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
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 (
80
class PackCommitBuilder(CommitBuilder):
81
"""A subclass of CommitBuilder to add texts with pack semantics.
83
Specifically this uses one knit object rather than one knit object per
84
added text, reducing memory and object pressure.
87
def __init__(self, repository, parents, config, timestamp=None,
88
timezone=None, committer=None, revprops=None,
90
CommitBuilder.__init__(self, repository, parents, config,
91
timestamp=timestamp, timezone=timezone, committer=committer,
92
revprops=revprops, revision_id=revision_id)
93
self._file_graph = graph.Graph(
94
repository._pack_collection.text_index.combined_index)
96
def _heads(self, file_id, revision_ids):
97
keys = [(file_id, revision_id) for revision_id in revision_ids]
98
return set([key[1] for key in self._file_graph.heads(keys)])
101
class PackRootCommitBuilder(RootCommitBuilder):
102
"""A subclass of RootCommitBuilder to add texts with pack semantics.
104
Specifically this uses one knit object rather than one knit object per
105
added text, reducing memory and object pressure.
108
def __init__(self, repository, parents, config, timestamp=None,
109
timezone=None, committer=None, revprops=None,
111
CommitBuilder.__init__(self, repository, parents, config,
112
timestamp=timestamp, timezone=timezone, committer=committer,
113
revprops=revprops, revision_id=revision_id)
114
self._file_graph = graph.Graph(
115
repository._pack_collection.text_index.combined_index)
117
def _heads(self, file_id, revision_ids):
118
keys = [(file_id, revision_id) for revision_id in revision_ids]
119
return set([key[1] for key in self._file_graph.heads(keys)])
123
"""An in memory proxy for a pack and its indices.
125
This is a base class that is not directly used, instead the classes
126
ExistingPack and NewPack are used.
129
# A map of index 'type' to the file extension and position in the
131
index_definitions = {
133
'revision': ('.rix', 0),
134
'inventory': ('.iix', 1),
136
'signature': ('.six', 3),
139
def __init__(self, revision_index, inventory_index, text_index,
140
signature_index, chk_index=None):
141
"""Create a pack instance.
143
:param revision_index: A GraphIndex for determining what revisions are
144
present in the Pack and accessing the locations of their texts.
145
:param inventory_index: A GraphIndex for determining what inventories are
146
present in the Pack and accessing the locations of their
148
:param text_index: A GraphIndex for determining what file texts
149
are present in the pack and accessing the locations of their
150
texts/deltas (via (fileid, revisionid) tuples).
151
:param signature_index: A GraphIndex for determining what signatures are
152
present in the Pack and accessing the locations of their texts.
153
:param chk_index: A GraphIndex for accessing content by CHK, if the
156
self.revision_index = revision_index
157
self.inventory_index = inventory_index
158
self.text_index = text_index
159
self.signature_index = signature_index
160
self.chk_index = chk_index
162
def access_tuple(self):
163
"""Return a tuple (transport, name) for the pack content."""
164
return self.pack_transport, self.file_name()
166
def _check_references(self):
167
"""Make sure our external references are present.
169
Packs are allowed to have deltas whose base is not in the pack, but it
170
must be present somewhere in this collection. It is not allowed to
171
have deltas based on a fallback repository.
172
(See <https://bugs.launchpad.net/bzr/+bug/288751>)
175
for (index_name, external_refs, index) in [
177
self._get_external_refs(self.text_index),
178
self._pack_collection.text_index.combined_index),
180
self._get_external_refs(self.inventory_index),
181
self._pack_collection.inventory_index.combined_index),
183
missing = external_refs.difference(
184
k for (idx, k, v, r) in
185
index.iter_entries(external_refs))
187
missing_items[index_name] = sorted(list(missing))
189
from pprint import pformat
190
raise errors.BzrCheckError(
191
"Newly created pack file %r has delta references to "
192
"items not in its repository:\n%s"
193
% (self, pformat(missing_items)))
196
"""Get the file name for the pack on disk."""
197
return self.name + '.pack'
199
def get_revision_count(self):
200
return self.revision_index.key_count()
202
def index_name(self, index_type, name):
203
"""Get the disk name of an index type for pack name 'name'."""
204
return name + Pack.index_definitions[index_type][0]
206
def index_offset(self, index_type):
207
"""Get the position in a index_size array for a given index type."""
208
return Pack.index_definitions[index_type][1]
210
def inventory_index_name(self, name):
211
"""The inv index is the name + .iix."""
212
return self.index_name('inventory', name)
214
def revision_index_name(self, name):
215
"""The revision index is the name + .rix."""
216
return self.index_name('revision', name)
218
def signature_index_name(self, name):
219
"""The signature index is the name + .six."""
220
return self.index_name('signature', name)
222
def text_index_name(self, name):
223
"""The text index is the name + .tix."""
224
return self.index_name('text', name)
226
def _replace_index_with_readonly(self, index_type):
227
setattr(self, index_type + '_index',
228
self.index_class(self.index_transport,
229
self.index_name(index_type, self.name),
230
self.index_sizes[self.index_offset(index_type)]))
233
class ExistingPack(Pack):
234
"""An in memory proxy for an existing .pack and its disk indices."""
236
def __init__(self, pack_transport, name, revision_index, inventory_index,
237
text_index, signature_index, chk_index=None):
238
"""Create an ExistingPack object.
240
:param pack_transport: The transport where the pack file resides.
241
:param name: The name of the pack on disk in the pack_transport.
243
Pack.__init__(self, revision_index, inventory_index, text_index,
244
signature_index, chk_index)
246
self.pack_transport = pack_transport
247
if None in (revision_index, inventory_index, text_index,
248
signature_index, name, pack_transport):
249
raise AssertionError()
251
def __eq__(self, other):
252
return self.__dict__ == other.__dict__
254
def __ne__(self, other):
255
return not self.__eq__(other)
258
return "<%s.%s object at 0x%x, %s, %s" % (
259
self.__class__.__module__, self.__class__.__name__, id(self),
260
self.pack_transport, self.name)
263
class ResumedPack(ExistingPack):
265
def __init__(self, name, revision_index, inventory_index, text_index,
266
signature_index, upload_transport, pack_transport, index_transport,
267
pack_collection, chk_index=None):
268
"""Create a ResumedPack object."""
269
ExistingPack.__init__(self, pack_transport, name, revision_index,
270
inventory_index, text_index, signature_index,
272
self.upload_transport = upload_transport
273
self.index_transport = index_transport
274
self.index_sizes = [None, None, None, None]
276
('revision', revision_index),
277
('inventory', inventory_index),
278
('text', text_index),
279
('signature', signature_index),
281
if chk_index is not None:
282
indices.append(('chk', chk_index))
283
self.index_sizes.append(None)
284
for index_type, index in indices:
285
offset = self.index_offset(index_type)
286
self.index_sizes[offset] = index._size
287
self.index_class = pack_collection._index_class
288
self._pack_collection = pack_collection
289
self._state = 'resumed'
290
# XXX: perhaps check that the .pack file exists?
292
def access_tuple(self):
293
if self._state == 'finished':
294
return Pack.access_tuple(self)
295
elif self._state == 'resumed':
296
return self.upload_transport, self.file_name()
298
raise AssertionError(self._state)
301
self.upload_transport.delete(self.file_name())
302
indices = [self.revision_index, self.inventory_index, self.text_index,
303
self.signature_index]
304
if self.chk_index is not None:
305
indices.append(self.chk_index)
306
for index in indices:
307
index._transport.delete(index._name)
310
self._check_references()
311
index_types = ['revision', 'inventory', 'text', 'signature']
312
if self.chk_index is not None:
313
index_types.append('chk')
314
for index_type in index_types:
315
old_name = self.index_name(index_type, self.name)
316
new_name = '../indices/' + old_name
317
self.upload_transport.rename(old_name, new_name)
318
self._replace_index_with_readonly(index_type)
319
new_name = '../packs/' + self.file_name()
320
self.upload_transport.rename(self.file_name(), new_name)
321
self._state = 'finished'
323
def _get_external_refs(self, index):
324
"""Return compression parents for this index that are not present.
326
This returns any compression parents that are referenced by this index,
327
which are not contained *in* this index. They may be present elsewhere.
329
return index.external_references(1)
333
"""An in memory proxy for a pack which is being created."""
335
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
336
"""Create a NewPack instance.
338
:param pack_collection: A PackCollection into which this is being inserted.
339
:param upload_suffix: An optional suffix to be given to any temporary
340
files created during the pack creation. e.g '.autopack'
341
:param file_mode: Unix permissions for newly created file.
343
# The relative locations of the packs are constrained, but all are
344
# passed in because the caller has them, so as to avoid object churn.
345
index_builder_class = pack_collection._index_builder_class
346
if pack_collection.chk_index is not None:
347
chk_index = index_builder_class(reference_lists=0)
351
# Revisions: parents list, no text compression.
352
index_builder_class(reference_lists=1),
353
# Inventory: We want to map compression only, but currently the
354
# knit code hasn't been updated enough to understand that, so we
355
# have a regular 2-list index giving parents and compression
357
index_builder_class(reference_lists=2),
358
# Texts: compression and per file graph, for all fileids - so two
359
# reference lists and two elements in the key tuple.
360
index_builder_class(reference_lists=2, key_elements=2),
361
# Signatures: Just blobs to store, no compression, no parents
363
index_builder_class(reference_lists=0),
364
# CHK based storage - just blobs, no compression or parents.
367
self._pack_collection = pack_collection
368
# When we make readonly indices, we need this.
369
self.index_class = pack_collection._index_class
370
# where should the new pack be opened
371
self.upload_transport = pack_collection._upload_transport
372
# where are indices written out to
373
self.index_transport = pack_collection._index_transport
374
# where is the pack renamed to when it is finished?
375
self.pack_transport = pack_collection._pack_transport
376
# What file mode to upload the pack and indices with.
377
self._file_mode = file_mode
378
# tracks the content written to the .pack file.
379
self._hash = osutils.md5()
380
# a tuple with the length in bytes of the indices, once the pack
381
# is finalised. (rev, inv, text, sigs, chk_if_in_use)
382
self.index_sizes = None
383
# How much data to cache when writing packs. Note that this is not
384
# synchronised with reads, because it's not in the transport layer, so
385
# is not safe unless the client knows it won't be reading from the pack
387
self._cache_limit = 0
388
# the temporary pack file name.
389
self.random_name = osutils.rand_chars(20) + upload_suffix
390
# when was this pack started ?
391
self.start_time = time.time()
392
# open an output stream for the data added to the pack.
393
self.write_stream = self.upload_transport.open_write_stream(
394
self.random_name, mode=self._file_mode)
395
if 'pack' in debug.debug_flags:
396
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
397
time.ctime(), self.upload_transport.base, self.random_name,
398
time.time() - self.start_time)
399
# A list of byte sequences to be written to the new pack, and the
400
# aggregate size of them. Stored as a list rather than separate
401
# variables so that the _write_data closure below can update them.
402
self._buffer = [[], 0]
403
# create a callable for adding data
405
# robertc says- this is a closure rather than a method on the object
406
# so that the variables are locals, and faster than accessing object
408
def _write_data(bytes, flush=False, _buffer=self._buffer,
409
_write=self.write_stream.write, _update=self._hash.update):
410
_buffer[0].append(bytes)
411
_buffer[1] += len(bytes)
413
if _buffer[1] > self._cache_limit or flush:
414
bytes = ''.join(_buffer[0])
418
# expose this on self, for the occasion when clients want to add data.
419
self._write_data = _write_data
420
# a pack writer object to serialise pack records.
421
self._writer = pack.ContainerWriter(self._write_data)
423
# what state is the pack in? (open, finished, aborted)
425
# no name until we finish writing the content
429
"""Cancel creating this pack."""
