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
# Only grab the output lines if we will be processing them
646
output_lines = bool(self.revision_ids)
647
inv_lines = self._copy_nodes_graph(inventory_index_map,
648
self.new_pack._writer, self.new_pack.inventory_index,
649
readv_group_iter, total_items, output_lines=output_lines)
650
if self.revision_ids:
651
self._process_inventory_lines(inv_lines)
653
# eat the iterator to cause it to execute.
655
self._text_filter = None
656
if 'pack' in debug.debug_flags:
657
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
658
time.ctime(), self._pack_collection._upload_transport.base,
659
self.new_pack.random_name,
660
self.new_pack.inventory_index.key_count(),
661
time.time() - self.new_pack.start_time)
663
def _copy_text_texts(self):
665
text_index_map, text_nodes = self._get_text_nodes()
666
if self._text_filter is not None:
667
# We could return the keys copied as part of the return value from
668
# _copy_nodes_graph but this doesn't work all that well with the
669
# need to get line output too, so we check separately, and as we're
670
# going to buffer everything anyway, we check beforehand, which
671
# saves reading knit data over the wire when we know there are
673
text_nodes = set(text_nodes)
674
present_text_keys = set(_node[1] for _node in text_nodes)
675
missing_text_keys = set(self._text_filter) - present_text_keys
676
if missing_text_keys:
677
# TODO: raise a specific error that can handle many missing
679
a_missing_key = missing_text_keys.pop()
680
raise errors.RevisionNotPresent(a_missing_key[1],
682
# copy text keys and adjust values
683
self.pb.update("Copying content texts", 3)
684
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
685
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
686
self.new_pack.text_index, readv_group_iter, total_items))
687
self._log_copied_texts()
689
def _check_references(self):
690
"""Make sure our external refereneces are present."""
691
external_refs = self.new_pack._external_compression_parents_of_texts()
693
index = self._pack_collection.text_index.combined_index
694
found_items = list(index.iter_entries(external_refs))
695
if len(found_items) != len(external_refs):
696
found_keys = set(k for idx, k, refs, value in found_items)
697
missing_items = external_refs - found_keys
698
missing_file_id, missing_revision_id = missing_items.pop()
699
raise errors.RevisionNotPresent(missing_revision_id,
702
def _create_pack_from_packs(self):
703
self.pb.update("Opening pack", 0, 5)
704
self.new_pack = self.open_pack()
705
new_pack = self.new_pack
706
# buffer data - we won't be reading-back during the pack creation and
707
# this makes a significant difference on sftp pushes.
708
new_pack.set_write_cache_size(1024*1024)
709
if 'pack' in debug.debug_flags:
710
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
711
for a_pack in self.packs]
712
if self.revision_ids is not None:
713
rev_count = len(self.revision_ids)
716
mutter('%s: create_pack: creating pack from source packs: '
717
'%s%s %s revisions wanted %s t=0',
718
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
719
plain_pack_list, rev_count)
720
self._copy_revision_texts()
721
self._copy_inventory_texts()
722
self._copy_text_texts()
723
# select signature keys
724
signature_filter = self._revision_keys # same keyspace
725
signature_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
726
self.packs, 'signature_index')[0]
727
signature_nodes = self._pack_collection._index_contents(signature_index_map,
729
# copy signature keys and adjust values
730
self.pb.update("Copying signature texts", 4)
731
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
732
new_pack.signature_index)
733
if 'pack' in debug.debug_flags:
734
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
735
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
736
new_pack.signature_index.key_count(),
737
time.time() - new_pack.start_time)
738
self._check_references()
739
if not self._use_pack(new_pack):
742
self.pb.update("Finishing pack", 5)
744
self._pack_collection.allocate(new_pack)
747
def _copy_nodes(self, nodes, index_map, writer, write_index):
748
"""Copy knit nodes between packs with no graph references."""
