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
37
from bzrlib.index import (
39
GraphIndexPrefixAdapter,
41
from bzrlib.knit import (
47
from bzrlib import tsort
54
revision as _mod_revision,
57
from bzrlib.decorators import needs_write_lock, only_raises
58
from bzrlib.btree_index import (
62
from bzrlib.index import (
66
from bzrlib.repofmt.knitrepo import KnitRepository
67
from bzrlib.repository import (
69
MetaDirRepositoryFormat,
74
from bzrlib.trace import (
81
class PackCommitBuilder(CommitBuilder):
82
"""A subclass of CommitBuilder to add texts with pack semantics.
84
Specifically this uses one knit object rather than one knit object per
85
added text, reducing memory and object pressure.
88
def __init__(self, repository, parents, config, timestamp=None,
89
timezone=None, committer=None, revprops=None,
91
CommitBuilder.__init__(self, repository, parents, config,
92
timestamp=timestamp, timezone=timezone, committer=committer,
93
revprops=revprops, revision_id=revision_id)
94
self._file_graph = graph.Graph(
95
repository._pack_collection.text_index.combined_index)
97
def _heads(self, file_id, revision_ids):
98
keys = [(file_id, revision_id) for revision_id in revision_ids]
99
return set([key[1] for key in self._file_graph.heads(keys)])
102
class PackRootCommitBuilder(RootCommitBuilder):
103
"""A subclass of RootCommitBuilder to add texts with pack semantics.
105
Specifically this uses one knit object rather than one knit object per
106
added text, reducing memory and object pressure.
109
def __init__(self, repository, parents, config, timestamp=None,
110
timezone=None, committer=None, revprops=None,
112
CommitBuilder.__init__(self, repository, parents, config,
113
timestamp=timestamp, timezone=timezone, committer=committer,
114
revprops=revprops, revision_id=revision_id)
115
self._file_graph = graph.Graph(
116
repository._pack_collection.text_index.combined_index)
118
def _heads(self, file_id, revision_ids):
119
keys = [(file_id, revision_id) for revision_id in revision_ids]
120
return set([key[1] for key in self._file_graph.heads(keys)])
124
"""An in memory proxy for a pack and its indices.
126
This is a base class that is not directly used, instead the classes
127
ExistingPack and NewPack are used.
130
# A map of index 'type' to the file extension and position in the
132
index_definitions = {
134
'revision': ('.rix', 0),
135
'inventory': ('.iix', 1),
137
'signature': ('.six', 3),
140
def __init__(self, revision_index, inventory_index, text_index,
141
signature_index, chk_index=None):
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.
154
:param chk_index: A GraphIndex for accessing content by CHK, if the
157
self.revision_index = revision_index
158
self.inventory_index = inventory_index
159
self.text_index = text_index
160
self.signature_index = signature_index
161
self.chk_index = chk_index
163
def access_tuple(self):
164
"""Return a tuple (transport, name) for the pack content."""
165
return self.pack_transport, self.file_name()
167
def _check_references(self):
168
"""Make sure our external references are present.
170
Packs are allowed to have deltas whose base is not in the pack, but it
171
must be present somewhere in this collection. It is not allowed to
172
have deltas based on a fallback repository.
173
(See <https://bugs.launchpad.net/bzr/+bug/288751>)
176
for (index_name, external_refs, index) in [
178
self._get_external_refs(self.text_index),
179
self._pack_collection.text_index.combined_index),
181
self._get_external_refs(self.inventory_index),
182
self._pack_collection.inventory_index.combined_index),
184
missing = external_refs.difference(
185
k for (idx, k, v, r) in
186
index.iter_entries(external_refs))
188
missing_items[index_name] = sorted(list(missing))
190
from pprint import pformat
191
raise errors.BzrCheckError(
192
"Newly created pack file %r has delta references to "
193
"items not in its repository:\n%s"
194
% (self, pformat(missing_items)))
197
"""Get the file name for the pack on disk."""
198
return self.name + '.pack'
200
def get_revision_count(self):
201
return self.revision_index.key_count()
203
def index_name(self, index_type, name):
204
"""Get the disk name of an index type for pack name 'name'."""
205
return name + Pack.index_definitions[index_type][0]
207
def index_offset(self, index_type):
208
"""Get the position in a index_size array for a given index type."""
209
return Pack.index_definitions[index_type][1]
211
def inventory_index_name(self, name):
212
"""The inv index is the name + .iix."""
213
return self.index_name('inventory', name)
215
def revision_index_name(self, name):
216
"""The revision index is the name + .rix."""
217
return self.index_name('revision', name)
219
def signature_index_name(self, name):
220
"""The signature index is the name + .six."""
221
return self.index_name('signature', name)
223
def text_index_name(self, name):
224
"""The text index is the name + .tix."""
225
return self.index_name('text', name)
227
def _replace_index_with_readonly(self, index_type):
228
setattr(self, index_type + '_index',
229
self.index_class(self.index_transport,
230
self.index_name(index_type, self.name),
231
self.index_sizes[self.index_offset(index_type)]))
234
class ExistingPack(Pack):
235
"""An in memory proxy for an existing .pack and its disk indices."""
237
def __init__(self, pack_transport, name, revision_index, inventory_index,
238
text_index, signature_index, chk_index=None):
239
"""Create an ExistingPack object.
241
:param pack_transport: The transport where the pack file resides.
242
:param name: The name of the pack on disk in the pack_transport.
244
Pack.__init__(self, revision_index, inventory_index, text_index,
245
signature_index, chk_index)
247
self.pack_transport = pack_transport
248
if None in (revision_index, inventory_index, text_index,
249
signature_index, name, pack_transport):
250
raise AssertionError()
252
def __eq__(self, other):
253
return self.__dict__ == other.__dict__
255
def __ne__(self, other):
256
return not self.__eq__(other)
259
return "<%s.%s object at 0x%x, %s, %s" % (
260
self.__class__.__module__, self.__class__.__name__, id(self),
261
self.pack_transport, self.name)
264
class ResumedPack(ExistingPack):
266
def __init__(self, name, revision_index, inventory_index, text_index,
267
signature_index, upload_transport, pack_transport, index_transport,
268
pack_collection, chk_index=None):
269
"""Create a ResumedPack object."""
270
ExistingPack.__init__(self, pack_transport, name, revision_index,
271
inventory_index, text_index, signature_index,
273
self.upload_transport = upload_transport
274
self.index_transport = index_transport
275
self.index_sizes = [None, None, None, None]
277
('revision', revision_index),
278
('inventory', inventory_index),
279
('text', text_index),
280
('signature', signature_index),
282
if chk_index is not None:
283
indices.append(('chk', chk_index))
284
self.index_sizes.append(None)
285
for index_type, index in indices:
286
offset = self.index_offset(index_type)
287
self.index_sizes[offset] = index._size
288
self.index_class = pack_collection._index_class
289
self._pack_collection = pack_collection
290
self._state = 'resumed'
291
# XXX: perhaps check that the .pack file exists?
293
def access_tuple(self):
294
if self._state == 'finished':
295
return Pack.access_tuple(self)
296
elif self._state == 'resumed':
297
return self.upload_transport, self.file_name()
299
raise AssertionError(self._state)
302
self.upload_transport.delete(self.file_name())
303
indices = [self.revision_index, self.inventory_index, self.text_index,
304
self.signature_index]
305
if self.chk_index is not None:
306
indices.append(self.chk_index)
307
for index in indices:
308
index._transport.delete(index._name)
311
self._check_references()
312
index_types = ['revision', 'inventory', 'text', 'signature']
313
if self.chk_index is not None:
314
index_types.append('chk')
315
for index_type in index_types:
316
old_name = self.index_name(index_type, self.name)
317
new_name = '../indices/' + old_name
318
self.upload_transport.rename(old_name, new_name)
319
self._replace_index_with_readonly(index_type)
320
new_name = '../packs/' + self.file_name()
321
self.upload_transport.rename(self.file_name(), new_name)
322
self._state = 'finished'
324
def _get_external_refs(self, index):
325
"""Return compression parents for this index that are not present.
327
This returns any compression parents that are referenced by this index,
328
which are not contained *in* this index. They may be present elsewhere.
330
return index.external_references(1)
334
"""An in memory proxy for a pack which is being created."""
336
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
337
"""Create a NewPack instance.
339
:param pack_collection: A PackCollection into which this is being inserted.
340
:param upload_suffix: An optional suffix to be given to any temporary
341
files created during the pack creation. e.g '.autopack'
342
:param file_mode: Unix permissions for newly created file.
344
# The relative locations of the packs are constrained, but all are
345
# passed in because the caller has them, so as to avoid object churn.
346
index_builder_class = pack_collection._index_builder_class
347
if pack_collection.chk_index is not None:
348
chk_index = index_builder_class(reference_lists=0)
352
# Revisions: parents list, no text compression.
353
index_builder_class(reference_lists=1),
354
# Inventory: We want to map compression only, but currently the
355
# knit code hasn't been updated enough to understand that, so we
356
# have a regular 2-list index giving parents and compression
358
index_builder_class(reference_lists=2),
359
# Texts: compression and per file graph, for all fileids - so two
360
# reference lists and two elements in the key tuple.
361
index_builder_class(reference_lists=2, key_elements=2),
362
# Signatures: Just blobs to store, no compression, no parents
364
index_builder_class(reference_lists=0),
365
# CHK based storage - just blobs, no compression or parents.
368
self._pack_collection = pack_collection
369
# When we make readonly indices, we need this.
370
self.index_class = pack_collection._index_class
371
# where should the new pack be opened
372
self.upload_transport = pack_collection._upload_transport
373
# where are indices written out to
374
self.index_transport = pack_collection._index_transport
375
# where is the pack renamed to when it is finished?
376
self.pack_transport = pack_collection._pack_transport
377
# What file mode to upload the pack and indices with.
378
self._file_mode = file_mode
379
# tracks the content written to the .pack file.
380
self._hash = osutils.md5()
381
# a tuple with the length in bytes of the indices, once the pack
382
# is finalised. (rev, inv, text, sigs, chk_if_in_use)
383
self.index_sizes = None
384
# How much data to cache when writing packs. Note that this is not
385
# synchronised with reads, because it's not in the transport layer, so
386
# is not safe unless the client knows it won't be reading from the pack
388
self._cache_limit = 0
389
# the temporary pack file name.
390
self.random_name = osutils.rand_chars(20) + upload_suffix
391
# when was this pack started ?
392
self.start_time = time.time()
393
# open an output stream for the data added to the pack.
394
self.write_stream = self.upload_transport.open_write_stream(
395
self.random_name, mode=self._file_mode)
396
if 'pack' in debug.debug_flags:
397
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
398
time.ctime(), self.upload_transport.base, self.random_name,
399
time.time() - self.start_time)
400
# A list of byte sequences to be written to the new pack, and the
401
# aggregate size of them. Stored as a list rather than separate
402
# variables so that the _write_data closure below can update them.
403
self._buffer = [[], 0]
404
# create a callable for adding data
406
# robertc says- this is a closure rather than a method on the object
407
# so that the variables are locals, and faster than accessing object
409
def _write_data(bytes, flush=False, _buffer=self._buffer,
410
_write=self.write_stream.write, _update=self._hash.update):
411
_buffer[0].append(bytes)
412
_buffer[1] += len(bytes)
414
if _buffer[1] > self._cache_limit or flush:
415
bytes = ''.join(_buffer[0])
419
# expose this on self, for the occasion when clients want to add data.
420
self._write_data = _write_data
421
# a pack writer object to serialise pack records.
422
self._writer = pack.ContainerWriter(self._write_data)
424
# what state is the pack in? (open, finished, aborted)
426
# no name until we finish writing the content
430
"""Cancel creating this pack."""