430
self._state = 'aborted'
431
self.write_stream.close()
432
# Remove the temporary pack file.
433
self.upload_transport.delete(self.random_name)
434
# The indices have no state on disk.
436
def access_tuple(self):
437
"""Return a tuple (transport, name) for the pack content."""
438
if self._state == 'finished':
439
return Pack.access_tuple(self)
440
elif self._state == 'open':
441
return self.upload_transport, self.random_name
443
raise AssertionError(self._state)
445
def data_inserted(self):
446
"""True if data has been added to this pack."""
447
return bool(self.get_revision_count() or
448
self.inventory_index.key_count() or
449
self.text_index.key_count() or
450
self.signature_index.key_count() or
451
(self.chk_index is not None and self.chk_index.key_count()))
453
def finish_content(self):
454
if self.name is not None:
458
self._write_data('', flush=True)
459
self.name = self._hash.hexdigest()
461
def finish(self, suspend=False):
462
"""Finish the new pack.
465
- finalises the content
466
- assigns a name (the md5 of the content, currently)
467
- writes out the associated indices
468
- renames the pack into place.
469
- stores the index size tuple for the pack in the index_sizes
472
self.finish_content()
474
self._check_references()
476
# XXX: It'd be better to write them all to temporary names, then
477
# rename them all into place, so that the window when only some are
478
# visible is smaller. On the other hand none will be seen until
479
# they're in the names list.
480
self.index_sizes = [None, None, None, None]
481
self._write_index('revision', self.revision_index, 'revision', suspend)
482
self._write_index('inventory', self.inventory_index, 'inventory',
484
self._write_index('text', self.text_index, 'file texts', suspend)
485
self._write_index('signature', self.signature_index,
486
'revision signatures', suspend)
487
if self.chk_index is not None:
488
self.index_sizes.append(None)
489
self._write_index('chk', self.chk_index,
490
'content hash bytes', suspend)
491
self.write_stream.close()
492
# Note that this will clobber an existing pack with the same name,
493
# without checking for hash collisions. While this is undesirable this
494
# is something that can be rectified in a subsequent release. One way
495
# to rectify it may be to leave the pack at the original name, writing
496
# its pack-names entry as something like 'HASH: index-sizes
497
# temporary-name'. Allocate that and check for collisions, if it is
498
# collision free then rename it into place. If clients know this scheme
499
# they can handle missing-file errors by:
500
# - try for HASH.pack
501
# - try for temporary-name
502
# - refresh the pack-list to see if the pack is now absent
503
new_name = self.name + '.pack'
505
new_name = '../packs/' + new_name
506
self.upload_transport.rename(self.random_name, new_name)
507
self._state = 'finished'
508
if 'pack' in debug.debug_flags:
509
# XXX: size might be interesting?
510
mutter('%s: create_pack: pack finished: %s%s->%s t+%6.3fs',
511
time.ctime(), self.upload_transport.base, self.random_name,
512
new_name, time.time() - self.start_time)
515
"""Flush any current data."""
517
bytes = ''.join(self._buffer[0])
518
self.write_stream.write(bytes)
519
self._hash.update(bytes)
520
self._buffer[:] = [[], 0]
522
def _get_external_refs(self, index):
523
return index._external_references()
525
def set_write_cache_size(self, size):
526
self._cache_limit = size
528
def _write_index(self, index_type, index, label, suspend=False):
529
"""Write out an index.
531
:param index_type: The type of index to write - e.g. 'revision'.
532
:param index: The index object to serialise.
533
:param label: What label to give the index e.g. 'revision'.
535
index_name = self.index_name(index_type, self.name)
537
transport = self.upload_transport
539
transport = self.index_transport
540
self.index_sizes[self.index_offset(index_type)] = transport.put_file(
541
index_name, index.finish(), mode=self._file_mode)
542
if 'pack' in debug.debug_flags:
543
# XXX: size might be interesting?
544
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
545
time.ctime(), label, self.upload_transport.base,
546
self.random_name, time.time() - self.start_time)
547
# Replace the writable index on this object with a readonly,
548
# presently unloaded index. We should alter
549
# the index layer to make its finish() error if add_node is
550
# subsequently used. RBC
551
self._replace_index_with_readonly(index_type)
554
class AggregateIndex(object):
555
"""An aggregated index for the RepositoryPackCollection.
557
AggregateIndex is reponsible for managing the PackAccess object,
558
Index-To-Pack mapping, and all indices list for a specific type of index
559
such as 'revision index'.
561
A CombinedIndex provides an index on a single key space built up
562
from several on-disk indices. The AggregateIndex builds on this
563
to provide a knit access layer, and allows having up to one writable
564
index within the collection.
566
# XXX: Probably 'can be written to' could/should be separated from 'acts
567
# like a knit index' -- mbp 20071024
569
def __init__(self, reload_func=None, flush_func=None):
570
"""Create an AggregateIndex.
572
:param reload_func: A function to call if we find we are missing an
573
index. Should have the form reload_func() => True if the list of
574
active pack files has changed.
576
self._reload_func = reload_func
577
self.index_to_pack = {}
578
self.combined_index = CombinedGraphIndex([], reload_func=reload_func)
579
self.data_access = _DirectPackAccess(self.index_to_pack,
580
reload_func=reload_func,
581
flush_func=flush_func)
582
self.add_callback = None
584
def replace_indices(self, index_to_pack, indices):
585
"""Replace the current mappings with fresh ones.
587
This should probably not be used eventually, rather incremental add and
588
removal of indices. It has been added during refactoring of existing
591
:param index_to_pack: A mapping from index objects to
592
(transport, name) tuples for the pack file data.
593
:param indices: A list of indices.
595
# refresh the revision pack map dict without replacing the instance.
596
self.index_to_pack.clear()
597
self.index_to_pack.update(index_to_pack)
598
# XXX: API break - clearly a 'replace' method would be good?
599
self.combined_index._indices[:] = indices
600
# the current add nodes callback for the current writable index if
602
self.add_callback = None
604
def add_index(self, index, pack):
605
"""Add index to the aggregate, which is an index for Pack pack.
607
Future searches on the aggregate index will seach this new index
608
before all previously inserted indices.
610
:param index: An Index for the pack.
611
:param pack: A Pack instance.
613
# expose it to the index map
614
self.index_to_pack[index] = pack.access_tuple()
615
# put it at the front of the linear index list
616
self.combined_index.insert_index(0, index)
618
def add_writable_index(self, index, pack):
619
"""Add an index which is able to have data added to it.
621
There can be at most one writable index at any time. Any
622
modifications made to the knit are put into this index.
624
:param index: An index from the pack parameter.
625
:param pack: A Pack instance.
627
if self.add_callback is not None:
628
raise AssertionError(
629
"%s already has a writable index through %s" % \
630
(self, self.add_callback))
631
# allow writing: queue writes to a new index
632
self.add_index(index, pack)
633
# Updates the index to packs mapping as a side effect,
634
self.data_access.set_writer(pack._writer, index, pack.access_tuple())
635
self.add_callback = index.add_nodes
638
"""Reset all the aggregate data to nothing."""
639
self.data_access.set_writer(None, None, (None, None))
640
self.index_to_pack.clear()
641
del self.combined_index._indices[:]
642
self.add_callback = None
644
def remove_index(self, index, pack):
645
"""Remove index from the indices used to answer queries.
647
:param index: An index from the pack parameter.
648
:param pack: A Pack instance.
650
del self.index_to_pack[index]
651
self.combined_index._indices.remove(index)
652
if (self.add_callback is not None and
653
getattr(index, 'add_nodes', None) == self.add_callback):
654
self.add_callback = None
655
self.data_access.set_writer(None, None, (None, None))
658
class Packer(object):
659
"""Create a pack from packs."""
661
def __init__(self, pack_collection, packs, suffix, revision_ids=None,
665
:param pack_collection: A RepositoryPackCollection object where the
666
new pack is being written to.
667
:param packs: The packs to combine.
668
:param suffix: The suffix to use on the temporary files for the pack.
669
:param revision_ids: Revision ids to limit the pack to.
670
:param reload_func: A function to call if a pack file/index goes
671
missing. The side effect of calling this function should be to
672
update self.packs. See also AggregateIndex
676
self.revision_ids = revision_ids
677
# The pack object we are creating.
679
self._pack_collection = pack_collection
680
self._reload_func = reload_func
681
# The index layer keys for the revisions being copied. None for 'all
683
self._revision_keys = None
684
# What text keys to copy. None for 'all texts'. This is set by
685
# _copy_inventory_texts
686
self._text_filter = None
689
def _extra_init(self):
690
"""A template hook to allow extending the constructor trivially."""
692
def _pack_map_and_index_list(self, index_attribute):
693
"""Convert a list of packs to an index pack map and index list.
695
:param index_attribute: The attribute that the desired index is found
697
:return: A tuple (map, list) where map contains the dict from
698
index:pack_tuple, and list contains the indices in the preferred
703
for pack_obj in self.packs:
704
index = getattr(pack_obj, index_attribute)
705
indices.append(index)
706
pack_map[index] = pack_obj
707
return pack_map, indices
709
def _index_contents(self, indices, key_filter=None):
710
"""Get an iterable of the index contents from a pack_map.
712
:param indices: The list of indices to query
713
:param key_filter: An optional filter to limit the keys returned.
715
all_index = CombinedGraphIndex(indices)
716
if key_filter is None:
717
return all_index.iter_all_entries()
719
return all_index.iter_entries(key_filter)
721
def pack(self, pb=None):
722
"""Create a new pack by reading data from other packs.
724
This does little more than a bulk copy of data. One key difference
725
is that data with the same item key across multiple packs is elided
726
from the output. The new pack is written into the current pack store
727
along with its indices, and the name added to the pack names. The
728
source packs are not altered and are not required to be in the current
731
:param pb: An optional progress bar to use. A nested bar is created if
733
:return: A Pack object, or None if nothing was copied.
735
# open a pack - using the same name as the last temporary file
736
# - which has already been flushed, so its safe.
737
# XXX: - duplicate code warning with start_write_group; fix before
738
# considering 'done'.
739
if self._pack_collection._new_pack is not None:
740
raise errors.BzrError('call to %s.pack() while another pack is'
742
% (self.__class__.__name__,))
743
if self.revision_ids is not None:
744
if len(self.revision_ids) == 0:
745
# silly fetch request.
748
self.revision_ids = frozenset(self.revision_ids)
749
self.revision_keys = frozenset((revid,) for revid in
752
self.pb = ui.ui_factory.nested_progress_bar()
756
return self._create_pack_from_packs()
762
"""Open a pack for the pack we are creating."""
763
new_pack = self._pack_collection.pack_factory(self._pack_collection,
764
upload_suffix=self.suffix,
765
file_mode=self._pack_collection.repo.bzrdir._get_file_mode())
766
# We know that we will process all nodes in order, and don't need to
767
# query, so don't combine any indices spilled to disk until we are done
768
new_pack.revision_index.set_optimize(combine_backing_indices=False)
769
new_pack.inventory_index.set_optimize(combine_backing_indices=False)
770
new_pack.text_index.set_optimize(combine_backing_indices=False)
771
new_pack.signature_index.set_optimize(combine_backing_indices=False)
774
def _update_pack_order(self, entries, index_to_pack_map):
775
"""Determine how we want our packs to be ordered.