749
pb = ui.ui_factory.nested_progress_bar()
751
return self._do_copy_nodes(nodes, index_map, writer,
756
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb):
757
# for record verification
758
knit_data = _KnitData(None)
759
# plan a readv on each source pack:
761
nodes = sorted(nodes)
762
# how to map this into knit.py - or knit.py into this?
763
# we don't want the typical knit logic, we want grouping by pack
764
# at this point - perhaps a helper library for the following code
765
# duplication points?
767
for index, key, value in nodes:
768
if index not in request_groups:
769
request_groups[index] = []
770
request_groups[index].append((key, value))
772
pb.update("Copied record", record_index, len(nodes))
773
for index, items in request_groups.iteritems():
774
pack_readv_requests = []
775
for key, value in items:
776
# ---- KnitGraphIndex.get_position
777
bits = value[1:].split(' ')
778
offset, length = int(bits[0]), int(bits[1])
779
pack_readv_requests.append((offset, length, (key, value[0])))
780
# linear scan up the pack
781
pack_readv_requests.sort()
783
transport, path = index_map[index]
784
reader = pack.make_readv_reader(transport, path,
785
[offset[0:2] for offset in pack_readv_requests])
786
for (names, read_func), (_1, _2, (key, eol_flag)) in \
787
izip(reader.iter_records(), pack_readv_requests):
788
raw_data = read_func(None)
789
# check the header only
790
df, _ = knit_data._parse_record_header(key[-1], raw_data)
792
pos, size = writer.add_bytes_record(raw_data, names)
793
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
794
pb.update("Copied record", record_index)
797
def _copy_nodes_graph(self, index_map, writer, write_index,
798
readv_group_iter, total_items, output_lines=False):
799
"""Copy knit nodes between packs.
801
:param output_lines: Return lines present in the copied data as
802
an iterator of line,version_id.
804
pb = ui.ui_factory.nested_progress_bar()
806
for result in self._do_copy_nodes_graph(index_map, writer,
807
write_index, output_lines, pb, readv_group_iter, total_items):
810
# Python 2.4 does not permit try:finally: in a generator.
816
def _do_copy_nodes_graph(self, index_map, writer, write_index,
817
output_lines, pb, readv_group_iter, total_items):
818
# for record verification
819
knit_data = _KnitData(None)
820
# for line extraction when requested (inventories only)
822
factory = knit.KnitPlainFactory()
824
pb.update("Copied record", record_index, total_items)
825
for index, readv_vector, node_vector in readv_group_iter:
827
transport, path = index_map[index]
828
reader = pack.make_readv_reader(transport, path, readv_vector)
829
for (names, read_func), (key, eol_flag, references) in \
830
izip(reader.iter_records(), node_vector):
831
raw_data = read_func(None)
834
# read the entire thing
835
content, _ = knit_data._parse_record(version_id, raw_data)
836
if len(references[-1]) == 0:
837
line_iterator = factory.get_fulltext_content(content)
839
line_iterator = factory.get_linedelta_content(content)
840
for line in line_iterator:
841
yield line, version_id
843
# check the header only
844
df, _ = knit_data._parse_record_header(version_id, raw_data)
846
pos, size = writer.add_bytes_record(raw_data, names)
847
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
848
pb.update("Copied record", record_index)
851
def _get_text_nodes(self):
852
text_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
853
self.packs, 'text_index')[0]
854
return text_index_map, self._pack_collection._index_contents(text_index_map,
857
def _least_readv_node_readv(self, nodes):
858
"""Generate request groups for nodes using the least readv's.
860
:param nodes: An iterable of graph index nodes.
861
:return: Total node count and an iterator of the data needed to perform
862
readvs to obtain the data for nodes. Each item yielded by the
863
iterator is a tuple with:
864
index, readv_vector, node_vector. readv_vector is a list ready to
865
hand to the transport readv method, and node_vector is a list of
866
(key, eol_flag, references) for the the node retrieved by the
867
matching readv_vector.
869
# group by pack so we do one readv per pack
870
nodes = sorted(nodes)
873
for index, key, value, references in nodes:
874
if index not in request_groups:
875
request_groups[index] = []
876
request_groups[index].append((key, value, references))
878
for index, items in request_groups.iteritems():
879
pack_readv_requests = []
880
for key, value, references in items:
881
# ---- KnitGraphIndex.get_position
882
bits = value[1:].split(' ')
883
offset, length = int(bits[0]), int(bits[1])
884
pack_readv_requests.append(
885
((offset, length), (key, value[0], references)))
886
# linear scan up the pack to maximum range combining.
887
pack_readv_requests.sort()
888
# split out the readv and the node data.
889
pack_readv = [readv for readv, node in pack_readv_requests]
890
node_vector = [node for readv, node in pack_readv_requests]
891
result.append((index, pack_readv, node_vector))
894
def _log_copied_texts(self):
895
if 'pack' in debug.debug_flags:
896
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
897
time.ctime(), self._pack_collection._upload_transport.base,
898
self.new_pack.random_name,
899
self.new_pack.text_index.key_count(),
900
time.time() - self.new_pack.start_time)
902
def _process_inventory_lines(self, inv_lines):
903
"""Use up the inv_lines generator and setup a text key filter."""