431
self._state = 'aborted'
432
self.write_stream.close()
433
# Remove the temporary pack file.
434
self.upload_transport.delete(self.random_name)
435
# The indices have no state on disk.
437
def access_tuple(self):
438
"""Return a tuple (transport, name) for the pack content."""
439
if self._state == 'finished':
440
return Pack.access_tuple(self)
441
elif self._state == 'open':
442
return self.upload_transport, self.random_name
444
raise AssertionError(self._state)
446
def data_inserted(self):
447
"""True if data has been added to this pack."""
448
return bool(self.get_revision_count() or
449
self.inventory_index.key_count() or
450
self.text_index.key_count() or
451
self.signature_index.key_count() or
452
(self.chk_index is not None and self.chk_index.key_count()))
454
def finish_content(self):
455
if self.name is not None:
459
self._write_data('', flush=True)
460
self.name = self._hash.hexdigest()
462
def finish(self, suspend=False):
463
"""Finish the new pack.
466
- finalises the content
467
- assigns a name (the md5 of the content, currently)
468
- writes out the associated indices
469
- renames the pack into place.
470
- stores the index size tuple for the pack in the index_sizes
473
self.finish_content()
475
self._check_references()
477
# XXX: It'd be better to write them all to temporary names, then
478
# rename them all into place, so that the window when only some are
479
# visible is smaller. On the other hand none will be seen until
480
# they're in the names list.
481
self.index_sizes = [None, None, None, None]
482
self._write_index('revision', self.revision_index, 'revision', suspend)
483
self._write_index('inventory', self.inventory_index, 'inventory',
485
self._write_index('text', self.text_index, 'file texts', suspend)
486
self._write_index('signature', self.signature_index,
487
'revision signatures', suspend)
488
if self.chk_index is not None:
489
self.index_sizes.append(None)
490
self._write_index('chk', self.chk_index,
491
'content hash bytes', suspend)
492
self.write_stream.close()
493
# Note that this will clobber an existing pack with the same name,
494
# without checking for hash collisions. While this is undesirable this
495
# is something that can be rectified in a subsequent release. One way
496
# to rectify it may be to leave the pack at the original name, writing
497
# its pack-names entry as something like 'HASH: index-sizes
498
# temporary-name'. Allocate that and check for collisions, if it is
499
# collision free then rename it into place. If clients know this scheme
500
# they can handle missing-file errors by:
501
# - try for HASH.pack
502
# - try for temporary-name
503
# - refresh the pack-list to see if the pack is now absent
504
new_name = self.name + '.pack'
506
new_name = '../packs/' + new_name
507
self.upload_transport.rename(self.random_name, new_name)
508
self._state = 'finished'
509
if 'pack' in debug.debug_flags:
510
# XXX: size might be interesting?
511
mutter('%s: create_pack: pack finished: %s%s->%s t+%6.3fs',
512
time.ctime(), self.upload_transport.base, self.random_name,
513
new_name, time.time() - self.start_time)
516
"""Flush any current data."""
518
bytes = ''.join(self._buffer[0])
519
self.write_stream.write(bytes)
520
self._hash.update(bytes)
521
self._buffer[:] = [[], 0]
523
def _get_external_refs(self, index):
524
return index._external_references()
526
def set_write_cache_size(self, size):
527
self._cache_limit = size
529
def _write_index(self, index_type, index, label, suspend=False):
530
"""Write out an index.
532
:param index_type: The type of index to write - e.g. 'revision'.
533
:param index: The index object to serialise.
534
:param label: What label to give the index e.g. 'revision'.
536
index_name = self.index_name(index_type, self.name)
538
transport = self.upload_transport
540
transport = self.index_transport
541
self.index_sizes[self.index_offset(index_type)] = transport.put_file(
542
index_name, index.finish(), mode=self._file_mode)
543
if 'pack' in debug.debug_flags:
544
# XXX: size might be interesting?
545
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
546
time.ctime(), label, self.upload_transport.base,
547
self.random_name, time.time() - self.start_time)
548
# Replace the writable index on this object with a readonly,
549
# presently unloaded index. We should alter
550
# the index layer to make its finish() error if add_node is
551
# subsequently used. RBC
552
self._replace_index_with_readonly(index_type)
555
class AggregateIndex(object):
556
"""An aggregated index for the RepositoryPackCollection.
558
AggregateIndex is reponsible for managing the PackAccess object,
559
Index-To-Pack mapping, and all indices list for a specific type of index
560
such as 'revision index'.
562
A CombinedIndex provides an index on a single key space built up
563
from several on-disk indices. The AggregateIndex builds on this
564
to provide a knit access layer, and allows having up to one writable
565
index within the collection.
567
# XXX: Probably 'can be written to' could/should be separated from 'acts
568
# like a knit index' -- mbp 20071024
570
def __init__(self, reload_func=None, flush_func=None):
571
"""Create an AggregateIndex.
573
:param reload_func: A function to call if we find we are missing an
574
index. Should have the form reload_func() => True if the list of
575
active pack files has changed.
577
self._reload_func = reload_func
578
self.index_to_pack = {}
579
self.combined_index = CombinedGraphIndex([], reload_func=reload_func)
580
self.data_access = _DirectPackAccess(self.index_to_pack,
581
reload_func=reload_func,
582
flush_func=flush_func)
583
self.add_callback = None
585
def replace_indices(self, index_to_pack, indices):
586
"""Replace the current mappings with fresh ones.
588
This should probably not be used eventually, rather incremental add and
589
removal of indices. It has been added during refactoring of existing
592
:param index_to_pack: A mapping from index objects to
593
(transport, name) tuples for the pack file data.
594
:param indices: A list of indices.
596
# refresh the revision pack map dict without replacing the instance.
597
self.index_to_pack.clear()
598
self.index_to_pack.update(index_to_pack)
599
# XXX: API break - clearly a 'replace' method would be good?
600
self.combined_index._indices[:] = indices
601
# the current add nodes callback for the current writable index if
603
self.add_callback = None
605
def add_index(self, index, pack):
606
"""Add index to the aggregate, which is an index for Pack pack.
608
Future searches on the aggregate index will seach this new index
609
before all previously inserted indices.
611
:param index: An Index for the pack.
612
:param pack: A Pack instance.
614
# expose it to the index map
615
self.index_to_pack[index] = pack.access_tuple()
616
# put it at the front of the linear index list
617
self.combined_index.insert_index(0, index)
619
def add_writable_index(self, index, pack):
620
"""Add an index which is able to have data added to it.
622
There can be at most one writable index at any time. Any
623
modifications made to the knit are put into this index.
625
:param index: An index from the pack parameter.
626
:param pack: A Pack instance.
628
if self.add_callback is not None:
629
raise AssertionError(
630
"%s already has a writable index through %s" % \
631
(self, self.add_callback))
632
# allow writing: queue writes to a new index
633
self.add_index(index, pack)
634
# Updates the index to packs mapping as a side effect,
635
self.data_access.set_writer(pack._writer, index, pack.access_tuple())
636
self.add_callback = index.add_nodes
639
"""Reset all the aggregate data to nothing."""
640
self.data_access.set_writer(None, None, (None, None))
641
self.index_to_pack.clear()
642
del self.combined_index._indices[:]
643
self.add_callback = None
645
def remove_index(self, index, pack):
646
"""Remove index from the indices used to answer queries.
648
:param index: An index from the pack parameter.
649
:param pack: A Pack instance.
651
del self.index_to_pack[index]
652
self.combined_index._indices.remove(index)
653
if (self.add_callback is not None and
654
getattr(index, 'add_nodes', None) == self.add_callback):
655
self.add_callback = None
656
self.data_access.set_writer(None, None, (None, None))
659
class Packer(object):
660
"""Create a pack from packs."""
662
def __init__(self, pack_collection, packs, suffix, revision_ids=None,
666
:param pack_collection: A RepositoryPackCollection object where the
667
new pack is being written to.
668
:param packs: The packs to combine.
669
:param suffix: The suffix to use on the temporary files for the pack.
670
:param revision_ids: Revision ids to limit the pack to.
671
:param reload_func: A function to call if a pack file/index goes
672
missing. The side effect of calling this function should be to
673
update self.packs. See also AggregateIndex
677
self.revision_ids = revision_ids
678
# The pack object we are creating.
680
self._pack_collection = pack_collection
681
self._reload_func = reload_func
682
# The index layer keys for the revisions being copied. None for 'all
684
self._revision_keys = None
685
# What text keys to copy. None for 'all texts'. This is set by
686
# _copy_inventory_texts
687
self._text_filter = None
690
def _extra_init(self):
691
"""A template hook to allow extending the constructor trivially."""
693
def _pack_map_and_index_list(self, index_attribute):
694
"""Convert a list of packs to an index pack map and index list.
696
:param index_attribute: The attribute that the desired index is found
698
:return: A tuple (map, list) where map contains the dict from
699
index:pack_tuple, and list contains the indices in the preferred
704
for pack_obj in self.packs:
705
index = getattr(pack_obj, index_attribute)
706
indices.append(index)
707
pack_map[index] = pack_obj
708
return pack_map, indices
710
def _index_contents(self, indices, key_filter=None):
711
"""Get an iterable of the index contents from a pack_map.
713
:param indices: The list of indices to query
714
:param key_filter: An optional filter to limit the keys returned.
716
all_index = CombinedGraphIndex(indices)
717
if key_filter is None:
718
return all_index.iter_all_entries()
720
return all_index.iter_entries(key_filter)
722
def pack(self, pb=None):
723
"""Create a new pack by reading data from other packs.
725
This does little more than a bulk copy of data. One key difference
726
is that data with the same item key across multiple packs is elided
727
from the output. The new pack is written into the current pack store
728
along with its indices, and the name added to the pack names. The
729
source packs are not altered and are not required to be in the current
732
:param pb: An optional progress bar to use. A nested bar is created if
734
:return: A Pack object, or None if nothing was copied.
736
# open a pack - using the same name as the last temporary file
737
# - which has already been flushed, so its safe.
738
# XXX: - duplicate code warning with start_write_group; fix before
739
# considering 'done'.
740
if self._pack_collection._new_pack is not None:
741
raise errors.BzrError('call to %s.pack() while another pack is'
743
% (self.__class__.__name__,))
744
if self.revision_ids is not None:
745
if len(self.revision_ids) == 0:
746
# silly fetch request.
749
self.revision_ids = frozenset(self.revision_ids)
750
self.revision_keys = frozenset((revid,) for revid in
753
self.pb = ui.ui_factory.nested_progress_bar()
757
return self._create_pack_from_packs()
763
"""Open a pack for the pack we are creating."""
764
new_pack = self._pack_collection.pack_factory(self._pack_collection,
765
upload_suffix=self.suffix,
766
file_mode=self._pack_collection.repo.bzrdir._get_file_mode())
767
# We know that we will process all nodes in order, and don't need to
768
# query, so don't combine any indices spilled to disk until we are done
769
new_pack.revision_index.set_optimize(combine_backing_indices=False)
770
new_pack.inventory_index.set_optimize(combine_backing_indices=False)
771
new_pack.text_index.set_optimize(combine_backing_indices=False)
772
new_pack.signature_index.set_optimize(combine_backing_indices=False)
775
def _update_pack_order(self, entries, index_to_pack_map):
776
"""Determine how we want our packs to be ordered.