777
This changes the sort order of the self.packs list so that packs unused
778
by 'entries' will be at the end of the list, so that future requests
779
can avoid probing them. Used packs will be at the front of the
780
self.packs list, in the order of their first use in 'entries'.
782
:param entries: A list of (index, ...) tuples
783
:param index_to_pack_map: A mapping from index objects to pack objects.
787
for entry in entries:
789
if index not in seen_indexes:
790
packs.append(index_to_pack_map[index])
791
seen_indexes.add(index)
792
if len(packs) == len(self.packs):
793
if 'pack' in debug.debug_flags:
794
mutter('Not changing pack list, all packs used.')
796
seen_packs = set(packs)
797
for pack in self.packs:
798
if pack not in seen_packs:
801
if 'pack' in debug.debug_flags:
802
old_names = [p.access_tuple()[1] for p in self.packs]
803
new_names = [p.access_tuple()[1] for p in packs]
804
mutter('Reordering packs\nfrom: %s\n to: %s',
805
old_names, new_names)
808
def _copy_revision_texts(self):
809
"""Copy revision data to the new pack."""
811
if self.revision_ids:
812
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
815
# select revision keys
816
revision_index_map, revision_indices = self._pack_map_and_index_list(
818
revision_nodes = self._index_contents(revision_indices, revision_keys)
819
revision_nodes = list(revision_nodes)
820
self._update_pack_order(revision_nodes, revision_index_map)
821
# copy revision keys and adjust values
822
self.pb.update("Copying revision texts", 1)
823
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
824
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
825
self.new_pack.revision_index, readv_group_iter, total_items))
826
if 'pack' in debug.debug_flags:
827
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
828
time.ctime(), self._pack_collection._upload_transport.base,
829
self.new_pack.random_name,
830
self.new_pack.revision_index.key_count(),
831
time.time() - self.new_pack.start_time)
832
self._revision_keys = revision_keys
834
def _copy_inventory_texts(self):
835
"""Copy the inventory texts to the new pack.
837
self._revision_keys is used to determine what inventories to copy.
839
Sets self._text_filter appropriately.
841
# select inventory keys
842
inv_keys = self._revision_keys # currently the same keyspace, and note that
843
# querying for keys here could introduce a bug where an inventory item
844
# is missed, so do not change it to query separately without cross
845
# checking like the text key check below.
846
inventory_index_map, inventory_indices = self._pack_map_and_index_list(
848
inv_nodes = self._index_contents(inventory_indices, inv_keys)
849
# copy inventory keys and adjust values
850
# XXX: Should be a helper function to allow different inv representation
852
self.pb.update("Copying inventory texts", 2)
853
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
854
# Only grab the output lines if we will be processing them
855
output_lines = bool(self.revision_ids)
856
inv_lines = self._copy_nodes_graph(inventory_index_map,
857
self.new_pack._writer, self.new_pack.inventory_index,
858
readv_group_iter, total_items, output_lines=output_lines)
859
if self.revision_ids:
860
self._process_inventory_lines(inv_lines)
862
# eat the iterator to cause it to execute.
864
self._text_filter = None
865
if 'pack' in debug.debug_flags:
866
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
867
time.ctime(), self._pack_collection._upload_transport.base,
868
self.new_pack.random_name,
869
self.new_pack.inventory_index.key_count(),
870
time.time() - self.new_pack.start_time)
872
def _copy_text_texts(self):
874
text_index_map, text_nodes = self._get_text_nodes()
875
if self._text_filter is not None:
876
# We could return the keys copied as part of the return value from
877
# _copy_nodes_graph but this doesn't work all that well with the
878
# need to get line output too, so we check separately, and as we're
879
# going to buffer everything anyway, we check beforehand, which
880
# saves reading knit data over the wire when we know there are
882
text_nodes = set(text_nodes)
883
present_text_keys = set(_node[1] for _node in text_nodes)
884
missing_text_keys = set(self._text_filter) - present_text_keys
885
if missing_text_keys:
886
# TODO: raise a specific error that can handle many missing
888
mutter("missing keys during fetch: %r", missing_text_keys)
889
a_missing_key = missing_text_keys.pop()
890
raise errors.RevisionNotPresent(a_missing_key[1],
892
# copy text keys and adjust values
893
self.pb.update("Copying content texts", 3)
894
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
895
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
896
self.new_pack.text_index, readv_group_iter, total_items))
897
self._log_copied_texts()
899
def _create_pack_from_packs(self):
900
self.pb.update("Opening pack", 0, 5)
901
self.new_pack = self.open_pack()
902
new_pack = self.new_pack
903
# buffer data - we won't be reading-back during the pack creation and
904
# this makes a significant difference on sftp pushes.
905
new_pack.set_write_cache_size(1024*1024)
906
if 'pack' in debug.debug_flags:
907
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
908
for a_pack in self.packs]
909
if self.revision_ids is not None:
910
rev_count = len(self.revision_ids)
913
mutter('%s: create_pack: creating pack from source packs: '
914
'%s%s %s revisions wanted %s t=0',
915
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
916
plain_pack_list, rev_count)
917
self._copy_revision_texts()
918
self._copy_inventory_texts()
919
self._copy_text_texts()
920
# select signature keys
921
signature_filter = self._revision_keys # same keyspace
922
signature_index_map, signature_indices = self._pack_map_and_index_list(
924
signature_nodes = self._index_contents(signature_indices,
926
# copy signature keys and adjust values
927
self.pb.update("Copying signature texts", 4)
928
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
929
new_pack.signature_index)
930
if 'pack' in debug.debug_flags:
931
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
932
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
933
new_pack.signature_index.key_count(),
934
time.time() - new_pack.start_time)
936
# NB XXX: how to check CHK references are present? perhaps by yielding
937
# the items? How should that interact with stacked repos?
938
if new_pack.chk_index is not None:
940
if 'pack' in debug.debug_flags:
941
mutter('%s: create_pack: chk content copied: %s%s %d items t+%6.3fs',
942
time.ctime(), self._pack_collection._upload_transport.base,
943
new_pack.random_name,
944
new_pack.chk_index.key_count(),
945
time.time() - new_pack.start_time)
946
new_pack._check_references()
947
if not self._use_pack(new_pack):
950
self.pb.update("Finishing pack", 5)
952
self._pack_collection.allocate(new_pack)
955
def _copy_chks(self, refs=None):
956
# XXX: Todo, recursive follow-pointers facility when fetching some
958
chk_index_map, chk_indices = self._pack_map_and_index_list(
960
chk_nodes = self._index_contents(chk_indices, refs)
962
# TODO: This isn't strictly tasteful as we are accessing some private
963
# variables (_serializer). Perhaps a better way would be to have
964
# Repository._deserialise_chk_node()
965
search_key_func = chk_map.search_key_registry.get(
966
self._pack_collection.repo._serializer.search_key_name)
967
def accumlate_refs(lines):
968
# XXX: move to a generic location
970
bytes = ''.join(lines)
971
node = chk_map._deserialise(bytes, ("unknown",), search_key_func)
972
new_refs.update(node.refs())
973
self._copy_nodes(chk_nodes, chk_index_map, self.new_pack._writer,
974
self.new_pack.chk_index, output_lines=accumlate_refs)
977
def _copy_nodes(self, nodes, index_map, writer, write_index,
979
"""Copy knit nodes between packs with no graph references.
981
:param output_lines: Output full texts of copied items.
983
pb = ui.ui_factory.nested_progress_bar()
985
return self._do_copy_nodes(nodes, index_map, writer,
986
write_index, pb, output_lines=output_lines)
990
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb,
992
# for record verification
993
knit = KnitVersionedFiles(None, None)
994
# plan a readv on each source pack:
996
nodes = sorted(nodes)
997
# how to map this into knit.py - or knit.py into this?
998
# we don't want the typical knit logic, we want grouping by pack
999
# at this point - perhaps a helper library for the following code
1000
# duplication points?
1002
for index, key, value in nodes:
1003
if index not in request_groups:
1004
request_groups[index] = []
1005
request_groups[index].append((key, value))
1007
pb.update("Copied record", record_index, len(nodes))
1008
for index, items in request_groups.iteritems():
1009
pack_readv_requests = []
1010
for key, value in items:
1011
# ---- KnitGraphIndex.get_position
1012
bits = value[1:].split(' ')
1013
offset, length = int(bits[0]), int(bits[1])
1014
pack_readv_requests.append((offset, length, (key, value[0])))
1015
# linear scan up the pack
1016
pack_readv_requests.sort()
1018
pack_obj = index_map[index]
1019
transport, path = pack_obj.access_tuple()
1021
reader = pack.make_readv_reader(transport, path,
1022
[offset[0:2] for offset in pack_readv_requests])
1023
except errors.NoSuchFile:
1024
if self._reload_func is not None:
1027
for (names, read_func), (_1, _2, (key, eol_flag)) in \
1028
izip(reader.iter_records(), pack_readv_requests):
1029
raw_data = read_func(None)
1030
# check the header only
1031
if output_lines is not None:
1032
output_lines(knit._parse_record(key[-1], raw_data)[0])
1034
df, _ = knit._parse_record_header(key, raw_data)
1036
pos, size = writer.add_bytes_record(raw_data, names)
1037
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
1038
pb.update("Copied record", record_index)
1041
def _copy_nodes_graph(self, index_map, writer, write_index,
1042
readv_group_iter, total_items, output_lines=False):
1043
"""Copy knit nodes between packs.
1045
:param output_lines: Return lines present in the copied data as
1046
an iterator of line,version_id.
1048
pb = ui.ui_factory.nested_progress_bar()
1050
for result in self._do_copy_nodes_graph(index_map, writer,
1051
write_index, output_lines, pb, readv_group_iter, total_items):
1054
# Python 2.4 does not permit try:finally: in a generator.
1060
def _do_copy_nodes_graph(self, index_map, writer, write_index,
1061
output_lines, pb, readv_group_iter, total_items):
1062
# for record verification
1063
knit = KnitVersionedFiles(None, None)
1064
# for line extraction when requested (inventories only)
1066
factory = KnitPlainFactory()
1068
pb.update("Copied record", record_index, total_items)
1069
for index, readv_vector, node_vector in readv_group_iter:
1071
pack_obj = index_map[index]
1072
transport, path = pack_obj.access_tuple()
1074
reader = pack.make_readv_reader(transport, path, readv_vector)
1075
except errors.NoSuchFile:
1076
if self._reload_func is not None:
1079
for (names, read_func), (key, eol_flag, references) in \
1080
izip(reader.iter_records(), node_vector):
1081
raw_data = read_func(None)
1083
# read the entire thing
1084
content, _ = knit._parse_record(key[-1], raw_data)
1085
if len(references[-1]) == 0:
1086
line_iterator = factory.get_fulltext_content(content)
1088
line_iterator = factory.get_linedelta_content(content)
1089
for line in line_iterator:
1092
# check the header only
1093
df, _ = knit._parse_record_header(key, raw_data)
1095
pos, size = writer.add_bytes_record(raw_data, names)
1096
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1097
pb.update("Copied record", record_index)
1100
def _get_text_nodes(self):
1101
text_index_map, text_indices = self._pack_map_and_index_list(
1103
return text_index_map, self._index_contents(text_indices,
1106
def _least_readv_node_readv(self, nodes):
1107
"""Generate request groups for nodes using the least readv's.
1109
:param nodes: An iterable of graph index nodes.
