1
# Copyright (C) 2005, 2006, 2007, 2008 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
from bzrlib.lazy_import import lazy_import
18
lazy_import(globals(), """
19
from itertools import izip
33
from bzrlib.index import (
37
GraphIndexPrefixAdapter,
40
from bzrlib.knit import (
46
from bzrlib import tsort
56
from bzrlib.decorators import needs_write_lock
57
from bzrlib.btree_index import (
61
from bzrlib.index import (
65
from bzrlib.repofmt.knitrepo import KnitRepository
66
from bzrlib.repository import (
68
MetaDirRepositoryFormat,
72
import bzrlib.revision as _mod_revision
73
from bzrlib.trace import (
79
class PackCommitBuilder(CommitBuilder):
80
"""A subclass of CommitBuilder to add texts with pack semantics.
82
Specifically this uses one knit object rather than one knit object per
83
added text, reducing memory and object pressure.
86
def __init__(self, repository, parents, config, timestamp=None,
87
timezone=None, committer=None, revprops=None,
89
CommitBuilder.__init__(self, repository, parents, config,
90
timestamp=timestamp, timezone=timezone, committer=committer,
91
revprops=revprops, revision_id=revision_id)
92
self._file_graph = graph.Graph(
93
repository._pack_collection.text_index.combined_index)
95
def _heads(self, file_id, revision_ids):
96
keys = [(file_id, revision_id) for revision_id in revision_ids]
97
return set([key[1] for key in self._file_graph.heads(keys)])
100
class PackRootCommitBuilder(RootCommitBuilder):
101
"""A subclass of RootCommitBuilder to add texts with pack semantics.
103
Specifically this uses one knit object rather than one knit object per
104
added text, reducing memory and object pressure.
107
def __init__(self, repository, parents, config, timestamp=None,
108
timezone=None, committer=None, revprops=None,
110
CommitBuilder.__init__(self, repository, parents, config,
111
timestamp=timestamp, timezone=timezone, committer=committer,
112
revprops=revprops, revision_id=revision_id)
113
self._file_graph = graph.Graph(
114
repository._pack_collection.text_index.combined_index)
116
def _heads(self, file_id, revision_ids):
117
keys = [(file_id, revision_id) for revision_id in revision_ids]
118
return set([key[1] for key in self._file_graph.heads(keys)])
122
"""An in memory proxy for a pack and its indices.
124
This is a base class that is not directly used, instead the classes
125
ExistingPack and NewPack are used.
128
def __init__(self, revision_index, inventory_index, text_index,
130
"""Create a pack instance.
132
:param revision_index: A GraphIndex for determining what revisions are
133
present in the Pack and accessing the locations of their texts.
134
:param inventory_index: A GraphIndex for determining what inventories are
135
present in the Pack and accessing the locations of their
137
:param text_index: A GraphIndex for determining what file texts
138
are present in the pack and accessing the locations of their
139
texts/deltas (via (fileid, revisionid) tuples).
140
:param signature_index: A GraphIndex for determining what signatures are
141
present in the Pack and accessing the locations of their texts.
143
self.revision_index = revision_index
144
self.inventory_index = inventory_index
145
self.text_index = text_index
146
self.signature_index = signature_index
148
def access_tuple(self):
149
"""Return a tuple (transport, name) for the pack content."""
150
return self.pack_transport, self.file_name()
153
"""Get the file name for the pack on disk."""
154
return self.name + '.pack'
156
def get_revision_count(self):
157
return self.revision_index.key_count()
159
def inventory_index_name(self, name):
160
"""The inv index is the name + .iix."""
161
return self.index_name('inventory', name)
163
def revision_index_name(self, name):
164
"""The revision index is the name + .rix."""
165
return self.index_name('revision', name)
167
def signature_index_name(self, name):
168
"""The signature index is the name + .six."""
169
return self.index_name('signature', name)
171
def text_index_name(self, name):
172
"""The text index is the name + .tix."""
173
return self.index_name('text', name)
175
def _external_compression_parents_of_texts(self):
178
for node in self.text_index.iter_all_entries():
180
refs.update(node[3][1])
184
class ExistingPack(Pack):
185
"""An in memory proxy for an existing .pack and its disk indices."""
187
def __init__(self, pack_transport, name, revision_index, inventory_index,
188
text_index, signature_index):
189
"""Create an ExistingPack object.
191
:param pack_transport: The transport where the pack file resides.
192
:param name: The name of the pack on disk in the pack_transport.
194
Pack.__init__(self, revision_index, inventory_index, text_index,
197
self.pack_transport = pack_transport
198
if None in (revision_index, inventory_index, text_index,
199
signature_index, name, pack_transport):
200
raise AssertionError()
202
def __eq__(self, other):
203
return self.__dict__ == other.__dict__
205
def __ne__(self, other):
206
return not self.__eq__(other)
209
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
210
id(self), self.pack_transport, self.name)
214
"""An in memory proxy for a pack which is being created."""
216
# A map of index 'type' to the file extension and position in the
218
index_definitions = {
219
'revision': ('.rix', 0),
220
'inventory': ('.iix', 1),
222
'signature': ('.six', 3),
225
def __init__(self, upload_transport, index_transport, pack_transport,
226
upload_suffix='', file_mode=None, index_builder_class=None,
228
"""Create a NewPack instance.
230
:param upload_transport: A writable transport for the pack to be
231
incrementally uploaded to.
232
:param index_transport: A writable transport for the pack's indices to
233
be written to when the pack is finished.
234
:param pack_transport: A writable transport for the pack to be renamed
235
to when the upload is complete. This *must* be the same as
236
upload_transport.clone('../packs').
237
:param upload_suffix: An optional suffix to be given to any temporary
238
files created during the pack creation. e.g '.autopack'
239
:param file_mode: An optional file mode to create the new files with.
240
:param index_builder_class: Required keyword parameter - the class of
241
index builder to use.
242
:param index_class: Required keyword parameter - the class of index
245
# The relative locations of the packs are constrained, but all are
246
# passed in because the caller has them, so as to avoid object churn.
248
# Revisions: parents list, no text compression.
249
index_builder_class(reference_lists=1),
250
# Inventory: We want to map compression only, but currently the
251
# knit code hasn't been updated enough to understand that, so we
252
# have a regular 2-list index giving parents and compression
254
index_builder_class(reference_lists=2),
255
# Texts: compression and per file graph, for all fileids - so two
256
# reference lists and two elements in the key tuple.
257
index_builder_class(reference_lists=2, key_elements=2),
258
# Signatures: Just blobs to store, no compression, no parents
260
index_builder_class(reference_lists=0),
262
# When we make readonly indices, we need this.
263
self.index_class = index_class
264
# where should the new pack be opened
265
self.upload_transport = upload_transport
266
# where are indices written out to
267
self.index_transport = index_transport
268
# where is the pack renamed to when it is finished?
269
self.pack_transport = pack_transport
270
# What file mode to upload the pack and indices with.
271
self._file_mode = file_mode
272
# tracks the content written to the .pack file.
273
self._hash = osutils.md5()
274
# a four-tuple with the length in bytes of the indices, once the pack
275
# is finalised. (rev, inv, text, sigs)
276
self.index_sizes = None
277
# How much data to cache when writing packs. Note that this is not
278
# synchronised with reads, because it's not in the transport layer, so
279
# is not safe unless the client knows it won't be reading from the pack
281
self._cache_limit = 0
282
# the temporary pack file name.
283
self.random_name = osutils.rand_chars(20) + upload_suffix
284
# when was this pack started ?
285
self.start_time = time.time()
286
# open an output stream for the data added to the pack.
287
self.write_stream = self.upload_transport.open_write_stream(
288
self.random_name, mode=self._file_mode)
289
if 'pack' in debug.debug_flags:
290
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
291
time.ctime(), self.upload_transport.base, self.random_name,
292
time.time() - self.start_time)
293
# A list of byte sequences to be written to the new pack, and the
294
# aggregate size of them. Stored as a list rather than separate
295
# variables so that the _write_data closure below can update them.
296
self._buffer = [[], 0]
297
# create a callable for adding data
299
# robertc says- this is a closure rather than a method on the object
300
# so that the variables are locals, and faster than accessing object
302
def _write_data(bytes, flush=False, _buffer=self._buffer,
303
_write=self.write_stream.write, _update=self._hash.update):
304
_buffer[0].append(bytes)
305
_buffer[1] += len(bytes)
307
if _buffer[1] > self._cache_limit or flush:
308
bytes = ''.join(_buffer[0])
312
# expose this on self, for the occasion when clients want to add data.
313
self._write_data = _write_data
314
# a pack writer object to serialise pack records.
315
self._writer = pack.ContainerWriter(self._write_data)
317
# what state is the pack in? (open, finished, aborted)
321
"""Cancel creating this pack."""
322
self._state = 'aborted'
323
self.write_stream.close()
324
# Remove the temporary pack file.
325
self.upload_transport.delete(self.random_name)
326
# The indices have no state on disk.
328
def access_tuple(self):
329
"""Return a tuple (transport, name) for the pack content."""
330
if self._state == 'finished':
331
return Pack.access_tuple(self)
332
elif self._state == 'open':
333
return self.upload_transport, self.random_name
335
raise AssertionError(self._state)
337
def data_inserted(self):
338
"""True if data has been added to this pack."""
339
return bool(self.get_revision_count() or
340
self.inventory_index.key_count() or
341
self.text_index.key_count() or
342
self.signature_index.key_count())
345
"""Finish the new pack.
348
- finalises the content
349
- assigns a name (the md5 of the content, currently)
350
- writes out the associated indices
351
- renames the pack into place.
