1
# Copyright (C) 2005, 2006, 2007 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
30
from bzrlib.index import (
35
GraphIndexPrefixAdapter,
37
from bzrlib.knit import KnitGraphIndex, _PackAccess, _KnitData
38
from bzrlib.osutils import rand_chars
39
from bzrlib.pack import ContainerWriter
40
from bzrlib.store import revision
41
from bzrlib import tsort
58
from bzrlib.decorators import needs_read_lock, needs_write_lock
59
from bzrlib.repofmt.knitrepo import KnitRepository
60
from bzrlib.repository import (
63
MetaDirRepositoryFormat,
66
import bzrlib.revision as _mod_revision
67
from bzrlib.store.revision.knit import KnitRevisionStore
68
from bzrlib.store.versioned import VersionedFileStore
69
from bzrlib.trace import mutter, note, warning
72
class PackCommitBuilder(CommitBuilder):
73
"""A subclass of CommitBuilder to add texts with pack semantics.
75
Specifically this uses one knit object rather than one knit object per
76
added text, reducing memory and object pressure.
79
def __init__(self, repository, parents, config, timestamp=None,
80
timezone=None, committer=None, revprops=None,
82
CommitBuilder.__init__(self, repository, parents, config,
83
timestamp=timestamp, timezone=timezone, committer=committer,
84
revprops=revprops, revision_id=revision_id)
85
self._file_graph = graph.Graph(
86
repository._pack_collection.text_index.combined_index)
88
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
89
return self.repository._pack_collection._add_text_to_weave(file_id,
90
self._new_revision_id, new_lines, parents, nostore_sha,
93
def _heads(self, file_id, revision_ids):
94
keys = [(file_id, revision_id) for revision_id in revision_ids]
95
return set([key[1] for key in self._file_graph.heads(keys)])
98
class PackRootCommitBuilder(RootCommitBuilder):
99
"""A subclass of RootCommitBuilder to add texts with pack semantics.
101
Specifically this uses one knit object rather than one knit object per
102
added text, reducing memory and object pressure.
105
def __init__(self, repository, parents, config, timestamp=None,
106
timezone=None, committer=None, revprops=None,
108
CommitBuilder.__init__(self, repository, parents, config,
109
timestamp=timestamp, timezone=timezone, committer=committer,
110
revprops=revprops, revision_id=revision_id)
111
self._file_graph = graph.Graph(
112
repository._pack_collection.text_index.combined_index)
114
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
115
return self.repository._pack_collection._add_text_to_weave(file_id,
116
self._new_revision_id, new_lines, parents, nostore_sha,
119
def _heads(self, file_id, revision_ids):
120
keys = [(file_id, revision_id) for revision_id in revision_ids]
121
return set([key[1] for key in self._file_graph.heads(keys)])
125
"""An in memory proxy for a pack and its indices.
127
This is a base class that is not directly used, instead the classes
128
ExistingPack and NewPack are used.
131
def __init__(self, revision_index, inventory_index, text_index,
133
"""Create a pack instance.
135
:param revision_index: A GraphIndex for determining what revisions are
136
present in the Pack and accessing the locations of their texts.
137
:param inventory_index: A GraphIndex for determining what inventories are
138
present in the Pack and accessing the locations of their
140
:param text_index: A GraphIndex for determining what file texts
141
are present in the pack and accessing the locations of their
142
texts/deltas (via (fileid, revisionid) tuples).
143
:param revision_index: A GraphIndex for determining what signatures are
144
present in the Pack and accessing the locations of their texts.
146
self.revision_index = revision_index
147
self.inventory_index = inventory_index
148
self.text_index = text_index
149
self.signature_index = signature_index
151
def access_tuple(self):
152
"""Return a tuple (transport, name) for the pack content."""
153
return self.pack_transport, self.file_name()
156
"""Get the file name for the pack on disk."""
157
return self.name + '.pack'
159
def get_revision_count(self):
160
return self.revision_index.key_count()
162
def inventory_index_name(self, name):
163
"""The inv index is the name + .iix."""
164
return self.index_name('inventory', name)
166
def revision_index_name(self, name):
167
"""The revision index is the name + .rix."""
168
return self.index_name('revision', name)
170
def signature_index_name(self, name):
171
"""The signature index is the name + .six."""
172
return self.index_name('signature', name)
174
def text_index_name(self, name):
175
"""The text index is the name + .tix."""
176
return self.index_name('text', name)
178
def _external_compression_parents_of_texts(self):
181
for node in self.text_index.iter_all_entries():
183
refs.update(node[3][1])
187
class ExistingPack(Pack):
188
"""An in memory proxy for an existing .pack and its disk indices."""
190
def __init__(self, pack_transport, name, revision_index, inventory_index,
191
text_index, signature_index):
192
"""Create an ExistingPack object.
194
:param pack_transport: The transport where the pack file resides.
195
:param name: The name of the pack on disk in the pack_transport.
197
Pack.__init__(self, revision_index, inventory_index, text_index,
200
self.pack_transport = pack_transport
201
assert None not in (revision_index, inventory_index, text_index,
202
signature_index, name, pack_transport)
204
def __eq__(self, other):
205
return self.__dict__ == other.__dict__
207
def __ne__(self, other):
208
return not self.__eq__(other)
211
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
212
id(self), self.transport, self.name)
216
"""An in memory proxy for a pack which is being created."""
218
# A map of index 'type' to the file extension and position in the
220
index_definitions = {
221
'revision': ('.rix', 0),
222
'inventory': ('.iix', 1),
224
'signature': ('.six', 3),
227
def __init__(self, upload_transport, index_transport, pack_transport,
228
upload_suffix='', file_mode=None):
229
"""Create a NewPack instance.
231
:param upload_transport: A writable transport for the pack to be
232
incrementally uploaded to.
233
:param index_transport: A writable transport for the pack's indices to
234
be written to when the pack is finished.
235
:param pack_transport: A writable transport for the pack to be renamed
236
to when the upload is complete. This *must* be the same as
237
upload_transport.clone('../packs').
238
:param upload_suffix: An optional suffix to be given to any temporary
239
files created during the pack creation. e.g '.autopack'
240
:param file_mode: An optional file mode to create the new files with.
242
# The relative locations of the packs are constrained, but all are
243
# passed in because the caller has them, so as to avoid object churn.
245
# Revisions: parents list, no text compression.
246
InMemoryGraphIndex(reference_lists=1),
247
# Inventory: We want to map compression only, but currently the
248
# knit code hasn't been updated enough to understand that, so we
249
# have a regular 2-list index giving parents and compression
251
InMemoryGraphIndex(reference_lists=2),
252
# Texts: compression and per file graph, for all fileids - so two
253
# reference lists and two elements in the key tuple.
254
InMemoryGraphIndex(reference_lists=2, key_elements=2),
255
# Signatures: Just blobs to store, no compression, no parents
257
InMemoryGraphIndex(reference_lists=0),
259
# where should the new pack be opened
260
self.upload_transport = upload_transport
261
# where are indices written out to
262
self.index_transport = index_transport
263
# where is the pack renamed to when it is finished?
264
self.pack_transport = pack_transport
265
# What file mode to upload the pack and indices with.
266
self._file_mode = file_mode
267
# tracks the content written to the .pack file.
268
self._hash = md5.new()
269
# a four-tuple with the length in bytes of the indices, once the pack
270
# is finalised. (rev, inv, text, sigs)
271
self.index_sizes = None
272
# How much data to cache when writing packs. Note that this is not
273
# synchronised with reads, because it's not in the transport layer, so
274
# is not safe unless the client knows it won't be reading from the pack
276
self._cache_limit = 0
277
# the temporary pack file name.
278
self.random_name = rand_chars(20) + upload_suffix
279
# when was this pack started ?
280
self.start_time = time.time()
281
# open an output stream for the data added to the pack.
282
self.write_stream = self.upload_transport.open_write_stream(
283
self.random_name, mode=self._file_mode)
284
if 'pack' in debug.debug_flags:
285
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
286
time.ctime(), self.upload_transport.base, self.random_name,
287
time.time() - self.start_time)
288
# A list of byte sequences to be written to the new pack, and the
289
# aggregate size of them. Stored as a list rather than separate
290
# variables so that the _write_data closure below can update them.
291
self._buffer = [[], 0]
292
# create a callable for adding data
294
# robertc says- this is a closure rather than a method on the object
295
# so that the variables are locals, and faster than accessing object
297
def _write_data(bytes, flush=False, _buffer=self._buffer,
298
_write=self.write_stream.write, _update=self._hash.update):
299
_buffer[0].append(bytes)
300
_buffer[1] += len(bytes)
302
if _buffer[1] > self._cache_limit or flush:
303
bytes = ''.join(_buffer[0])
307
# expose this on self, for the occasion when clients want to add data.
308
self._write_data = _write_data
309
# a pack writer object to serialise pack records.
310
self._writer = pack.ContainerWriter(self._write_data)
312
# what state is the pack in? (open, finished, aborted)
316
"""Cancel creating this pack."""
317
self._state = 'aborted'
318
self.write_stream.close()
319
# Remove the temporary pack file.
320
self.upload_transport.delete(self.random_name)
321
# The indices have no state on disk.
323
def access_tuple(self):
324
"""Return a tuple (transport, name) for the pack content."""
325
assert self._state in ('open', 'finished')
326
if self._state == 'finished':
327
return Pack.access_tuple(self)
329
return self.upload_transport, self.random_name
331
def data_inserted(self):
332
"""True if data has been added to this pack."""
333
return bool(self.get_revision_count() or
334
self.inventory_index.key_count() or
335
self.text_index.key_count() or
336
self.signature_index.key_count())
339
"""Finish the new pack.