1110
:return: Total node count and an iterator of the data needed to perform
1111
readvs to obtain the data for nodes. Each item yielded by the
1112
iterator is a tuple with:
1113
index, readv_vector, node_vector. readv_vector is a list ready to
1114
hand to the transport readv method, and node_vector is a list of
1115
(key, eol_flag, references) for the the node retrieved by the
1116
matching readv_vector.
1118
# group by pack so we do one readv per pack
1119
nodes = sorted(nodes)
1122
for index, key, value, references in nodes:
1123
if index not in request_groups:
1124
request_groups[index] = []
1125
request_groups[index].append((key, value, references))
1127
for index, items in request_groups.iteritems():
1128
pack_readv_requests = []
1129
for key, value, references in items:
1130
# ---- KnitGraphIndex.get_position
1131
bits = value[1:].split(' ')
1132
offset, length = int(bits[0]), int(bits[1])
1133
pack_readv_requests.append(
1134
((offset, length), (key, value[0], references)))
1135
# linear scan up the pack to maximum range combining.
1136
pack_readv_requests.sort()
1137
# split out the readv and the node data.
1138
pack_readv = [readv for readv, node in pack_readv_requests]
1139
node_vector = [node for readv, node in pack_readv_requests]
1140
result.append((index, pack_readv, node_vector))
1141
return total, result
1143
def _log_copied_texts(self):
1144
if 'pack' in debug.debug_flags:
1145
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
1146
time.ctime(), self._pack_collection._upload_transport.base,
1147
self.new_pack.random_name,
1148
self.new_pack.text_index.key_count(),
1149
time.time() - self.new_pack.start_time)
1151
def _process_inventory_lines(self, inv_lines):
1152
"""Use up the inv_lines generator and setup a text key filter."""
1153
repo = self._pack_collection.repo
1154
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
1155
inv_lines, self.revision_keys)
1157
for fileid, file_revids in fileid_revisions.iteritems():
1158
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
1159
self._text_filter = text_filter
1161
def _revision_node_readv(self, revision_nodes):
1162
"""Return the total revisions and the readv's to issue.
1164
:param revision_nodes: The revision index contents for the packs being
1165
incorporated into the new pack.
1166
:return: As per _least_readv_node_readv.
1168
return self._least_readv_node_readv(revision_nodes)
1170
def _use_pack(self, new_pack):
1171
"""Return True if new_pack should be used.
1173
:param new_pack: The pack that has just been created.
1174
:return: True if the pack should be used.
1176
return new_pack.data_inserted()
1179
class OptimisingPacker(Packer):
1180
"""A packer which spends more time to create better disk layouts."""
1182
def _revision_node_readv(self, revision_nodes):
1183
"""Return the total revisions and the readv's to issue.
1185
This sort places revisions in topological order with the ancestors
1188
:param revision_nodes: The revision index contents for the packs being
1189
incorporated into the new pack.
1190
:return: As per _least_readv_node_readv.
1192
# build an ancestors dict
1195
for index, key, value, references in revision_nodes:
1196
ancestors[key] = references[0]
1197
by_key[key] = (index, value, references)
1198
order = tsort.topo_sort(ancestors)
1200
# Single IO is pathological, but it will work as a starting point.
1202
for key in reversed(order):
1203
index, value, references = by_key[key]
1204
# ---- KnitGraphIndex.get_position
1205
bits = value[1:].split(' ')
1206
offset, length = int(bits[0]), int(bits[1])
1208
(index, [(offset, length)], [(key, value[0], references)]))
1209
# TODO: combine requests in the same index that are in ascending order.
1210
return total, requests
1212
def open_pack(self):
1213
"""Open a pack for the pack we are creating."""
1214
new_pack = super(OptimisingPacker, self).open_pack()
1215
# Turn on the optimization flags for all the index builders.
1216
new_pack.revision_index.set_optimize(for_size=True)
1217
new_pack.inventory_index.set_optimize(for_size=True)
1218
new_pack.text_index.set_optimize(for_size=True)
1219
new_pack.signature_index.set_optimize(for_size=True)
1223
class ReconcilePacker(Packer):
1224
"""A packer which regenerates indices etc as it copies.
1226
This is used by ``bzr reconcile`` to cause parent text pointers to be
1230
def _extra_init(self):
1231
self._data_changed = False
1233
def _process_inventory_lines(self, inv_lines):
1234
"""Generate a text key reference map rather for reconciling with."""
1235
repo = self._pack_collection.repo
1236
refs = repo._find_text_key_references_from_xml_inventory_lines(
1238
self._text_refs = refs
1239
# during reconcile we:
1240
# - convert unreferenced texts to full texts
1241
# - correct texts which reference a text not copied to be full texts
1242
# - copy all others as-is but with corrected parents.
1243
# - so at this point we don't know enough to decide what becomes a full
1245
self._text_filter = None
1247
def _copy_text_texts(self):
1248
"""generate what texts we should have and then copy."""
1249
self.pb.update("Copying content texts", 3)
1250
# we have three major tasks here:
1251
# 1) generate the ideal index
1252
repo = self._pack_collection.repo
1253
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1254
_1, key, _2, refs in
1255
self.new_pack.revision_index.iter_all_entries()])
1256
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1257
# 2) generate a text_nodes list that contains all the deltas that can
1258
# be used as-is, with corrected parents.
1261
discarded_nodes = []
1262
NULL_REVISION = _mod_revision.NULL_REVISION
1263
text_index_map, text_nodes = self._get_text_nodes()
1264
for node in text_nodes:
1270
ideal_parents = tuple(ideal_index[node[1]])
1272
discarded_nodes.append(node)
1273
self._data_changed = True
1275
if ideal_parents == (NULL_REVISION,):
1277
if ideal_parents == node[3][0]:
1279
ok_nodes.append(node)
1280
elif ideal_parents[0:1] == node[3][0][0:1]:
1281
# the left most parent is the same, or there are no parents
1282
# today. Either way, we can preserve the representation as
1283
# long as we change the refs to be inserted.
1284
self._data_changed = True
1285
ok_nodes.append((node[0], node[1], node[2],
1286
(ideal_parents, node[3][1])))
1287
self._data_changed = True
1289
# Reinsert this text completely
1290
bad_texts.append((node[1], ideal_parents))
1291
self._data_changed = True
1292
# we're finished with some data.
1295
# 3) bulk copy the ok data
1296
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1297
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1298
self.new_pack.text_index, readv_group_iter, total_items))
1299
# 4) adhoc copy all the other texts.
1300
# We have to topologically insert all texts otherwise we can fail to
1301
# reconcile when parts of a single delta chain are preserved intact,
1302
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1303
# reinserted, and if d3 has incorrect parents it will also be
1304
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1305
# copied), so we will try to delta, but d2 is not currently able to be
1306
# extracted because it's basis d1 is not present. Topologically sorting
1307
# addresses this. The following generates a sort for all the texts that
1308
# are being inserted without having to reference the entire text key
1309
# space (we only topo sort the revisions, which is smaller).
1310
topo_order = tsort.topo_sort(ancestors)
1311
rev_order = dict(zip(topo_order, range(len(topo_order))))
1312
bad_texts.sort(key=lambda key:rev_order.get(key[0][1], 0))
1313
transaction = repo.get_transaction()
1314
file_id_index = GraphIndexPrefixAdapter(
1315
self.new_pack.text_index,
1317
add_nodes_callback=self.new_pack.text_index.add_nodes)
1318
data_access = _DirectPackAccess(
1319
{self.new_pack.text_index:self.new_pack.access_tuple()})
1320
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1321
self.new_pack.access_tuple())
1322
output_texts = KnitVersionedFiles(
1323
_KnitGraphIndex(self.new_pack.text_index,
1324
add_callback=self.new_pack.text_index.add_nodes,
1325
deltas=True, parents=True, is_locked=repo.is_locked),
1326
data_access=data_access, max_delta_chain=200)
1327
for key, parent_keys in bad_texts:
1328
# We refer to the new pack to delta data being output.
1329
# A possible improvement would be to catch errors on short reads
1330
# and only flush then.
1331
self.new_pack.flush()
1333
for parent_key in parent_keys:
1334
if parent_key[0] != key[0]:
1335
# Graph parents must match the fileid
1336
raise errors.BzrError('Mismatched key parent %r:%r' %
1338
parents.append(parent_key[1])
1339
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1340
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1341
output_texts.add_lines(key, parent_keys, text_lines,
1342
random_id=True, check_content=False)
1343
# 5) check that nothing inserted has a reference outside the keyspace.
1344
missing_text_keys = self.new_pack.text_index._external_references()
1345
if missing_text_keys:
1346
raise errors.BzrCheckError('Reference to missing compression parents %r'
1347
% (missing_text_keys,))
1348
self._log_copied_texts()
1350
def _use_pack(self, new_pack):
1351
"""Override _use_pack to check for reconcile having changed content."""
1352
# XXX: we might be better checking this at the copy time.
1353
original_inventory_keys = set()
1354
inv_index = self._pack_collection.inventory_index.combined_index
1355
for entry in inv_index.iter_all_entries():
1356
original_inventory_keys.add(entry[1])
1357
new_inventory_keys = set()
1358
for entry in new_pack.inventory_index.iter_all_entries():
1359
new_inventory_keys.add(entry[1])
1360
if new_inventory_keys != original_inventory_keys:
1361
self._data_changed = True
1362
return new_pack.data_inserted() and self._data_changed
1365
class RepositoryPackCollection(object):
1366
"""Management of packs within a repository.
1368
:ivar _names: map of {pack_name: (index_size,)}
1371
pack_factory = NewPack
1372
resumed_pack_factory = ResumedPack
1374
def __init__(self, repo, transport, index_transport, upload_transport,
1375
pack_transport, index_builder_class, index_class,
1377
"""Create a new RepositoryPackCollection.
1379
:param transport: Addresses the repository base directory
1380
(typically .bzr/repository/).
1381
:param index_transport: Addresses the directory containing indices.
1382
:param upload_transport: Addresses the directory into which packs are written
1383
while they're being created.
1384
:param pack_transport: Addresses the directory of existing complete packs.
1385
:param index_builder_class: The index builder class to use.
1386
:param index_class: The index class to use.
1387
:param use_chk_index: Whether to setup and manage a CHK index.
1389
# XXX: This should call self.reset()
1391
self.transport = transport
1392
self._index_transport = index_transport
1393
self._upload_transport = upload_transport
1394
self._pack_transport = pack_transport
1395
self._index_builder_class = index_builder_class
1396
self._index_class = index_class
1397
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3,
1402
self._packs_by_name = {}
1403
# the previous pack-names content
1404
self._packs_at_load = None
1405
# when a pack is being created by this object, the state of that pack.
1406
self._new_pack = None
1407
# aggregated revision index data
1408
flush = self._flush_new_pack
1409
self.revision_index = AggregateIndex(self.reload_pack_names, flush)
1410
self.inventory_index = AggregateIndex(self.reload_pack_names, flush)
1411
self.text_index = AggregateIndex(self.reload_pack_names, flush)
1412
self.signature_index = AggregateIndex(self.reload_pack_names, flush)
1414
self.chk_index = AggregateIndex(self.reload_pack_names, flush)
1416
# used to determine if we're using a chk_index elsewhere.
1417
self.chk_index = None
1419
self._resumed_packs = []
1421
def add_pack_to_memory(self, pack):
1422
"""Make a Pack object available to the repository to satisfy queries.
1424
:param pack: A Pack object.