904
repo = self._pack_collection.repo
905
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
906
inv_lines, self.revision_ids)
908
for fileid, file_revids in fileid_revisions.iteritems():
909
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
910
self._text_filter = text_filter
912
def _revision_node_readv(self, revision_nodes):
913
"""Return the total revisions and the readv's to issue.
915
:param revision_nodes: The revision index contents for the packs being
916
incorporated into the new pack.
917
:return: As per _least_readv_node_readv.
919
return self._least_readv_node_readv(revision_nodes)
921
def _use_pack(self, new_pack):
922
"""Return True if new_pack should be used.
924
:param new_pack: The pack that has just been created.
925
:return: True if the pack should be used.
927
return new_pack.data_inserted()
930
class OptimisingPacker(Packer):
931
"""A packer which spends more time to create better disk layouts."""
933
def _revision_node_readv(self, revision_nodes):
934
"""Return the total revisions and the readv's to issue.
936
This sort places revisions in topological order with the ancestors
939
:param revision_nodes: The revision index contents for the packs being
940
incorporated into the new pack.
941
:return: As per _least_readv_node_readv.
943
# build an ancestors dict
946
for index, key, value, references in revision_nodes:
947
ancestors[key] = references[0]
948
by_key[key] = (index, value, references)
949
order = tsort.topo_sort(ancestors)
951
# Single IO is pathological, but it will work as a starting point.
953
for key in reversed(order):
954
index, value, references = by_key[key]
955
# ---- KnitGraphIndex.get_position
956
bits = value[1:].split(' ')
957
offset, length = int(bits[0]), int(bits[1])
959
(index, [(offset, length)], [(key, value[0], references)]))
960
# TODO: combine requests in the same index that are in ascending order.
961
return total, requests
964
class ReconcilePacker(Packer):
965
"""A packer which regenerates indices etc as it copies.
967
This is used by ``bzr reconcile`` to cause parent text pointers to be
971
def _extra_init(self):
972
self._data_changed = False
974
def _process_inventory_lines(self, inv_lines):
975
"""Generate a text key reference map rather for reconciling with."""
976
repo = self._pack_collection.repo
977
refs = repo._find_text_key_references_from_xml_inventory_lines(
979
self._text_refs = refs
980
# during reconcile we:
981
# - convert unreferenced texts to full texts
982
# - correct texts which reference a text not copied to be full texts
983
# - copy all others as-is but with corrected parents.
984
# - so at this point we don't know enough to decide what becomes a full
986
self._text_filter = None
988
def _copy_text_texts(self):
989
"""generate what texts we should have and then copy."""
990
self.pb.update("Copying content texts", 3)
991
# we have three major tasks here:
992
# 1) generate the ideal index
993
repo = self._pack_collection.repo
994
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
996
self.new_pack.revision_index.iter_all_entries()])
997
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
998
# 2) generate a text_nodes list that contains all the deltas that can
999
# be used as-is, with corrected parents.
1002
discarded_nodes = []
1003
NULL_REVISION = _mod_revision.NULL_REVISION
1004
text_index_map, text_nodes = self._get_text_nodes()
1005
for node in text_nodes:
1011
ideal_parents = tuple(ideal_index[node[1]])
1013
discarded_nodes.append(node)
1014
self._data_changed = True
1016
if ideal_parents == (NULL_REVISION,):
1018
if ideal_parents == node[3][0]:
1020
ok_nodes.append(node)
1021
elif ideal_parents[0:1] == node[3][0][0:1]:
1022
# the left most parent is the same, or there are no parents
1023
# today. Either way, we can preserve the representation as
1024
# long as we change the refs to be inserted.
1025
self._data_changed = True
1026
ok_nodes.append((node[0], node[1], node[2],
1027
(ideal_parents, node[3][1])))
1028
self._data_changed = True
1030
# Reinsert this text completely
1031
bad_texts.append((node[1], ideal_parents))
1032
self._data_changed = True
1033
# we're finished with some data.
1036
# 3) bulk copy the ok data
1037
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1038
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1039
self.new_pack.text_index, readv_group_iter, total_items))