778
This changes the sort order of the self.packs list so that packs unused
779
by 'entries' will be at the end of the list, so that future requests
780
can avoid probing them. Used packs will be at the front of the
781
self.packs list, in the order of their first use in 'entries'.
783
:param entries: A list of (index, ...) tuples
784
:param index_to_pack_map: A mapping from index objects to pack objects.
788
for entry in entries:
790
if index not in seen_indexes:
791
packs.append(index_to_pack_map[index])
792
seen_indexes.add(index)
793
if len(packs) == len(self.packs):
794
if 'pack' in debug.debug_flags:
795
mutter('Not changing pack list, all packs used.')
797
seen_packs = set(packs)
798
for pack in self.packs:
799
if pack not in seen_packs:
802
if 'pack' in debug.debug_flags:
803
old_names = [p.access_tuple()[1] for p in self.packs]
804
new_names = [p.access_tuple()[1] for p in packs]
805
mutter('Reordering packs\nfrom: %s\n to: %s',
806
old_names, new_names)
809
def _copy_revision_texts(self):
810
"""Copy revision data to the new pack."""
812
if self.revision_ids:
813
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
816
# select revision keys
817
revision_index_map, revision_indices = self._pack_map_and_index_list(
819
revision_nodes = self._index_contents(revision_indices, revision_keys)
820
revision_nodes = list(revision_nodes)
821
self._update_pack_order(revision_nodes, revision_index_map)
822
# copy revision keys and adjust values
823
self.pb.update("Copying revision texts", 1)
824
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
825
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
826
self.new_pack.revision_index, readv_group_iter, total_items))
827
if 'pack' in debug.debug_flags:
828
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
829
time.ctime(), self._pack_collection._upload_transport.base,
830
self.new_pack.random_name,
831
self.new_pack.revision_index.key_count(),
832
time.time() - self.new_pack.start_time)
833
self._revision_keys = revision_keys
835
def _copy_inventory_texts(self):
836
"""Copy the inventory texts to the new pack.
838
self._revision_keys is used to determine what inventories to copy.
840
Sets self._text_filter appropriately.
842
# select inventory keys
843
inv_keys = self._revision_keys # currently the same keyspace, and note that
844
# querying for keys here could introduce a bug where an inventory item
845
# is missed, so do not change it to query separately without cross
846
# checking like the text key check below.
847
inventory_index_map, inventory_indices = self._pack_map_and_index_list(
849
inv_nodes = self._index_contents(inventory_indices, inv_keys)
850
# copy inventory keys and adjust values
851
# XXX: Should be a helper function to allow different inv representation
853
self.pb.update("Copying inventory texts", 2)
854
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
855
# Only grab the output lines if we will be processing them
856
output_lines = bool(self.revision_ids)
857
inv_lines = self._copy_nodes_graph(inventory_index_map,
858
self.new_pack._writer, self.new_pack.inventory_index,
859
readv_group_iter, total_items, output_lines=output_lines)
860
if self.revision_ids:
861
self._process_inventory_lines(inv_lines)
863
# eat the iterator to cause it to execute.
865
self._text_filter = None
866
if 'pack' in debug.debug_flags:
867
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
868
time.ctime(), self._pack_collection._upload_transport.base,
869
self.new_pack.random_name,
870
self.new_pack.inventory_index.key_count(),
871
time.time() - self.new_pack.start_time)
873
def _copy_text_texts(self):
875
text_index_map, text_nodes = self._get_text_nodes()
876
if self._text_filter is not None:
877
# We could return the keys copied as part of the return value from
878
# _copy_nodes_graph but this doesn't work all that well with the
879
# need to get line output too, so we check separately, and as we're
880
# going to buffer everything anyway, we check beforehand, which
881
# saves reading knit data over the wire when we know there are
883
text_nodes = set(text_nodes)
884
present_text_keys = set(_node[1] for _node in text_nodes)
885
missing_text_keys = set(self._text_filter) - present_text_keys
886
if missing_text_keys:
887
# TODO: raise a specific error that can handle many missing
889
mutter("missing keys during fetch: %r", missing_text_keys)
890
a_missing_key = missing_text_keys.pop()
891
raise errors.RevisionNotPresent(a_missing_key[1],
893
# copy text keys and adjust values
894
self.pb.update("Copying content texts", 3)
895
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
896
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
897
self.new_pack.text_index, readv_group_iter, total_items))
898
self._log_copied_texts()
900
def _create_pack_from_packs(self):
901
self.pb.update("Opening pack", 0, 5)
902
self.new_pack = self.open_pack()
903
new_pack = self.new_pack
904
# buffer data - we won't be reading-back during the pack creation and
905
# this makes a significant difference on sftp pushes.
906
new_pack.set_write_cache_size(1024*1024)
907
if 'pack' in debug.debug_flags:
908
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
909
for a_pack in self.packs]
910
if self.revision_ids is not None:
911
rev_count = len(self.revision_ids)
914
mutter('%s: create_pack: creating pack from source packs: '
915
'%s%s %s revisions wanted %s t=0',
916
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
917
plain_pack_list, rev_count)
918
self._copy_revision_texts()
919
self._copy_inventory_texts()
920
self._copy_text_texts()
921
# select signature keys
922
signature_filter = self._revision_keys # same keyspace
923
signature_index_map, signature_indices = self._pack_map_and_index_list(
925
signature_nodes = self._index_contents(signature_indices,
927
# copy signature keys and adjust values
928
self.pb.update("Copying signature texts", 4)
929
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
930
new_pack.signature_index)
931
if 'pack' in debug.debug_flags:
932
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
933
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
934
new_pack.signature_index.key_count(),
935
time.time() - new_pack.start_time)
937
# NB XXX: how to check CHK references are present? perhaps by yielding
938
# the items? How should that interact with stacked repos?
939
if new_pack.chk_index is not None:
941
if 'pack' in debug.debug_flags:
942
mutter('%s: create_pack: chk content copied: %s%s %d items t+%6.3fs',
943
time.ctime(), self._pack_collection._upload_transport.base,
944
new_pack.random_name,
945
new_pack.chk_index.key_count(),
946
time.time() - new_pack.start_time)
947
new_pack._check_references()
948
if not self._use_pack(new_pack):
951
self.pb.update("Finishing pack", 5)
953
self._pack_collection.allocate(new_pack)
956
def _copy_chks(self, refs=None):
957
# XXX: Todo, recursive follow-pointers facility when fetching some
959
chk_index_map, chk_indices = self._pack_map_and_index_list(
961
chk_nodes = self._index_contents(chk_indices, refs)
963
# TODO: This isn't strictly tasteful as we are accessing some private
964
# variables (_serializer). Perhaps a better way would be to have
965
# Repository._deserialise_chk_node()
966
search_key_func = chk_map.search_key_registry.get(
967
self._pack_collection.repo._serializer.search_key_name)
968
def accumlate_refs(lines):
969
# XXX: move to a generic location
971
bytes = ''.join(lines)
972
node = chk_map._deserialise(bytes, ("unknown",), search_key_func)
973
new_refs.update(node.refs())
974
self._copy_nodes(chk_nodes, chk_index_map, self.new_pack._writer,
975
self.new_pack.chk_index, output_lines=accumlate_refs)
978
def _copy_nodes(self, nodes, index_map, writer, write_index,
980
"""Copy knit nodes between packs with no graph references.
982
:param output_lines: Output full texts of copied items.
984
pb = ui.ui_factory.nested_progress_bar()
986
return self._do_copy_nodes(nodes, index_map, writer,
987
write_index, pb, output_lines=output_lines)
991
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb,
993
# for record verification
994
knit = KnitVersionedFiles(None, None)
995
# plan a readv on each source pack:
997
nodes = sorted(nodes)
998
# how to map this into knit.py - or knit.py into this?
999
# we don't want the typical knit logic, we want grouping by pack
1000
# at this point - perhaps a helper library for the following code
1001
# duplication points?
1003
for index, key, value in nodes:
1004
if index not in request_groups:
1005
request_groups[index] = []
1006
request_groups[index].append((key, value))
1008
pb.update("Copied record", record_index, len(nodes))
1009
for index, items in request_groups.iteritems():
1010
pack_readv_requests = []
1011
for key, value in items:
1012
# ---- KnitGraphIndex.get_position
1013
bits = value[1:].split(' ')
1014
offset, length = int(bits[0]), int(bits[1])
1015
pack_readv_requests.append((offset, length, (key, value[0])))
1016
# linear scan up the pack
1017
pack_readv_requests.sort()
1019
pack_obj = index_map[index]
1020
transport, path = pack_obj.access_tuple()
1022
reader = pack.make_readv_reader(transport, path,
1023
[offset[0:2] for offset in pack_readv_requests])
1024
except errors.NoSuchFile:
1025
if self._reload_func is not None:
1028
for (names, read_func), (_1, _2, (key, eol_flag)) in \
1029
izip(reader.iter_records(), pack_readv_requests):
1030
raw_data = read_func(None)
1031
# check the header only
1032
if output_lines is not None:
1033
output_lines(knit._parse_record(key[-1], raw_data)[0])
1035
df, _ = knit._parse_record_header(key, raw_data)
1037
pos, size = writer.add_bytes_record(raw_data, names)
1038
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
1039
pb.update("Copied record", record_index)
1042
def _copy_nodes_graph(self, index_map, writer, write_index,
1043
readv_group_iter, total_items, output_lines=False):
1044
"""Copy knit nodes between packs.
1046
:param output_lines: Return lines present in the copied data as
1047
an iterator of line,version_id.
1049
pb = ui.ui_factory.nested_progress_bar()
1051
for result in self._do_copy_nodes_graph(index_map, writer,
1052
write_index, output_lines, pb, readv_group_iter, total_items):
1055
# Python 2.4 does not permit try:finally: in a generator.
1061
def _do_copy_nodes_graph(self, index_map, writer, write_index,
1062
output_lines, pb, readv_group_iter, total_items):
1063
# for record verification
1064
knit = KnitVersionedFiles(None, None)
1065
# for line extraction when requested (inventories only)
1067
factory = KnitPlainFactory()
1069
pb.update("Copied record", record_index, total_items)
1070
for index, readv_vector, node_vector in readv_group_iter:
1072
pack_obj = index_map[index]
1073
transport, path = pack_obj.access_tuple()
1075
reader = pack.make_readv_reader(transport, path, readv_vector)
1076
except errors.NoSuchFile:
1077
if self._reload_func is not None:
1080
for (names, read_func), (key, eol_flag, references) in \
1081
izip(reader.iter_records(), node_vector):
1082
raw_data = read_func(None)
1084
# read the entire thing
1085
content, _ = knit._parse_record(key[-1], raw_data)
1086
if len(references[-1]) == 0:
1087
line_iterator = factory.get_fulltext_content(content)
1089
line_iterator = factory.get_linedelta_content(content)
1090
for line in line_iterator:
1093
# check the header only
1094
df, _ = knit._parse_record_header(key, raw_data)
1096
pos, size = writer.add_bytes_record(raw_data, names)
1097
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1098
pb.update("Copied record", record_index)
1101
def _get_text_nodes(self):
1102
text_index_map, text_indices = self._pack_map_and_index_list(
1104
return text_index_map, self._index_contents(text_indices,
1107
def _least_readv_node_readv(self, nodes):
1108
"""Generate request groups for nodes using the least readv's.
1110
:param nodes: An iterable of graph index nodes.
1111
:return: Total node count and an iterator of the data needed to perform
1112
readvs to obtain the data for nodes. Each item yielded by the
1113
iterator is a tuple with:
1114
index, readv_vector, node_vector. readv_vector is a list ready to
1115
hand to the transport readv method, and node_vector is a list of
1116
(key, eol_flag, references) for the the node retrieved by the
1117
matching readv_vector.