1426
if pack.name in self._packs_by_name:
1427
raise AssertionError(
1428
'pack %s already in _packs_by_name' % (pack.name,))
1429
self.packs.append(pack)
1430
self._packs_by_name[pack.name] = pack
1431
self.revision_index.add_index(pack.revision_index, pack)
1432
self.inventory_index.add_index(pack.inventory_index, pack)
1433
self.text_index.add_index(pack.text_index, pack)
1434
self.signature_index.add_index(pack.signature_index, pack)
1435
if self.chk_index is not None:
1436
self.chk_index.add_index(pack.chk_index, pack)
1438
def all_packs(self):
1439
"""Return a list of all the Pack objects this repository has.
1441
Note that an in-progress pack being created is not returned.
1443
:return: A list of Pack objects for all the packs in the repository.
1446
for name in self.names():
1447
result.append(self.get_pack_by_name(name))
1451
"""Pack the pack collection incrementally.
1453
This will not attempt global reorganisation or recompression,
1454
rather it will just ensure that the total number of packs does
1455
not grow without bound. It uses the _max_pack_count method to
1456
determine if autopacking is needed, and the pack_distribution
1457
method to determine the number of revisions in each pack.
1459
If autopacking takes place then the packs name collection will have
1460
been flushed to disk - packing requires updating the name collection
1461
in synchronisation with certain steps. Otherwise the names collection
1464
:return: Something evaluating true if packing took place.
1468
return self._do_autopack()
1469
except errors.RetryAutopack:
1470
# If we get a RetryAutopack exception, we should abort the
1471
# current action, and retry.
1474
def _do_autopack(self):
1475
# XXX: Should not be needed when the management of indices is sane.
1476
total_revisions = self.revision_index.combined_index.key_count()
1477
total_packs = len(self._names)
1478
if self._max_pack_count(total_revisions) >= total_packs:
1480
# determine which packs need changing
1481
pack_distribution = self.pack_distribution(total_revisions)
1483
for pack in self.all_packs():
1484
revision_count = pack.get_revision_count()
1485
if revision_count == 0:
1486
# revision less packs are not generated by normal operation,
1487
# only by operations like sign-my-commits, and thus will not
1488
# tend to grow rapdily or without bound like commit containing
1489
# packs do - leave them alone as packing them really should
1490
# group their data with the relevant commit, and that may
1491
# involve rewriting ancient history - which autopack tries to
1492
# avoid. Alternatively we could not group the data but treat
1493
# each of these as having a single revision, and thus add
1494
# one revision for each to the total revision count, to get
1495
# a matching distribution.
1497
existing_packs.append((revision_count, pack))
1498
pack_operations = self.plan_autopack_combinations(
1499
existing_packs, pack_distribution)
1500
num_new_packs = len(pack_operations)
1501
num_old_packs = sum([len(po[1]) for po in pack_operations])
1502
num_revs_affected = sum([po[0] for po in pack_operations])
1503
mutter('Auto-packing repository %s, which has %d pack files, '
1504
'containing %d revisions. Packing %d files into %d affecting %d'
1505
' revisions', self, total_packs, total_revisions, num_old_packs,
1506
num_new_packs, num_revs_affected)
1507
result = self._execute_pack_operations(pack_operations,
1508
reload_func=self._restart_autopack)
1509
mutter('Auto-packing repository %s completed', self)
1512
def _execute_pack_operations(self, pack_operations, _packer_class=Packer,
1514
"""Execute a series of pack operations.
1516
:param pack_operations: A list of [revision_count, packs_to_combine].
1517
:param _packer_class: The class of packer to use (default: Packer).
1518
:return: The new pack names.
1520
for revision_count, packs in pack_operations:
1521
# we may have no-ops from the setup logic
1524
packer = _packer_class(self, packs, '.autopack',
1525
reload_func=reload_func)
1528
except errors.RetryWithNewPacks:
1529
# An exception is propagating out of this context, make sure
1530
# this packer has cleaned up. Packer() doesn't set its new_pack
1531
# state into the RepositoryPackCollection object, so we only
1532
# have access to it directly here.
1533
if packer.new_pack is not None:
1534
packer.new_pack.abort()
1537
self._remove_pack_from_memory(pack)
1538
# record the newly available packs and stop advertising the old
1540
result = self._save_pack_names(clear_obsolete_packs=True)
1541
# Move the old packs out of the way now they are no longer referenced.
1542
for revision_count, packs in pack_operations:
1543
self._obsolete_packs(packs)
1546
def _flush_new_pack(self):
1547
if self._new_pack is not None:
1548
self._new_pack.flush()
1550
def lock_names(self):
1551
"""Acquire the mutex around the pack-names index.
1553
This cannot be used in the middle of a read-only transaction on the
1556
self.repo.control_files.lock_write()
1558
def _already_packed(self):
1559
"""Is the collection already packed?"""
1560
return not (self.repo._format.pack_compresses or (len(self._names) > 1))
1562
def pack(self, hint=None):
1563
"""Pack the pack collection totally."""
1564
self.ensure_loaded()
1565
total_packs = len(self._names)
1566
if self._already_packed():
1568
total_revisions = self.revision_index.combined_index.key_count()
1569
# XXX: the following may want to be a class, to pack with a given
1571
mutter('Packing repository %s, which has %d pack files, '
1572
'containing %d revisions with hint %r.', self, total_packs,
1573
total_revisions, hint)
1574
# determine which packs need changing
1575
pack_operations = [[0, []]]
1576
for pack in self.all_packs():
1577
if hint is None or pack.name in hint:
1578
# Either no hint was provided (so we are packing everything),
1579
# or this pack was included in the hint.
1580
pack_operations[-1][0] += pack.get_revision_count()
1581
pack_operations[-1][1].append(pack)
1582
self._execute_pack_operations(pack_operations, OptimisingPacker)
1584
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1585
"""Plan a pack operation.
1587
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1589
:param pack_distribution: A list with the number of revisions desired
1592
if len(existing_packs) <= len(pack_distribution):
1594
existing_packs.sort(reverse=True)
1595
pack_operations = [[0, []]]
1596
# plan out what packs to keep, and what to reorganise
1597
while len(existing_packs):
1598
# take the largest pack, and if its less than the head of the
1599
# distribution chart we will include its contents in the new pack
1600
# for that position. If its larger, we remove its size from the
1601
# distribution chart
1602
next_pack_rev_count, next_pack = existing_packs.pop(0)
1603
if next_pack_rev_count >= pack_distribution[0]:
1604
# this is already packed 'better' than this, so we can
1605
# not waste time packing it.
1606
while next_pack_rev_count > 0:
1607
next_pack_rev_count -= pack_distribution[0]
1608
if next_pack_rev_count >= 0:
1610
del pack_distribution[0]
1612
# didn't use that entire bucket up
1613
pack_distribution[0] = -next_pack_rev_count
1615
# add the revisions we're going to add to the next output pack
1616
pack_operations[-1][0] += next_pack_rev_count
1617
# allocate this pack to the next pack sub operation
1618
pack_operations[-1][1].append(next_pack)
1619
if pack_operations[-1][0] >= pack_distribution[0]:
1620
# this pack is used up, shift left.
1621
del pack_distribution[0]
1622
pack_operations.append([0, []])
1623
# Now that we know which pack files we want to move, shove them all
1624
# into a single pack file.
1626
final_pack_list = []
1627
for num_revs, pack_files in pack_operations:
1628
final_rev_count += num_revs
1629
final_pack_list.extend(pack_files)
1630
if len(final_pack_list) == 1:
1631
raise AssertionError('We somehow generated an autopack with a'
1632
' single pack file being moved.')
1634
return [[final_rev_count, final_pack_list]]
1636
def ensure_loaded(self):
1637
"""Ensure we have read names from disk.
1639
:return: True if the disk names had not been previously read.
1641
# NB: if you see an assertion error here, its probably access against
1642
# an unlocked repo. Naughty.
1643
if not self.repo.is_locked():
1644
raise errors.ObjectNotLocked(self.repo)
1645
if self._names is None:
1647
self._packs_at_load = set()
1648
for index, key, value in self._iter_disk_pack_index():
1650
self._names[name] = self._parse_index_sizes(value)
1651
self._packs_at_load.add((key, value))
1655
# populate all the metadata.
1659
def _parse_index_sizes(self, value):
1660
"""Parse a string of index sizes."""
1661
return tuple([int(digits) for digits in value.split(' ')])
1663
def get_pack_by_name(self, name):
1664
"""Get a Pack object by name.
1666
:param name: The name of the pack - e.g. '123456'
1667
:return: A Pack object.
1670
return self._packs_by_name[name]
1672
rev_index = self._make_index(name, '.rix')
1673
inv_index = self._make_index(name, '.iix')
1674
txt_index = self._make_index(name, '.tix')
1675
sig_index = self._make_index(name, '.six')
1676
if self.chk_index is not None:
1677
chk_index = self._make_index(name, '.cix')
1680
result = ExistingPack(self._pack_transport, name, rev_index,
1681
inv_index, txt_index, sig_index, chk_index)
1682
self.add_pack_to_memory(result)
1685
def _resume_pack(self, name):
1686
"""Get a suspended Pack object by name.
1688
:param name: The name of the pack - e.g. '123456'
1689
:return: A Pack object.
1691
if not re.match('[a-f0-9]{32}', name):
1692
# Tokens should be md5sums of the suspended pack file, i.e. 32 hex
1694
raise errors.UnresumableWriteGroup(
1695
self.repo, [name], 'Malformed write group token')
1697
rev_index = self._make_index(name, '.rix', resume=True)
1698
inv_index = self._make_index(name, '.iix', resume=True)
1699
txt_index = self._make_index(name, '.tix', resume=True)
1700
sig_index = self._make_index(name, '.six', resume=True)
1701
if self.chk_index is not None:
1702
chk_index = self._make_index(name, '.cix', resume=True)
1705
result = self.resumed_pack_factory(name, rev_index, inv_index,
1706
txt_index, sig_index, self._upload_transport,
1707
self._pack_transport, self._index_transport, self,
1708
chk_index=chk_index)
1709
except errors.NoSuchFile, e:
1710
raise errors.UnresumableWriteGroup(self.repo, [name], str(e))
1711
self.add_pack_to_memory(result)
1712
self._resumed_packs.append(result)
1715
def allocate(self, a_new_pack):
1716
"""Allocate name in the list of packs.
1718
:param a_new_pack: A NewPack instance to be added to the collection of
1719
packs for this repository.
1721
self.ensure_loaded()
1722
if a_new_pack.name in self._names:
1723
raise errors.BzrError(
1724
'Pack %r already exists in %s' % (a_new_pack.name, self))
1725
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1726
self.add_pack_to_memory(a_new_pack)
1728
def _iter_disk_pack_index(self):
1729
"""Iterate over the contents of the pack-names index.
1731
This is used when loading the list from disk, and before writing to
1732
detect updates from others during our write operation.
1733
:return: An iterator of the index contents.
1735
return self._index_class(self.transport, 'pack-names', None
1736
).iter_all_entries()
1738
def _make_index(self, name, suffix, resume=False):
1739
size_offset = self._suffix_offsets[suffix]
1740
index_name = name + suffix
1742
transport = self._upload_transport
1743
index_size = transport.stat(index_name).st_size
1745
transport = self._index_transport
1746
index_size = self._names[name][size_offset]
1747
return self._index_class(transport, index_name, index_size)
1749
def _max_pack_count(self, total_revisions):
1750
"""Return the maximum number of packs to use for total revisions.
1752
:param total_revisions: The total number of revisions in the
1755
if not total_revisions:
1757
digits = str(total_revisions)
1759
for digit in digits:
1760
result += int(digit)
1764
"""Provide an order to the underlying names."""