342
- finalises the content
343
- assigns a name (the md5 of the content, currently)
344
- writes out the associated indices
345
- renames the pack into place.
346
- stores the index size tuple for the pack in the index_sizes
351
self._write_data('', flush=True)
352
self.name = self._hash.hexdigest()
354
# XXX: It'd be better to write them all to temporary names, then
355
# rename them all into place, so that the window when only some are
356
# visible is smaller. On the other hand none will be seen until
357
# they're in the names list.
358
self.index_sizes = [None, None, None, None]
359
self._write_index('revision', self.revision_index, 'revision')
360
self._write_index('inventory', self.inventory_index, 'inventory')
361
self._write_index('text', self.text_index, 'file texts')
362
self._write_index('signature', self.signature_index,
363
'revision signatures')
364
self.write_stream.close()
365
# Note that this will clobber an existing pack with the same name,
366
# without checking for hash collisions. While this is undesirable this
367
# is something that can be rectified in a subsequent release. One way
368
# to rectify it may be to leave the pack at the original name, writing
369
# its pack-names entry as something like 'HASH: index-sizes
370
# temporary-name'. Allocate that and check for collisions, if it is
371
# collision free then rename it into place. If clients know this scheme
372
# they can handle missing-file errors by:
373
# - try for HASH.pack
374
# - try for temporary-name
375
# - refresh the pack-list to see if the pack is now absent
376
self.upload_transport.rename(self.random_name,
377
'../packs/' + self.name + '.pack')
378
self._state = 'finished'
379
if 'pack' in debug.debug_flags:
380
# XXX: size might be interesting?
381
mutter('%s: create_pack: pack renamed into place: %s%s->%s%s t+%6.3fs',
382
time.ctime(), self.upload_transport.base, self.random_name,
383
self.pack_transport, self.name,
384
time.time() - self.start_time)
387
"""Flush any current data."""
389
bytes = ''.join(self._buffer[0])
390
self.write_stream.write(bytes)
391
self._hash.update(bytes)
392
self._buffer[:] = [[], 0]
394
def index_name(self, index_type, name):
395
"""Get the disk name of an index type for pack name 'name'."""
396
return name + NewPack.index_definitions[index_type][0]
398
def index_offset(self, index_type):
399
"""Get the position in a index_size array for a given index type."""
400
return NewPack.index_definitions[index_type][1]
402
def _replace_index_with_readonly(self, index_type):
403
setattr(self, index_type + '_index',
404
GraphIndex(self.index_transport,
405
self.index_name(index_type, self.name),
406
self.index_sizes[self.index_offset(index_type)]))
408
def set_write_cache_size(self, size):
409
self._cache_limit = size
411
def _write_index(self, index_type, index, label):
412
"""Write out an index.
414
:param index_type: The type of index to write - e.g. 'revision'.
415
:param index: The index object to serialise.
416
:param label: What label to give the index e.g. 'revision'.
418
index_name = self.index_name(index_type, self.name)
419
self.index_sizes[self.index_offset(index_type)] = \
420
self.index_transport.put_file(index_name, index.finish(),
421
mode=self._file_mode)
422
if 'pack' in debug.debug_flags:
423
# XXX: size might be interesting?
424
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
425
time.ctime(), label, self.upload_transport.base,
426
self.random_name, time.time() - self.start_time)
427
# Replace the writable index on this object with a readonly,
428
# presently unloaded index. We should alter
429
# the index layer to make its finish() error if add_node is
430
# subsequently used. RBC
431
self._replace_index_with_readonly(index_type)
434
class AggregateIndex(object):
435
"""An aggregated index for the RepositoryPackCollection.
437
AggregateIndex is reponsible for managing the PackAccess object,
438
Index-To-Pack mapping, and all indices list for a specific type of index
439
such as 'revision index'.
441
A CombinedIndex provides an index on a single key space built up
442
from several on-disk indices. The AggregateIndex builds on this
443
to provide a knit access layer, and allows having up to one writable
444
index within the collection.
446
# XXX: Probably 'can be written to' could/should be separated from 'acts
447
# like a knit index' -- mbp 20071024
450
"""Create an AggregateIndex."""
451
self.index_to_pack = {}
452
self.combined_index = CombinedGraphIndex([])
453
self.knit_access = _PackAccess(self.index_to_pack)
455
def replace_indices(self, index_to_pack, indices):
456
"""Replace the current mappings with fresh ones.
458
This should probably not be used eventually, rather incremental add and
459
removal of indices. It has been added during refactoring of existing
462
:param index_to_pack: A mapping from index objects to
463
(transport, name) tuples for the pack file data.
464
:param indices: A list of indices.
466
# refresh the revision pack map dict without replacing the instance.
467
self.index_to_pack.clear()
468
self.index_to_pack.update(index_to_pack)
469
# XXX: API break - clearly a 'replace' method would be good?
470
self.combined_index._indices[:] = indices
471
# the current add nodes callback for the current writable index if
473
self.add_callback = None
475
def add_index(self, index, pack):
476
"""Add index to the aggregate, which is an index for Pack pack.
478
Future searches on the aggregate index will seach this new index
479
before all previously inserted indices.
481
:param index: An Index for the pack.
482
:param pack: A Pack instance.
484
# expose it to the index map
485
self.index_to_pack[index] = pack.access_tuple()
486
# put it at the front of the linear index list
487
self.combined_index.insert_index(0, index)
489
def add_writable_index(self, index, pack):
490
"""Add an index which is able to have data added to it.
492
There can be at most one writable index at any time. Any
493
modifications made to the knit are put into this index.
495
:param index: An index from the pack parameter.
496
:param pack: A Pack instance.
498
assert self.add_callback is None, \
499
"%s already has a writable index through %s" % \
500
(self, self.add_callback)
501
# allow writing: queue writes to a new index
502
self.add_index(index, pack)
503
# Updates the index to packs mapping as a side effect,
504
self.knit_access.set_writer(pack._writer, index, pack.access_tuple())
505
self.add_callback = index.add_nodes
508
"""Reset all the aggregate data to nothing."""
509
self.knit_access.set_writer(None, None, (None, None))
510
self.index_to_pack.clear()
511
del self.combined_index._indices[:]
512
self.add_callback = None
514
def remove_index(self, index, pack):
515
"""Remove index from the indices used to answer queries.
517
:param index: An index from the pack parameter.
518
:param pack: A Pack instance.
520
del self.index_to_pack[index]
521
self.combined_index._indices.remove(index)
522
if (self.add_callback is not None and
523
getattr(index, 'add_nodes', None) == self.add_callback):
524
self.add_callback = None
525
self.knit_access.set_writer(None, None, (None, None))
528
class Packer(object):
529
"""Create a pack from packs."""
531
def __init__(self, pack_collection, packs, suffix, revision_ids=None):
534
:param pack_collection: A RepositoryPackCollection object where the
535
new pack is being written to.
536
:param packs: The packs to combine.
537
:param suffix: The suffix to use on the temporary files for the pack.
538
:param revision_ids: Revision ids to limit the pack to.
542
self.revision_ids = revision_ids
543
# The pack object we are creating.
545
self._pack_collection = pack_collection
546
# The index layer keys for the revisions being copied. None for 'all
548
self._revision_keys = None
549
# What text keys to copy. None for 'all texts'. This is set by
550
# _copy_inventory_texts
551
self._text_filter = None
554
def _extra_init(self):
555
"""A template hook to allow extending the constructor trivially."""
557
def pack(self, pb=None):
558
"""Create a new pack by reading data from other packs.
560
This does little more than a bulk copy of data. One key difference
561
is that data with the same item key across multiple packs is elided
562
from the output. The new pack is written into the current pack store
563
along with its indices, and the name added to the pack names. The
564
source packs are not altered and are not required to be in the current
567
:param pb: An optional progress bar to use. A nested bar is created if
569
:return: A Pack object, or None if nothing was copied.
571
# open a pack - using the same name as the last temporary file
572
# - which has already been flushed, so its safe.
573
# XXX: - duplicate code warning with start_write_group; fix before
574
# considering 'done'.
575
if self._pack_collection._new_pack is not None:
576
raise errors.BzrError('call to create_pack_from_packs while '
577
'another pack is being written.')
578
if self.revision_ids is not None:
579
if len(self.revision_ids) == 0:
580
# silly fetch request.
583
self.revision_ids = frozenset(self.revision_ids)
585
self.pb = ui.ui_factory.nested_progress_bar()
589
return self._create_pack_from_packs()
595
"""Open a pack for the pack we are creating."""
596
return NewPack(self._pack_collection._upload_transport,
597
self._pack_collection._index_transport,
598
self._pack_collection._pack_transport, upload_suffix=self.suffix,
599
file_mode=self._pack_collection.repo.control_files._file_mode)
601
def _copy_revision_texts(self):
602
"""Copy revision data to the new pack."""
604
if self.revision_ids:
605
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
608
# select revision keys
609
revision_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
610
self.packs, 'revision_index')[0]
611
revision_nodes = self._pack_collection._index_contents(revision_index_map, revision_keys)
612
# copy revision keys and adjust values
613
self.pb.update("Copying revision texts", 1)
614
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
615
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
616
self.new_pack.revision_index, readv_group_iter, total_items))
617
if 'pack' in debug.debug_flags:
618
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
619
time.ctime(), self._pack_collection._upload_transport.base,
620
self.new_pack.random_name,
621
self.new_pack.revision_index.key_count(),
622
time.time() - self.new_pack.start_time)
623
self._revision_keys = revision_keys
625
def _copy_inventory_texts(self):
626
"""Copy the inventory texts to the new pack.
628
self._revision_keys is used to determine what inventories to copy.
630
Sets self._text_filter appropriately.
