
2609
def _get_matching_bzrdir(self):
2610
return bzrdir.format_registry.make_bzrdir('1.9')
2612
def _ignore_setting_bzrdir(self, format):
2615
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2617
def get_format_string(self):
2618
"""See RepositoryFormat.get_format_string()."""
2619
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2621
def get_format_description(self):
2622
"""See RepositoryFormat.get_format_description()."""
2623
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2625
def check_conversion_target(self, target_format):
2629
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2630
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2632
1.6-rich-root with B+Tree indices.
2635
repository_class = KnitPackRepository
2636
_commit_builder_class = PackRootCommitBuilder
2637
rich_root_data = True
2638
supports_tree_reference = False # no subtrees
2639
supports_external_lookups = True
2640
# What index classes to use
2641
index_builder_class = BTreeBuilder
2642
index_class = BTreeGraphIndex
2645
def _serializer(self):
2646
return xml6.serializer_v6
2648
def _get_matching_bzrdir(self):
2649
return bzrdir.format_registry.make_bzrdir(
2652
def _ignore_setting_bzrdir(self, format):
2655
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2657
def check_conversion_target(self, target_format):
2658
if not target_format.rich_root_data:
2659
raise errors.BadConversionTarget(
2660
'Does not support rich root data.', target_format)
2662
def get_format_string(self):
2663
"""See RepositoryFormat.get_format_string()."""
2664
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2666
def get_format_description(self):
2667
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2670
class RepositoryFormatPackDevelopment2(RepositoryFormatPack):
2671
"""A no-subtrees development repository.
2673
This format should be retained until the second release after bzr 1.7.
2675
This is pack-1.6.1 with B+Tree indices.
2678
repository_class = KnitPackRepository
2679
_commit_builder_class = PackCommitBuilder
2680
supports_external_lookups = True
2681
# What index classes to use
2682
index_builder_class = BTreeBuilder
2683
index_class = BTreeGraphIndex
2684
# Set to true to get the fast-commit code path tested until a really fast
2685
# format lands in trunk. Not actually fast in this format.
2689
def _serializer(self):
2690
return xml5.serializer_v5
2692
def _get_matching_bzrdir(self):
2693
return bzrdir.format_registry.make_bzrdir('development2')
2695
def _ignore_setting_bzrdir(self, format):
2698
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2700
def get_format_string(self):
2701
"""See RepositoryFormat.get_format_string()."""
2702
return "Bazaar development format 2 (needs bzr.dev from before 1.8)\n"
2704
def get_format_description(self):
2705
"""See RepositoryFormat.get_format_description()."""
2706
return ("Development repository format, currently the same as "
2707
"1.6.1 with B+Trees.\n")
2709
def check_conversion_target(self, target_format):
2713
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2714
"""A subtrees development repository.
2716
This format should be retained until the second release after bzr 1.7.
2718
1.6.1-subtree[as it might have been] with B+Tree indices.
2721
repository_class = KnitPackRepository
2722
_commit_builder_class = PackRootCommitBuilder
2723
rich_root_data = True
2724
supports_tree_reference = True
2725
supports_external_lookups = True
2726
# What index classes to use
2727
index_builder_class = BTreeBuilder
2728
index_class = BTreeGraphIndex
2731
def _serializer(self):
2732
return xml7.serializer_v7
2734
def _get_matching_bzrdir(self):
2735
return bzrdir.format_registry.make_bzrdir(
2736
'development2-subtree')
2738
def _ignore_setting_bzrdir(self, format):
2741
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2743
def check_conversion_target(self, target_format):
2744
if not target_format.rich_root_data:
2745
raise errors.BadConversionTarget(
2746
'Does not support rich root data.', target_format)
2747
if not getattr(target_format, 'supports_tree_reference', False):
2748
raise errors.BadConversionTarget(
2749
'Does not support nested trees', target_format)
2751
def get_format_string(self):
2752
"""See RepositoryFormat.get_format_string()."""
2753
return ("Bazaar development format 2 with subtree support "
2754
"(needs bzr.dev from before 1.8)\n")
2756
def get_format_description(self):
2757
"""See RepositoryFormat.get_format_description()."""
2758
return ("Development repository format, currently the same as "
2759
"1.6.1-subtree with B+Tree indices.\n")
added
removed
bzrlib/repofmt/pack_repo.py