1765
return sorted(self._names.keys())
1767
def _obsolete_packs(self, packs):
1768
"""Move a number of packs which have been obsoleted out of the way.
1770
Each pack and its associated indices are moved out of the way.
1772
Note: for correctness this function should only be called after a new
1773
pack names index has been written without these pack names, and with
1774
the names of packs that contain the data previously available via these
1777
:param packs: The packs to obsolete.
1778
:param return: None.
1781
pack.pack_transport.rename(pack.file_name(),
1782
'../obsolete_packs/' + pack.file_name())
1783
# TODO: Probably needs to know all possible indices for this pack
1784
# - or maybe list the directory and move all indices matching this
1785
# name whether we recognize it or not?
1786
suffixes = ['.iix', '.six', '.tix', '.rix']
1787
if self.chk_index is not None:
1788
suffixes.append('.cix')
1789
for suffix in suffixes:
1790
self._index_transport.rename(pack.name + suffix,
1791
'../obsolete_packs/' + pack.name + suffix)
1793
def pack_distribution(self, total_revisions):
1794
"""Generate a list of the number of revisions to put in each pack.
1796
:param total_revisions: The total number of revisions in the
1799
if total_revisions == 0:
1801
digits = reversed(str(total_revisions))
1803
for exponent, count in enumerate(digits):
1804
size = 10 ** exponent
1805
for pos in range(int(count)):
1807
return list(reversed(result))
1809
def _pack_tuple(self, name):
1810
"""Return a tuple with the transport and file name for a pack name."""
1811
return self._pack_transport, name + '.pack'
1813
def _remove_pack_from_memory(self, pack):
1814
"""Remove pack from the packs accessed by this repository.
1816
Only affects memory state, until self._save_pack_names() is invoked.
1818
self._names.pop(pack.name)
1819
self._packs_by_name.pop(pack.name)
1820
self._remove_pack_indices(pack)
1821
self.packs.remove(pack)
1823
def _remove_pack_indices(self, pack):
1824
"""Remove the indices for pack from the aggregated indices."""
1825
self.revision_index.remove_index(pack.revision_index, pack)
1826
self.inventory_index.remove_index(pack.inventory_index, pack)
1827
self.text_index.remove_index(pack.text_index, pack)
1828
self.signature_index.remove_index(pack.signature_index, pack)
1829
if self.chk_index is not None:
1830
self.chk_index.remove_index(pack.chk_index, pack)
1833
"""Clear all cached data."""
1834
# cached revision data
1835
self.revision_index.clear()
1836
# cached signature data
1837
self.signature_index.clear()
1838
# cached file text data
1839
self.text_index.clear()
1840
# cached inventory data
1841
self.inventory_index.clear()
1843
if self.chk_index is not None:
1844
self.chk_index.clear()
1845
# remove the open pack
1846
self._new_pack = None
1847
# information about packs.
1850
self._packs_by_name = {}
1851
self._packs_at_load = None
1853
def _unlock_names(self):
1854
"""Release the mutex around the pack-names index."""
1855
self.repo.control_files.unlock()
1857
def _diff_pack_names(self):
1858
"""Read the pack names from disk, and compare it to the one in memory.
1860
:return: (disk_nodes, deleted_nodes, new_nodes)
1861
disk_nodes The final set of nodes that should be referenced
1862
deleted_nodes Nodes which have been removed from when we started
1863
new_nodes Nodes that are newly introduced
1865
# load the disk nodes across
1867
for index, key, value in self._iter_disk_pack_index():
1868
disk_nodes.add((key, value))
1870
# do a two-way diff against our original content
1871
current_nodes = set()
1872
for name, sizes in self._names.iteritems():
1874
((name, ), ' '.join(str(size) for size in sizes)))
1876
# Packs no longer present in the repository, which were present when we
1877
# locked the repository
1878
deleted_nodes = self._packs_at_load - current_nodes
1879
# Packs which this process is adding
1880
new_nodes = current_nodes - self._packs_at_load
1882
# Update the disk_nodes set to include the ones we are adding, and
1883
# remove the ones which were removed by someone else
1884
disk_nodes.difference_update(deleted_nodes)
1885
disk_nodes.update(new_nodes)
1887
return disk_nodes, deleted_nodes, new_nodes
1889
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1890
"""Given the correct set of pack files, update our saved info.
1892
:return: (removed, added, modified)
1893
removed pack names removed from self._names
1894
added pack names added to self._names
1895
modified pack names that had changed value
1900
## self._packs_at_load = disk_nodes
1901
new_names = dict(disk_nodes)
1902
# drop no longer present nodes
1903
for pack in self.all_packs():
1904
if (pack.name,) not in new_names:
1905
removed.append(pack.name)
1906
self._remove_pack_from_memory(pack)
1907
# add new nodes/refresh existing ones
1908
for key, value in disk_nodes:
1910
sizes = self._parse_index_sizes(value)
1911
if name in self._names:
1913
if sizes != self._names[name]:
1914
# the pack for name has had its indices replaced - rare but
1915
# important to handle. XXX: probably can never happen today
1916
# because the three-way merge code above does not handle it
1917
# - you may end up adding the same key twice to the new
1918
# disk index because the set values are the same, unless
1919
# the only index shows up as deleted by the set difference
1920
# - which it may. Until there is a specific test for this,
1921
# assume its broken. RBC 20071017.
1922
self._remove_pack_from_memory(self.get_pack_by_name(name))
1923
self._names[name] = sizes
1924
self.get_pack_by_name(name)
1925
modified.append(name)
1928
self._names[name] = sizes
1929
self.get_pack_by_name(name)
1931
return removed, added, modified
1933
def _save_pack_names(self, clear_obsolete_packs=False):
1934
"""Save the list of packs.
1936
This will take out the mutex around the pack names list for the
1937
duration of the method call. If concurrent updates have been made, a
1938
three-way merge between the current list and the current in memory list
1941
:param clear_obsolete_packs: If True, clear out the contents of the
1942
obsolete_packs directory.
1943
:return: A list of the names saved that were not previously on disk.
1947
builder = self._index_builder_class()
1948
disk_nodes, deleted_nodes, new_nodes = self._diff_pack_names()
1949
# TODO: handle same-name, index-size-changes here -
1950
# e.g. use the value from disk, not ours, *unless* we're the one
1952
for key, value in disk_nodes:
1953
builder.add_node(key, value)
1954
self.transport.put_file('pack-names', builder.finish(),
1955
mode=self.repo.bzrdir._get_file_mode())
1956
# move the baseline forward
1957
self._packs_at_load = disk_nodes
1958
if clear_obsolete_packs:
1959
self._clear_obsolete_packs()
1961
self._unlock_names()
1962
# synchronise the memory packs list with what we just wrote:
1963
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1964
return [new_node[0][0] for new_node in new_nodes]
1966
def reload_pack_names(self):
1967
"""Sync our pack listing with what is present in the repository.
1969
This should be called when we find out that something we thought was
1970
present is now missing. This happens when another process re-packs the
1973
:return: True if the in-memory list of packs has been altered at all.
1975
# The ensure_loaded call is to handle the case where the first call
1976
# made involving the collection was to reload_pack_names, where we
1977
# don't have a view of disk contents. Its a bit of a bandaid, and
1978
# causes two reads of pack-names, but its a rare corner case not struck
1979
# with regular push/pull etc.
1980
first_read = self.ensure_loaded()
1983
# out the new value.
1984
disk_nodes, _, _ = self._diff_pack_names()
1985
self._packs_at_load = disk_nodes
1987
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1988
if removed or added or modified:
1992
def _restart_autopack(self):
1993
"""Reload the pack names list, and restart the autopack code."""
1994
if not self.reload_pack_names():
1995
# Re-raise the original exception, because something went missing
1996
# and a restart didn't find it
1998
raise errors.RetryAutopack(self.repo, False, sys.exc_info())
2000
def _clear_obsolete_packs(self):
2001
"""Delete everything from the obsolete-packs directory.
2003
obsolete_pack_transport = self.transport.clone('obsolete_packs')
2004
for filename in obsolete_pack_transport.list_dir('.'):
2006
obsolete_pack_transport.delete(filename)
2007
except (errors.PathError, errors.TransportError), e:
2008
warning("couldn't delete obsolete pack, skipping it:\n%s" % (e,))
2010
def _start_write_group(self):
2011
# Do not permit preparation for writing if we're not in a 'write lock'.
2012
if not self.repo.is_write_locked():
2013
raise errors.NotWriteLocked(self)
2014
self._new_pack = self.pack_factory(self, upload_suffix='.pack',
2015
file_mode=self.repo.bzrdir._get_file_mode())
2016
# allow writing: queue writes to a new index
2017
self.revision_index.add_writable_index(self._new_pack.revision_index,
2019
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
2021
self.text_index.add_writable_index(self._new_pack.text_index,
2023
self._new_pack.text_index.set_optimize(combine_backing_indices=False)
2024
self.signature_index.add_writable_index(self._new_pack.signature_index,
2026
if self.chk_index is not None:
2027
self.chk_index.add_writable_index(self._new_pack.chk_index,
2029
self.repo.chk_bytes._index._add_callback = self.chk_index.add_callback
2030
self._new_pack.chk_index.set_optimize(combine_backing_indices=False)
2032
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
2033
self.repo.revisions._index._add_callback = self.revision_index.add_callback
2034
self.repo.signatures._index._add_callback = self.signature_index.add_callback
2035
self.repo.texts._index._add_callback = self.text_index.add_callback
2037
def _abort_write_group(self):
2038
# FIXME: just drop the transient index.
2039
# forget what names there are
2040
if self._new_pack is not None:
2042
self._new_pack.abort()
2044
# XXX: If we aborted while in the middle of finishing the write
2045
# group, _remove_pack_indices can fail because the indexes are
2046
# already gone. If they're not there we shouldn't fail in this
2047
# case. -- mbp 20081113
2048
self._remove_pack_indices(self._new_pack)
2049
self._new_pack = None
2050
for resumed_pack in self._resumed_packs:
2052
resumed_pack.abort()
2054
# See comment in previous finally block.
2056
self._remove_pack_indices(resumed_pack)
2059
del self._resumed_packs[:]
2061
def _remove_resumed_pack_indices(self):
2062
for resumed_pack in self._resumed_packs:
2063
self._remove_pack_indices(resumed_pack)
2064
del self._resumed_packs[:]
2066
def _check_new_inventories(self):
2067
"""Detect missing inventories in this write group.
2069
:returns: set of missing keys. Note that not every missing key is
2070
guaranteed to be reported.