632
# select inventory keys
633
inv_keys = self._revision_keys # currently the same keyspace, and note that
634
# querying for keys here could introduce a bug where an inventory item
635
# is missed, so do not change it to query separately without cross
636
# checking like the text key check below.
637
inventory_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
638
self.packs, 'inventory_index')[0]
639
inv_nodes = self._pack_collection._index_contents(inventory_index_map, inv_keys)
640
# copy inventory keys and adjust values
641
# XXX: Should be a helper function to allow different inv representation
643
self.pb.update("Copying inventory texts", 2)
644
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
645
inv_lines = self._copy_nodes_graph(inventory_index_map,
646
self.new_pack._writer, self.new_pack.inventory_index,
647
readv_group_iter, total_items, output_lines=True)
648
if self.revision_ids:
649
self._process_inventory_lines(inv_lines)
651
# eat the iterator to cause it to execute.
653
self._text_filter = None
654
if 'pack' in debug.debug_flags:
655
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
656
time.ctime(), self._pack_collection._upload_transport.base,
657
self.new_pack.random_name,
658
self.new_pack.inventory_index.key_count(),
659
time.time() - new_pack.start_time)
661
def _copy_text_texts(self):
663
text_index_map, text_nodes = self._get_text_nodes()
664
if self._text_filter is not None:
665
# We could return the keys copied as part of the return value from
666
# _copy_nodes_graph but this doesn't work all that well with the
667
# need to get line output too, so we check separately, and as we're
668
# going to buffer everything anyway, we check beforehand, which
669
# saves reading knit data over the wire when we know there are
671
text_nodes = set(text_nodes)
672
present_text_keys = set(_node[1] for _node in text_nodes)
673
missing_text_keys = set(self._text_filter) - present_text_keys
674
if missing_text_keys:
675
# TODO: raise a specific error that can handle many missing
677
a_missing_key = missing_text_keys.pop()
678
raise errors.RevisionNotPresent(a_missing_key[1],
680
# copy text keys and adjust values
681
self.pb.update("Copying content texts", 3)
682
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
683
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
684
self.new_pack.text_index, readv_group_iter, total_items))
685
self._log_copied_texts()
687
def _check_references(self):
688
"""Make sure our external refereneces are present."""
689
external_refs = self.new_pack._external_compression_parents_of_texts()
691
index = self._pack_collection.text_index.combined_index
692
found_items = list(index.iter_entries(external_refs))
693
if len(found_items) != len(external_refs):
694
found_keys = set(k for idx, k, refs, value in found_items)
695
missing_items = external_refs - found_keys
696
missing_file_id, missing_revision_id = missing_items.pop()
697
raise errors.RevisionNotPresent(missing_revision_id,
700
def _create_pack_from_packs(self):
701
self.pb.update("Opening pack", 0, 5)
702
self.new_pack = self.open_pack()
703
new_pack = self.new_pack
704
# buffer data - we won't be reading-back during the pack creation and
705
# this makes a significant difference on sftp pushes.
706
new_pack.set_write_cache_size(1024*1024)
707
if 'pack' in debug.debug_flags:
708
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
709
for a_pack in self.packs]
710
if self.revision_ids is not None:
711
rev_count = len(self.revision_ids)
714
mutter('%s: create_pack: creating pack from source packs: '
715
'%s%s %s revisions wanted %s t=0',
716
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
717
plain_pack_list, rev_count)
718
self._copy_revision_texts()
719
self._copy_inventory_texts()
720
self._copy_text_texts()
721
# select signature keys
722
signature_filter = self._revision_keys # same keyspace
723
signature_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
724
self.packs, 'signature_index')[0]
725
signature_nodes = self._pack_collection._index_contents(signature_index_map,
727
# copy signature keys and adjust values
728
self.pb.update("Copying signature texts", 4)
729
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
730
new_pack.signature_index)
731
if 'pack' in debug.debug_flags:
732
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
733
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
734
new_pack.signature_index.key_count(),
735
time.time() - new_pack.start_time)
736
self._check_references()
737
if not self._use_pack(new_pack):
740
self.pb.update("Finishing pack", 5)
742
self._pack_collection.allocate(new_pack)
745
def _copy_nodes(self, nodes, index_map, writer, write_index):
746
"""Copy knit nodes between packs with no graph references."""
747
pb = ui.ui_factory.nested_progress_bar()
749
return self._do_copy_nodes(nodes, index_map, writer,
754
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb):
755
# for record verification
756
knit_data = _KnitData(None)
757
# plan a readv on each source pack:
759
nodes = sorted(nodes)
760
# how to map this into knit.py - or knit.py into this?
761
# we don't want the typical knit logic, we want grouping by pack
762
# at this point - perhaps a helper library for the following code
763
# duplication points?
765
for index, key, value in nodes:
766
if index not in request_groups:
767
request_groups[index] = []
768
request_groups[index].append((key, value))
770
pb.update("Copied record", record_index, len(nodes))
771
for index, items in request_groups.iteritems():
772
pack_readv_requests = []
773
for key, value in items:
774
# ---- KnitGraphIndex.get_position
775
bits = value[1:].split(' ')
776
offset, length = int(bits[0]), int(bits[1])
777
pack_readv_requests.append((offset, length, (key, value[0])))
778
# linear scan up the pack
779
pack_readv_requests.sort()
781
transport, path = index_map[index]
782
reader = pack.make_readv_reader(transport, path,
783
[offset[0:2] for offset in pack_readv_requests])
784
for (names, read_func), (_1, _2, (key, eol_flag)) in \
785
izip(reader.iter_records(), pack_readv_requests):
786
raw_data = read_func(None)
787
# check the header only
788
df, _ = knit_data._parse_record_header(key[-1], raw_data)
790
pos, size = writer.add_bytes_record(raw_data, names)
791
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
792
pb.update("Copied record", record_index)
795
def _copy_nodes_graph(self, index_map, writer, write_index,
796
readv_group_iter, total_items, output_lines=False):
797
"""Copy knit nodes between packs.
799
:param output_lines: Return lines present in the copied data as
800
an iterator of line,version_id.
802
pb = ui.ui_factory.nested_progress_bar()
804
for result in self._do_copy_nodes_graph(index_map, writer,
805
write_index, output_lines, pb, readv_group_iter, total_items):
808
# Python 2.4 does not permit try:finally: in a generator.
814
def _do_copy_nodes_graph(self, index_map, writer, write_index,
815
output_lines, pb, readv_group_iter, total_items):
816
# for record verification
817
knit_data = _KnitData(None)
818
# for line extraction when requested (inventories only)
820
factory = knit.KnitPlainFactory()
822
pb.update("Copied record", record_index, total_items)
823
for index, readv_vector, node_vector in readv_group_iter:
825
transport, path = index_map[index]
826
reader = pack.make_readv_reader(transport, path, readv_vector)
827
for (names, read_func), (key, eol_flag, references) in \
828
izip(reader.iter_records(), node_vector):
829
raw_data = read_func(None)
832
# read the entire thing
833
content, _ = knit_data._parse_record(version_id, raw_data)
834
if len(references[-1]) == 0:
835
line_iterator = factory.get_fulltext_content(content)
837
line_iterator = factory.get_linedelta_content(content)
838
for line in line_iterator:
839
yield line, version_id
841
# check the header only
842
df, _ = knit_data._parse_record_header(version_id, raw_data)
844
pos, size = writer.add_bytes_record(raw_data, names)
845
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
846
pb.update("Copied record", record_index)
849
def _get_text_nodes(self):
850
text_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
851
self.packs, 'text_index')[0]
852
return text_index_map, self._pack_collection._index_contents(text_index_map,
855
def _least_readv_node_readv(self, nodes):
856
"""Generate request groups for nodes using the least readv's.
858
:param nodes: An iterable of graph index nodes.
859
:return: Total node count and an iterator of the data needed to perform
860
readvs to obtain the data for nodes. Each item yielded by the
861
iterator is a tuple with:
862
index, readv_vector, node_vector. readv_vector is a list ready to
863
hand to the transport readv method, and node_vector is a list of
864
(key, eol_flag, references) for the the node retrieved by the
865
matching readv_vector.
867
# group by pack so we do one readv per pack
868
nodes = sorted(nodes)
871
for index, key, value, references in nodes:
872
if index not in request_groups:
873
request_groups[index] = []
874
request_groups[index].append((key, value, references))
876
for index, items in request_groups.iteritems():
877
pack_readv_requests = []
878
for key, value, references in items:
879
# ---- KnitGraphIndex.get_position
880
bits = value[1:].split(' ')
881
offset, length = int(bits[0]), int(bits[1])
882
pack_readv_requests.append(
883
((offset, length), (key, value[0], references)))
884
# linear scan up the pack to maximum range combining.
885
pack_readv_requests.sort()
886
# split out the readv and the node data.
887
pack_readv = [readv for readv, node in pack_readv_requests]
888
node_vector = [node for readv, node in pack_readv_requests]
889
result.append((index, pack_readv, node_vector))
892
def _log_copied_texts(self):
893
if 'pack' in debug.debug_flags:
894
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
895
time.ctime(), self._pack_collection._upload_transport.base,
896
self.new_pack.random_name,
897
self.new_pack.text_index.key_count(),
898
time.time() - self.new_pack.start_time)
900
def _process_inventory_lines(self, inv_lines):
901
"""Use up the inv_lines generator and setup a text key filter."""
902
repo = self._pack_collection.repo
903
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
904
inv_lines, self.revision_ids)
906
for fileid, file_revids in fileid_revisions.iteritems():
907
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
908
self._text_filter = text_filter
910
def _revision_node_readv(self, revision_nodes):
911
"""Return the total revisions and the readv's to issue.