1040
# 4) adhoc copy all the other texts.
1041
# We have to topologically insert all texts otherwise we can fail to
1042
# reconcile when parts of a single delta chain are preserved intact,
1043
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1044
# reinserted, and if d3 has incorrect parents it will also be
1045
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1046
# copied), so we will try to delta, but d2 is not currently able to be
1047
# extracted because it's basis d1 is not present. Topologically sorting
1048
# addresses this. The following generates a sort for all the texts that
1049
# are being inserted without having to reference the entire text key
1050
# space (we only topo sort the revisions, which is smaller).
1051
topo_order = tsort.topo_sort(ancestors)
1052
rev_order = dict(zip(topo_order, range(len(topo_order))))
1053
bad_texts.sort(key=lambda key:rev_order[key[0][1]])
1054
transaction = repo.get_transaction()
1055
file_id_index = GraphIndexPrefixAdapter(
1056
self.new_pack.text_index,
1058
add_nodes_callback=self.new_pack.text_index.add_nodes)
1059
knit_index = KnitGraphIndex(file_id_index,
1060
add_callback=file_id_index.add_nodes,
1061
deltas=True, parents=True)
1062
output_knit = knit.KnitVersionedFile('reconcile-texts',
1063
self._pack_collection.transport,
1065
access_method=_PackAccess(
1066
{self.new_pack.text_index:self.new_pack.access_tuple()},
1067
(self.new_pack._writer, self.new_pack.text_index)),
1068
factory=knit.KnitPlainFactory())
1069
for key, parent_keys in bad_texts:
1070
# We refer to the new pack to delta data being output.
1071
# A possible improvement would be to catch errors on short reads
1072
# and only flush then.
1073
self.new_pack.flush()
1075
for parent_key in parent_keys:
1076
if parent_key[0] != key[0]:
1077
# Graph parents must match the fileid
1078
raise errors.BzrError('Mismatched key parent %r:%r' %
1080
parents.append(parent_key[1])
1081
source_weave = repo.weave_store.get_weave(key[0], transaction)
1082
text_lines = source_weave.get_lines(key[1])
1083
# adapt the 'knit' to the current file_id.
1084
file_id_index = GraphIndexPrefixAdapter(
1085
self.new_pack.text_index,
1087
add_nodes_callback=self.new_pack.text_index.add_nodes)
1088
knit_index._graph_index = file_id_index
1089
knit_index._add_callback = file_id_index.add_nodes
1090
output_knit.add_lines_with_ghosts(
1091
key[1], parents, text_lines, random_id=True, check_content=False)
1092
# 5) check that nothing inserted has a reference outside the keyspace.
1093
missing_text_keys = self.new_pack._external_compression_parents_of_texts()
1094
if missing_text_keys:
1095
raise errors.BzrError('Reference to missing compression parents %r'
1097
self._log_copied_texts()
1099
def _use_pack(self, new_pack):
1100
"""Override _use_pack to check for reconcile having changed content."""
1101
# XXX: we might be better checking this at the copy time.
1102
original_inventory_keys = set()
1103
inv_index = self._pack_collection.inventory_index.combined_index
1104
for entry in inv_index.iter_all_entries():
1105
original_inventory_keys.add(entry[1])
1106
new_inventory_keys = set()
1107
for entry in new_pack.inventory_index.iter_all_entries():
1108
new_inventory_keys.add(entry[1])
1109
if new_inventory_keys != original_inventory_keys:
1110
self._data_changed = True
1111
return new_pack.data_inserted() and self._data_changed
1114
class RepositoryPackCollection(object):
1115
"""Management of packs within a repository."""
1117
def __init__(self, repo, transport, index_transport, upload_transport,
1119
"""Create a new RepositoryPackCollection.
1121
:param transport: Addresses the repository base directory
1122
(typically .bzr/repository/).
1123
:param index_transport: Addresses the directory containing indices.
1124
:param upload_transport: Addresses the directory into which packs are written
1125
while they're being created.
1126
:param pack_transport: Addresses the directory of existing complete packs.
1129
self.transport = transport
1130
self._index_transport = index_transport
1131
self._upload_transport = upload_transport
1132
self._pack_transport = pack_transport
1133
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
1136
self._packs_by_name = {}
1137
# the previous pack-names content
1138
self._packs_at_load = None
1139
# when a pack is being created by this object, the state of that pack.
1140
self._new_pack = None
1141
# aggregated revision index data
1142
self.revision_index = AggregateIndex()
1143
self.inventory_index = AggregateIndex()
1144
self.text_index = AggregateIndex()
1145
self.signature_index = AggregateIndex()
1147
def add_pack_to_memory(self, pack):
1148
"""Make a Pack object available to the repository to satisfy queries.