352
- stores the index size tuple for the pack in the index_sizes
357
self._write_data('', flush=True)
358
self.name = self._hash.hexdigest()
360
# XXX: It'd be better to write them all to temporary names, then
361
# rename them all into place, so that the window when only some are
362
# visible is smaller. On the other hand none will be seen until
363
# they're in the names list.
364
self.index_sizes = [None, None, None, None]
365
self._write_index('revision', self.revision_index, 'revision')
366
self._write_index('inventory', self.inventory_index, 'inventory')
367
self._write_index('text', self.text_index, 'file texts')
368
self._write_index('signature', self.signature_index,
369
'revision signatures')
370
self.write_stream.close()
371
# Note that this will clobber an existing pack with the same name,
372
# without checking for hash collisions. While this is undesirable this
373
# is something that can be rectified in a subsequent release. One way
374
# to rectify it may be to leave the pack at the original name, writing
375
# its pack-names entry as something like 'HASH: index-sizes
376
# temporary-name'. Allocate that and check for collisions, if it is
377
# collision free then rename it into place. If clients know this scheme
378
# they can handle missing-file errors by:
379
# - try for HASH.pack
380
# - try for temporary-name
381
# - refresh the pack-list to see if the pack is now absent
382
self.upload_transport.rename(self.random_name,
383
'../packs/' + self.name + '.pack')
384
self._state = 'finished'
385
if 'pack' in debug.debug_flags:
386
# XXX: size might be interesting?
387
mutter('%s: create_pack: pack renamed into place: %s%s->%s%s t+%6.3fs',
388
time.ctime(), self.upload_transport.base, self.random_name,
389
self.pack_transport, self.name,
390
time.time() - self.start_time)
393
"""Flush any current data."""
395
bytes = ''.join(self._buffer[0])
396
self.write_stream.write(bytes)
397
self._hash.update(bytes)
398
self._buffer[:] = [[], 0]
400
def index_name(self, index_type, name):
401
"""Get the disk name of an index type for pack name 'name'."""
402
return name + NewPack.index_definitions[index_type][0]
404
def index_offset(self, index_type):
405
"""Get the position in a index_size array for a given index type."""
406
return NewPack.index_definitions[index_type][1]
408
def _replace_index_with_readonly(self, index_type):
409
setattr(self, index_type + '_index',
410
self.index_class(self.index_transport,
411
self.index_name(index_type, self.name),
412
self.index_sizes[self.index_offset(index_type)]))
414
def set_write_cache_size(self, size):
415
self._cache_limit = size
417
def _write_index(self, index_type, index, label):
418
"""Write out an index.
420
:param index_type: The type of index to write - e.g. 'revision'.
421
:param index: The index object to serialise.
422
:param label: What label to give the index e.g. 'revision'.
424
index_name = self.index_name(index_type, self.name)
425
self.index_sizes[self.index_offset(index_type)] = \
426
self.index_transport.put_file(index_name, index.finish(),
427
mode=self._file_mode)
428
if 'pack' in debug.debug_flags:
429
# XXX: size might be interesting?
430
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
431
time.ctime(), label, self.upload_transport.base,
432
self.random_name, time.time() - self.start_time)
433
# Replace the writable index on this object with a readonly,
434
# presently unloaded index. We should alter
435
# the index layer to make its finish() error if add_node is
436
# subsequently used. RBC
437
self._replace_index_with_readonly(index_type)
440
class AggregateIndex(object):
441
"""An aggregated index for the RepositoryPackCollection.
443
AggregateIndex is reponsible for managing the PackAccess object,
444
Index-To-Pack mapping, and all indices list for a specific type of index
445
such as 'revision index'.
447
A CombinedIndex provides an index on a single key space built up
448
from several on-disk indices. The AggregateIndex builds on this
449
to provide a knit access layer, and allows having up to one writable
450
index within the collection.
452
# XXX: Probably 'can be written to' could/should be separated from 'acts
453
# like a knit index' -- mbp 20071024
455
def __init__(self, reload_func=None):
456
"""Create an AggregateIndex.
458
:param reload_func: A function to call if we find we are missing an
459
index. Should have the form reload_func() => True if the list of
460
active pack files has changed.
462
self._reload_func = reload_func
463
self.index_to_pack = {}
464
self.combined_index = CombinedGraphIndex([], reload_func=reload_func)
465
self.data_access = _DirectPackAccess(self.index_to_pack)
466
self.add_callback = None
468
def replace_indices(self, index_to_pack, indices):
469
"""Replace the current mappings with fresh ones.
471
This should probably not be used eventually, rather incremental add and
472
removal of indices. It has been added during refactoring of existing
475
:param index_to_pack: A mapping from index objects to
476
(transport, name) tuples for the pack file data.
477
:param indices: A list of indices.
479
# refresh the revision pack map dict without replacing the instance.
480
self.index_to_pack.clear()
481
self.index_to_pack.update(index_to_pack)
482
# XXX: API break - clearly a 'replace' method would be good?
483
self.combined_index._indices[:] = indices
484
# the current add nodes callback for the current writable index if
486
self.add_callback = None
488
def add_index(self, index, pack):
489
"""Add index to the aggregate, which is an index for Pack pack.
491
Future searches on the aggregate index will seach this new index
492
before all previously inserted indices.
494
:param index: An Index for the pack.
495
:param pack: A Pack instance.
497
# expose it to the index map
498
self.index_to_pack[index] = pack.access_tuple()
499
# put it at the front of the linear index list
500
self.combined_index.insert_index(0, index)
502
def add_writable_index(self, index, pack):
503
"""Add an index which is able to have data added to it.
505
There can be at most one writable index at any time. Any
506
modifications made to the knit are put into this index.
508
:param index: An index from the pack parameter.
509
:param pack: A Pack instance.
511
if self.add_callback is not None:
512
raise AssertionError(
513
"%s already has a writable index through %s" % \
514
(self, self.add_callback))
515
# allow writing: queue writes to a new index
516
self.add_index(index, pack)
517
# Updates the index to packs mapping as a side effect,
518
self.data_access.set_writer(pack._writer, index, pack.access_tuple())
519
self.add_callback = index.add_nodes
522
"""Reset all the aggregate data to nothing."""
523
self.data_access.set_writer(None, None, (None, None))
524
self.index_to_pack.clear()
525
del self.combined_index._indices[:]
526
self.add_callback = None
528
def remove_index(self, index, pack):
529
"""Remove index from the indices used to answer queries.
531
:param index: An index from the pack parameter.
532
:param pack: A Pack instance.
534
del self.index_to_pack[index]
535
self.combined_index._indices.remove(index)
536
if (self.add_callback is not None and
537
getattr(index, 'add_nodes', None) == self.add_callback):
538
self.add_callback = None
539
self.data_access.set_writer(None, None, (None, None))
542
class Packer(object):
543
"""Create a pack from packs."""
545
def __init__(self, pack_collection, packs, suffix, revision_ids=None):
548
:param pack_collection: A RepositoryPackCollection object where the
549
new pack is being written to.
550
:param packs: The packs to combine.
551
:param suffix: The suffix to use on the temporary files for the pack.
552
:param revision_ids: Revision ids to limit the pack to.
556
self.revision_ids = revision_ids
557
# The pack object we are creating.
559
self._pack_collection = pack_collection
560
# The index layer keys for the revisions being copied. None for 'all
562
self._revision_keys = None
563
# What text keys to copy. None for 'all texts'. This is set by
564
# _copy_inventory_texts
565
self._text_filter = None
568
def _extra_init(self):
569
"""A template hook to allow extending the constructor trivially."""
571
def pack(self, pb=None):
572
"""Create a new pack by reading data from other packs.
574
This does little more than a bulk copy of data. One key difference
575
is that data with the same item key across multiple packs is elided
576
from the output. The new pack is written into the current pack store
577
along with its indices, and the name added to the pack names. The
578
source packs are not altered and are not required to be in the current
581
:param pb: An optional progress bar to use. A nested bar is created if
583
:return: A Pack object, or None if nothing was copied.
585
# open a pack - using the same name as the last temporary file
586
# - which has already been flushed, so its safe.
587
# XXX: - duplicate code warning with start_write_group; fix before
588
# considering 'done'.
589
if self._pack_collection._new_pack is not None:
590
raise errors.BzrError('call to create_pack_from_packs while '
591
'another pack is being written.')
592
if self.revision_ids is not None:
593
if len(self.revision_ids) == 0:
594
# silly fetch request.
597
self.revision_ids = frozenset(self.revision_ids)
598
self.revision_keys = frozenset((revid,) for revid in
601
self.pb = ui.ui_factory.nested_progress_bar()
605
return self._create_pack_from_packs()
611
"""Open a pack for the pack we are creating."""
612
return NewPack(self._pack_collection._upload_transport,
613
self._pack_collection._index_transport,
614
self._pack_collection._pack_transport, upload_suffix=self.suffix,
615
file_mode=self._pack_collection.repo.bzrdir._get_file_mode(),
616
index_builder_class=self._pack_collection._index_builder_class,
617
index_class=self._pack_collection._index_class)
619
def _copy_revision_texts(self):
620
"""Copy revision data to the new pack."""
622
if self.revision_ids:
623
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
626
# select revision keys
627
revision_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
628
self.packs, 'revision_index')[0]
629
revision_nodes = self._pack_collection._index_contents(revision_index_map, revision_keys)
630
# copy revision keys and adjust values
631
self.pb.update("Copying revision texts", 1)
632
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
633
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
634
self.new_pack.revision_index, readv_group_iter, total_items))
635
if 'pack' in debug.debug_flags:
636
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
637
time.ctime(), self._pack_collection._upload_transport.base,
638
self.new_pack.random_name,
639
self.new_pack.revision_index.key_count(),
640
time.time() - self.new_pack.start_time)
641
self._revision_keys = revision_keys
643
def _copy_inventory_texts(self):
644
"""Copy the inventory texts to the new pack.