913
:param revision_nodes: The revision index contents for the packs being
914
incorporated into the new pack.
915
:return: As per _least_readv_node_readv.
917
return self._least_readv_node_readv(revision_nodes)
919
def _use_pack(self, new_pack):
920
"""Return True if new_pack should be used.
922
:param new_pack: The pack that has just been created.
923
:return: True if the pack should be used.
925
return new_pack.data_inserted()
928
class OptimisingPacker(Packer):
929
"""A packer which spends more time to create better disk layouts."""
931
def _revision_node_readv(self, revision_nodes):
932
"""Return the total revisions and the readv's to issue.
934
This sort places revisions in topological order with the ancestors
937
:param revision_nodes: The revision index contents for the packs being
938
incorporated into the new pack.
939
:return: As per _least_readv_node_readv.
941
# build an ancestors dict
944
for index, key, value, references in revision_nodes:
945
ancestors[key] = references[0]
946
by_key[key] = (index, value, references)
947
order = tsort.topo_sort(ancestors)
949
# Single IO is pathological, but it will work as a starting point.
951
for key in reversed(order):
952
index, value, references = by_key[key]
953
# ---- KnitGraphIndex.get_position
954
bits = value[1:].split(' ')
955
offset, length = int(bits[0]), int(bits[1])
957
(index, [(offset, length)], [(key, value[0], references)]))
958
# TODO: combine requests in the same index that are in ascending order.
959
return total, requests
962
class ReconcilePacker(Packer):
963
"""A packer which regenerates indices etc as it copies.
965
This is used by ``bzr reconcile`` to cause parent text pointers to be
969
def _extra_init(self):
970
self._data_changed = False
972
def _process_inventory_lines(self, inv_lines):
973
"""Generate a text key reference map rather for reconciling with."""
974
repo = self._pack_collection.repo
975
refs = repo._find_text_key_references_from_xml_inventory_lines(
977
self._text_refs = refs
978
# during reconcile we:
979
# - convert unreferenced texts to full texts
980
# - correct texts which reference a text not copied to be full texts
981
# - copy all others as-is but with corrected parents.
982
# - so at this point we don't know enough to decide what becomes a full
984
self._text_filter = None
986
def _copy_text_texts(self):
987
"""generate what texts we should have and then copy."""
988
self.pb.update("Copying content texts", 3)
989
# we have three major tasks here:
990
# 1) generate the ideal index
991
repo = self._pack_collection.repo
992
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
994
self.new_pack.revision_index.iter_all_entries()])
995
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
996
# 2) generate a text_nodes list that contains all the deltas that can
997
# be used as-is, with corrected parents.
1000
discarded_nodes = []
1001
NULL_REVISION = _mod_revision.NULL_REVISION
1002
text_index_map, text_nodes = self._get_text_nodes()
1003
for node in text_nodes:
1009
ideal_parents = tuple(ideal_index[node[1]])
1011
discarded_nodes.append(node)
1012
self._data_changed = True
1014
if ideal_parents == (NULL_REVISION,):
1016
if ideal_parents == node[3][0]:
1018
ok_nodes.append(node)
1019
elif ideal_parents[0:1] == node[3][0][0:1]:
1020
# the left most parent is the same, or there are no parents
1021
# today. Either way, we can preserve the representation as
1022
# long as we change the refs to be inserted.
1023
self._data_changed = True
1024
ok_nodes.append((node[0], node[1], node[2],
1025
(ideal_parents, node[3][1])))
1026
self._data_changed = True
1028
# Reinsert this text completely
1029
bad_texts.append((node[1], ideal_parents))
1030
self._data_changed = True
1031
# we're finished with some data.
1034
# 3) bulk copy the ok data
1035
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1036
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1037
self.new_pack.text_index, readv_group_iter, total_items))
1038
# 4) adhoc copy all the other texts.
1039
# We have to topologically insert all texts otherwise we can fail to
1040
# reconcile when parts of a single delta chain are preserved intact,
1041
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1042
# reinserted, and if d3 has incorrect parents it will also be
1043
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1044
# copied), so we will try to delta, but d2 is not currently able to be
1045
# extracted because it's basis d1 is not present. Topologically sorting
1046
# addresses this. The following generates a sort for all the texts that
1047
# are being inserted without having to reference the entire text key
1048
# space (we only topo sort the revisions, which is smaller).
1049
topo_order = tsort.topo_sort(ancestors)
1050
rev_order = dict(zip(topo_order, range(len(topo_order))))
1051
bad_texts.sort(key=lambda key:rev_order[key[0][1]])
1052
transaction = repo.get_transaction()
1053
file_id_index = GraphIndexPrefixAdapter(
1054
self.new_pack.text_index,
1056
add_nodes_callback=self.new_pack.text_index.add_nodes)
1057
knit_index = KnitGraphIndex(file_id_index,
1058
add_callback=file_id_index.add_nodes,
1059
deltas=True, parents=True)
1060
output_knit = knit.KnitVersionedFile('reconcile-texts',
1061
self._pack_collection.transport,
1064
access_method=_PackAccess(
1065
{self.new_pack.text_index:self.new_pack.access_tuple()},
1066
(self.new_pack._writer, self.new_pack.text_index)),
1067
factory=knit.KnitPlainFactory())
1068
for key, parent_keys in bad_texts:
1069
# We refer to the new pack to delta data being output.
1070
# A possible improvement would be to catch errors on short reads
1071
# and only flush then.
1072
self.new_pack.flush()
1074
for parent_key in parent_keys:
1075
if parent_key[0] != key[0]:
1076
# Graph parents must match the fileid
1077
raise errors.BzrError('Mismatched key parent %r:%r' %
1079
parents.append(parent_key[1])
1080
source_weave = repo.weave_store.get_weave(key[0], transaction)
1081
text_lines = source_weave.get_lines(key[1])
1082
# adapt the 'knit' to the current file_id.
1083
file_id_index = GraphIndexPrefixAdapter(
1084
self.new_pack.text_index,
1086
add_nodes_callback=self.new_pack.text_index.add_nodes)
1087
knit_index._graph_index = file_id_index
1088
knit_index._add_callback = file_id_index.add_nodes
1089
output_knit.add_lines_with_ghosts(
1090
key[1], parents, text_lines, random_id=True, check_content=False)
1091
# 5) check that nothing inserted has a reference outside the keyspace.
1092
missing_text_keys = self.new_pack._external_compression_parents_of_texts()
1093
if missing_text_keys:
1094
raise errors.BzrError('Reference to missing compression parents %r'
1096
self._log_copied_texts()
1098
def _use_pack(self, new_pack):
1099
"""Override _use_pack to check for reconcile having changed content."""
1100
# XXX: we might be better checking this at the copy time.
1101
original_inventory_keys = set()
1102
inv_index = self._pack_collection.inventory_index.combined_index
1103
for entry in inv_index.iter_all_entries():
1104
original_inventory_keys.add(entry[1])
1105
new_inventory_keys = set()
1106
for entry in new_pack.inventory_index.iter_all_entries():
1107
new_inventory_keys.add(entry[1])
1108
if new_inventory_keys != original_inventory_keys:
1109
self._data_changed = True
1110
return new_pack.data_inserted() and self._data_changed
1113
class RepositoryPackCollection(object):
1114
"""Management of packs within a repository."""
1116
def __init__(self, repo, transport, index_transport, upload_transport,
1118
"""Create a new RepositoryPackCollection.
1120
:param transport: Addresses the repository base directory
1121
(typically .bzr/repository/).
1122
:param index_transport: Addresses the directory containing indices.
1123
:param upload_transport: Addresses the directory into which packs are written
1124
while they're being created.
1125
:param pack_transport: Addresses the directory of existing complete packs.
1128
self.transport = transport
1129
self._index_transport = index_transport
1130
self._upload_transport = upload_transport
1131
self._pack_transport = pack_transport
1132
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
1135
self._packs_by_name = {}
1136
# the previous pack-names content
1137
self._packs_at_load = None
1138
# when a pack is being created by this object, the state of that pack.
1139
self._new_pack = None
1140
# aggregated revision index data
1141
self.revision_index = AggregateIndex()
1142
self.inventory_index = AggregateIndex()
1143
self.text_index = AggregateIndex()
1144
self.signature_index = AggregateIndex()
1146
def add_pack_to_memory(self, pack):
1147
"""Make a Pack object available to the repository to satisfy queries.
1149
:param pack: A Pack object.
1151
assert pack.name not in self._packs_by_name
1152
self.packs.append(pack)
1153
self._packs_by_name[pack.name] = pack
1154
self.revision_index.add_index(pack.revision_index, pack)
1155
self.inventory_index.add_index(pack.inventory_index, pack)
1156
self.text_index.add_index(pack.text_index, pack)
1157
self.signature_index.add_index(pack.signature_index, pack)
1159
def _add_text_to_weave(self, file_id, revision_id, new_lines, parents,
1160
nostore_sha, random_revid):
1161
file_id_index = GraphIndexPrefixAdapter(
1162
self.text_index.combined_index,
1164
add_nodes_callback=self.text_index.add_callback)
1165
self.repo._text_knit._index._graph_index = file_id_index
1166
self.repo._text_knit._index._add_callback = file_id_index.add_nodes
1167
return self.repo._text_knit.add_lines_with_ghosts(
1168
revision_id, parents, new_lines, nostore_sha=nostore_sha,
1169
random_id=random_revid, check_content=False)[0:2]
1171
def all_packs(self):
1172
"""Return a list of all the Pack objects this repository has.
1174
Note that an in-progress pack being created is not returned.
1176
:return: A list of Pack objects for all the packs in the repository.
1179
for name in self.names():
1180
result.append(self.get_pack_by_name(name))
1184
"""Pack the pack collection incrementally.