1119
# group by pack so we do one readv per pack
1120
nodes = sorted(nodes)
1123
for index, key, value, references in nodes:
1124
if index not in request_groups:
1125
request_groups[index] = []
1126
request_groups[index].append((key, value, references))
1128
for index, items in request_groups.iteritems():
1129
pack_readv_requests = []
1130
for key, value, references in items:
1131
# ---- KnitGraphIndex.get_position
1132
bits = value[1:].split(' ')
1133
offset, length = int(bits[0]), int(bits[1])
1134
pack_readv_requests.append(
1135
((offset, length), (key, value[0], references)))
1136
# linear scan up the pack to maximum range combining.
1137
pack_readv_requests.sort()
1138
# split out the readv and the node data.
1139
pack_readv = [readv for readv, node in pack_readv_requests]
1140
node_vector = [node for readv, node in pack_readv_requests]
1141
result.append((index, pack_readv, node_vector))
1142
return total, result
1144
def _log_copied_texts(self):
1145
if 'pack' in debug.debug_flags:
1146
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
1147
time.ctime(), self._pack_collection._upload_transport.base,
1148
self.new_pack.random_name,
1149
self.new_pack.text_index.key_count(),
1150
time.time() - self.new_pack.start_time)
1152
def _process_inventory_lines(self, inv_lines):
1153
"""Use up the inv_lines generator and setup a text key filter."""
1154
repo = self._pack_collection.repo
1155
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
1156
inv_lines, self.revision_keys)
1158
for fileid, file_revids in fileid_revisions.iteritems():
1159
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
1160
self._text_filter = text_filter
1162
def _revision_node_readv(self, revision_nodes):
1163
"""Return the total revisions and the readv's to issue.
1165
:param revision_nodes: The revision index contents for the packs being
1166
incorporated into the new pack.
1167
:return: As per _least_readv_node_readv.
1169
return self._least_readv_node_readv(revision_nodes)
1171
def _use_pack(self, new_pack):
1172
"""Return True if new_pack should be used.
1174
:param new_pack: The pack that has just been created.
1175
:return: True if the pack should be used.
1177
return new_pack.data_inserted()
1180
class OptimisingPacker(Packer):
1181
"""A packer which spends more time to create better disk layouts."""
1183
def _revision_node_readv(self, revision_nodes):
1184
"""Return the total revisions and the readv's to issue.
1186
This sort places revisions in topological order with the ancestors
1189
:param revision_nodes: The revision index contents for the packs being
1190
incorporated into the new pack.
1191
:return: As per _least_readv_node_readv.
1193
# build an ancestors dict
1196
for index, key, value, references in revision_nodes:
1197
ancestors[key] = references[0]
1198
by_key[key] = (index, value, references)
1199
order = tsort.topo_sort(ancestors)
1201
# Single IO is pathological, but it will work as a starting point.
1203
for key in reversed(order):
1204
index, value, references = by_key[key]
1205
# ---- KnitGraphIndex.get_position
1206
bits = value[1:].split(' ')
1207
offset, length = int(bits[0]), int(bits[1])
1209
(index, [(offset, length)], [(key, value[0], references)]))
1210
# TODO: combine requests in the same index that are in ascending order.
1211
return total, requests
1213
def open_pack(self):
1214
"""Open a pack for the pack we are creating."""
1215
new_pack = super(OptimisingPacker, self).open_pack()
1216
# Turn on the optimization flags for all the index builders.
1217
new_pack.revision_index.set_optimize(for_size=True)
1218
new_pack.inventory_index.set_optimize(for_size=True)
1219
new_pack.text_index.set_optimize(for_size=True)
1220
new_pack.signature_index.set_optimize(for_size=True)
1224
class ReconcilePacker(Packer):
1225
"""A packer which regenerates indices etc as it copies.
1227
This is used by ``bzr reconcile`` to cause parent text pointers to be
1231
def _extra_init(self):
1232
self._data_changed = False
1234
def _process_inventory_lines(self, inv_lines):
1235
"""Generate a text key reference map rather for reconciling with."""
1236
repo = self._pack_collection.repo
1237
refs = repo._find_text_key_references_from_xml_inventory_lines(
1239
self._text_refs = refs
1240
# during reconcile we:
1241
# - convert unreferenced texts to full texts
1242
# - correct texts which reference a text not copied to be full texts
1243
# - copy all others as-is but with corrected parents.
1244
# - so at this point we don't know enough to decide what becomes a full
1246
self._text_filter = None
1248
def _copy_text_texts(self):
1249
"""generate what texts we should have and then copy."""
1250
self.pb.update("Copying content texts", 3)
1251
# we have three major tasks here:
1252
# 1) generate the ideal index
1253
repo = self._pack_collection.repo
1254
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1255
_1, key, _2, refs in
1256
self.new_pack.revision_index.iter_all_entries()])
1257
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1258
# 2) generate a text_nodes list that contains all the deltas that can
1259
# be used as-is, with corrected parents.
1262
discarded_nodes = []
1263
NULL_REVISION = _mod_revision.NULL_REVISION
1264
text_index_map, text_nodes = self._get_text_nodes()
1265
for node in text_nodes:
1271
ideal_parents = tuple(ideal_index[node[1]])
1273
discarded_nodes.append(node)
1274
self._data_changed = True
1276
if ideal_parents == (NULL_REVISION,):
1278
if ideal_parents == node[3][0]:
1280
ok_nodes.append(node)
1281
elif ideal_parents[0:1] == node[3][0][0:1]:
1282
# the left most parent is the same, or there are no parents
1283
# today. Either way, we can preserve the representation as
1284
# long as we change the refs to be inserted.
1285
self._data_changed = True
1286
ok_nodes.append((node[0], node[1], node[2],
1287
(ideal_parents, node[3][1])))
1288
self._data_changed = True
1290
# Reinsert this text completely
1291
bad_texts.append((node[1], ideal_parents))
1292
self._data_changed = True
1293
# we're finished with some data.
1296
# 3) bulk copy the ok data
1297
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1298
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1299
self.new_pack.text_index, readv_group_iter, total_items))
1300
# 4) adhoc copy all the other texts.
1301
# We have to topologically insert all texts otherwise we can fail to
1302
# reconcile when parts of a single delta chain are preserved intact,
1303
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1304
# reinserted, and if d3 has incorrect parents it will also be
1305
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1306
# copied), so we will try to delta, but d2 is not currently able to be
1307
# extracted because it's basis d1 is not present. Topologically sorting
1308
# addresses this. The following generates a sort for all the texts that
1309
# are being inserted without having to reference the entire text key
1310
# space (we only topo sort the revisions, which is smaller).
1311
topo_order = tsort.topo_sort(ancestors)
1312
rev_order = dict(zip(topo_order, range(len(topo_order))))
1313
bad_texts.sort(key=lambda key:rev_order.get(key[0][1], 0))
1314
transaction = repo.get_transaction()
1315
file_id_index = GraphIndexPrefixAdapter(
1316
self.new_pack.text_index,
1318
add_nodes_callback=self.new_pack.text_index.add_nodes)
1319
data_access = _DirectPackAccess(
1320
{self.new_pack.text_index:self.new_pack.access_tuple()})
1321
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1322
self.new_pack.access_tuple())
1323
output_texts = KnitVersionedFiles(
1324
_KnitGraphIndex(self.new_pack.text_index,
1325
add_callback=self.new_pack.text_index.add_nodes,
1326
deltas=True, parents=True, is_locked=repo.is_locked),
1327
data_access=data_access, max_delta_chain=200)
1328
for key, parent_keys in bad_texts:
1329
# We refer to the new pack to delta data being output.
1330
# A possible improvement would be to catch errors on short reads
1331
# and only flush then.
1332
self.new_pack.flush()
1334
for parent_key in parent_keys:
1335
if parent_key[0] != key[0]:
1336
# Graph parents must match the fileid
1337
raise errors.BzrError('Mismatched key parent %r:%r' %
1339
parents.append(parent_key[1])
1340
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1341
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1342
output_texts.add_lines(key, parent_keys, text_lines,
1343
random_id=True, check_content=False)
1344
# 5) check that nothing inserted has a reference outside the keyspace.
1345
missing_text_keys = self.new_pack.text_index._external_references()
1346
if missing_text_keys:
1347
raise errors.BzrCheckError('Reference to missing compression parents %r'
1348
% (missing_text_keys,))
1349
self._log_copied_texts()
1351
def _use_pack(self, new_pack):
1352
"""Override _use_pack to check for reconcile having changed content."""
1353
# XXX: we might be better checking this at the copy time.
1354
original_inventory_keys = set()
1355
inv_index = self._pack_collection.inventory_index.combined_index
1356
for entry in inv_index.iter_all_entries():
1357
original_inventory_keys.add(entry[1])
1358
new_inventory_keys = set()
1359
for entry in new_pack.inventory_index.iter_all_entries():
1360
new_inventory_keys.add(entry[1])
1361
if new_inventory_keys != original_inventory_keys:
1362
self._data_changed = True
1363
return new_pack.data_inserted() and self._data_changed
1366
class RepositoryPackCollection(object):
1367
"""Management of packs within a repository.
1369
:ivar _names: map of {pack_name: (index_size,)}
1372
pack_factory = NewPack
1373
resumed_pack_factory = ResumedPack
1375
def __init__(self, repo, transport, index_transport, upload_transport,
1376
pack_transport, index_builder_class, index_class,
1378
"""Create a new RepositoryPackCollection.
1380
:param transport: Addresses the repository base directory
1381
(typically .bzr/repository/).
1382
:param index_transport: Addresses the directory containing indices.
1383
:param upload_transport: Addresses the directory into which packs are written
1384
while they're being created.
1385
:param pack_transport: Addresses the directory of existing complete packs.
1386
:param index_builder_class: The index builder class to use.
1387
:param index_class: The index class to use.
1388
:param use_chk_index: Whether to setup and manage a CHK index.
1390
# XXX: This should call self.reset()
1392
self.transport = transport
1393
self._index_transport = index_transport
1394
self._upload_transport = upload_transport
1395
self._pack_transport = pack_transport
1396
self._index_builder_class = index_builder_class
1397
self._index_class = index_class
1398
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3,
1403
self._packs_by_name = {}
1404
# the previous pack-names content
1405
self._packs_at_load = None
1406
# when a pack is being created by this object, the state of that pack.
1407
self._new_pack = None
1408
# aggregated revision index data
1409
flush = self._flush_new_pack
1410
self.revision_index = AggregateIndex(self.reload_pack_names, flush)
1411
self.inventory_index = AggregateIndex(self.reload_pack_names, flush)
1412
self.text_index = AggregateIndex(self.reload_pack_names, flush)
1413
self.signature_index = AggregateIndex(self.reload_pack_names, flush)
1415
self.chk_index = AggregateIndex(self.reload_pack_names, flush)
1417
# used to determine if we're using a chk_index elsewhere.
1418
self.chk_index = None
1420
self._resumed_packs = []
1422
def add_pack_to_memory(self, pack):
1423
"""Make a Pack object available to the repository to satisfy queries.
1425
:param pack: A Pack object.
1427
if pack.name in self._packs_by_name:
1428
raise AssertionError(
1429
'pack %s already in _packs_by_name' % (pack.name,))
1430
self.packs.append(pack)
1431
self._packs_by_name[pack.name] = pack
1432
self.revision_index.add_index(pack.revision_index, pack)
1433
self.inventory_index.add_index(pack.inventory_index, pack)
1434
self.text_index.add_index(pack.text_index, pack)
1435
self.signature_index.add_index(pack.signature_index, pack)
1436
if self.chk_index is not None:
1437
self.chk_index.add_index(pack.chk_index, pack)
1439
def all_packs(self):
1440
"""Return a list of all the Pack objects this repository has.