646
self._revision_keys is used to determine what inventories to copy.
648
Sets self._text_filter appropriately.
650
# select inventory keys
651
inv_keys = self._revision_keys # currently the same keyspace, and note that
652
# querying for keys here could introduce a bug where an inventory item
653
# is missed, so do not change it to query separately without cross
654
# checking like the text key check below.
655
inventory_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
656
self.packs, 'inventory_index')[0]
657
inv_nodes = self._pack_collection._index_contents(inventory_index_map, inv_keys)
658
# copy inventory keys and adjust values
659
# XXX: Should be a helper function to allow different inv representation
661
self.pb.update("Copying inventory texts", 2)
662
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
663
# Only grab the output lines if we will be processing them
664
output_lines = bool(self.revision_ids)
665
inv_lines = self._copy_nodes_graph(inventory_index_map,
666
self.new_pack._writer, self.new_pack.inventory_index,
667
readv_group_iter, total_items, output_lines=output_lines)
668
if self.revision_ids:
669
self._process_inventory_lines(inv_lines)
671
# eat the iterator to cause it to execute.
673
self._text_filter = None
674
if 'pack' in debug.debug_flags:
675
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
676
time.ctime(), self._pack_collection._upload_transport.base,
677
self.new_pack.random_name,
678
self.new_pack.inventory_index.key_count(),
679
time.time() - self.new_pack.start_time)
681
def _copy_text_texts(self):
683
text_index_map, text_nodes = self._get_text_nodes()
684
if self._text_filter is not None:
685
# We could return the keys copied as part of the return value from
686
# _copy_nodes_graph but this doesn't work all that well with the
687
# need to get line output too, so we check separately, and as we're
688
# going to buffer everything anyway, we check beforehand, which
689
# saves reading knit data over the wire when we know there are
691
text_nodes = set(text_nodes)
692
present_text_keys = set(_node[1] for _node in text_nodes)
693
missing_text_keys = set(self._text_filter) - present_text_keys
694
if missing_text_keys:
695
# TODO: raise a specific error that can handle many missing
697
a_missing_key = missing_text_keys.pop()
698
raise errors.RevisionNotPresent(a_missing_key[1],
700
# copy text keys and adjust values
701
self.pb.update("Copying content texts", 3)
702
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
703
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
704
self.new_pack.text_index, readv_group_iter, total_items))
705
self._log_copied_texts()
707
def _check_references(self):
708
"""Make sure our external refereneces are present."""
709
external_refs = self.new_pack._external_compression_parents_of_texts()
711
self._pack_collection._check_references_present(external_refs)
713
def _create_pack_from_packs(self):
714
self.pb.update("Opening pack", 0, 5)
715
self.new_pack = self.open_pack()
716
new_pack = self.new_pack
717
# buffer data - we won't be reading-back during the pack creation and
718
# this makes a significant difference on sftp pushes.
719
new_pack.set_write_cache_size(1024*1024)
720
if 'pack' in debug.debug_flags:
721
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
722
for a_pack in self.packs]
723
if self.revision_ids is not None:
724
rev_count = len(self.revision_ids)
727
mutter('%s: create_pack: creating pack from source packs: '
728
'%s%s %s revisions wanted %s t=0',
729
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
730
plain_pack_list, rev_count)
731
self._copy_revision_texts()
732
self._copy_inventory_texts()
733
self._copy_text_texts()
734
# select signature keys
735
signature_filter = self._revision_keys # same keyspace
736
signature_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
737
self.packs, 'signature_index')[0]
738
signature_nodes = self._pack_collection._index_contents(signature_index_map,
740
# copy signature keys and adjust values
741
self.pb.update("Copying signature texts", 4)
742
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
743
new_pack.signature_index)
744
if 'pack' in debug.debug_flags:
745
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
746
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
747
new_pack.signature_index.key_count(),
748
time.time() - new_pack.start_time)
749
self._check_references()
750
if not self._use_pack(new_pack):
753
self.pb.update("Finishing pack", 5)
755
self._pack_collection.allocate(new_pack)
758
def _copy_nodes(self, nodes, index_map, writer, write_index):
759
"""Copy knit nodes between packs with no graph references."""
760
pb = ui.ui_factory.nested_progress_bar()
762
return self._do_copy_nodes(nodes, index_map, writer,
767
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb):
768
# for record verification
769
knit = KnitVersionedFiles(None, None)
770
# plan a readv on each source pack:
772
nodes = sorted(nodes)
773
# how to map this into knit.py - or knit.py into this?
774
# we don't want the typical knit logic, we want grouping by pack
775
# at this point - perhaps a helper library for the following code
776
# duplication points?
778
for index, key, value in nodes:
779
if index not in request_groups:
780
request_groups[index] = []
781
request_groups[index].append((key, value))
783
pb.update("Copied record", record_index, len(nodes))
784
for index, items in request_groups.iteritems():
785
pack_readv_requests = []
786
for key, value in items:
787
# ---- KnitGraphIndex.get_position
788
bits = value[1:].split(' ')
789
offset, length = int(bits[0]), int(bits[1])
790
pack_readv_requests.append((offset, length, (key, value[0])))
791
# linear scan up the pack
792
pack_readv_requests.sort()
794
transport, path = index_map[index]
795
reader = pack.make_readv_reader(transport, path,
796
[offset[0:2] for offset in pack_readv_requests])
797
for (names, read_func), (_1, _2, (key, eol_flag)) in \
798
izip(reader.iter_records(), pack_readv_requests):
799
raw_data = read_func(None)
800
# check the header only
801
df, _ = knit._parse_record_header(key, raw_data)
803
pos, size = writer.add_bytes_record(raw_data, names)
804
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
805
pb.update("Copied record", record_index)
808
def _copy_nodes_graph(self, index_map, writer, write_index,
809
readv_group_iter, total_items, output_lines=False):
810
"""Copy knit nodes between packs.
812
:param output_lines: Return lines present in the copied data as
813
an iterator of line,version_id.
815
pb = ui.ui_factory.nested_progress_bar()
817
for result in self._do_copy_nodes_graph(index_map, writer,
818
write_index, output_lines, pb, readv_group_iter, total_items):
821
# Python 2.4 does not permit try:finally: in a generator.
827
def _do_copy_nodes_graph(self, index_map, writer, write_index,
828
output_lines, pb, readv_group_iter, total_items):
829
# for record verification
830
knit = KnitVersionedFiles(None, None)
831
# for line extraction when requested (inventories only)
833
factory = KnitPlainFactory()
835
pb.update("Copied record", record_index, total_items)
836
for index, readv_vector, node_vector in readv_group_iter:
838
transport, path = index_map[index]
839
reader = pack.make_readv_reader(transport, path, readv_vector)
840
for (names, read_func), (key, eol_flag, references) in \
841
izip(reader.iter_records(), node_vector):
842
raw_data = read_func(None)
844
# read the entire thing
845
content, _ = knit._parse_record(key[-1], raw_data)
846
if len(references[-1]) == 0:
847
line_iterator = factory.get_fulltext_content(content)
849
line_iterator = factory.get_linedelta_content(content)
850
for line in line_iterator:
853
# check the header only
854
df, _ = knit._parse_record_header(key, raw_data)
856
pos, size = writer.add_bytes_record(raw_data, names)
857
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
858
pb.update("Copied record", record_index)
861
def _get_text_nodes(self):
862
text_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
863
self.packs, 'text_index')[0]
864
return text_index_map, self._pack_collection._index_contents(text_index_map,
867
def _least_readv_node_readv(self, nodes):
868
"""Generate request groups for nodes using the least readv's.
870
:param nodes: An iterable of graph index nodes.
871
:return: Total node count and an iterator of the data needed to perform
872
readvs to obtain the data for nodes. Each item yielded by the
873
iterator is a tuple with:
874
index, readv_vector, node_vector. readv_vector is a list ready to
875
hand to the transport readv method, and node_vector is a list of
876
(key, eol_flag, references) for the the node retrieved by the
877
matching readv_vector.
879
# group by pack so we do one readv per pack
880
nodes = sorted(nodes)
883
for index, key, value, references in nodes:
884
if index not in request_groups:
885
request_groups[index] = []
886
request_groups[index].append((key, value, references))
888
for index, items in request_groups.iteritems():
889
pack_readv_requests = []
890
for key, value, references in items:
891
# ---- KnitGraphIndex.get_position
892
bits = value[1:].split(' ')
893
offset, length = int(bits[0]), int(bits[1])
894
pack_readv_requests.append(
895
((offset, length), (key, value[0], references)))
896
# linear scan up the pack to maximum range combining.
897
pack_readv_requests.sort()
898
# split out the readv and the node data.
899
pack_readv = [readv for readv, node in pack_readv_requests]
900
node_vector = [node for readv, node in pack_readv_requests]
901
result.append((index, pack_readv, node_vector))
904
def _log_copied_texts(self):
905
if 'pack' in debug.debug_flags:
906
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
907
time.ctime(), self._pack_collection._upload_transport.base,
908
self.new_pack.random_name,
909
self.new_pack.text_index.key_count(),
910
time.time() - self.new_pack.start_time)
912
def _process_inventory_lines(self, inv_lines):
913
"""Use up the inv_lines generator and setup a text key filter."""
914
repo = self._pack_collection.repo
915
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
916
inv_lines, self.revision_keys)
918
for fileid, file_revids in fileid_revisions.iteritems():
919
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
920
self._text_filter = text_filter
922
def _revision_node_readv(self, revision_nodes):
923
"""Return the total revisions and the readv's to issue.