2072
# The base implementation does no checks. GCRepositoryPackCollection
2076
def _commit_write_group(self):
2078
for prefix, versioned_file in (
2079
('revisions', self.repo.revisions),
2080
('inventories', self.repo.inventories),
2081
('texts', self.repo.texts),
2082
('signatures', self.repo.signatures),
2084
missing = versioned_file.get_missing_compression_parent_keys()
2085
all_missing.update([(prefix,) + key for key in missing])
2087
raise errors.BzrCheckError(
2088
"Repository %s has missing compression parent(s) %r "
2089
% (self.repo, sorted(all_missing)))
2090
all_missing = self._check_new_inventories()
2092
raise errors.BzrCheckError(
2093
"Repository %s missing keys for new revisions %r "
2094
% (self.repo, sorted(all_missing)))
2095
self._remove_pack_indices(self._new_pack)
2096
any_new_content = False
2097
if self._new_pack.data_inserted():
2098
# get all the data to disk and read to use
2099
self._new_pack.finish()
2100
self.allocate(self._new_pack)
2101
self._new_pack = None
2102
any_new_content = True
2104
self._new_pack.abort()
2105
self._new_pack = None
2106
for resumed_pack in self._resumed_packs:
2107
# XXX: this is a pretty ugly way to turn the resumed pack into a
2108
# properly committed pack.
2109
self._names[resumed_pack.name] = None
2110
self._remove_pack_from_memory(resumed_pack)
2111
resumed_pack.finish()
2112
self.allocate(resumed_pack)
2113
any_new_content = True
2114
del self._resumed_packs[:]
2116
result = self.autopack()
2118
# when autopack takes no steps, the names list is still
2120
return self._save_pack_names()
2124
def _suspend_write_group(self):
2125
tokens = [pack.name for pack in self._resumed_packs]
2126
self._remove_pack_indices(self._new_pack)
2127
if self._new_pack.data_inserted():
2128
# get all the data to disk and read to use
2129
self._new_pack.finish(suspend=True)
2130
tokens.append(self._new_pack.name)
2131
self._new_pack = None
2133
self._new_pack.abort()
2134
self._new_pack = None
2135
self._remove_resumed_pack_indices()
2138
def _resume_write_group(self, tokens):
2139
for token in tokens:
2140
self._resume_pack(token)
2143
class KnitPackRepository(KnitRepository):
2144
"""Repository with knit objects stored inside pack containers.
2146
The layering for a KnitPackRepository is:
2148
Graph | HPSS | Repository public layer |
2149
===================================================
2150
Tuple based apis below, string based, and key based apis above
2151
---------------------------------------------------
2153
Provides .texts, .revisions etc
2154
This adapts the N-tuple keys to physical knit records which only have a
2155
single string identifier (for historical reasons), which in older formats
2156
was always the revision_id, and in the mapped code for packs is always
2157
the last element of key tuples.
2158
---------------------------------------------------
2160
A separate GraphIndex is used for each of the
2161
texts/inventories/revisions/signatures contained within each individual
2162
pack file. The GraphIndex layer works in N-tuples and is unaware of any
2164
===================================================
2168
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
2170
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2171
_commit_builder_class, _serializer)
2172
index_transport = self._transport.clone('indices')
2173
self._pack_collection = RepositoryPackCollection(self, self._transport,
2175
self._transport.clone('upload'),
2176
self._transport.clone('packs'),
2177
_format.index_builder_class,
2178
_format.index_class,
2179
use_chk_index=self._format.supports_chks,
2181
self.inventories = KnitVersionedFiles(
2182
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2183
add_callback=self._pack_collection.inventory_index.add_callback,
2184
deltas=True, parents=True, is_locked=self.is_locked),
2185
data_access=self._pack_collection.inventory_index.data_access,
2186
max_delta_chain=200)
2187
self.revisions = KnitVersionedFiles(
2188
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2189
add_callback=self._pack_collection.revision_index.add_callback,
2190
deltas=False, parents=True, is_locked=self.is_locked,
2191
track_external_parent_refs=True),
2192
data_access=self._pack_collection.revision_index.data_access,
2194
self.signatures = KnitVersionedFiles(
2195
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2196
add_callback=self._pack_collection.signature_index.add_callback,
2197
deltas=False, parents=False, is_locked=self.is_locked),
2198
data_access=self._pack_collection.signature_index.data_access,
2200
self.texts = KnitVersionedFiles(
2201
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2202
add_callback=self._pack_collection.text_index.add_callback,
2203
deltas=True, parents=True, is_locked=self.is_locked),
2204
data_access=self._pack_collection.text_index.data_access,
2205
max_delta_chain=200)
2206
if _format.supports_chks:
2207
# No graph, no compression:- references from chks are between
2208
# different objects not temporal versions of the same; and without
2209
# some sort of temporal structure knit compression will just fail.
2210
self.chk_bytes = KnitVersionedFiles(
2211
_KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2212
add_callback=self._pack_collection.chk_index.add_callback,
2213
deltas=False, parents=False, is_locked=self.is_locked),
2214
data_access=self._pack_collection.chk_index.data_access,
2217
self.chk_bytes = None
2218
# True when the repository object is 'write locked' (as opposed to the
2219
# physical lock only taken out around changes to the pack-names list.)
2220
# Another way to represent this would be a decorator around the control
2221
# files object that presents logical locks as physical ones - if this
2222
# gets ugly consider that alternative design. RBC 20071011
2223
self._write_lock_count = 0
2224
self._transaction = None
2226
self._reconcile_does_inventory_gc = True
2227
self._reconcile_fixes_text_parents = True
2228
self._reconcile_backsup_inventory = False
2230
def _warn_if_deprecated(self):
2231
# This class isn't deprecated, but one sub-format is
2232
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2233
from bzrlib import repository
2234
if repository._deprecation_warning_done:
2236
repository._deprecation_warning_done = True
2237
warning("Format %s for %s is deprecated - please use"
2238
" 'bzr upgrade --1.6.1-rich-root'"
2239
% (self._format, self.bzrdir.transport.base))
2241
def _abort_write_group(self):
2242
self.revisions._index._key_dependencies.clear()
2243
self._pack_collection._abort_write_group()
2245
def _get_source(self, to_format):
2246
if to_format.network_name() == self._format.network_name():
2247
return KnitPackStreamSource(self, to_format)
2248
return super(KnitPackRepository, self)._get_source(to_format)
2250
def _make_parents_provider(self):
2251
return graph.CachingParentsProvider(self)
2253
def _refresh_data(self):
2254
if not self.is_locked():
2256
self._pack_collection.reload_pack_names()
2258
def _start_write_group(self):
2259
self._pack_collection._start_write_group()
2261
def _commit_write_group(self):
2262
hint = self._pack_collection._commit_write_group()
2263
self.revisions._index._key_dependencies.clear()
2266
def suspend_write_group(self):
2267
# XXX check self._write_group is self.get_transaction()?
2268
tokens = self._pack_collection._suspend_write_group()
2269
self.revisions._index._key_dependencies.clear()
2270
self._write_group = None
2273
def _resume_write_group(self, tokens):
2274
self._start_write_group()
2276
self._pack_collection._resume_write_group(tokens)
2277
except errors.UnresumableWriteGroup:
2278
self._abort_write_group()
2280
for pack in self._pack_collection._resumed_packs:
2281
self.revisions._index.scan_unvalidated_index(pack.revision_index)
2283
def get_transaction(self):
2284
if self._write_lock_count:
2285
return self._transaction
2287
return self.control_files.get_transaction()
2289
def is_locked(self):
2290
return self._write_lock_count or self.control_files.is_locked()
2292
def is_write_locked(self):
2293
return self._write_lock_count
2295
def lock_write(self, token=None):
2296
locked = self.is_locked()
2297
if not self._write_lock_count and locked:
2298
raise errors.ReadOnlyError(self)
2299
self._write_lock_count += 1
2300
if self._write_lock_count == 1:
2301
self._transaction = transactions.WriteTransaction()
2303
for repo in self._fallback_repositories:
2304
# Writes don't affect fallback repos
2306
self._refresh_data()
2308
def lock_read(self):
2309
locked = self.is_locked()
2310
if self._write_lock_count:
2311
self._write_lock_count += 1
2313
self.control_files.lock_read()
2315
for repo in self._fallback_repositories:
2317
self._refresh_data()
2319
def leave_lock_in_place(self):
2320
# not supported - raise an error
2321
raise NotImplementedError(self.leave_lock_in_place)
2323
def dont_leave_lock_in_place(self):
2324
# not supported - raise an error
2325
raise NotImplementedError(self.dont_leave_lock_in_place)
2328
def pack(self, hint=None):
2329
"""Compress the data within the repository.
2331
This will pack all the data to a single pack. In future it may
2332
recompress deltas or do other such expensive operations.
2334
self._pack_collection.pack(hint=hint)
2337
def reconcile(self, other=None, thorough=False):
2338
"""Reconcile this repository."""
2339
from bzrlib.reconcile import PackReconciler
2340
reconciler = PackReconciler(self, thorough=thorough)
2341
reconciler.reconcile()
2344
def _reconcile_pack(self, collection, packs, extension, revs, pb):
2345
packer = ReconcilePacker(collection, packs, extension, revs)
2346
return packer.pack(pb)
2349
if self._write_lock_count == 1 and self._write_group is not None:
2350
self.abort_write_group()
2351
self._transaction = None
2352
self._write_lock_count = 0
2353
raise errors.BzrError(
2354
'Must end write group before releasing write lock on %s'
2356
if self._write_lock_count:
2357
self._write_lock_count -= 1
2358
if not self._write_lock_count:
2359
transaction = self._transaction
2360
self._transaction = None
2361
transaction.finish()
2363
self.control_files.unlock()
2365
if not self.is_locked():
2366
for repo in self._fallback_repositories:
2370
class KnitPackStreamSource(StreamSource):
2371
"""A StreamSource used to transfer data between same-format KnitPack repos.
2373
This source assumes:
2374
1) Same serialization format for all objects
2375
2) Same root information
2376
3) XML format inventories
2377
4) Atomic inserts (so we can stream inventory texts before text
2382
def __init__(self, from_repository, to_format):
2383
super(KnitPackStreamSource, self).__init__(from_repository, to_format)
2384
self._text_keys = None
2385
self._text_fetch_order = 'unordered'
2387
def _get_filtered_inv_stream(self, revision_ids):
2388
from_repo = self.from_repository
2389
parent_ids = from_repo._find_parent_ids_of_revisions(revision_ids)
2390
parent_keys = [(p,) for p in parent_ids]
2391
find_text_keys = from_repo._find_text_key_references_from_xml_inventory_lines
2392
parent_text_keys = set(find_text_keys(
2393
from_repo._inventory_xml_lines_for_keys(parent_keys)))
2394
content_text_keys = set()
2395
knit = KnitVersionedFiles(None, None)
2396
factory = KnitPlainFactory()
2397
def find_text_keys_from_content(record):
2398
if record.storage_kind not in ('knit-delta-gz', 'knit-ft-gz'):
2399
raise ValueError("Unknown content storage kind for"
2400
" inventory text: %s" % (record.storage_kind,))
2401
# It's a knit record, it has a _raw_record field (even if it was
2402
# reconstituted from a network stream).
2403
raw_data = record._raw_record
2404
# read the entire thing
2405
revision_id = record.key[-1]
2406
content, _ = knit._parse_record(revision_id, raw_data)
2407
if record.storage_kind == 'knit-delta-gz':
2408
line_iterator = factory.get_linedelta_content(content)
2409
elif record.storage_kind == 'knit-ft-gz':
2410
line_iterator = factory.get_fulltext_content(content)
2411
content_text_keys.update(find_text_keys(
2412
[(line, revision_id) for line in line_iterator]))
2413
revision_keys = [(r,) for r in revision_ids]
2414
def _filtered_inv_stream():
2415
source_vf = from_repo.inventories
2416
stream = source_vf.get_record_stream(revision_keys,
2418
for record in stream:
2419
if record.storage_kind == 'absent':
2420
raise errors.NoSuchRevision(from_repo, record.key)
2421
find_text_keys_from_content(record)
2423
self._text_keys = content_text_keys - parent_text_keys
2424
return ('inventories', _filtered_inv_stream())
2426
def _get_text_stream(self):
2427
# Note: We know we don't have to handle adding root keys, because both
2428
# the source and target are the identical network name.
2429
text_stream = self.from_repository.texts.get_record_stream(
2430
self._text_keys, self._text_fetch_order, False)
2431
return ('texts', text_stream)
2433
def get_stream(self, search):
2434
revision_ids = search.get_keys()
2435
for stream_info in self._fetch_revision_texts(revision_ids):
2437
self._revision_keys = [(rev_id,) for rev_id in revision_ids]
2438
yield self._get_filtered_inv_stream(revision_ids)
2439
yield self._get_text_stream()
2443
class RepositoryFormatPack(MetaDirRepositoryFormat):
2444
"""Format logic for pack structured repositories.