1442
Note that an in-progress pack being created is not returned.
1444
:return: A list of Pack objects for all the packs in the repository.
1447
for name in self.names():
1448
result.append(self.get_pack_by_name(name))
1452
"""Pack the pack collection incrementally.
1454
This will not attempt global reorganisation or recompression,
1455
rather it will just ensure that the total number of packs does
1456
not grow without bound. It uses the _max_pack_count method to
1457
determine if autopacking is needed, and the pack_distribution
1458
method to determine the number of revisions in each pack.
1460
If autopacking takes place then the packs name collection will have
1461
been flushed to disk - packing requires updating the name collection
1462
in synchronisation with certain steps. Otherwise the names collection
1465
:return: Something evaluating true if packing took place.
1469
return self._do_autopack()
1470
except errors.RetryAutopack:
1471
# If we get a RetryAutopack exception, we should abort the
1472
# current action, and retry.
1475
def _do_autopack(self):
1476
# XXX: Should not be needed when the management of indices is sane.
1477
total_revisions = self.revision_index.combined_index.key_count()
1478
total_packs = len(self._names)
1479
if self._max_pack_count(total_revisions) >= total_packs:
1481
# determine which packs need changing
1482
pack_distribution = self.pack_distribution(total_revisions)
1484
for pack in self.all_packs():
1485
revision_count = pack.get_revision_count()
1486
if revision_count == 0:
1487
# revision less packs are not generated by normal operation,
1488
# only by operations like sign-my-commits, and thus will not
1489
# tend to grow rapdily or without bound like commit containing
1490
# packs do - leave them alone as packing them really should
1491
# group their data with the relevant commit, and that may
1492
# involve rewriting ancient history - which autopack tries to
1493
# avoid. Alternatively we could not group the data but treat
1494
# each of these as having a single revision, and thus add
1495
# one revision for each to the total revision count, to get
1496
# a matching distribution.
1498
existing_packs.append((revision_count, pack))
1499
pack_operations = self.plan_autopack_combinations(
1500
existing_packs, pack_distribution)
1501
num_new_packs = len(pack_operations)
1502
num_old_packs = sum([len(po[1]) for po in pack_operations])
1503
num_revs_affected = sum([po[0] for po in pack_operations])
1504
mutter('Auto-packing repository %s, which has %d pack files, '
1505
'containing %d revisions. Packing %d files into %d affecting %d'
1506
' revisions', self, total_packs, total_revisions, num_old_packs,
1507
num_new_packs, num_revs_affected)
1508
result = self._execute_pack_operations(pack_operations,
1509
reload_func=self._restart_autopack)
1510
mutter('Auto-packing repository %s completed', self)
1513
def _execute_pack_operations(self, pack_operations, _packer_class=Packer,
1515
"""Execute a series of pack operations.
1517
:param pack_operations: A list of [revision_count, packs_to_combine].
1518
:param _packer_class: The class of packer to use (default: Packer).
1519
:return: The new pack names.
1521
for revision_count, packs in pack_operations:
1522
# we may have no-ops from the setup logic
1525
packer = _packer_class(self, packs, '.autopack',
1526
reload_func=reload_func)
1529
except errors.RetryWithNewPacks:
1530
# An exception is propagating out of this context, make sure
1531
# this packer has cleaned up. Packer() doesn't set its new_pack
1532
# state into the RepositoryPackCollection object, so we only
1533
# have access to it directly here.
1534
if packer.new_pack is not None:
1535
packer.new_pack.abort()
1538
self._remove_pack_from_memory(pack)
1539
# record the newly available packs and stop advertising the old
1541
result = self._save_pack_names(clear_obsolete_packs=True)
1542
# Move the old packs out of the way now they are no longer referenced.
1543
for revision_count, packs in pack_operations:
1544
self._obsolete_packs(packs)
1547
def _flush_new_pack(self):
1548
if self._new_pack is not None:
1549
self._new_pack.flush()
1551
def lock_names(self):
1552
"""Acquire the mutex around the pack-names index.
1554
This cannot be used in the middle of a read-only transaction on the
1557
self.repo.control_files.lock_write()
1559
def _already_packed(self):
1560
"""Is the collection already packed?"""
1561
return not (self.repo._format.pack_compresses or (len(self._names) > 1))
1563
def pack(self, hint=None):
1564
"""Pack the pack collection totally."""
1565
self.ensure_loaded()
1566
total_packs = len(self._names)
1567
if self._already_packed():
1569
total_revisions = self.revision_index.combined_index.key_count()
1570
# XXX: the following may want to be a class, to pack with a given
1572
mutter('Packing repository %s, which has %d pack files, '
1573
'containing %d revisions with hint %r.', self, total_packs,
1574
total_revisions, hint)
1575
# determine which packs need changing
1576
pack_operations = [[0, []]]
1577
for pack in self.all_packs():
1578
if hint is None or pack.name in hint:
1579
# Either no hint was provided (so we are packing everything),
1580
# or this pack was included in the hint.
1581
pack_operations[-1][0] += pack.get_revision_count()
1582
pack_operations[-1][1].append(pack)
1583
self._execute_pack_operations(pack_operations, OptimisingPacker)
1585
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1586
"""Plan a pack operation.
1588
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1590
:param pack_distribution: A list with the number of revisions desired
1593
if len(existing_packs) <= len(pack_distribution):
1595
existing_packs.sort(reverse=True)
1596
pack_operations = [[0, []]]
1597
# plan out what packs to keep, and what to reorganise
1598
while len(existing_packs):
1599
# take the largest pack, and if its less than the head of the
1600
# distribution chart we will include its contents in the new pack
1601
# for that position. If its larger, we remove its size from the
1602
# distribution chart
1603
next_pack_rev_count, next_pack = existing_packs.pop(0)
1604
if next_pack_rev_count >= pack_distribution[0]:
1605
# this is already packed 'better' than this, so we can
1606
# not waste time packing it.
1607
while next_pack_rev_count > 0:
1608
next_pack_rev_count -= pack_distribution[0]
1609
if next_pack_rev_count >= 0:
1611
del pack_distribution[0]
1613
# didn't use that entire bucket up
1614
pack_distribution[0] = -next_pack_rev_count
1616
# add the revisions we're going to add to the next output pack
1617
pack_operations[-1][0] += next_pack_rev_count
1618
# allocate this pack to the next pack sub operation
1619
pack_operations[-1][1].append(next_pack)
1620
if pack_operations[-1][0] >= pack_distribution[0]:
1621
# this pack is used up, shift left.
1622
del pack_distribution[0]
1623
pack_operations.append([0, []])
1624
# Now that we know which pack files we want to move, shove them all
1625
# into a single pack file.
1627
final_pack_list = []
1628
for num_revs, pack_files in pack_operations:
1629
final_rev_count += num_revs
1630
final_pack_list.extend(pack_files)
1631
if len(final_pack_list) == 1:
1632
raise AssertionError('We somehow generated an autopack with a'
1633
' single pack file being moved.')
1635
return [[final_rev_count, final_pack_list]]
1637
def ensure_loaded(self):
1638
"""Ensure we have read names from disk.
1640
:return: True if the disk names had not been previously read.
1642
# NB: if you see an assertion error here, its probably access against
1643
# an unlocked repo. Naughty.
1644
if not self.repo.is_locked():
1645
raise errors.ObjectNotLocked(self.repo)
1646
if self._names is None:
1648
self._packs_at_load = set()
1649
for index, key, value in self._iter_disk_pack_index():
1651
self._names[name] = self._parse_index_sizes(value)
1652
self._packs_at_load.add((key, value))
1656
# populate all the metadata.
1660
def _parse_index_sizes(self, value):
1661
"""Parse a string of index sizes."""
1662
return tuple([int(digits) for digits in value.split(' ')])
1664
def get_pack_by_name(self, name):
1665
"""Get a Pack object by name.
1667
:param name: The name of the pack - e.g. '123456'
1668
:return: A Pack object.
1671
return self._packs_by_name[name]
1673
rev_index = self._make_index(name, '.rix')
1674
inv_index = self._make_index(name, '.iix')
1675
txt_index = self._make_index(name, '.tix')
1676
sig_index = self._make_index(name, '.six')
1677
if self.chk_index is not None:
1678
chk_index = self._make_index(name, '.cix')
1681
result = ExistingPack(self._pack_transport, name, rev_index,
1682
inv_index, txt_index, sig_index, chk_index)
1683
self.add_pack_to_memory(result)
1686
def _resume_pack(self, name):
1687
"""Get a suspended Pack object by name.
1689
:param name: The name of the pack - e.g. '123456'
1690
:return: A Pack object.
1692
if not re.match('[a-f0-9]{32}', name):
1693
# Tokens should be md5sums of the suspended pack file, i.e. 32 hex
1695
raise errors.UnresumableWriteGroup(
1696
self.repo, [name], 'Malformed write group token')
1698
rev_index = self._make_index(name, '.rix', resume=True)
1699
inv_index = self._make_index(name, '.iix', resume=True)
1700
txt_index = self._make_index(name, '.tix', resume=True)
1701
sig_index = self._make_index(name, '.six', resume=True)
1702
if self.chk_index is not None:
1703
chk_index = self._make_index(name, '.cix', resume=True)
1706
result = self.resumed_pack_factory(name, rev_index, inv_index,
1707
txt_index, sig_index, self._upload_transport,
1708
self._pack_transport, self._index_transport, self,
1709
chk_index=chk_index)
1710
except errors.NoSuchFile, e:
1711
raise errors.UnresumableWriteGroup(self.repo, [name], str(e))
1712
self.add_pack_to_memory(result)
1713
self._resumed_packs.append(result)
1716
def allocate(self, a_new_pack):
1717
"""Allocate name in the list of packs.
1719
:param a_new_pack: A NewPack instance to be added to the collection of
1720
packs for this repository.
1722
self.ensure_loaded()
1723
if a_new_pack.name in self._names:
1724
raise errors.BzrError(
1725
'Pack %r already exists in %s' % (a_new_pack.name, self))
1726
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1727
self.add_pack_to_memory(a_new_pack)
1729
def _iter_disk_pack_index(self):
1730
"""Iterate over the contents of the pack-names index.
1732
This is used when loading the list from disk, and before writing to
1733
detect updates from others during our write operation.
1734
:return: An iterator of the index contents.
1736
return self._index_class(self.transport, 'pack-names', None
1737
).iter_all_entries()
1739
def _make_index(self, name, suffix, resume=False):
1740
size_offset = self._suffix_offsets[suffix]
1741
index_name = name + suffix
1743
transport = self._upload_transport
1744
index_size = transport.stat(index_name).st_size
1746
transport = self._index_transport
1747
index_size = self._names[name][size_offset]
1748
return self._index_class(transport, index_name, index_size)
1750
def _max_pack_count(self, total_revisions):
1751
"""Return the maximum number of packs to use for total revisions.
1753
:param total_revisions: The total number of revisions in the
1756
if not total_revisions:
1758
digits = str(total_revisions)
1760
for digit in digits:
1761
result += int(digit)
1765
"""Provide an order to the underlying names."""
1766
return sorted(self._names.keys())
1768
def _obsolete_packs(self, packs):
1769
"""Move a number of packs which have been obsoleted out of the way.
1771
Each pack and its associated indices are moved out of the way.