1186
This will not attempt global reorganisation or recompression,
1187
rather it will just ensure that the total number of packs does
1188
not grow without bound. It uses the _max_pack_count method to
1189
determine if autopacking is needed, and the pack_distribution
1190
method to determine the number of revisions in each pack.
1192
If autopacking takes place then the packs name collection will have
1193
been flushed to disk - packing requires updating the name collection
1194
in synchronisation with certain steps. Otherwise the names collection
1197
:return: True if packing took place.
1199
# XXX: Should not be needed when the management of indices is sane.
1200
total_revisions = self.revision_index.combined_index.key_count()
1201
total_packs = len(self._names)
1202
if self._max_pack_count(total_revisions) >= total_packs:
1204
# XXX: the following may want to be a class, to pack with a given
1206
mutter('Auto-packing repository %s, which has %d pack files, '
1207
'containing %d revisions into %d packs.', self, total_packs,
1208
total_revisions, self._max_pack_count(total_revisions))
1209
# determine which packs need changing
1210
pack_distribution = self.pack_distribution(total_revisions)
1212
for pack in self.all_packs():
1213
revision_count = pack.get_revision_count()
1214
if revision_count == 0:
1215
# revision less packs are not generated by normal operation,
1216
# only by operations like sign-my-commits, and thus will not
1217
# tend to grow rapdily or without bound like commit containing
1218
# packs do - leave them alone as packing them really should
1219
# group their data with the relevant commit, and that may
1220
# involve rewriting ancient history - which autopack tries to
1221
# avoid. Alternatively we could not group the data but treat
1222
# each of these as having a single revision, and thus add
1223
# one revision for each to the total revision count, to get
1224
# a matching distribution.
1226
existing_packs.append((revision_count, pack))
1227
pack_operations = self.plan_autopack_combinations(
1228
existing_packs, pack_distribution)
1229
self._execute_pack_operations(pack_operations)
1232
def _execute_pack_operations(self, pack_operations, _packer_class=Packer):
1233
"""Execute a series of pack operations.
1235
:param pack_operations: A list of [revision_count, packs_to_combine].
1236
:param _packer_class: The class of packer to use (default: Packer).
1239
for revision_count, packs in pack_operations:
1240
# we may have no-ops from the setup logic
1243
_packer_class(self, packs, '.autopack').pack()
1245
self._remove_pack_from_memory(pack)
1246
# record the newly available packs and stop advertising the old
1248
self._save_pack_names(clear_obsolete_packs=True)
1249
# Move the old packs out of the way now they are no longer referenced.
1250
for revision_count, packs in pack_operations:
1251
self._obsolete_packs(packs)
1253
def lock_names(self):
1254
"""Acquire the mutex around the pack-names index.
1256
This cannot be used in the middle of a read-only transaction on the
1259
self.repo.control_files.lock_write()
1262
"""Pack the pack collection totally."""
1263
self.ensure_loaded()
1264
total_packs = len(self._names)
1266
# This is arguably wrong because we might not be optimal, but for
1267
# now lets leave it in. (e.g. reconcile -> one pack. But not
1270
total_revisions = self.revision_index.combined_index.key_count()
1271
# XXX: the following may want to be a class, to pack with a given
1273
mutter('Packing repository %s, which has %d pack files, '
1274
'containing %d revisions into 1 packs.', self, total_packs,
1276
# determine which packs need changing
1277
pack_distribution = [1]
1278
pack_operations = [[0, []]]
1279
for pack in self.all_packs():
1280
pack_operations[-1][0] += pack.get_revision_count()
1281
pack_operations[-1][1].append(pack)
1282
self._execute_pack_operations(pack_operations, OptimisingPacker)
1284
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1285
"""Plan a pack operation.
1287
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1289
:param pack_distribution: A list with the number of revisions desired
1292
if len(existing_packs) <= len(pack_distribution):
1294
existing_packs.sort(reverse=True)
1295
pack_operations = [[0, []]]
1296
# plan out what packs to keep, and what to reorganise
1297
while len(existing_packs):
1298
# take the largest pack, and if its less than the head of the
1299
# distribution chart we will include its contents in the new pack for
1300
# that position. If its larger, we remove its size from the
1301
# distribution chart
1302
next_pack_rev_count, next_pack = existing_packs.pop(0)
1303
if next_pack_rev_count >= pack_distribution[0]:
1304
# this is already packed 'better' than this, so we can
1305
# not waste time packing it.
1306
while next_pack_rev_count > 0:
1307
next_pack_rev_count -= pack_distribution[0]
1308
if next_pack_rev_count >= 0:
1310
del pack_distribution[0]
1312
# didn't use that entire bucket up
1313
pack_distribution[0] = -next_pack_rev_count
1315
# add the revisions we're going to add to the next output pack
1316
pack_operations[-1][0] += next_pack_rev_count
1317
# allocate this pack to the next pack sub operation
1318
pack_operations[-1][1].append(next_pack)
1319
if pack_operations[-1][0] >= pack_distribution[0]:
1320
# this pack is used up, shift left.
1321
del pack_distribution[0]
1322
pack_operations.append([0, []])
1324
return pack_operations
1326
def ensure_loaded(self):
1327
# NB: if you see an assertion error here, its probably access against
1328
# an unlocked repo. Naughty.
1329
if not self.repo.is_locked():
1330
raise errors.ObjectNotLocked(self.repo)
1331
if self._names is None:
1333
self._packs_at_load = set()
1334
for index, key, value in self._iter_disk_pack_index():
1336
self._names[name] = self._parse_index_sizes(value)
1337
self._packs_at_load.add((key, value))
1338
# populate all the metadata.
1341
def _parse_index_sizes(self, value):
1342
"""Parse a string of index sizes."""
1343
return tuple([int(digits) for digits in value.split(' ')])
1345
def get_pack_by_name(self, name):
1346
"""Get a Pack object by name.
1348
:param name: The name of the pack - e.g. '123456'
1349
:return: A Pack object.
1352
return self._packs_by_name[name]
1354
rev_index = self._make_index(name, '.rix')
1355
inv_index = self._make_index(name, '.iix')
1356
txt_index = self._make_index(name, '.tix')
1357
sig_index = self._make_index(name, '.six')
1358
result = ExistingPack(self._pack_transport, name, rev_index,
1359
inv_index, txt_index, sig_index)
1360
self.add_pack_to_memory(result)
1363
def allocate(self, a_new_pack):
1364
"""Allocate name in the list of packs.
1366
:param a_new_pack: A NewPack instance to be added to the collection of
1367
packs for this repository.
1369
self.ensure_loaded()
1370
if a_new_pack.name in self._names:
1371
raise errors.BzrError(
1372
'Pack %r already exists in %s' % (a_new_pack.name, self))
1373
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1374
self.add_pack_to_memory(a_new_pack)
1376
def _iter_disk_pack_index(self):
1377
"""Iterate over the contents of the pack-names index.
1379
This is used when loading the list from disk, and before writing to
1380
detect updates from others during our write operation.
1381
:return: An iterator of the index contents.
1383
return GraphIndex(self.transport, 'pack-names', None
1384
).iter_all_entries()
1386
def _make_index(self, name, suffix):
1387
size_offset = self._suffix_offsets[suffix]
1388
index_name = name + suffix
1389
index_size = self._names[name][size_offset]
1391
self._index_transport, index_name, index_size)
1393
def _max_pack_count(self, total_revisions):
1394
"""Return the maximum number of packs to use for total revisions.
1396
:param total_revisions: The total number of revisions in the
1399
if not total_revisions:
1401
digits = str(total_revisions)
1403
for digit in digits:
1404
result += int(digit)
1408
"""Provide an order to the underlying names."""
1409
return sorted(self._names.keys())
1411
def _obsolete_packs(self, packs):
1412
"""Move a number of packs which have been obsoleted out of the way.
1414
Each pack and its associated indices are moved out of the way.
1416
Note: for correctness this function should only be called after a new
1417
pack names index has been written without these pack names, and with
1418
the names of packs that contain the data previously available via these
1421
:param packs: The packs to obsolete.
1422
:param return: None.
1425
pack.pack_transport.rename(pack.file_name(),
1426
'../obsolete_packs/' + pack.file_name())
1427
# TODO: Probably needs to know all possible indices for this pack
1428
# - or maybe list the directory and move all indices matching this
1429
# name whether we recognize it or not?
1430
for suffix in ('.iix', '.six', '.tix', '.rix'):
1431
self._index_transport.rename(pack.name + suffix,
1432
'../obsolete_packs/' + pack.name + suffix)
1434
def pack_distribution(self, total_revisions):
1435
"""Generate a list of the number of revisions to put in each pack.
1437
:param total_revisions: The total number of revisions in the
1440
if total_revisions == 0:
1442
digits = reversed(str(total_revisions))
1444
for exponent, count in enumerate(digits):
1445
size = 10 ** exponent
1446
for pos in range(int(count)):
1448
return list(reversed(result))
1450
def _pack_tuple(self, name):
1451
"""Return a tuple with the transport and file name for a pack name."""
1452
return self._pack_transport, name + '.pack'
1454
def _remove_pack_from_memory(self, pack):
1455
"""Remove pack from the packs accessed by this repository.
1457
Only affects memory state, until self._save_pack_names() is invoked.
1459
self._names.pop(pack.name)
1460
self._packs_by_name.pop(pack.name)
1461
self._remove_pack_indices(pack)
1463
def _remove_pack_indices(self, pack):
1464
"""Remove the indices for pack from the aggregated indices."""