925
:param revision_nodes: The revision index contents for the packs being
926
incorporated into the new pack.
927
:return: As per _least_readv_node_readv.
929
return self._least_readv_node_readv(revision_nodes)
931
def _use_pack(self, new_pack):
932
"""Return True if new_pack should be used.
934
:param new_pack: The pack that has just been created.
935
:return: True if the pack should be used.
937
return new_pack.data_inserted()
940
class OptimisingPacker(Packer):
941
"""A packer which spends more time to create better disk layouts."""
943
def _revision_node_readv(self, revision_nodes):
944
"""Return the total revisions and the readv's to issue.
946
This sort places revisions in topological order with the ancestors
949
:param revision_nodes: The revision index contents for the packs being
950
incorporated into the new pack.
951
:return: As per _least_readv_node_readv.
953
# build an ancestors dict
956
for index, key, value, references in revision_nodes:
957
ancestors[key] = references[0]
958
by_key[key] = (index, value, references)
959
order = tsort.topo_sort(ancestors)
961
# Single IO is pathological, but it will work as a starting point.
963
for key in reversed(order):
964
index, value, references = by_key[key]
965
# ---- KnitGraphIndex.get_position
966
bits = value[1:].split(' ')
967
offset, length = int(bits[0]), int(bits[1])
969
(index, [(offset, length)], [(key, value[0], references)]))
970
# TODO: combine requests in the same index that are in ascending order.
971
return total, requests
974
"""Open a pack for the pack we are creating."""
975
new_pack = super(OptimisingPacker, self).open_pack()
976
# Turn on the optimization flags for all the index builders.
977
new_pack.revision_index.set_optimize(for_size=True)
978
new_pack.inventory_index.set_optimize(for_size=True)
979
new_pack.text_index.set_optimize(for_size=True)
980
new_pack.signature_index.set_optimize(for_size=True)
984
class ReconcilePacker(Packer):
985
"""A packer which regenerates indices etc as it copies.
987
This is used by ``bzr reconcile`` to cause parent text pointers to be
991
def _extra_init(self):
992
self._data_changed = False
994
def _process_inventory_lines(self, inv_lines):
995
"""Generate a text key reference map rather for reconciling with."""
996
repo = self._pack_collection.repo
997
refs = repo._find_text_key_references_from_xml_inventory_lines(
999
self._text_refs = refs
1000
# during reconcile we:
1001
# - convert unreferenced texts to full texts
1002
# - correct texts which reference a text not copied to be full texts
1003
# - copy all others as-is but with corrected parents.
1004
# - so at this point we don't know enough to decide what becomes a full
1006
self._text_filter = None
1008
def _copy_text_texts(self):
1009
"""generate what texts we should have and then copy."""
1010
self.pb.update("Copying content texts", 3)
1011
# we have three major tasks here:
1012
# 1) generate the ideal index
1013
repo = self._pack_collection.repo
1014
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1015
_1, key, _2, refs in
1016
self.new_pack.revision_index.iter_all_entries()])
1017
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1018
# 2) generate a text_nodes list that contains all the deltas that can
1019
# be used as-is, with corrected parents.
1022
discarded_nodes = []
1023
NULL_REVISION = _mod_revision.NULL_REVISION
1024
text_index_map, text_nodes = self._get_text_nodes()
1025
for node in text_nodes:
1031
ideal_parents = tuple(ideal_index[node[1]])
1033
discarded_nodes.append(node)
1034
self._data_changed = True
1036
if ideal_parents == (NULL_REVISION,):
1038
if ideal_parents == node[3][0]:
1040
ok_nodes.append(node)
1041
elif ideal_parents[0:1] == node[3][0][0:1]:
1042
# the left most parent is the same, or there are no parents
1043
# today. Either way, we can preserve the representation as
1044
# long as we change the refs to be inserted.
1045
self._data_changed = True
1046
ok_nodes.append((node[0], node[1], node[2],
1047
(ideal_parents, node[3][1])))
1048
self._data_changed = True
1050
# Reinsert this text completely
1051
bad_texts.append((node[1], ideal_parents))
1052
self._data_changed = True
1053
# we're finished with some data.
1056
# 3) bulk copy the ok data
1057
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1058
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1059
self.new_pack.text_index, readv_group_iter, total_items))
1060
# 4) adhoc copy all the other texts.
1061
# We have to topologically insert all texts otherwise we can fail to
1062
# reconcile when parts of a single delta chain are preserved intact,
1063
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1064
# reinserted, and if d3 has incorrect parents it will also be
1065
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1066
# copied), so we will try to delta, but d2 is not currently able to be
1067
# extracted because it's basis d1 is not present. Topologically sorting
1068
# addresses this. The following generates a sort for all the texts that
1069
# are being inserted without having to reference the entire text key
1070
# space (we only topo sort the revisions, which is smaller).
1071
topo_order = tsort.topo_sort(ancestors)
1072
rev_order = dict(zip(topo_order, range(len(topo_order))))
1073
bad_texts.sort(key=lambda key:rev_order[key[0][1]])
1074
transaction = repo.get_transaction()
1075
file_id_index = GraphIndexPrefixAdapter(
1076
self.new_pack.text_index,
1078
add_nodes_callback=self.new_pack.text_index.add_nodes)
1079
data_access = _DirectPackAccess(
1080
{self.new_pack.text_index:self.new_pack.access_tuple()})
1081
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1082
self.new_pack.access_tuple())
1083
output_texts = KnitVersionedFiles(
1084
_KnitGraphIndex(self.new_pack.text_index,
1085
add_callback=self.new_pack.text_index.add_nodes,
1086
deltas=True, parents=True, is_locked=repo.is_locked),
1087
data_access=data_access, max_delta_chain=200)
1088
for key, parent_keys in bad_texts:
1089
# We refer to the new pack to delta data being output.
1090
# A possible improvement would be to catch errors on short reads
1091
# and only flush then.
1092
self.new_pack.flush()
1094
for parent_key in parent_keys:
1095
if parent_key[0] != key[0]:
1096
# Graph parents must match the fileid
1097
raise errors.BzrError('Mismatched key parent %r:%r' %
1099
parents.append(parent_key[1])
1100
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1101
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1102
output_texts.add_lines(key, parent_keys, text_lines,
1103
random_id=True, check_content=False)
1104
# 5) check that nothing inserted has a reference outside the keyspace.
1105
missing_text_keys = self.new_pack._external_compression_parents_of_texts()
1106
if missing_text_keys:
1107
raise errors.BzrError('Reference to missing compression parents %r'
1108
% (missing_text_keys,))
1109
self._log_copied_texts()
1111
def _use_pack(self, new_pack):
1112
"""Override _use_pack to check for reconcile having changed content."""
1113
# XXX: we might be better checking this at the copy time.
1114
original_inventory_keys = set()
1115
inv_index = self._pack_collection.inventory_index.combined_index
1116
for entry in inv_index.iter_all_entries():
1117
original_inventory_keys.add(entry[1])
1118
new_inventory_keys = set()
1119
for entry in new_pack.inventory_index.iter_all_entries():
1120
new_inventory_keys.add(entry[1])
1121
if new_inventory_keys != original_inventory_keys:
1122
self._data_changed = True
1123
return new_pack.data_inserted() and self._data_changed
1126
class RepositoryPackCollection(object):
1127
"""Management of packs within a repository.
1129
:ivar _names: map of {pack_name: (index_size,)}
1132
def __init__(self, repo, transport, index_transport, upload_transport,
1133
pack_transport, index_builder_class, index_class):
1134
"""Create a new RepositoryPackCollection.
1136
:param transport: Addresses the repository base directory
1137
(typically .bzr/repository/).
1138
:param index_transport: Addresses the directory containing indices.
1139
:param upload_transport: Addresses the directory into which packs are written
1140
while they're being created.
1141
:param pack_transport: Addresses the directory of existing complete packs.
1142
:param index_builder_class: The index builder class to use.
1143
:param index_class: The index class to use.
1146
self.transport = transport
1147
self._index_transport = index_transport
1148
self._upload_transport = upload_transport
1149
self._pack_transport = pack_transport
1150
self._index_builder_class = index_builder_class
1151
self._index_class = index_class
1152
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
1155
self._packs_by_name = {}
1156
# the previous pack-names content
1157
self._packs_at_load = None
1158
# when a pack is being created by this object, the state of that pack.
1159
self._new_pack = None
1160
# aggregated revision index data
1161
self.revision_index = AggregateIndex(self.reload_pack_names)
1162
self.inventory_index = AggregateIndex(self.reload_pack_names)
1163
self.text_index = AggregateIndex(self.reload_pack_names)
1164
self.signature_index = AggregateIndex(self.reload_pack_names)
1166
def add_pack_to_memory(self, pack):
1167
"""Make a Pack object available to the repository to satisfy queries.
1169
:param pack: A Pack object.
1171
if pack.name in self._packs_by_name:
1172
raise AssertionError()
1173
self.packs.append(pack)
1174
self._packs_by_name[pack.name] = pack
1175
self.revision_index.add_index(pack.revision_index, pack)
1176
self.inventory_index.add_index(pack.inventory_index, pack)
1177
self.text_index.add_index(pack.text_index, pack)
1178
self.signature_index.add_index(pack.signature_index, pack)
1180
def all_packs(self):
1181
"""Return a list of all the Pack objects this repository has.
1183
Note that an in-progress pack being created is not returned.
1185
:return: A list of Pack objects for all the packs in the repository.
1188
for name in self.names():
1189
result.append(self.get_pack_by_name(name))
1193
"""Pack the pack collection incrementally.