1773
Note: for correctness this function should only be called after a new
1774
pack names index has been written without these pack names, and with
1775
the names of packs that contain the data previously available via these
1778
:param packs: The packs to obsolete.
1779
:param return: None.
1782
pack.pack_transport.rename(pack.file_name(),
1783
'../obsolete_packs/' + pack.file_name())
1784
# TODO: Probably needs to know all possible indices for this pack
1785
# - or maybe list the directory and move all indices matching this
1786
# name whether we recognize it or not?
1787
suffixes = ['.iix', '.six', '.tix', '.rix']
1788
if self.chk_index is not None:
1789
suffixes.append('.cix')
1790
for suffix in suffixes:
1791
self._index_transport.rename(pack.name + suffix,
1792
'../obsolete_packs/' + pack.name + suffix)
1794
def pack_distribution(self, total_revisions):
1795
"""Generate a list of the number of revisions to put in each pack.
1797
:param total_revisions: The total number of revisions in the
1800
if total_revisions == 0:
1802
digits = reversed(str(total_revisions))
1804
for exponent, count in enumerate(digits):
1805
size = 10 ** exponent
1806
for pos in range(int(count)):
1808
return list(reversed(result))
1810
def _pack_tuple(self, name):
1811
"""Return a tuple with the transport and file name for a pack name."""
1812
return self._pack_transport, name + '.pack'
1814
def _remove_pack_from_memory(self, pack):
1815
"""Remove pack from the packs accessed by this repository.
1817
Only affects memory state, until self._save_pack_names() is invoked.
1819
self._names.pop(pack.name)
1820
self._packs_by_name.pop(pack.name)
1821
self._remove_pack_indices(pack)
1822
self.packs.remove(pack)
1824
def _remove_pack_indices(self, pack):
1825
"""Remove the indices for pack from the aggregated indices."""
1826
self.revision_index.remove_index(pack.revision_index, pack)
1827
self.inventory_index.remove_index(pack.inventory_index, pack)
1828
self.text_index.remove_index(pack.text_index, pack)
1829
self.signature_index.remove_index(pack.signature_index, pack)
1830
if self.chk_index is not None:
1831
self.chk_index.remove_index(pack.chk_index, pack)
1834
"""Clear all cached data."""
1835
# cached revision data
1836
self.revision_index.clear()
1837
# cached signature data
1838
self.signature_index.clear()
1839
# cached file text data
1840
self.text_index.clear()
1841
# cached inventory data
1842
self.inventory_index.clear()
1844
if self.chk_index is not None:
1845
self.chk_index.clear()
1846
# remove the open pack
1847
self._new_pack = None
1848
# information about packs.
1851
self._packs_by_name = {}
1852
self._packs_at_load = None
1854
def _unlock_names(self):
1855
"""Release the mutex around the pack-names index."""
1856
self.repo.control_files.unlock()
1858
def _diff_pack_names(self):
1859
"""Read the pack names from disk, and compare it to the one in memory.
1861
:return: (disk_nodes, deleted_nodes, new_nodes)
1862
disk_nodes The final set of nodes that should be referenced
1863
deleted_nodes Nodes which have been removed from when we started
1864
new_nodes Nodes that are newly introduced
1866
# load the disk nodes across
1868
for index, key, value in self._iter_disk_pack_index():
1869
disk_nodes.add((key, value))
1871
# do a two-way diff against our original content
1872
current_nodes = set()
1873
for name, sizes in self._names.iteritems():
1875
((name, ), ' '.join(str(size) for size in sizes)))
1877
# Packs no longer present in the repository, which were present when we
1878
# locked the repository
1879
deleted_nodes = self._packs_at_load - current_nodes
1880
# Packs which this process is adding
1881
new_nodes = current_nodes - self._packs_at_load
1883
# Update the disk_nodes set to include the ones we are adding, and
1884
# remove the ones which were removed by someone else
1885
disk_nodes.difference_update(deleted_nodes)
1886
disk_nodes.update(new_nodes)
1888
return disk_nodes, deleted_nodes, new_nodes
1890
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1891
"""Given the correct set of pack files, update our saved info.
1893
:return: (removed, added, modified)
1894
removed pack names removed from self._names
1895
added pack names added to self._names
1896
modified pack names that had changed value
1901
## self._packs_at_load = disk_nodes
1902
new_names = dict(disk_nodes)
1903
# drop no longer present nodes
1904
for pack in self.all_packs():
1905
if (pack.name,) not in new_names:
1906
removed.append(pack.name)
1907
self._remove_pack_from_memory(pack)
1908
# add new nodes/refresh existing ones
1909
for key, value in disk_nodes:
1911
sizes = self._parse_index_sizes(value)
1912
if name in self._names:
1914
if sizes != self._names[name]:
1915
# the pack for name has had its indices replaced - rare but
1916
# important to handle. XXX: probably can never happen today
1917
# because the three-way merge code above does not handle it
1918
# - you may end up adding the same key twice to the new
1919
# disk index because the set values are the same, unless
1920
# the only index shows up as deleted by the set difference
1921
# - which it may. Until there is a specific test for this,
1922
# assume its broken. RBC 20071017.
1923
self._remove_pack_from_memory(self.get_pack_by_name(name))
1924
self._names[name] = sizes
1925
self.get_pack_by_name(name)
1926
modified.append(name)
1929
self._names[name] = sizes
1930
self.get_pack_by_name(name)
1932
return removed, added, modified
1934
def _save_pack_names(self, clear_obsolete_packs=False):
1935
"""Save the list of packs.
1937
This will take out the mutex around the pack names list for the
1938
duration of the method call. If concurrent updates have been made, a
1939
three-way merge between the current list and the current in memory list
1942
:param clear_obsolete_packs: If True, clear out the contents of the
1943
obsolete_packs directory.
1944
:return: A list of the names saved that were not previously on disk.
1948
builder = self._index_builder_class()
1949
disk_nodes, deleted_nodes, new_nodes = self._diff_pack_names()
1950
# TODO: handle same-name, index-size-changes here -
1951
# e.g. use the value from disk, not ours, *unless* we're the one
1953
for key, value in disk_nodes:
1954
builder.add_node(key, value)
1955
self.transport.put_file('pack-names', builder.finish(),
1956
mode=self.repo.bzrdir._get_file_mode())
1957
# move the baseline forward
1958
self._packs_at_load = disk_nodes
1959
if clear_obsolete_packs:
1960
self._clear_obsolete_packs()
1962
self._unlock_names()
1963
# synchronise the memory packs list with what we just wrote:
1964
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1965
return [new_node[0][0] for new_node in new_nodes]
1967
def reload_pack_names(self):
1968
"""Sync our pack listing with what is present in the repository.
1970
This should be called when we find out that something we thought was
1971
present is now missing. This happens when another process re-packs the
1974
:return: True if the in-memory list of packs has been altered at all.
1976
# The ensure_loaded call is to handle the case where the first call
1977
# made involving the collection was to reload_pack_names, where we
1978
# don't have a view of disk contents. Its a bit of a bandaid, and
1979
# causes two reads of pack-names, but its a rare corner case not struck
1980
# with regular push/pull etc.
1981
first_read = self.ensure_loaded()
1984
# out the new value.
1985
disk_nodes, _, _ = self._diff_pack_names()
1986
self._packs_at_load = disk_nodes
1988
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1989
if removed or added or modified:
1993
def _restart_autopack(self):
1994
"""Reload the pack names list, and restart the autopack code."""
1995
if not self.reload_pack_names():
1996
# Re-raise the original exception, because something went missing
1997
# and a restart didn't find it
1999
raise errors.RetryAutopack(self.repo, False, sys.exc_info())
2001
def _clear_obsolete_packs(self):
2002
"""Delete everything from the obsolete-packs directory.
2004
obsolete_pack_transport = self.transport.clone('obsolete_packs')
2005
for filename in obsolete_pack_transport.list_dir('.'):
2007
obsolete_pack_transport.delete(filename)
2008
except (errors.PathError, errors.TransportError), e:
2009
warning("couldn't delete obsolete pack, skipping it:\n%s" % (e,))
2011
def _start_write_group(self):
2012
# Do not permit preparation for writing if we're not in a 'write lock'.
2013
if not self.repo.is_write_locked():
2014
raise errors.NotWriteLocked(self)
2015
self._new_pack = self.pack_factory(self, upload_suffix='.pack',
2016
file_mode=self.repo.bzrdir._get_file_mode())
2017
# allow writing: queue writes to a new index
2018
self.revision_index.add_writable_index(self._new_pack.revision_index,
2020
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
2022
self.text_index.add_writable_index(self._new_pack.text_index,
2024
self._new_pack.text_index.set_optimize(combine_backing_indices=False)
2025
self.signature_index.add_writable_index(self._new_pack.signature_index,
2027
if self.chk_index is not None:
2028
self.chk_index.add_writable_index(self._new_pack.chk_index,
2030
self.repo.chk_bytes._index._add_callback = self.chk_index.add_callback
2031
self._new_pack.chk_index.set_optimize(combine_backing_indices=False)
2033
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
2034
self.repo.revisions._index._add_callback = self.revision_index.add_callback
2035
self.repo.signatures._index._add_callback = self.signature_index.add_callback
2036
self.repo.texts._index._add_callback = self.text_index.add_callback
2038
def _abort_write_group(self):
2039
# FIXME: just drop the transient index.
2040
# forget what names there are
2041
if self._new_pack is not None:
2043
self._new_pack.abort()
2045
# XXX: If we aborted while in the middle of finishing the write
2046
# group, _remove_pack_indices can fail because the indexes are
2047
# already gone. If they're not there we shouldn't fail in this
2048
# case. -- mbp 20081113
2049
self._remove_pack_indices(self._new_pack)
2050
self._new_pack = None
2051
for resumed_pack in self._resumed_packs:
2053
resumed_pack.abort()
2055
# See comment in previous finally block.
2057
self._remove_pack_indices(resumed_pack)
2060
del self._resumed_packs[:]
2062
def _remove_resumed_pack_indices(self):
2063
for resumed_pack in self._resumed_packs:
2064
self._remove_pack_indices(resumed_pack)
2065
del self._resumed_packs[:]
2067
def _check_new_inventories(self):
2068
"""Detect missing inventories in this write group.
2070
:returns: list of strs, summarising any problems found. If the list is
2071
empty no problems were found.
2073
# The base implementation does no checks. GCRepositoryPackCollection
2077
def _commit_write_group(self):
2079
for prefix, versioned_file in (
2080
('revisions', self.repo.revisions),
2081
('inventories', self.repo.inventories),
2082
('texts', self.repo.texts),
2083
('signatures', self.repo.signatures),
2085
missing = versioned_file.get_missing_compression_parent_keys()
2086
all_missing.update([(prefix,) + key for key in missing])
2088
raise errors.BzrCheckError(
2089
"Repository %s has missing compression parent(s) %r "
2090
% (self.repo, sorted(all_missing)))
2091
problems = self._check_new_inventories()
2093
problems_summary = '\n'.join(problems)
2094
raise errors.BzrCheckError(
2095
"Cannot add revision(s) to repository: " + problems_summary)
2096
self._remove_pack_indices(self._new_pack)
2097
any_new_content = False
2098
if self._new_pack.data_inserted():
2099
# get all the data to disk and read to use
2100
self._new_pack.finish()
2101
self.allocate(self._new_pack)
2102
self._new_pack = None
2103
any_new_content = True
2105
self._new_pack.abort()
2106
self._new_pack = None
2107
for resumed_pack in self._resumed_packs:
2108
# XXX: this is a pretty ugly way to turn the resumed pack into a
2109
# properly committed pack.
2110
self._names[resumed_pack.name] = None
2111
self._remove_pack_from_memory(resumed_pack)
2112
resumed_pack.finish()
2113
self.allocate(resumed_pack)
2114
any_new_content = True
2115
del self._resumed_packs[:]
2117
result = self.autopack()
2119
# when autopack takes no steps, the names list is still
2121
return self._save_pack_names()
2125
def _suspend_write_group(self):
2126
tokens = [pack.name for pack in self._resumed_packs]
2127
self._remove_pack_indices(self._new_pack)
2128
if self._new_pack.data_inserted():
2129
# get all the data to disk and read to use
2130
self._new_pack.finish(suspend=True)
2131
tokens.append(self._new_pack.name)
2132
self._new_pack = None
2134
self._new_pack.abort()
2135
self._new_pack = None
2136
self._remove_resumed_pack_indices()
2139
def _resume_write_group(self, tokens):
2140
for token in tokens:
2141
self._resume_pack(token)
2144
class KnitPackRepository(KnitRepository):
2145
"""Repository with knit objects stored inside pack containers.