1465
self.revision_index.remove_index(pack.revision_index, pack)
1466
self.inventory_index.remove_index(pack.inventory_index, pack)
1467
self.text_index.remove_index(pack.text_index, pack)
1468
self.signature_index.remove_index(pack.signature_index, pack)
1471
"""Clear all cached data."""
1472
# cached revision data
1473
self.repo._revision_knit = None
1474
self.revision_index.clear()
1475
# cached signature data
1476
self.repo._signature_knit = None
1477
self.signature_index.clear()
1478
# cached file text data
1479
self.text_index.clear()
1480
self.repo._text_knit = None
1481
# cached inventory data
1482
self.inventory_index.clear()
1483
# remove the open pack
1484
self._new_pack = None
1485
# information about packs.
1488
self._packs_by_name = {}
1489
self._packs_at_load = None
1491
def _make_index_map(self, index_suffix):
1492
"""Return information on existing indices.
1494
:param suffix: Index suffix added to pack name.
1496
:returns: (pack_map, indices) where indices is a list of GraphIndex
1497
objects, and pack_map is a mapping from those objects to the
1498
pack tuple they describe.
1500
# TODO: stop using this; it creates new indices unnecessarily.
1501
self.ensure_loaded()
1502
suffix_map = {'.rix': 'revision_index',
1503
'.six': 'signature_index',
1504
'.iix': 'inventory_index',
1505
'.tix': 'text_index',
1507
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1508
suffix_map[index_suffix])
1510
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1511
"""Convert a list of packs to an index pack map and index list.
1513
:param packs: The packs list to process.
1514
:param index_attribute: The attribute that the desired index is found
1516
:return: A tuple (map, list) where map contains the dict from
1517
index:pack_tuple, and lsit contains the indices in the same order
1523
index = getattr(pack, index_attribute)
1524
indices.append(index)
1525
pack_map[index] = (pack.pack_transport, pack.file_name())
1526
return pack_map, indices
1528
def _index_contents(self, pack_map, key_filter=None):
1529
"""Get an iterable of the index contents from a pack_map.
1531
:param pack_map: A map from indices to pack details.
1532
:param key_filter: An optional filter to limit the
1535
indices = [index for index in pack_map.iterkeys()]
1536
all_index = CombinedGraphIndex(indices)
1537
if key_filter is None:
1538
return all_index.iter_all_entries()
1540
return all_index.iter_entries(key_filter)
1542
def _unlock_names(self):
1543
"""Release the mutex around the pack-names index."""
1544
self.repo.control_files.unlock()
1546
def _save_pack_names(self, clear_obsolete_packs=False):
1547
"""Save the list of packs.
1549
This will take out the mutex around the pack names list for the
1550
duration of the method call. If concurrent updates have been made, a
1551
three-way merge between the current list and the current in memory list
1554
:param clear_obsolete_packs: If True, clear out the contents of the
1555
obsolete_packs directory.
1559
builder = GraphIndexBuilder()
1560
# load the disk nodes across
1562
for index, key, value in self._iter_disk_pack_index():
1563
disk_nodes.add((key, value))
1564
# do a two-way diff against our original content
1565
current_nodes = set()
1566
for name, sizes in self._names.iteritems():
1568
((name, ), ' '.join(str(size) for size in sizes)))
1569
deleted_nodes = self._packs_at_load - current_nodes
1570
new_nodes = current_nodes - self._packs_at_load
1571
disk_nodes.difference_update(deleted_nodes)
1572
disk_nodes.update(new_nodes)
1573
# TODO: handle same-name, index-size-changes here -
1574
# e.g. use the value from disk, not ours, *unless* we're the one
1576
for key, value in disk_nodes:
1577
builder.add_node(key, value)
1578
self.transport.put_file('pack-names', builder.finish(),
1579
mode=self.repo.control_files._file_mode)
1580
# move the baseline forward
1581
self._packs_at_load = disk_nodes
1582
# now clear out the obsolete packs directory
1583
if clear_obsolete_packs:
1584
self.transport.clone('obsolete_packs').delete_multi(
1585
self.transport.list_dir('obsolete_packs'))
1587
self._unlock_names()
1588
# synchronise the memory packs list with what we just wrote:
1589
new_names = dict(disk_nodes)
1590
# drop no longer present nodes
1591
for pack in self.all_packs():
1592
if (pack.name,) not in new_names:
1593
self._remove_pack_from_memory(pack)
1594
# add new nodes/refresh existing ones
1595
for key, value in disk_nodes:
1597
sizes = self._parse_index_sizes(value)
1598
if name in self._names:
1600
if sizes != self._names[name]:
1601
# the pack for name has had its indices replaced - rare but
1602
# important to handle. XXX: probably can never happen today
1603
# because the three-way merge code above does not handle it
1604
# - you may end up adding the same key twice to the new
1605
# disk index because the set values are the same, unless
1606
# the only index shows up as deleted by the set difference
1607
# - which it may. Until there is a specific test for this,
1608
# assume its broken. RBC 20071017.
1609
self._remove_pack_from_memory(self.get_pack_by_name(name))
1610
self._names[name] = sizes
1611
self.get_pack_by_name(name)
1614
self._names[name] = sizes
1615
self.get_pack_by_name(name)
1617
def _start_write_group(self):
1618
# Do not permit preparation for writing if we're not in a 'write lock'.
1619
if not self.repo.is_write_locked():
1620
raise errors.NotWriteLocked(self)
1621
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1622
self._pack_transport, upload_suffix='.pack',
1623
file_mode=self.repo.control_files._file_mode)
1624
# allow writing: queue writes to a new index
1625
self.revision_index.add_writable_index(self._new_pack.revision_index,
1627
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
1629
self.text_index.add_writable_index(self._new_pack.text_index,
1631
self.signature_index.add_writable_index(self._new_pack.signature_index,
1634
# reused revision and signature knits may need updating
1636
# "Hysterical raisins. client code in bzrlib grabs those knits outside
1637
# of write groups and then mutates it inside the write group."
1638
if self.repo._revision_knit is not None:
1639
self.repo._revision_knit._index._add_callback = \
1640
self.revision_index.add_callback
1641
if self.repo._signature_knit is not None:
1642
self.repo._signature_knit._index._add_callback = \
1643
self.signature_index.add_callback
1644
# create a reused knit object for text addition in commit.
1645
self.repo._text_knit = self.repo.weave_store.get_weave_or_empty(
1648
def _abort_write_group(self):
1649
# FIXME: just drop the transient index.
1650
# forget what names there are
1651
if self._new_pack is not None:
1652
self._new_pack.abort()
1653
self._remove_pack_indices(self._new_pack)
1654
self._new_pack = None
1655
self.repo._text_knit = None
1657
def _commit_write_group(self):
1658
self._remove_pack_indices(self._new_pack)
1659
if self._new_pack.data_inserted():
1660
# get all the data to disk and read to use
1661
self._new_pack.finish()
1662
self.allocate(self._new_pack)
1663
self._new_pack = None
1664
if not self.autopack():
1665
# when autopack takes no steps, the names list is still
1667
self._save_pack_names()
1669
self._new_pack.abort()
1670
self._new_pack = None
1671
self.repo._text_knit = None
1674
class KnitPackRevisionStore(KnitRevisionStore):
1675
"""An object to adapt access from RevisionStore's to use KnitPacks.
1677
This class works by replacing the original RevisionStore.
1678
We need to do this because the KnitPackRevisionStore is less
1679
isolated in its layering - it uses services from the repo.
1682
def __init__(self, repo, transport, revisionstore):
1683
"""Create a KnitPackRevisionStore on repo with revisionstore.
1685
This will store its state in the Repository, use the
1686
indices to provide a KnitGraphIndex,
1687
and at the end of transactions write new indices.
1689
KnitRevisionStore.__init__(self, revisionstore.versioned_file_store)
1691
self._serializer = revisionstore._serializer
1692
self.transport = transport
1694
def get_revision_file(self, transaction):
1695
"""Get the revision versioned file object."""
1696
if getattr(self.repo, '_revision_knit', None) is not None:
1697
return self.repo._revision_knit
1698
self.repo._pack_collection.ensure_loaded()
1699
add_callback = self.repo._pack_collection.revision_index.add_callback
1700
# setup knit specific objects
1701
knit_index = KnitGraphIndex(
1702
self.repo._pack_collection.revision_index.combined_index,
1703
add_callback=add_callback)
1704
self.repo._revision_knit = knit.KnitVersionedFile(
1705
'revisions', self.transport.clone('..'),
1706
self.repo.control_files._file_mode,
1707
create=False, access_mode=self.repo._access_mode(),
1708
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1709
access_method=self.repo._pack_collection.revision_index.knit_access)
1710
return self.repo._revision_knit
1712
def get_signature_file(self, transaction):
1713
"""Get the signature versioned file object."""
1714
if getattr(self.repo, '_signature_knit', None) is not None:
1715
return self.repo._signature_knit
1716
self.repo._pack_collection.ensure_loaded()
1717
add_callback = self.repo._pack_collection.signature_index.add_callback
1718
# setup knit specific objects
1719
knit_index = KnitGraphIndex(
1720
self.repo._pack_collection.signature_index.combined_index,
1721
add_callback=add_callback, parents=False)
1722
self.repo._signature_knit = knit.KnitVersionedFile(
1723
'signatures', self.transport.clone('..'),
1724
self.repo.control_files._file_mode,
1725
create=False, access_mode=self.repo._access_mode(),
1726
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1727
access_method=self.repo._pack_collection.signature_index.knit_access)
1728
return self.repo._signature_knit
1731
class KnitPackTextStore(VersionedFileStore):
1732
"""Presents a TextStore abstraction on top of packs.
1734
This class works by replacing the original VersionedFileStore.
1735
We need to do this because the KnitPackRevisionStore is less
1736
isolated in its layering - it uses services from the repo and shares them
1737
with all the data written in a single write group.