1195
This will not attempt global reorganisation or recompression,
1196
rather it will just ensure that the total number of packs does
1197
not grow without bound. It uses the _max_pack_count method to
1198
determine if autopacking is needed, and the pack_distribution
1199
method to determine the number of revisions in each pack.
1201
If autopacking takes place then the packs name collection will have
1202
been flushed to disk - packing requires updating the name collection
1203
in synchronisation with certain steps. Otherwise the names collection
1206
:return: True if packing took place.
1208
# XXX: Should not be needed when the management of indices is sane.
1209
total_revisions = self.revision_index.combined_index.key_count()
1210
total_packs = len(self._names)
1211
if self._max_pack_count(total_revisions) >= total_packs:
1213
# XXX: the following may want to be a class, to pack with a given
1215
mutter('Auto-packing repository %s, which has %d pack files, '
1216
'containing %d revisions into %d packs.', self, total_packs,
1217
total_revisions, self._max_pack_count(total_revisions))
1218
# determine which packs need changing
1219
pack_distribution = self.pack_distribution(total_revisions)
1221
for pack in self.all_packs():
1222
revision_count = pack.get_revision_count()
1223
if revision_count == 0:
1224
# revision less packs are not generated by normal operation,
1225
# only by operations like sign-my-commits, and thus will not
1226
# tend to grow rapdily or without bound like commit containing
1227
# packs do - leave them alone as packing them really should
1228
# group their data with the relevant commit, and that may
1229
# involve rewriting ancient history - which autopack tries to
1230
# avoid. Alternatively we could not group the data but treat
1231
# each of these as having a single revision, and thus add
1232
# one revision for each to the total revision count, to get
1233
# a matching distribution.
1235
existing_packs.append((revision_count, pack))
1236
pack_operations = self.plan_autopack_combinations(
1237
existing_packs, pack_distribution)
1238
self._execute_pack_operations(pack_operations)
1241
def _execute_pack_operations(self, pack_operations, _packer_class=Packer):
1242
"""Execute a series of pack operations.
1244
:param pack_operations: A list of [revision_count, packs_to_combine].
1245
:param _packer_class: The class of packer to use (default: Packer).
1248
for revision_count, packs in pack_operations:
1249
# we may have no-ops from the setup logic
1252
_packer_class(self, packs, '.autopack').pack()
1254
self._remove_pack_from_memory(pack)
1255
# record the newly available packs and stop advertising the old
1257
self._save_pack_names(clear_obsolete_packs=True)
1258
# Move the old packs out of the way now they are no longer referenced.
1259
for revision_count, packs in pack_operations:
1260
self._obsolete_packs(packs)
1262
def lock_names(self):
1263
"""Acquire the mutex around the pack-names index.
1265
This cannot be used in the middle of a read-only transaction on the
1268
self.repo.control_files.lock_write()
1271
"""Pack the pack collection totally."""
1272
self.ensure_loaded()
1273
total_packs = len(self._names)
1275
# This is arguably wrong because we might not be optimal, but for
1276
# now lets leave it in. (e.g. reconcile -> one pack. But not
1279
total_revisions = self.revision_index.combined_index.key_count()
1280
# XXX: the following may want to be a class, to pack with a given
1282
mutter('Packing repository %s, which has %d pack files, '
1283
'containing %d revisions into 1 packs.', self, total_packs,
1285
# determine which packs need changing
1286
pack_distribution = [1]
1287
pack_operations = [[0, []]]
1288
for pack in self.all_packs():
1289
pack_operations[-1][0] += pack.get_revision_count()
1290
pack_operations[-1][1].append(pack)
1291
self._execute_pack_operations(pack_operations, OptimisingPacker)
1293
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1294
"""Plan a pack operation.
1296
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1298
:param pack_distribution: A list with the number of revisions desired
1301
if len(existing_packs) <= len(pack_distribution):
1303
existing_packs.sort(reverse=True)
1304
pack_operations = [[0, []]]
1305
# plan out what packs to keep, and what to reorganise
1306
while len(existing_packs):
1307
# take the largest pack, and if its less than the head of the
1308
# distribution chart we will include its contents in the new pack
1309
# for that position. If its larger, we remove its size from the
1310
# distribution chart
1311
next_pack_rev_count, next_pack = existing_packs.pop(0)
1312
if next_pack_rev_count >= pack_distribution[0]:
1313
# this is already packed 'better' than this, so we can
1314
# not waste time packing it.
1315
while next_pack_rev_count > 0:
1316
next_pack_rev_count -= pack_distribution[0]
1317
if next_pack_rev_count >= 0:
1319
del pack_distribution[0]
1321
# didn't use that entire bucket up
1322
pack_distribution[0] = -next_pack_rev_count
1324
# add the revisions we're going to add to the next output pack
1325
pack_operations[-1][0] += next_pack_rev_count
1326
# allocate this pack to the next pack sub operation
1327
pack_operations[-1][1].append(next_pack)
1328
if pack_operations[-1][0] >= pack_distribution[0]:
1329
# this pack is used up, shift left.
1330
del pack_distribution[0]
1331
pack_operations.append([0, []])
1332
# Now that we know which pack files we want to move, shove them all
1333
# into a single pack file.
1335
final_pack_list = []
1336
for num_revs, pack_files in pack_operations:
1337
final_rev_count += num_revs
1338
final_pack_list.extend(pack_files)
1339
if len(final_pack_list) == 1:
1340
raise AssertionError('We somehow generated an autopack with a'
1341
' single pack file being moved.')
1343
return [[final_rev_count, final_pack_list]]
1345
def ensure_loaded(self):
1346
# NB: if you see an assertion error here, its probably access against
1347
# an unlocked repo. Naughty.
1348
if not self.repo.is_locked():
1349
raise errors.ObjectNotLocked(self.repo)
1350
if self._names is None:
1352
self._packs_at_load = set()
1353
for index, key, value in self._iter_disk_pack_index():
1355
self._names[name] = self._parse_index_sizes(value)
1356
self._packs_at_load.add((key, value))
1357
# populate all the metadata.
1360
def _parse_index_sizes(self, value):
1361
"""Parse a string of index sizes."""
1362
return tuple([int(digits) for digits in value.split(' ')])
1364
def get_pack_by_name(self, name):
1365
"""Get a Pack object by name.
1367
:param name: The name of the pack - e.g. '123456'
1368
:return: A Pack object.
1371
return self._packs_by_name[name]
1373
rev_index = self._make_index(name, '.rix')
1374
inv_index = self._make_index(name, '.iix')
1375
txt_index = self._make_index(name, '.tix')
1376
sig_index = self._make_index(name, '.six')
1377
result = ExistingPack(self._pack_transport, name, rev_index,
1378
inv_index, txt_index, sig_index)
1379
self.add_pack_to_memory(result)
1382
def allocate(self, a_new_pack):
1383
"""Allocate name in the list of packs.
1385
:param a_new_pack: A NewPack instance to be added to the collection of
1386
packs for this repository.
1388
self.ensure_loaded()
1389
if a_new_pack.name in self._names:
1390
raise errors.BzrError(
1391
'Pack %r already exists in %s' % (a_new_pack.name, self))
1392
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1393
self.add_pack_to_memory(a_new_pack)
1395
def _iter_disk_pack_index(self):
1396
"""Iterate over the contents of the pack-names index.
1398
This is used when loading the list from disk, and before writing to
1399
detect updates from others during our write operation.
1400
:return: An iterator of the index contents.
1402
return self._index_class(self.transport, 'pack-names', None
1403
).iter_all_entries()
1405
def _make_index(self, name, suffix):
1406
size_offset = self._suffix_offsets[suffix]
1407
index_name = name + suffix
1408
index_size = self._names[name][size_offset]
1409
return self._index_class(
1410
self._index_transport, index_name, index_size)
1412
def _max_pack_count(self, total_revisions):
1413
"""Return the maximum number of packs to use for total revisions.
1415
:param total_revisions: The total number of revisions in the
1418
if not total_revisions:
1420
digits = str(total_revisions)
1422
for digit in digits:
1423
result += int(digit)
1427
"""Provide an order to the underlying names."""
1428
return sorted(self._names.keys())
1430
def _obsolete_packs(self, packs):
1431
"""Move a number of packs which have been obsoleted out of the way.
1433
Each pack and its associated indices are moved out of the way.
1435
Note: for correctness this function should only be called after a new
1436
pack names index has been written without these pack names, and with
1437
the names of packs that contain the data previously available via these
1440
:param packs: The packs to obsolete.
1441
:param return: None.
1444
pack.pack_transport.rename(pack.file_name(),
1445
'../obsolete_packs/' + pack.file_name())
1446
# TODO: Probably needs to know all possible indices for this pack
1447
# - or maybe list the directory and move all indices matching this
1448
# name whether we recognize it or not?
1449
for suffix in ('.iix', '.six', '.tix', '.rix'):
1450
self._index_transport.rename(pack.name + suffix,
1451
'../obsolete_packs/' + pack.name + suffix)
1453
def pack_distribution(self, total_revisions):
1454
"""Generate a list of the number of revisions to put in each pack.
1456
:param total_revisions: The total number of revisions in the
1459
if total_revisions == 0:
1461
digits = reversed(str(total_revisions))
1463
for exponent, count in enumerate(digits):
1464
size = 10 ** exponent
1465
for pos in range(int(count)):
1467
return list(reversed(result))
1469
def _pack_tuple(self, name):
1470
"""Return a tuple with the transport and file name for a pack name."""
1471
return self._pack_transport, name + '.pack'
1473
def _remove_pack_from_memory(self, pack):
1474
"""Remove pack from the packs accessed by this repository.
1476
Only affects memory state, until self._save_pack_names() is invoked.