2147
The layering for a KnitPackRepository is:
2149
Graph | HPSS | Repository public layer |
2150
===================================================
2151
Tuple based apis below, string based, and key based apis above
2152
---------------------------------------------------
2154
Provides .texts, .revisions etc
2155
This adapts the N-tuple keys to physical knit records which only have a
2156
single string identifier (for historical reasons), which in older formats
2157
was always the revision_id, and in the mapped code for packs is always
2158
the last element of key tuples.
2159
---------------------------------------------------
2161
A separate GraphIndex is used for each of the
2162
texts/inventories/revisions/signatures contained within each individual
2163
pack file. The GraphIndex layer works in N-tuples and is unaware of any
2165
===================================================
2169
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
2171
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2172
_commit_builder_class, _serializer)
2173
index_transport = self._transport.clone('indices')
2174
self._pack_collection = RepositoryPackCollection(self, self._transport,
2176
self._transport.clone('upload'),
2177
self._transport.clone('packs'),
2178
_format.index_builder_class,
2179
_format.index_class,
2180
use_chk_index=self._format.supports_chks,
2182
self.inventories = KnitVersionedFiles(
2183
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2184
add_callback=self._pack_collection.inventory_index.add_callback,
2185
deltas=True, parents=True, is_locked=self.is_locked),
2186
data_access=self._pack_collection.inventory_index.data_access,
2187
max_delta_chain=200)
2188
self.revisions = KnitVersionedFiles(
2189
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2190
add_callback=self._pack_collection.revision_index.add_callback,
2191
deltas=False, parents=True, is_locked=self.is_locked,
2192
track_external_parent_refs=True),
2193
data_access=self._pack_collection.revision_index.data_access,
2195
self.signatures = KnitVersionedFiles(
2196
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2197
add_callback=self._pack_collection.signature_index.add_callback,
2198
deltas=False, parents=False, is_locked=self.is_locked),
2199
data_access=self._pack_collection.signature_index.data_access,
2201
self.texts = KnitVersionedFiles(
2202
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2203
add_callback=self._pack_collection.text_index.add_callback,
2204
deltas=True, parents=True, is_locked=self.is_locked),
2205
data_access=self._pack_collection.text_index.data_access,
2206
max_delta_chain=200)
2207
if _format.supports_chks:
2208
# No graph, no compression:- references from chks are between
2209
# different objects not temporal versions of the same; and without
2210
# some sort of temporal structure knit compression will just fail.
2211
self.chk_bytes = KnitVersionedFiles(
2212
_KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2213
add_callback=self._pack_collection.chk_index.add_callback,
2214
deltas=False, parents=False, is_locked=self.is_locked),
2215
data_access=self._pack_collection.chk_index.data_access,
2218
self.chk_bytes = None
2219
# True when the repository object is 'write locked' (as opposed to the
2220
# physical lock only taken out around changes to the pack-names list.)
2221
# Another way to represent this would be a decorator around the control
2222
# files object that presents logical locks as physical ones - if this
2223
# gets ugly consider that alternative design. RBC 20071011
2224
self._write_lock_count = 0
2225
self._transaction = None
2227
self._reconcile_does_inventory_gc = True
2228
self._reconcile_fixes_text_parents = True
2229
self._reconcile_backsup_inventory = False
2231
def _warn_if_deprecated(self):
2232
# This class isn't deprecated, but one sub-format is
2233
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2234
from bzrlib import repository
2235
if repository._deprecation_warning_done:
2237
repository._deprecation_warning_done = True
2238
warning("Format %s for %s is deprecated - please use"
2239
" 'bzr upgrade --1.6.1-rich-root'"
2240
% (self._format, self.bzrdir.transport.base))
2242
def _abort_write_group(self):
2243
self.revisions._index._key_dependencies.clear()
2244
self._pack_collection._abort_write_group()
2246
def _get_source(self, to_format):
2247
if to_format.network_name() == self._format.network_name():
2248
return KnitPackStreamSource(self, to_format)
2249
return super(KnitPackRepository, self)._get_source(to_format)
2251
def _make_parents_provider(self):
2252
return graph.CachingParentsProvider(self)
2254
def _refresh_data(self):
2255
if not self.is_locked():
2257
self._pack_collection.reload_pack_names()
2259
def _start_write_group(self):
2260
self._pack_collection._start_write_group()
2262
def _commit_write_group(self):
2263
hint = self._pack_collection._commit_write_group()
2264
self.revisions._index._key_dependencies.clear()
2267
def suspend_write_group(self):
2268
# XXX check self._write_group is self.get_transaction()?
2269
tokens = self._pack_collection._suspend_write_group()
2270
self.revisions._index._key_dependencies.clear()
2271
self._write_group = None
2274
def _resume_write_group(self, tokens):
2275
self._start_write_group()
2277
self._pack_collection._resume_write_group(tokens)
2278
except errors.UnresumableWriteGroup:
2279
self._abort_write_group()
2281
for pack in self._pack_collection._resumed_packs:
2282
self.revisions._index.scan_unvalidated_index(pack.revision_index)
2284
def get_transaction(self):
2285
if self._write_lock_count:
2286
return self._transaction
2288
return self.control_files.get_transaction()
2290
def is_locked(self):
2291
return self._write_lock_count or self.control_files.is_locked()
2293
def is_write_locked(self):
2294
return self._write_lock_count
2296
def lock_write(self, token=None):
2297
locked = self.is_locked()
2298
if not self._write_lock_count and locked:
2299
raise errors.ReadOnlyError(self)
2300
self._write_lock_count += 1
2301
if self._write_lock_count == 1:
2302
self._transaction = transactions.WriteTransaction()
2304
if 'relock' in debug.debug_flags and self._prev_lock == 'w':
2305
note('%r was write locked again', self)
2306
self._prev_lock = 'w'
2307
for repo in self._fallback_repositories:
2308
# Writes don't affect fallback repos
2310
self._refresh_data()
2312
def lock_read(self):
2313
locked = self.is_locked()
2314
if self._write_lock_count:
2315
self._write_lock_count += 1
2317
self.control_files.lock_read()
2319
if 'relock' in debug.debug_flags and self._prev_lock == 'r':
2320
note('%r was read locked again', self)
2321
self._prev_lock = 'r'
2322
for repo in self._fallback_repositories:
2324
self._refresh_data()
2326
def leave_lock_in_place(self):
2327
# not supported - raise an error
2328
raise NotImplementedError(self.leave_lock_in_place)
2330
def dont_leave_lock_in_place(self):
2331
# not supported - raise an error
2332
raise NotImplementedError(self.dont_leave_lock_in_place)
2335
def pack(self, hint=None):
2336
"""Compress the data within the repository.
2338
This will pack all the data to a single pack. In future it may
2339
recompress deltas or do other such expensive operations.
2341
self._pack_collection.pack(hint=hint)
2344
def reconcile(self, other=None, thorough=False):
2345
"""Reconcile this repository."""
2346
from bzrlib.reconcile import PackReconciler
2347
reconciler = PackReconciler(self, thorough=thorough)
2348
reconciler.reconcile()
2351
def _reconcile_pack(self, collection, packs, extension, revs, pb):
2352
packer = ReconcilePacker(collection, packs, extension, revs)
2353
return packer.pack(pb)
2355
@only_raises(errors.LockNotHeld, errors.LockBroken)
2357
if self._write_lock_count == 1 and self._write_group is not None:
2358
self.abort_write_group()
2359
self._transaction = None
2360
self._write_lock_count = 0
2361
raise errors.BzrError(
2362
'Must end write group before releasing write lock on %s'
2364
if self._write_lock_count:
2365
self._write_lock_count -= 1
2366
if not self._write_lock_count:
2367
transaction = self._transaction
2368
self._transaction = None
2369
transaction.finish()
2371
self.control_files.unlock()
2373
if not self.is_locked():
2374
for repo in self._fallback_repositories:
2378
class KnitPackStreamSource(StreamSource):
2379
"""A StreamSource used to transfer data between same-format KnitPack repos.
2381
This source assumes:
2382
1) Same serialization format for all objects
2383
2) Same root information
2384
3) XML format inventories
2385
4) Atomic inserts (so we can stream inventory texts before text
2390
def __init__(self, from_repository, to_format):
2391
super(KnitPackStreamSource, self).__init__(from_repository, to_format)
2392
self._text_keys = None
2393
self._text_fetch_order = 'unordered'
2395
def _get_filtered_inv_stream(self, revision_ids):
2396
from_repo = self.from_repository
2397
parent_ids = from_repo._find_parent_ids_of_revisions(revision_ids)
2398
parent_keys = [(p,) for p in parent_ids]
2399
find_text_keys = from_repo._find_text_key_references_from_xml_inventory_lines
2400
parent_text_keys = set(find_text_keys(
2401
from_repo._inventory_xml_lines_for_keys(parent_keys)))
2402
content_text_keys = set()
2403
knit = KnitVersionedFiles(None, None)
2404
factory = KnitPlainFactory()
2405
def find_text_keys_from_content(record):
2406
if record.storage_kind not in ('knit-delta-gz', 'knit-ft-gz'):
2407
raise ValueError("Unknown content storage kind for"
2408
" inventory text: %s" % (record.storage_kind,))
2409
# It's a knit record, it has a _raw_record field (even if it was
2410
# reconstituted from a network stream).
2411
raw_data = record._raw_record
2412
# read the entire thing
2413
revision_id = record.key[-1]
2414
content, _ = knit._parse_record(revision_id, raw_data)
2415
if record.storage_kind == 'knit-delta-gz':
2416
line_iterator = factory.get_linedelta_content(content)
2417
elif record.storage_kind == 'knit-ft-gz':
2418
line_iterator = factory.get_fulltext_content(content)
2419
content_text_keys.update(find_text_keys(
2420
[(line, revision_id) for line in line_iterator]))
2421
revision_keys = [(r,) for r in revision_ids]
2422
def _filtered_inv_stream():
2423
source_vf = from_repo.inventories
2424
stream = source_vf.get_record_stream(revision_keys,
2426
for record in stream:
2427
if record.storage_kind == 'absent':
2428
raise errors.NoSuchRevision(from_repo, record.key)
2429
find_text_keys_from_content(record)
2431
self._text_keys = content_text_keys - parent_text_keys
2432
return ('inventories', _filtered_inv_stream())
2434
def _get_text_stream(self):
2435
# Note: We know we don't have to handle adding root keys, because both
2436
# the source and target are the identical network name.
2437
text_stream = self.from_repository.texts.get_record_stream(
2438
self._text_keys, self._text_fetch_order, False)
2439
return ('texts', text_stream)
2441
def get_stream(self, search):
2442
revision_ids = search.get_keys()
2443
for stream_info in self._fetch_revision_texts(revision_ids):
2445
self._revision_keys = [(rev_id,) for rev_id in revision_ids]
2446
yield self._get_filtered_inv_stream(revision_ids)
2447
yield self._get_text_stream()
2451
class RepositoryFormatPack(MetaDirRepositoryFormat):
2452
"""Format logic for pack structured repositories.