1740
def __init__(self, repo, transport, weavestore):
1741
"""Create a KnitPackTextStore on repo with weavestore.
1743
This will store its state in the Repository, use the
1744
indices FileNames to provide a KnitGraphIndex,
1745
and at the end of transactions write new indices.
1747
# don't call base class constructor - it's not suitable.
1748
# no transient data stored in the transaction
1750
self._precious = False
1752
self.transport = transport
1753
self.weavestore = weavestore
1754
# XXX for check() which isn't updated yet
1755
self._transport = weavestore._transport
1757
def get_weave_or_empty(self, file_id, transaction):
1758
"""Get a 'Knit' backed by the .tix indices.
1760
The transaction parameter is ignored.
1762
self.repo._pack_collection.ensure_loaded()
1763
add_callback = self.repo._pack_collection.text_index.add_callback
1764
# setup knit specific objects
1765
file_id_index = GraphIndexPrefixAdapter(
1766
self.repo._pack_collection.text_index.combined_index,
1767
(file_id, ), 1, add_nodes_callback=add_callback)
1768
knit_index = KnitGraphIndex(file_id_index,
1769
add_callback=file_id_index.add_nodes,
1770
deltas=True, parents=True)
1771
return knit.KnitVersionedFile('text:' + file_id,
1772
self.transport.clone('..'),
1775
access_method=self.repo._pack_collection.text_index.knit_access,
1776
factory=knit.KnitPlainFactory())
1778
get_weave = get_weave_or_empty
1781
"""Generate a list of the fileids inserted, for use by check."""
1782
self.repo._pack_collection.ensure_loaded()
1784
for index, key, value, refs in \
1785
self.repo._pack_collection.text_index.combined_index.iter_all_entries():
1790
class InventoryKnitThunk(object):
1791
"""An object to manage thunking get_inventory_weave to pack based knits."""
1793
def __init__(self, repo, transport):
1794
"""Create an InventoryKnitThunk for repo at transport.
1796
This will store its state in the Repository, use the
1797
indices FileNames to provide a KnitGraphIndex,
1798
and at the end of transactions write a new index..
1801
self.transport = transport
1803
def get_weave(self):
1804
"""Get a 'Knit' that contains inventory data."""
1805
self.repo._pack_collection.ensure_loaded()
1806
add_callback = self.repo._pack_collection.inventory_index.add_callback
1807
# setup knit specific objects
1808
knit_index = KnitGraphIndex(
1809
self.repo._pack_collection.inventory_index.combined_index,
1810
add_callback=add_callback, deltas=True, parents=True)
1811
return knit.KnitVersionedFile(
1812
'inventory', self.transport.clone('..'),
1813
self.repo.control_files._file_mode,
1814
create=False, access_mode=self.repo._access_mode(),
1815
index=knit_index, delta=True, factory=knit.KnitPlainFactory(),
1816
access_method=self.repo._pack_collection.inventory_index.knit_access)
1819
class KnitPackRepository(KnitRepository):
1820
"""Experimental graph-knit using repository."""
1822
def __init__(self, _format, a_bzrdir, control_files, _revision_store,
1823
control_store, text_store, _commit_builder_class, _serializer):
1824
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
1825
_revision_store, control_store, text_store, _commit_builder_class,
1827
index_transport = control_files._transport.clone('indices')
1828
self._pack_collection = RepositoryPackCollection(self, control_files._transport,
1830
control_files._transport.clone('upload'),
1831
control_files._transport.clone('packs'))
1832
self._revision_store = KnitPackRevisionStore(self, index_transport, self._revision_store)
1833
self.weave_store = KnitPackTextStore(self, index_transport, self.weave_store)
1834
self._inv_thunk = InventoryKnitThunk(self, index_transport)
1835
# True when the repository object is 'write locked' (as opposed to the
1836
# physical lock only taken out around changes to the pack-names list.)
1837
# Another way to represent this would be a decorator around the control
1838
# files object that presents logical locks as physical ones - if this
1839
# gets ugly consider that alternative design. RBC 20071011
1840
self._write_lock_count = 0
1841
self._transaction = None
1843
self._reconcile_does_inventory_gc = True
1844
self._reconcile_fixes_text_parents = True
1845
self._reconcile_backsup_inventory = False
1847
def _abort_write_group(self):
1848
self._pack_collection._abort_write_group()
1850
def _access_mode(self):
1851
"""Return 'w' or 'r' for depending on whether a write lock is active.
1853
This method is a helper for the Knit-thunking support objects.
1855
if self.is_write_locked():
1859
def _find_inconsistent_revision_parents(self):
1860
"""Find revisions with incorrectly cached parents.
1862
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1863
parents-in-revision).
1865
if not self.is_locked():
1866
raise errors.ObjectNotLocked(self)
1867
pb = ui.ui_factory.nested_progress_bar()
1870
revision_nodes = self._pack_collection.revision_index \
1871
.combined_index.iter_all_entries()
1872
index_positions = []
1873
# Get the cached index values for all revisions, and also the location
1874
# in each index of the revision text so we can perform linear IO.
1875
for index, key, value, refs in revision_nodes:
1876
pos, length = value[1:].split(' ')
1877
index_positions.append((index, int(pos), key[0],
1878
tuple(parent[0] for parent in refs[0])))
1879
pb.update("Reading revision index.", 0, 0)
1880
index_positions.sort()
1881
batch_count = len(index_positions) / 1000 + 1
1882
pb.update("Checking cached revision graph.", 0, batch_count)
1883
for offset in xrange(batch_count):
1884
pb.update("Checking cached revision graph.", offset)
1885
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1888
rev_ids = [item[2] for item in to_query]
1889
revs = self.get_revisions(rev_ids)
1890
for revision, item in zip(revs, to_query):
1891
index_parents = item[3]
1892
rev_parents = tuple(revision.parent_ids)
1893
if index_parents != rev_parents:
1894
result.append((revision.revision_id, index_parents, rev_parents))
1899
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
1900
def get_parents(self, revision_ids):
1901
"""See graph._StackedParentsProvider.get_parents."""
1902
parent_map = self.get_parent_map(revision_ids)
1903
return [parent_map.get(r, None) for r in revision_ids]
1905
def get_parent_map(self, keys):
1906
"""See graph._StackedParentsProvider.get_parent_map
1908
This implementation accesses the combined revision index to provide
1911
self._pack_collection.ensure_loaded()
1912
index = self._pack_collection.revision_index.combined_index
1914
if _mod_revision.NULL_REVISION in keys:
1915
keys.discard(_mod_revision.NULL_REVISION)
1916
found_parents = {_mod_revision.NULL_REVISION:()}
1919
search_keys = set((revision_id,) for revision_id in keys)
1920
for index, key, value, refs in index.iter_entries(search_keys):
1923
parents = (_mod_revision.NULL_REVISION,)
1925
parents = tuple(parent[0] for parent in parents)
1926
found_parents[key[0]] = parents
1927
return found_parents
1929
def has_revisions(self, revision_ids):
1930
"""See Repository.has_revisions()."""
1931
revision_ids = set(revision_ids)
1932
result = revision_ids.intersection(
1933
set([None, _mod_revision.NULL_REVISION]))
1934
revision_ids.difference_update(result)
1935
index = self._pack_collection.revision_index.combined_index
1936
keys = [(revision_id,) for revision_id in revision_ids]
1937
result.update(node[1][0] for node in index.iter_entries(keys))
1940
def _make_parents_provider(self):
1941
return graph.CachingParentsProvider(self)
1943
def _refresh_data(self):
1944
if self._write_lock_count == 1 or (
1945
self.control_files._lock_count == 1 and
1946
self.control_files._lock_mode == 'r'):
1947
# forget what names there are
1948
self._pack_collection.reset()
1949
# XXX: Better to do an in-memory merge when acquiring a new lock -
1950
# factor out code from _save_pack_names.
1951
self._pack_collection.ensure_loaded()
1953
def _start_write_group(self):
1954
self._pack_collection._start_write_group()
1956
def _commit_write_group(self):
1957
return self._pack_collection._commit_write_group()
1959
def get_inventory_weave(self):
1960
return self._inv_thunk.get_weave()
1962
def get_transaction(self):
1963
if self._write_lock_count:
1964
return self._transaction
1966
return self.control_files.get_transaction()
1968
def is_locked(self):
1969
return self._write_lock_count or self.control_files.is_locked()
1971
def is_write_locked(self):
1972
return self._write_lock_count
1974
def lock_write(self, token=None):
1975
if not self._write_lock_count and self.is_locked():
1976
raise errors.ReadOnlyError(self)
1977
self._write_lock_count += 1
1978
if self._write_lock_count == 1:
1979
from bzrlib import transactions
1980
self._transaction = transactions.WriteTransaction()
1981
self._refresh_data()
1983
def lock_read(self):
1984
if self._write_lock_count:
1985
self._write_lock_count += 1
1987
self.control_files.lock_read()
1988
self._refresh_data()
1990
def leave_lock_in_place(self):
1991
# not supported - raise an error
1992
raise NotImplementedError(self.leave_lock_in_place)
1994
def dont_leave_lock_in_place(self):
1995
# not supported - raise an error
1996
raise NotImplementedError(self.dont_leave_lock_in_place)
2000
"""Compress the data within the repository.
2002
This will pack all the data to a single pack. In future it may
2003
recompress deltas or do other such expensive operations.
2005
self._pack_collection.pack()
2008
def reconcile(self, other=None, thorough=False):
2009
"""Reconcile this repository."""