1478
self._names.pop(pack.name)
1479
self._packs_by_name.pop(pack.name)
1480
self._remove_pack_indices(pack)
1481
self.packs.remove(pack)
1483
def _remove_pack_indices(self, pack):
1484
"""Remove the indices for pack from the aggregated indices."""
1485
self.revision_index.remove_index(pack.revision_index, pack)
1486
self.inventory_index.remove_index(pack.inventory_index, pack)
1487
self.text_index.remove_index(pack.text_index, pack)
1488
self.signature_index.remove_index(pack.signature_index, pack)
1491
"""Clear all cached data."""
1492
# cached revision data
1493
self.repo._revision_knit = None
1494
self.revision_index.clear()
1495
# cached signature data
1496
self.repo._signature_knit = None
1497
self.signature_index.clear()
1498
# cached file text data
1499
self.text_index.clear()
1500
self.repo._text_knit = None
1501
# cached inventory data
1502
self.inventory_index.clear()
1503
# remove the open pack
1504
self._new_pack = None
1505
# information about packs.
1508
self._packs_by_name = {}
1509
self._packs_at_load = None
1511
def _make_index_map(self, index_suffix):
1512
"""Return information on existing indices.
1514
:param suffix: Index suffix added to pack name.
1516
:returns: (pack_map, indices) where indices is a list of GraphIndex
1517
objects, and pack_map is a mapping from those objects to the
1518
pack tuple they describe.
1520
# TODO: stop using this; it creates new indices unnecessarily.
1521
self.ensure_loaded()
1522
suffix_map = {'.rix': 'revision_index',
1523
'.six': 'signature_index',
1524
'.iix': 'inventory_index',
1525
'.tix': 'text_index',
1527
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1528
suffix_map[index_suffix])
1530
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1531
"""Convert a list of packs to an index pack map and index list.
1533
:param packs: The packs list to process.
1534
:param index_attribute: The attribute that the desired index is found
1536
:return: A tuple (map, list) where map contains the dict from
1537
index:pack_tuple, and lsit contains the indices in the same order
1543
index = getattr(pack, index_attribute)
1544
indices.append(index)
1545
pack_map[index] = (pack.pack_transport, pack.file_name())
1546
return pack_map, indices
1548
def _index_contents(self, pack_map, key_filter=None):
1549
"""Get an iterable of the index contents from a pack_map.
1551
:param pack_map: A map from indices to pack details.
1552
:param key_filter: An optional filter to limit the
1555
indices = [index for index in pack_map.iterkeys()]
1556
all_index = CombinedGraphIndex(indices)
1557
if key_filter is None:
1558
return all_index.iter_all_entries()
1560
return all_index.iter_entries(key_filter)
1562
def _unlock_names(self):
1563
"""Release the mutex around the pack-names index."""
1564
self.repo.control_files.unlock()
1566
def _diff_pack_names(self):
1567
"""Read the pack names from disk, and compare it to the one in memory.
1569
:return: (disk_nodes, deleted_nodes, new_nodes)
1570
disk_nodes The final set of nodes that should be referenced
1571
deleted_nodes Nodes which have been removed from when we started
1572
new_nodes Nodes that are newly introduced
1574
# load the disk nodes across
1576
for index, key, value in self._iter_disk_pack_index():
1577
disk_nodes.add((key, value))
1579
# do a two-way diff against our original content
1580
current_nodes = set()
1581
for name, sizes in self._names.iteritems():
1583
((name, ), ' '.join(str(size) for size in sizes)))
1585
# Packs no longer present in the repository, which were present when we
1586
# locked the repository
1587
deleted_nodes = self._packs_at_load - current_nodes
1588
# Packs which this process is adding
1589
new_nodes = current_nodes - self._packs_at_load
1591
# Update the disk_nodes set to include the ones we are adding, and
1592
# remove the ones which were removed by someone else
1593
disk_nodes.difference_update(deleted_nodes)
1594
disk_nodes.update(new_nodes)
1596
return disk_nodes, deleted_nodes, new_nodes
1598
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1599
"""Given the correct set of pack files, update our saved info.
1601
:return: (removed, added, modified)
1602
removed pack names removed from self._names
1603
added pack names added to self._names
1604
modified pack names that had changed value
1609
## self._packs_at_load = disk_nodes
1610
new_names = dict(disk_nodes)
1611
# drop no longer present nodes
1612
for pack in self.all_packs():
1613
if (pack.name,) not in new_names:
1614
removed.append(pack.name)
1615
self._remove_pack_from_memory(pack)
1616
# add new nodes/refresh existing ones
1617
for key, value in disk_nodes:
1619
sizes = self._parse_index_sizes(value)
1620
if name in self._names:
1622
if sizes != self._names[name]:
1623
# the pack for name has had its indices replaced - rare but
1624
# important to handle. XXX: probably can never happen today
1625
# because the three-way merge code above does not handle it
1626
# - you may end up adding the same key twice to the new
1627
# disk index because the set values are the same, unless
1628
# the only index shows up as deleted by the set difference
1629
# - which it may. Until there is a specific test for this,
1630
# assume its broken. RBC 20071017.
1631
self._remove_pack_from_memory(self.get_pack_by_name(name))
1632
self._names[name] = sizes
1633
self.get_pack_by_name(name)
1634
modified.append(name)
1637
self._names[name] = sizes
1638
self.get_pack_by_name(name)
1640
return removed, added, modified
1642
def _save_pack_names(self, clear_obsolete_packs=False):
1643
"""Save the list of packs.
1645
This will take out the mutex around the pack names list for the
1646
duration of the method call. If concurrent updates have been made, a
1647
three-way merge between the current list and the current in memory list
1650
:param clear_obsolete_packs: If True, clear out the contents of the
1651
obsolete_packs directory.
1655
builder = self._index_builder_class()
1656
disk_nodes, deleted_nodes, new_nodes = self._diff_pack_names()
1657
# TODO: handle same-name, index-size-changes here -
1658
# e.g. use the value from disk, not ours, *unless* we're the one
1660
for key, value in disk_nodes:
1661
builder.add_node(key, value)
1662
self.transport.put_file('pack-names', builder.finish(),
1663
mode=self.repo.bzrdir._get_file_mode())
1664
# move the baseline forward
1665
self._packs_at_load = disk_nodes
1666
if clear_obsolete_packs:
1667
self._clear_obsolete_packs()
1669
self._unlock_names()
1670
# synchronise the memory packs list with what we just wrote:
1671
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1673
def reload_pack_names(self):
1674
"""Sync our pack listing with what is present in the repository.
1676
This should be called when we find out that something we thought was
1677
present is now missing. This happens when another process re-packs the
1680
# This is functionally similar to _save_pack_names, but we don't write
1681
# out the new value.
1682
disk_nodes, _, _ = self._diff_pack_names()
1683
self._packs_at_load = disk_nodes
1685
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1686
if removed or added or modified:
1690
def _clear_obsolete_packs(self):
1691
"""Delete everything from the obsolete-packs directory.
1693
obsolete_pack_transport = self.transport.clone('obsolete_packs')
1694
for filename in obsolete_pack_transport.list_dir('.'):
1696
obsolete_pack_transport.delete(filename)
1697
except (errors.PathError, errors.TransportError), e:
1698
warning("couldn't delete obsolete pack, skipping it:\n%s" % (e,))
1700
def _start_write_group(self):
1701
# Do not permit preparation for writing if we're not in a 'write lock'.
1702
if not self.repo.is_write_locked():
1703
raise errors.NotWriteLocked(self)
1704
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1705
self._pack_transport, upload_suffix='.pack',
1706
file_mode=self.repo.bzrdir._get_file_mode(),
1707
index_builder_class=self._index_builder_class,
1708
index_class=self._index_class)
1709
# allow writing: queue writes to a new index
1710
self.revision_index.add_writable_index(self._new_pack.revision_index,
1712
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
1714
self.text_index.add_writable_index(self._new_pack.text_index,
1716
self.signature_index.add_writable_index(self._new_pack.signature_index,
1719
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
1720
self.repo.revisions._index._add_callback = self.revision_index.add_callback
1721
self.repo.signatures._index._add_callback = self.signature_index.add_callback
1722
self.repo.texts._index._add_callback = self.text_index.add_callback
1724
def _abort_write_group(self):
1725
# FIXME: just drop the transient index.
1726
# forget what names there are
1727
if self._new_pack is not None:
1728
self._new_pack.abort()
1729
self._remove_pack_indices(self._new_pack)
1730
self._new_pack = None
1731
self.repo._text_knit = None
1733
def _commit_write_group(self):
1734
self._remove_pack_indices(self._new_pack)
1735
if self._new_pack.data_inserted():
1736
# get all the data to disk and read to use
1737
self._new_pack.finish()
1738
self.allocate(self._new_pack)
1739
self._new_pack = None
1740
if not self.autopack():
1741
# when autopack takes no steps, the names list is still
1743
self._save_pack_names()
1745
self._new_pack.abort()
1746
self._new_pack = None
1747
self.repo._text_knit = None
1749
def _check_references_present(self, external_refs):
1750
index = self.text_index.combined_index
1751
found_items = list(index.iter_entries(external_refs))
1752
if len(found_items) != len(external_refs):
1753
found_keys = set(k for idx, k, refs, value in found_items)
1754
missing_items = external_refs - found_keys
1755
missing_file_id, missing_revision_id = missing_items.pop()
1756
raise errors.RevisionNotPresent(missing_revision_id,
1760
class KnitPackRepository(KnitRepository):
1761
"""Repository with knit objects stored inside pack containers.