2454
This repository format has:
2455
- a list of packs in pack-names
2456
- packs in packs/NAME.pack
2457
- indices in indices/NAME.{iix,six,tix,rix}
2458
- knit deltas in the packs, knit indices mapped to the indices.
2459
- thunk objects to support the knits programming API.
2460
- a format marker of its own
2461
- an optional 'shared-storage' flag
2462
- an optional 'no-working-trees' flag
2466
# Set this attribute in derived classes to control the repository class
2467
# created by open and initialize.
2468
repository_class = None
2469
# Set this attribute in derived classes to control the
2470
# _commit_builder_class that the repository objects will have passed to
2471
# their constructor.
2472
_commit_builder_class = None
2473
# Set this attribute in derived clases to control the _serializer that the
2474
# repository objects will have passed to their constructor.
2476
# Packs are not confused by ghosts.
2477
supports_ghosts = True
2478
# External references are not supported in pack repositories yet.
2479
supports_external_lookups = False
2480
# Most pack formats do not use chk lookups.
2481
supports_chks = False
2482
# What index classes to use
2483
index_builder_class = None
2485
_fetch_uses_deltas = True
2488
def initialize(self, a_bzrdir, shared=False):
2489
"""Create a pack based repository.
2491
:param a_bzrdir: bzrdir to contain the new repository; must already
2493
:param shared: If true the repository will be initialized as a shared
2496
mutter('creating repository in %s.', a_bzrdir.transport.base)
2497
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
2498
builder = self.index_builder_class()
2499
files = [('pack-names', builder.finish())]
2500
utf8_files = [('format', self.get_format_string())]
2502
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
2503
return self.open(a_bzrdir=a_bzrdir, _found=True)
2505
def open(self, a_bzrdir, _found=False, _override_transport=None):
2506
"""See RepositoryFormat.open().
2508
:param _override_transport: INTERNAL USE ONLY. Allows opening the
2509
repository at a slightly different url
2510
than normal. I.e. during 'upgrade'.
2513
format = RepositoryFormat.find_format(a_bzrdir)
2514
if _override_transport is not None:
2515
repo_transport = _override_transport
2517
repo_transport = a_bzrdir.get_repository_transport(None)
2518
control_files = lockable_files.LockableFiles(repo_transport,
2519
'lock', lockdir.LockDir)
2520
return self.repository_class(_format=self,
2522
control_files=control_files,
2523
_commit_builder_class=self._commit_builder_class,
2524
_serializer=self._serializer)
2527
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2528
"""A no-subtrees parameterized Pack repository.
2530
This format was introduced in 0.92.
2533
repository_class = KnitPackRepository
2534
_commit_builder_class = PackCommitBuilder
2536
def _serializer(self):
2537
return xml5.serializer_v5
2538
# What index classes to use
2539
index_builder_class = InMemoryGraphIndex
2540
index_class = GraphIndex
2542
def _get_matching_bzrdir(self):
2543
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2545
def _ignore_setting_bzrdir(self, format):
2548
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2550
def get_format_string(self):
2551
"""See RepositoryFormat.get_format_string()."""
2552
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2554
def get_format_description(self):
2555
"""See RepositoryFormat.get_format_description()."""
2556
return "Packs containing knits without subtree support"
2559
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2560
"""A subtrees parameterized Pack repository.
2562
This repository format uses the xml7 serializer to get:
2563
- support for recording full info about the tree root
2564
- support for recording tree-references
2566
This format was introduced in 0.92.
2569
repository_class = KnitPackRepository
2570
_commit_builder_class = PackRootCommitBuilder
2571
rich_root_data = True
2572
supports_tree_reference = True
2574
def _serializer(self):
2575
return xml7.serializer_v7
2576
# What index classes to use
2577
index_builder_class = InMemoryGraphIndex
2578
index_class = GraphIndex
2580
def _get_matching_bzrdir(self):
2581
return bzrdir.format_registry.make_bzrdir(
2582
'pack-0.92-subtree')
2584
def _ignore_setting_bzrdir(self, format):
2587
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2589
def get_format_string(self):
2590
"""See RepositoryFormat.get_format_string()."""
2591
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2593
def get_format_description(self):
2594
"""See RepositoryFormat.get_format_description()."""
2595
return "Packs containing knits with subtree support\n"
2598
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2599
"""A rich-root, no subtrees parameterized Pack repository.
2601
This repository format uses the xml6 serializer to get:
2602
- support for recording full info about the tree root
2604
This format was introduced in 1.0.
2607
repository_class = KnitPackRepository
2608
_commit_builder_class = PackRootCommitBuilder
2609
rich_root_data = True
2610
supports_tree_reference = False
2612
def _serializer(self):
2613
return xml6.serializer_v6
2614
# What index classes to use
2615
index_builder_class = InMemoryGraphIndex
2616
index_class = GraphIndex
2618
def _get_matching_bzrdir(self):
2619
return bzrdir.format_registry.make_bzrdir(
2622
def _ignore_setting_bzrdir(self, format):
2625
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2627
def get_format_string(self):
2628
"""See RepositoryFormat.get_format_string()."""
2629
return ("Bazaar pack repository format 1 with rich root"
2630
" (needs bzr 1.0)\n")
2632
def get_format_description(self):
2633
"""See RepositoryFormat.get_format_description()."""
2634
return "Packs containing knits with rich root support\n"
2637
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2638
"""Repository that supports external references to allow stacking.
2642
Supports external lookups, which results in non-truncated ghosts after
2643
reconcile compared to pack-0.92 formats.
2646
repository_class = KnitPackRepository
2647
_commit_builder_class = PackCommitBuilder
2648
supports_external_lookups = True
2649
# What index classes to use
2650
index_builder_class = InMemoryGraphIndex
2651
index_class = GraphIndex
2654
def _serializer(self):
2655
return xml5.serializer_v5
2657
def _get_matching_bzrdir(self):
2658
return bzrdir.format_registry.make_bzrdir('1.6')
2660
def _ignore_setting_bzrdir(self, format):
2663
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2665
def get_format_string(self):
2666
"""See RepositoryFormat.get_format_string()."""
2667
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2669
def get_format_description(self):
2670
"""See RepositoryFormat.get_format_description()."""
2671
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2674
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2675
"""A repository with rich roots and stacking.
2677
New in release 1.6.1.
2679
Supports stacking on other repositories, allowing data to be accessed
2680
without being stored locally.
2683
repository_class = KnitPackRepository
2684
_commit_builder_class = PackRootCommitBuilder
2685
rich_root_data = True
2686
supports_tree_reference = False # no subtrees
2687
supports_external_lookups = True
2688
# What index classes to use
2689
index_builder_class = InMemoryGraphIndex
2690
index_class = GraphIndex
2693
def _serializer(self):
2694
return xml6.serializer_v6
2696
def _get_matching_bzrdir(self):
2697
return bzrdir.format_registry.make_bzrdir(
2700
def _ignore_setting_bzrdir(self, format):
2703
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2705
def get_format_string(self):
2706
"""See RepositoryFormat.get_format_string()."""
2707
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2709
def get_format_description(self):
2710
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2713
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2714
"""A repository with rich roots and external references.
2718
Supports external lookups, which results in non-truncated ghosts after
2719
reconcile compared to pack-0.92 formats.
2721
This format was deprecated because the serializer it uses accidentally
2722
supported subtrees, when the format was not intended to. This meant that
2723
someone could accidentally fetch from an incorrect repository.
2726
repository_class = KnitPackRepository
2727
_commit_builder_class = PackRootCommitBuilder
2728
rich_root_data = True
2729
supports_tree_reference = False # no subtrees
2731
supports_external_lookups = True
2732
# What index classes to use
2733
index_builder_class = InMemoryGraphIndex
2734
index_class = GraphIndex
2737
def _serializer(self):
2738
return xml7.serializer_v7
2740
def _get_matching_bzrdir(self):
2741
matching = bzrdir.format_registry.make_bzrdir(
2743
matching.repository_format = self
2746
def _ignore_setting_bzrdir(self, format):
2749
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2751
def get_format_string(self):
2752
"""See RepositoryFormat.get_format_string()."""
2753
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2755
def get_format_description(self):
2756
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2760
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2761
"""A repository with stacking and btree indexes,
2762
without rich roots or subtrees.
2764
This is equivalent to pack-1.6 with B+Tree indices.
2767
repository_class = KnitPackRepository
2768
_commit_builder_class = PackCommitBuilder
2769
supports_external_lookups = True
2770
# What index classes to use
2771
index_builder_class = BTreeBuilder
2772
index_class = BTreeGraphIndex
2775
def _serializer(self):
2776
return xml5.serializer_v5
2778
def _get_matching_bzrdir(self):
2779
return bzrdir.format_registry.make_bzrdir('1.9')
2781
def _ignore_setting_bzrdir(self, format):
2784
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2786
def get_format_string(self):
2787
"""See RepositoryFormat.get_format_string()."""
2788
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2790
def get_format_description(self):
2791
"""See RepositoryFormat.get_format_description()."""
2792
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2795
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2796
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2798
1.6-rich-root with B+Tree indices.
2801
repository_class = KnitPackRepository
2802
_commit_builder_class = PackRootCommitBuilder
2803
rich_root_data = True
2804
supports_tree_reference = False # no subtrees
2805
supports_external_lookups = True
2806
# What index classes to use
2807
index_builder_class = BTreeBuilder
2808
index_class = BTreeGraphIndex
2811
def _serializer(self):
2812
return xml6.serializer_v6
2814
def _get_matching_bzrdir(self):
2815
return bzrdir.format_registry.make_bzrdir(
2818
def _ignore_setting_bzrdir(self, format):
2821
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2823
def get_format_string(self):
2824
"""See RepositoryFormat.get_format_string()."""
2825
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2827
def get_format_description(self):
2828
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2831
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2832
"""A subtrees development repository.
2834
This format should be retained until the second release after bzr 1.7.
2836
1.6.1-subtree[as it might have been] with B+Tree indices.
2838
This is [now] retained until we have a CHK based subtree format in
2842
repository_class = KnitPackRepository
2843
_commit_builder_class = PackRootCommitBuilder
2844
rich_root_data = True
2845
supports_tree_reference = True
2846
supports_external_lookups = True
2847
# What index classes to use
2848
index_builder_class = BTreeBuilder
2849
index_class = BTreeGraphIndex
2852
def _serializer(self):
2853
return xml7.serializer_v7
2855
def _get_matching_bzrdir(self):
2856
return bzrdir.format_registry.make_bzrdir(
2857
'development-subtree')
2859
def _ignore_setting_bzrdir(self, format):
2862
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2864
def get_format_string(self):
2865
"""See RepositoryFormat.get_format_string()."""
2866
return ("Bazaar development format 2 with subtree support "
2867
"(needs bzr.dev from before 1.8)\n")
2869
def get_format_description(self):
2870
"""See RepositoryFormat.get_format_description()."""
2871
return ("Development repository format, currently the same as "
2872
"1.6.1-subtree with B+Tree indices.\n")