2010
from bzrlib.reconcile import PackReconciler
2011
reconciler = PackReconciler(self, thorough=thorough)
2012
reconciler.reconcile()
2016
if self._write_lock_count == 1 and self._write_group is not None:
2017
self.abort_write_group()
2018
self._transaction = None
2019
self._write_lock_count = 0
2020
raise errors.BzrError(
2021
'Must end write group before releasing write lock on %s'
2023
if self._write_lock_count:
2024
self._write_lock_count -= 1
2025
if not self._write_lock_count:
2026
transaction = self._transaction
2027
self._transaction = None
2028
transaction.finish()
2030
self.control_files.unlock()
2033
class RepositoryFormatPack(MetaDirRepositoryFormat):
2034
"""Format logic for pack structured repositories.
2036
This repository format has:
2037
- a list of packs in pack-names
2038
- packs in packs/NAME.pack
2039
- indices in indices/NAME.{iix,six,tix,rix}
2040
- knit deltas in the packs, knit indices mapped to the indices.
2041
- thunk objects to support the knits programming API.
2042
- a format marker of its own
2043
- an optional 'shared-storage' flag
2044
- an optional 'no-working-trees' flag
2048
# Set this attribute in derived classes to control the repository class
2049
# created by open and initialize.
2050
repository_class = None
2051
# Set this attribute in derived classes to control the
2052
# _commit_builder_class that the repository objects will have passed to
2053
# their constructor.
2054
_commit_builder_class = None
2055
# Set this attribute in derived clases to control the _serializer that the
2056
# repository objects will have passed to their constructor.
2058
# External references are not supported in pack repositories yet.
2059
supports_external_lookups = False
2061
def _get_control_store(self, repo_transport, control_files):
2062
"""Return the control store for this repository."""
2063
return VersionedFileStore(
2066
file_mode=control_files._file_mode,
2067
versionedfile_class=knit.KnitVersionedFile,
2068
versionedfile_kwargs={'factory': knit.KnitPlainFactory()},
2071
def _get_revision_store(self, repo_transport, control_files):
2072
"""See RepositoryFormat._get_revision_store()."""
2073
versioned_file_store = VersionedFileStore(
2075
file_mode=control_files._file_mode,
2078
versionedfile_class=knit.KnitVersionedFile,
2079
versionedfile_kwargs={'delta': False,
2080
'factory': knit.KnitPlainFactory(),
2084
return KnitRevisionStore(versioned_file_store)
2086
def _get_text_store(self, transport, control_files):
2087
"""See RepositoryFormat._get_text_store()."""
2088
return self._get_versioned_file_store('knits',
2091
versionedfile_class=knit.KnitVersionedFile,
2092
versionedfile_kwargs={
2093
'create_parent_dir': True,
2094
'delay_create': True,
2095
'dir_mode': control_files._dir_mode,
2099
def initialize(self, a_bzrdir, shared=False):
2100
"""Create a pack based repository.
2102
:param a_bzrdir: bzrdir to contain the new repository; must already
2104
:param shared: If true the repository will be initialized as a shared
2107
mutter('creating repository in %s.', a_bzrdir.transport.base)
2108
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
2109
builder = GraphIndexBuilder()
2110
files = [('pack-names', builder.finish())]
2111
utf8_files = [('format', self.get_format_string())]
2113
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
2114
return self.open(a_bzrdir=a_bzrdir, _found=True)
2116
def open(self, a_bzrdir, _found=False, _override_transport=None):
2117
"""See RepositoryFormat.open().
2119
:param _override_transport: INTERNAL USE ONLY. Allows opening the
2120
repository at a slightly different url
2121
than normal. I.e. during 'upgrade'.
2124
format = RepositoryFormat.find_format(a_bzrdir)
2125
assert format.__class__ == self.__class__
2126
if _override_transport is not None:
2127
repo_transport = _override_transport
2129
repo_transport = a_bzrdir.get_repository_transport(None)
2130
control_files = lockable_files.LockableFiles(repo_transport,
2131
'lock', lockdir.LockDir)
2132
text_store = self._get_text_store(repo_transport, control_files)
2133
control_store = self._get_control_store(repo_transport, control_files)
2134
_revision_store = self._get_revision_store(repo_transport, control_files)
2135
return self.repository_class(_format=self,
2137
control_files=control_files,
2138
_revision_store=_revision_store,
2139
control_store=control_store,
2140
text_store=text_store,
2141
_commit_builder_class=self._commit_builder_class,
2142
_serializer=self._serializer)
2145
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2146
"""A no-subtrees parameterized Pack repository.
2148
This format was introduced in 0.92.
2151
repository_class = KnitPackRepository
2152
_commit_builder_class = PackCommitBuilder
2153
_serializer = xml5.serializer_v5
2155
def _get_matching_bzrdir(self):
2156
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2158
def _ignore_setting_bzrdir(self, format):
2161
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2163
def get_format_string(self):
2164
"""See RepositoryFormat.get_format_string()."""
2165
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2167
def get_format_description(self):
2168
"""See RepositoryFormat.get_format_description()."""
2169
return "Packs containing knits without subtree support"
2171
def check_conversion_target(self, target_format):
2175
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2176
"""A subtrees parameterized Pack repository.
2178
This repository format uses the xml7 serializer to get:
2179
- support for recording full info about the tree root
2180
- support for recording tree-references
2182
This format was introduced in 0.92.
2185
repository_class = KnitPackRepository
2186
_commit_builder_class = PackRootCommitBuilder
2187
rich_root_data = True
2188
supports_tree_reference = True
2189
_serializer = xml7.serializer_v7
2191
def _get_matching_bzrdir(self):
2192
return bzrdir.format_registry.make_bzrdir(
2193
'pack-0.92-subtree')
2195
def _ignore_setting_bzrdir(self, format):
2198
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2200
def check_conversion_target(self, target_format):
2201
if not target_format.rich_root_data:
2202
raise errors.BadConversionTarget(
2203
'Does not support rich root data.', target_format)
2204
if not getattr(target_format, 'supports_tree_reference', False):
2205
raise errors.BadConversionTarget(
2206
'Does not support nested trees', target_format)
2208
def get_format_string(self):
2209
"""See RepositoryFormat.get_format_string()."""
2210
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2212
def get_format_description(self):
2213
"""See RepositoryFormat.get_format_description()."""
2214
return "Packs containing knits with subtree support\n"
2217
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2218
"""A rich-root, no subtrees parameterized Pack repository.
2220
This repository format uses the xml6 serializer to get:
2221
- support for recording full info about the tree root
2223
This format was introduced in 1.0.
2226
repository_class = KnitPackRepository
2227
_commit_builder_class = PackRootCommitBuilder
2228
rich_root_data = True
2229
supports_tree_reference = False
2230
_serializer = xml6.serializer_v6
2232
def _get_matching_bzrdir(self):
2233
return bzrdir.format_registry.make_bzrdir(
2236
def _ignore_setting_bzrdir(self, format):
2239
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2241
def check_conversion_target(self, target_format):
2242
if not target_format.rich_root_data:
2243
raise errors.BadConversionTarget(
2244
'Does not support rich root data.', target_format)
2246
def get_format_string(self):
2247
"""See RepositoryFormat.get_format_string()."""
2248
return ("Bazaar pack repository format 1 with rich root"
2249
" (needs bzr 1.0)\n")
2251
def get_format_description(self):
2252
"""See RepositoryFormat.get_format_description()."""
2253
return "Packs containing knits with rich root support\n"
2256
class RepositoryFormatPackDevelopment0(RepositoryFormatPack):
2257
"""A no-subtrees development repository.
2259
This format should be retained until the second release after bzr 1.0.
2261
No changes to the disk behaviour from pack-0.92.
2264
repository_class = KnitPackRepository
2265
_commit_builder_class = PackCommitBuilder
2266
_serializer = xml5.serializer_v5
2268
def _get_matching_bzrdir(self):
2269
return bzrdir.format_registry.make_bzrdir('development0')
2271
def _ignore_setting_bzrdir(self, format):
2274
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2276
def get_format_string(self):
2277
"""See RepositoryFormat.get_format_string()."""
2278
return "Bazaar development format 0 (needs bzr.dev from before 1.3)\n"
2280
def get_format_description(self):
2281
"""See RepositoryFormat.get_format_description()."""
2282
return ("Development repository format, currently the same as "
2285
def check_conversion_target(self, target_format):
2289
class RepositoryFormatPackDevelopment0Subtree(RepositoryFormatPack):
2290
"""A subtrees development repository.
2292
This format should be retained until the second release after bzr 1.0.
2294
No changes to the disk behaviour from pack-0.92-subtree.
2297
repository_class = KnitPackRepository
2298
_commit_builder_class = PackRootCommitBuilder
2299
rich_root_data = True
2300
supports_tree_reference = True
2301
_serializer = xml7.serializer_v7
2303
def _get_matching_bzrdir(self):
2304
return bzrdir.format_registry.make_bzrdir(
2305
'development0-subtree')
2307
def _ignore_setting_bzrdir(self, format):
2310
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2312
def check_conversion_target(self, target_format):
2313
if not target_format.rich_root_data:
2314
raise errors.BadConversionTarget(
2315
'Does not support rich root data.', target_format)
2316
if not getattr(target_format, 'supports_tree_reference', False):
2317
raise errors.BadConversionTarget(
2318
'Does not support nested trees', target_format)
2320
def get_format_string(self):
2321
"""See RepositoryFormat.get_format_string()."""
2322
return ("Bazaar development format 0 with subtree support "
2323
"(needs bzr.dev from before 1.3)\n")
2325
def get_format_description(self):
2326
"""See RepositoryFormat.get_format_description()."""
2327
return ("Development repository format, currently the same as "
2328
"pack-0.92-subtree\n")
added
removed
bzrlib/repofmt/pack_repo.py