1763
The layering for a KnitPackRepository is:
1765
Graph | HPSS | Repository public layer |
1766
===================================================
1767
Tuple based apis below, string based, and key based apis above
1768
---------------------------------------------------
1770
Provides .texts, .revisions etc
1771
This adapts the N-tuple keys to physical knit records which only have a
1772
single string identifier (for historical reasons), which in older formats
1773
was always the revision_id, and in the mapped code for packs is always
1774
the last element of key tuples.
1775
---------------------------------------------------
1777
A separate GraphIndex is used for each of the
1778
texts/inventories/revisions/signatures contained within each individual
1779
pack file. The GraphIndex layer works in N-tuples and is unaware of any
1781
===================================================
1785
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
1787
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
1788
_commit_builder_class, _serializer)
1789
index_transport = self._transport.clone('indices')
1790
self._pack_collection = RepositoryPackCollection(self, self._transport,
1792
self._transport.clone('upload'),
1793
self._transport.clone('packs'),
1794
_format.index_builder_class,
1795
_format.index_class)
1796
self.inventories = KnitVersionedFiles(
1797
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
1798
add_callback=self._pack_collection.inventory_index.add_callback,
1799
deltas=True, parents=True, is_locked=self.is_locked),
1800
data_access=self._pack_collection.inventory_index.data_access,
1801
max_delta_chain=200)
1802
self.revisions = KnitVersionedFiles(
1803
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
1804
add_callback=self._pack_collection.revision_index.add_callback,
1805
deltas=False, parents=True, is_locked=self.is_locked),
1806
data_access=self._pack_collection.revision_index.data_access,
1808
self.signatures = KnitVersionedFiles(
1809
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
1810
add_callback=self._pack_collection.signature_index.add_callback,
1811
deltas=False, parents=False, is_locked=self.is_locked),
1812
data_access=self._pack_collection.signature_index.data_access,
1814
self.texts = KnitVersionedFiles(
1815
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
1816
add_callback=self._pack_collection.text_index.add_callback,
1817
deltas=True, parents=True, is_locked=self.is_locked),
1818
data_access=self._pack_collection.text_index.data_access,
1819
max_delta_chain=200)
1820
# True when the repository object is 'write locked' (as opposed to the
1821
# physical lock only taken out around changes to the pack-names list.)
1822
# Another way to represent this would be a decorator around the control
1823
# files object that presents logical locks as physical ones - if this
1824
# gets ugly consider that alternative design. RBC 20071011
1825
self._write_lock_count = 0
1826
self._transaction = None
1828
self._reconcile_does_inventory_gc = True
1829
self._reconcile_fixes_text_parents = True
1830
self._reconcile_backsup_inventory = False
1831
self._fetch_order = 'unordered'
1833
def _warn_if_deprecated(self):
1834
# This class isn't deprecated, but one sub-format is
1835
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
1836
from bzrlib import repository
1837
if repository._deprecation_warning_done:
1839
repository._deprecation_warning_done = True
1840
warning("Format %s for %s is deprecated - please use"
1841
" 'bzr upgrade --1.6.1-rich-root'"
1842
% (self._format, self.bzrdir.transport.base))
1844
def _abort_write_group(self):
1845
self._pack_collection._abort_write_group()
1847
def _find_inconsistent_revision_parents(self):
1848
"""Find revisions with incorrectly cached parents.
1850
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1851
parents-in-revision).
1853
if not self.is_locked():
1854
raise errors.ObjectNotLocked(self)
1855
pb = ui.ui_factory.nested_progress_bar()
1858
revision_nodes = self._pack_collection.revision_index \
1859
.combined_index.iter_all_entries()
1860
index_positions = []
1861
# Get the cached index values for all revisions, and also the location
1862
# in each index of the revision text so we can perform linear IO.
1863
for index, key, value, refs in revision_nodes:
1864
pos, length = value[1:].split(' ')
1865
index_positions.append((index, int(pos), key[0],
1866
tuple(parent[0] for parent in refs[0])))
1867
pb.update("Reading revision index.", 0, 0)
1868
index_positions.sort()
1869
batch_count = len(index_positions) / 1000 + 1
1870
pb.update("Checking cached revision graph.", 0, batch_count)
1871
for offset in xrange(batch_count):
1872
pb.update("Checking cached revision graph.", offset)
1873
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1876
rev_ids = [item[2] for item in to_query]
1877
revs = self.get_revisions(rev_ids)
1878
for revision, item in zip(revs, to_query):
1879
index_parents = item[3]
1880
rev_parents = tuple(revision.parent_ids)
1881
if index_parents != rev_parents:
1882
result.append((revision.revision_id, index_parents, rev_parents))
1887
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
1888
def get_parents(self, revision_ids):
1889
"""See graph._StackedParentsProvider.get_parents."""
1890
parent_map = self.get_parent_map(revision_ids)
1891
return [parent_map.get(r, None) for r in revision_ids]
1893
def _make_parents_provider(self):
1894
return graph.CachingParentsProvider(self)
1896
def _refresh_data(self):
1897
if self._write_lock_count == 1 or (
1898
self.control_files._lock_count == 1 and
1899
self.control_files._lock_mode == 'r'):
1900
# forget what names there are
1901
self._pack_collection.reset()
1902
# XXX: Better to do an in-memory merge when acquiring a new lock -
1903
# factor out code from _save_pack_names.
1904
self._pack_collection.ensure_loaded()
1906
def _start_write_group(self):
1907
self._pack_collection._start_write_group()
1909
def _commit_write_group(self):
1910
return self._pack_collection._commit_write_group()
1912
def get_transaction(self):
1913
if self._write_lock_count:
1914
return self._transaction
1916
return self.control_files.get_transaction()
1918
def is_locked(self):
1919
return self._write_lock_count or self.control_files.is_locked()
1921
def is_write_locked(self):
1922
return self._write_lock_count
1924
def lock_write(self, token=None):
1925
if not self._write_lock_count and self.is_locked():
1926
raise errors.ReadOnlyError(self)
1927
self._write_lock_count += 1
1928
if self._write_lock_count == 1:
1929
self._transaction = transactions.WriteTransaction()
1930
for repo in self._fallback_repositories:
1931
# Writes don't affect fallback repos
1933
self._refresh_data()
1935
def lock_read(self):
1936
if self._write_lock_count:
1937
self._write_lock_count += 1
1939
self.control_files.lock_read()
1940
for repo in self._fallback_repositories:
1941
# Writes don't affect fallback repos
1943
self._refresh_data()
1945
def leave_lock_in_place(self):
1946
# not supported - raise an error
1947
raise NotImplementedError(self.leave_lock_in_place)
1949
def dont_leave_lock_in_place(self):
1950
# not supported - raise an error
1951
raise NotImplementedError(self.dont_leave_lock_in_place)
1955
"""Compress the data within the repository.
1957
This will pack all the data to a single pack. In future it may
1958
recompress deltas or do other such expensive operations.
1960
self._pack_collection.pack()
1963
def reconcile(self, other=None, thorough=False):
1964
"""Reconcile this repository."""
1965
from bzrlib.reconcile import PackReconciler
1966
reconciler = PackReconciler(self, thorough=thorough)
1967
reconciler.reconcile()
1971
if self._write_lock_count == 1 and self._write_group is not None:
1972
self.abort_write_group()
1973
self._transaction = None
1974
self._write_lock_count = 0
1975
raise errors.BzrError(
1976
'Must end write group before releasing write lock on %s'
1978
if self._write_lock_count:
1979
self._write_lock_count -= 1
1980
if not self._write_lock_count:
1981
transaction = self._transaction
1982
self._transaction = None
1983
transaction.finish()
1984
for repo in self._fallback_repositories:
1987
self.control_files.unlock()
1988
for repo in self._fallback_repositories:
1992
class RepositoryFormatPack(MetaDirRepositoryFormat):
1993
"""Format logic for pack structured repositories.
1995
This repository format has:
1996
- a list of packs in pack-names
1997
- packs in packs/NAME.pack
1998
- indices in indices/NAME.{iix,six,tix,rix}
1999
- knit deltas in the packs, knit indices mapped to the indices.
2000
- thunk objects to support the knits programming API.
2001
- a format marker of its own
2002
- an optional 'shared-storage' flag
2003
- an optional 'no-working-trees' flag
2007
# Set this attribute in derived classes to control the repository class
2008
# created by open and initialize.
2009
repository_class = None
2010
# Set this attribute in derived classes to control the
2011
# _commit_builder_class that the repository objects will have passed to
2012
# their constructor.
2013
_commit_builder_class = None
2014
# Set this attribute in derived clases to control the _serializer that the
2015
# repository objects will have passed to their constructor.
2017
# External references are not supported in pack repositories yet.
2018
supports_external_lookups = False
2019
# What index classes to use
2020
index_builder_class = None
2023
def initialize(self, a_bzrdir, shared=False):
2024
"""Create a pack based repository.
2026
:param a_bzrdir: bzrdir to contain the new repository; must already
2028
:param shared: If true the repository will be initialized as a shared
2031
mutter('creating repository in %s.', a_bzrdir.transport.base)
2032
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
2033
builder = self.index_builder_class()
2034
files = [('pack-names', builder.finish())]
2035
utf8_files = [('format', self.get_format_string())]
2037
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
2038
return self.open(a_bzrdir=a_bzrdir, _found=True)
2040
def open(self, a_bzrdir, _found=False, _override_transport=None):
2041
"""See RepositoryFormat.open().
2043
:param _override_transport: INTERNAL USE ONLY. Allows opening the
2044
repository at a slightly different url
2045
than normal. I.e. during 'upgrade'.