2446
This repository format has:
2447
- a list of packs in pack-names
2448
- packs in packs/NAME.pack
2449
- indices in indices/NAME.{iix,six,tix,rix}
2450
- knit deltas in the packs, knit indices mapped to the indices.
2451
- thunk objects to support the knits programming API.
2452
- a format marker of its own
2453
- an optional 'shared-storage' flag
2454
- an optional 'no-working-trees' flag
2458
# Set this attribute in derived classes to control the repository class
2459
# created by open and initialize.
2460
repository_class = None
2461
# Set this attribute in derived classes to control the
2462
# _commit_builder_class that the repository objects will have passed to
2463
# their constructor.
2464
_commit_builder_class = None
2465
# Set this attribute in derived clases to control the _serializer that the
2466
# repository objects will have passed to their constructor.
2468
# Packs are not confused by ghosts.
2469
supports_ghosts = True
2470
# External references are not supported in pack repositories yet.
2471
supports_external_lookups = False
2472
# Most pack formats do not use chk lookups.
2473
supports_chks = False
2474
# What index classes to use
2475
index_builder_class = None
2477
_fetch_uses_deltas = True
2480
def initialize(self, a_bzrdir, shared=False):
2481
"""Create a pack based repository.
2483
:param a_bzrdir: bzrdir to contain the new repository; must already
2485
:param shared: If true the repository will be initialized as a shared
2488
mutter('creating repository in %s.', a_bzrdir.transport.base)
2489
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
2490
builder = self.index_builder_class()
2491
files = [('pack-names', builder.finish())]
2492
utf8_files = [('format', self.get_format_string())]
2494
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
2495
return self.open(a_bzrdir=a_bzrdir, _found=True)
2497
def open(self, a_bzrdir, _found=False, _override_transport=None):
2498
"""See RepositoryFormat.open().
2500
:param _override_transport: INTERNAL USE ONLY. Allows opening the
2501
repository at a slightly different url
2502
than normal. I.e. during 'upgrade'.
2505
format = RepositoryFormat.find_format(a_bzrdir)
2506
if _override_transport is not None:
2507
repo_transport = _override_transport
2509
repo_transport = a_bzrdir.get_repository_transport(None)
2510
control_files = lockable_files.LockableFiles(repo_transport,
2511
'lock', lockdir.LockDir)
2512
return self.repository_class(_format=self,
2514
control_files=control_files,
2515
_commit_builder_class=self._commit_builder_class,
2516
_serializer=self._serializer)
2519
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2520
"""A no-subtrees parameterized Pack repository.
2522
This format was introduced in 0.92.
2525
repository_class = KnitPackRepository
2526
_commit_builder_class = PackCommitBuilder
2528
def _serializer(self):
2529
return xml5.serializer_v5
2530
# What index classes to use
2531
index_builder_class = InMemoryGraphIndex
2532
index_class = GraphIndex
2534
def _get_matching_bzrdir(self):
2535
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2537
def _ignore_setting_bzrdir(self, format):
2540
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2542
def get_format_string(self):
2543
"""See RepositoryFormat.get_format_string()."""
2544
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2546
def get_format_description(self):
2547
"""See RepositoryFormat.get_format_description()."""
2548
return "Packs containing knits without subtree support"
2551
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2552
"""A subtrees parameterized Pack repository.
2554
This repository format uses the xml7 serializer to get:
2555
- support for recording full info about the tree root
2556
- support for recording tree-references
2558
This format was introduced in 0.92.
2561
repository_class = KnitPackRepository
2562
_commit_builder_class = PackRootCommitBuilder
2563
rich_root_data = True
2564
supports_tree_reference = True
2566
def _serializer(self):
2567
return xml7.serializer_v7
2568
# What index classes to use
2569
index_builder_class = InMemoryGraphIndex
2570
index_class = GraphIndex
2572
def _get_matching_bzrdir(self):
2573
return bzrdir.format_registry.make_bzrdir(
2574
'pack-0.92-subtree')
2576
def _ignore_setting_bzrdir(self, format):
2579
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2581
def get_format_string(self):
2582
"""See RepositoryFormat.get_format_string()."""
2583
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2585
def get_format_description(self):
2586
"""See RepositoryFormat.get_format_description()."""
2587
return "Packs containing knits with subtree support\n"
2590
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2591
"""A rich-root, no subtrees parameterized Pack repository.
2593
This repository format uses the xml6 serializer to get:
2594
- support for recording full info about the tree root
2596
This format was introduced in 1.0.
2599
repository_class = KnitPackRepository
2600
_commit_builder_class = PackRootCommitBuilder
2601
rich_root_data = True
2602
supports_tree_reference = False
2604
def _serializer(self):
2605
return xml6.serializer_v6
2606
# What index classes to use
2607
index_builder_class = InMemoryGraphIndex
2608
index_class = GraphIndex
2610
def _get_matching_bzrdir(self):
2611
return bzrdir.format_registry.make_bzrdir(
2614
def _ignore_setting_bzrdir(self, format):
2617
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2619
def get_format_string(self):
2620
"""See RepositoryFormat.get_format_string()."""
2621
return ("Bazaar pack repository format 1 with rich root"
2622
" (needs bzr 1.0)\n")
2624
def get_format_description(self):
2625
"""See RepositoryFormat.get_format_description()."""
2626
return "Packs containing knits with rich root support\n"
2629
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2630
"""Repository that supports external references to allow stacking.
2634
Supports external lookups, which results in non-truncated ghosts after
2635
reconcile compared to pack-0.92 formats.
2638
repository_class = KnitPackRepository
2639
_commit_builder_class = PackCommitBuilder
2640
supports_external_lookups = True
2641
# What index classes to use
2642
index_builder_class = InMemoryGraphIndex
2643
index_class = GraphIndex
2646
def _serializer(self):
2647
return xml5.serializer_v5
2649
def _get_matching_bzrdir(self):
2650
return bzrdir.format_registry.make_bzrdir('1.6')
2652
def _ignore_setting_bzrdir(self, format):
2655
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2657
def get_format_string(self):
2658
"""See RepositoryFormat.get_format_string()."""
2659
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2661
def get_format_description(self):
2662
"""See RepositoryFormat.get_format_description()."""
2663
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2666
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2667
"""A repository with rich roots and stacking.
2669
New in release 1.6.1.
2671
Supports stacking on other repositories, allowing data to be accessed
2672
without being stored locally.
2675
repository_class = KnitPackRepository
2676
_commit_builder_class = PackRootCommitBuilder
2677
rich_root_data = True
2678
supports_tree_reference = False # no subtrees
2679
supports_external_lookups = True
2680
# What index classes to use
2681
index_builder_class = InMemoryGraphIndex
2682
index_class = GraphIndex
2685
def _serializer(self):
2686
return xml6.serializer_v6
2688
def _get_matching_bzrdir(self):
2689
return bzrdir.format_registry.make_bzrdir(
2692
def _ignore_setting_bzrdir(self, format):
2695
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2697
def get_format_string(self):
2698
"""See RepositoryFormat.get_format_string()."""
2699
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2701
def get_format_description(self):
2702
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2705
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2706
"""A repository with rich roots and external references.
2710
Supports external lookups, which results in non-truncated ghosts after
2711
reconcile compared to pack-0.92 formats.
2713
This format was deprecated because the serializer it uses accidentally
2714
supported subtrees, when the format was not intended to. This meant that
2715
someone could accidentally fetch from an incorrect repository.
2718
repository_class = KnitPackRepository
2719
_commit_builder_class = PackRootCommitBuilder
2720
rich_root_data = True
2721
supports_tree_reference = False # no subtrees
2723
supports_external_lookups = True
2724
# What index classes to use
2725
index_builder_class = InMemoryGraphIndex
2726
index_class = GraphIndex
2729
def _serializer(self):
2730
return xml7.serializer_v7
2732
def _get_matching_bzrdir(self):
2733
matching = bzrdir.format_registry.make_bzrdir(
2735
matching.repository_format = self
2738
def _ignore_setting_bzrdir(self, format):
2741
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2743
def get_format_string(self):
2744
"""See RepositoryFormat.get_format_string()."""
2745
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2747
def get_format_description(self):
2748
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2752
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2753
"""A repository with stacking and btree indexes,
2754
without rich roots or subtrees.
2756
This is equivalent to pack-1.6 with B+Tree indices.
2759
repository_class = KnitPackRepository
2760
_commit_builder_class = PackCommitBuilder
2761
supports_external_lookups = True
2762
# What index classes to use
2763
index_builder_class = BTreeBuilder
2764
index_class = BTreeGraphIndex
2767
def _serializer(self):
2768
return xml5.serializer_v5
2770
def _get_matching_bzrdir(self):
2771
return bzrdir.format_registry.make_bzrdir('1.9')
2773
def _ignore_setting_bzrdir(self, format):
2776
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2778
def get_format_string(self):
2779
"""See RepositoryFormat.get_format_string()."""
2780
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2782
def get_format_description(self):
2783
"""See RepositoryFormat.get_format_description()."""
2784
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2787
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2788
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2790
1.6-rich-root with B+Tree indices.
2793
repository_class = KnitPackRepository
2794
_commit_builder_class = PackRootCommitBuilder
2795
rich_root_data = True
2796
supports_tree_reference = False # no subtrees
2797
supports_external_lookups = True
2798
# What index classes to use
2799
index_builder_class = BTreeBuilder
2800
index_class = BTreeGraphIndex
2803
def _serializer(self):
2804
return xml6.serializer_v6
2806
def _get_matching_bzrdir(self):
2807
return bzrdir.format_registry.make_bzrdir(
2810
def _ignore_setting_bzrdir(self, format):
2813
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2815
def get_format_string(self):
2816
"""See RepositoryFormat.get_format_string()."""
2817
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2819
def get_format_description(self):
2820
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2823
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2824
"""A subtrees development repository.
2826
This format should be retained until the second release after bzr 1.7.
2828
1.6.1-subtree[as it might have been] with B+Tree indices.
2830
This is [now] retained until we have a CHK based subtree format in
2834
repository_class = KnitPackRepository
2835
_commit_builder_class = PackRootCommitBuilder
2836
rich_root_data = True
2837
supports_tree_reference = True
2838
supports_external_lookups = True
2839
# What index classes to use
2840
index_builder_class = BTreeBuilder
2841
index_class = BTreeGraphIndex
2844
def _serializer(self):
2845
return xml7.serializer_v7
2847
def _get_matching_bzrdir(self):
2848
return bzrdir.format_registry.make_bzrdir(
2849
'development-subtree')
2851
def _ignore_setting_bzrdir(self, format):
2854
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2856
def get_format_string(self):
2857
"""See RepositoryFormat.get_format_string()."""
2858
return ("Bazaar development format 2 with subtree support "
2859
"(needs bzr.dev from before 1.8)\n")
2861
def get_format_description(self):
2862
"""See RepositoryFormat.get_format_description()."""
2863
return ("Development repository format, currently the same as "
2864
"1.6.1-subtree with B+Tree indices.\n")
added
removed
bzrlib/repofmt/pack_repo.py