2048
format = RepositoryFormat.find_format(a_bzrdir)
2049
if _override_transport is not None:
2050
repo_transport = _override_transport
2052
repo_transport = a_bzrdir.get_repository_transport(None)
2053
control_files = lockable_files.LockableFiles(repo_transport,
2054
'lock', lockdir.LockDir)
2055
return self.repository_class(_format=self,
2057
control_files=control_files,
2058
_commit_builder_class=self._commit_builder_class,
2059
_serializer=self._serializer)
2062
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2063
"""A no-subtrees parameterized Pack repository.
2065
This format was introduced in 0.92.
2068
repository_class = KnitPackRepository
2069
_commit_builder_class = PackCommitBuilder
2071
def _serializer(self):
2072
return xml5.serializer_v5
2073
# What index classes to use
2074
index_builder_class = InMemoryGraphIndex
2075
index_class = GraphIndex
2077
def _get_matching_bzrdir(self):
2078
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2080
def _ignore_setting_bzrdir(self, format):
2083
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2085
def get_format_string(self):
2086
"""See RepositoryFormat.get_format_string()."""
2087
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2089
def get_format_description(self):
2090
"""See RepositoryFormat.get_format_description()."""
2091
return "Packs containing knits without subtree support"
2093
def check_conversion_target(self, target_format):
2097
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2098
"""A subtrees parameterized Pack repository.
2100
This repository format uses the xml7 serializer to get:
2101
- support for recording full info about the tree root
2102
- support for recording tree-references
2104
This format was introduced in 0.92.
2107
repository_class = KnitPackRepository
2108
_commit_builder_class = PackRootCommitBuilder
2109
rich_root_data = True
2110
supports_tree_reference = True
2112
def _serializer(self):
2113
return xml7.serializer_v7
2114
# What index classes to use
2115
index_builder_class = InMemoryGraphIndex
2116
index_class = GraphIndex
2118
def _get_matching_bzrdir(self):
2119
return bzrdir.format_registry.make_bzrdir(
2120
'pack-0.92-subtree')
2122
def _ignore_setting_bzrdir(self, format):
2125
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2127
def check_conversion_target(self, target_format):
2128
if not target_format.rich_root_data:
2129
raise errors.BadConversionTarget(
2130
'Does not support rich root data.', target_format)
2131
if not getattr(target_format, 'supports_tree_reference', False):
2132
raise errors.BadConversionTarget(
2133
'Does not support nested trees', target_format)
2135
def get_format_string(self):
2136
"""See RepositoryFormat.get_format_string()."""
2137
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2139
def get_format_description(self):
2140
"""See RepositoryFormat.get_format_description()."""
2141
return "Packs containing knits with subtree support\n"
2144
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2145
"""A rich-root, no subtrees parameterized Pack repository.
2147
This repository format uses the xml6 serializer to get:
2148
- support for recording full info about the tree root
2150
This format was introduced in 1.0.
2153
repository_class = KnitPackRepository
2154
_commit_builder_class = PackRootCommitBuilder
2155
rich_root_data = True
2156
supports_tree_reference = False
2158
def _serializer(self):
2159
return xml6.serializer_v6
2160
# What index classes to use
2161
index_builder_class = InMemoryGraphIndex
2162
index_class = GraphIndex
2164
def _get_matching_bzrdir(self):
2165
return bzrdir.format_registry.make_bzrdir(
2168
def _ignore_setting_bzrdir(self, format):
2171
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2173
def check_conversion_target(self, target_format):
2174
if not target_format.rich_root_data:
2175
raise errors.BadConversionTarget(
2176
'Does not support rich root data.', target_format)
2178
def get_format_string(self):
2179
"""See RepositoryFormat.get_format_string()."""
2180
return ("Bazaar pack repository format 1 with rich root"
2181
" (needs bzr 1.0)\n")
2183
def get_format_description(self):
2184
"""See RepositoryFormat.get_format_description()."""
2185
return "Packs containing knits with rich root support\n"
2188
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2189
"""Repository that supports external references to allow stacking.
2193
Supports external lookups, which results in non-truncated ghosts after
2194
reconcile compared to pack-0.92 formats.
2197
repository_class = KnitPackRepository
2198
_commit_builder_class = PackCommitBuilder
2199
supports_external_lookups = True
2200
# What index classes to use
2201
index_builder_class = InMemoryGraphIndex
2202
index_class = GraphIndex
2205
def _serializer(self):
2206
return xml5.serializer_v5
2208
def _get_matching_bzrdir(self):
2209
return bzrdir.format_registry.make_bzrdir('1.6')
2211
def _ignore_setting_bzrdir(self, format):
2214
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2216
def get_format_string(self):
2217
"""See RepositoryFormat.get_format_string()."""
2218
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2220
def get_format_description(self):
2221
"""See RepositoryFormat.get_format_description()."""
2222
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2224
def check_conversion_target(self, target_format):
2228
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2229
"""A repository with rich roots and stacking.
2231
New in release 1.6.1.
2233
Supports stacking on other repositories, allowing data to be accessed
2234
without being stored locally.
2237
repository_class = KnitPackRepository
2238
_commit_builder_class = PackRootCommitBuilder
2239
rich_root_data = True
2240
supports_tree_reference = False # no subtrees
2241
supports_external_lookups = True
2242
# What index classes to use
2243
index_builder_class = InMemoryGraphIndex
2244
index_class = GraphIndex
2247
def _serializer(self):
2248
return xml6.serializer_v6
2250
def _get_matching_bzrdir(self):
2251
return bzrdir.format_registry.make_bzrdir(
2254
def _ignore_setting_bzrdir(self, format):
2257
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2259
def check_conversion_target(self, target_format):
2260
if not target_format.rich_root_data:
2261
raise errors.BadConversionTarget(
2262
'Does not support rich root data.', target_format)
2264
def get_format_string(self):
2265
"""See RepositoryFormat.get_format_string()."""
2266
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2268
def get_format_description(self):
2269
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2272
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2273
"""A repository with rich roots and external references.
2277
Supports external lookups, which results in non-truncated ghosts after
2278
reconcile compared to pack-0.92 formats.
2280
This format was deprecated because the serializer it uses accidentally
2281
supported subtrees, when the format was not intended to. This meant that
2282
someone could accidentally fetch from an incorrect repository.
2285
repository_class = KnitPackRepository
2286
_commit_builder_class = PackRootCommitBuilder
2287
rich_root_data = True
2288
supports_tree_reference = False # no subtrees
2290
supports_external_lookups = True
2291
# What index classes to use
2292
index_builder_class = InMemoryGraphIndex
2293
index_class = GraphIndex
2296
def _serializer(self):
2297
return xml7.serializer_v7
2299
def _get_matching_bzrdir(self):
2300
return bzrdir.format_registry.make_bzrdir(
2303
def _ignore_setting_bzrdir(self, format):
2306
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2308
def check_conversion_target(self, target_format):
2309
if not target_format.rich_root_data:
2310
raise errors.BadConversionTarget(
2311
'Does not support rich root data.', target_format)
2313
def get_format_string(self):
2314
"""See RepositoryFormat.get_format_string()."""
2315
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2317
def get_format_description(self):
2318
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2322
class RepositoryFormatPackDevelopment2(RepositoryFormatPack):
2323
"""A no-subtrees development repository.
2325
This format should be retained until the second release after bzr 1.7.
2327
This is pack-1.6.1 with B+Tree indices.
2330
repository_class = KnitPackRepository
2331
_commit_builder_class = PackCommitBuilder
2332
supports_external_lookups = True
2333
# What index classes to use
2334
index_builder_class = BTreeBuilder
2335
index_class = BTreeGraphIndex
2338
def _serializer(self):
2339
return xml5.serializer_v5
2341
def _get_matching_bzrdir(self):
2342
return bzrdir.format_registry.make_bzrdir('development2')
2344
def _ignore_setting_bzrdir(self, format):
2347
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2349
def get_format_string(self):
2350
"""See RepositoryFormat.get_format_string()."""
2351
return "Bazaar development format 2 (needs bzr.dev from before 1.8)\n"
2353
def get_format_description(self):
2354
"""See RepositoryFormat.get_format_description()."""
2355
return ("Development repository format, currently the same as "
2356
"1.6.1 with B+Trees.\n")
2358
def check_conversion_target(self, target_format):
2362
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2363
"""A subtrees development repository.
2365
This format should be retained until the second release after bzr 1.7.
2367
1.6.1-subtree[as it might have been] with B+Tree indices.
2370
repository_class = KnitPackRepository
2371
_commit_builder_class = PackRootCommitBuilder
2372
rich_root_data = True
2373
supports_tree_reference = True
2374
supports_external_lookups = True
2375
# What index classes to use
2376
index_builder_class = BTreeBuilder
2377
index_class = BTreeGraphIndex
2380
def _serializer(self):
2381
return xml7.serializer_v7
2383
def _get_matching_bzrdir(self):
2384
return bzrdir.format_registry.make_bzrdir(
2385
'development2-subtree')
2387
def _ignore_setting_bzrdir(self, format):
2390
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2392
def check_conversion_target(self, target_format):
2393
if not target_format.rich_root_data:
2394
raise errors.BadConversionTarget(
2395
'Does not support rich root data.', target_format)
2396
if not getattr(target_format, 'supports_tree_reference', False):
2397
raise errors.BadConversionTarget(
2398
'Does not support nested trees', target_format)
2400
def get_format_string(self):
2401
"""See RepositoryFormat.get_format_string()."""
2402
return ("Bazaar development format 2 with subtree support "
2403
"(needs bzr.dev from before 1.8)\n")
2405
def get_format_description(self):
2406
"""See RepositoryFormat.get_format_description()."""
2407
return ("Development repository format, currently the same as "
2408
"1.6.1-subtree with B+Tree indices.\n")
added
removed
bzrlib/repofmt/pack_repo.py