1150
:param pack: A Pack object.
1152
assert pack.name not in self._packs_by_name
1153
self.packs.append(pack)
1154
self._packs_by_name[pack.name] = pack
1155
self.revision_index.add_index(pack.revision_index, pack)
1156
self.inventory_index.add_index(pack.inventory_index, pack)
1157
self.text_index.add_index(pack.text_index, pack)
1158
self.signature_index.add_index(pack.signature_index, pack)
1160
def _add_text_to_weave(self, file_id, revision_id, new_lines, parents,
1161
nostore_sha, random_revid):
1162
file_id_index = GraphIndexPrefixAdapter(
1163
self.text_index.combined_index,
1165
add_nodes_callback=self.text_index.add_callback)
1166
self.repo._text_knit._index._graph_index = file_id_index
1167
self.repo._text_knit._index._add_callback = file_id_index.add_nodes
1168
return self.repo._text_knit.add_lines_with_ghosts(
1169
revision_id, parents, new_lines, nostore_sha=nostore_sha,
1170
random_id=random_revid, check_content=False)[0:2]
1172
def all_packs(self):
1173
"""Return a list of all the Pack objects this repository has.
1175
Note that an in-progress pack being created is not returned.
1177
:return: A list of Pack objects for all the packs in the repository.
1180
for name in self.names():
1181
result.append(self.get_pack_by_name(name))
1185
"""Pack the pack collection incrementally.
1187
This will not attempt global reorganisation or recompression,
1188
rather it will just ensure that the total number of packs does
1189
not grow without bound. It uses the _max_pack_count method to
1190
determine if autopacking is needed, and the pack_distribution
1191
method to determine the number of revisions in each pack.
1193
If autopacking takes place then the packs name collection will have
1194
been flushed to disk - packing requires updating the name collection
1195
in synchronisation with certain steps. Otherwise the names collection
1198
:return: True if packing took place.
1200
# XXX: Should not be needed when the management of indices is sane.
1201
total_revisions = self.revision_index.combined_index.key_count()
1202
total_packs = len(self._names)
1203
if self._max_pack_count(total_revisions) >= total_packs:
1205
# XXX: the following may want to be a class, to pack with a given
1207
mutter('Auto-packing repository %s, which has %d pack files, '
1208
'containing %d revisions into %d packs.', self, total_packs,
1209
total_revisions, self._max_pack_count(total_revisions))
1210
# determine which packs need changing
1211
pack_distribution = self.pack_distribution(total_revisions)
1213
for pack in self.all_packs():
1214
revision_count = pack.get_revision_count()
1215
if revision_count == 0:
1216
# revision less packs are not generated by normal operation,
1217
# only by operations like sign-my-commits, and thus will not
1218
# tend to grow rapdily or without bound like commit containing
1219
# packs do - leave them alone as packing them really should
1220
# group their data with the relevant commit, and that may
1221
# involve rewriting ancient history - which autopack tries to
1222
# avoid. Alternatively we could not group the data but treat
1223
# each of these as having a single revision, and thus add
1224
# one revision for each to the total revision count, to get
1225
# a matching distribution.
1227
existing_packs.append((revision_count, pack))
1228
pack_operations = self.plan_autopack_combinations(
1229
existing_packs, pack_distribution)
1230
self._execute_pack_operations(pack_operations)
1233
def _execute_pack_operations(self, pack_operations, _packer_class=Packer):
1234
"""Execute a series of pack operations.
1236
:param pack_operations: A list of [revision_count, packs_to_combine].
1237
:param _packer_class: The class of packer to use (default: Packer).
1240
for revision_count, packs in pack_operations:
1241
# we may have no-ops from the setup logic
1244
_packer_class(self, packs, '.autopack').pack()
1246
self._remove_pack_from_memory(pack)
1247
# record the newly available packs and stop advertising the old
1249
self._save_pack_names(clear_obsolete_packs=True)
1250
# Move the old packs out of the way now they are no longer referenced.
1251
for revision_count, packs in pack_operations:
1252
self._obsolete_packs(packs)
1254
def lock_names(self):
1255
"""Acquire the mutex around the pack-names index.
1257
This cannot be used in the middle of a read-only transaction on the
1260
self.repo.control_files.lock_write()
1263
"""Pack the pack collection totally."""
1264
self.ensure_loaded()
1265
total_packs = len(self._names)
1267
# This is arguably wrong because we might not be optimal, but for
1268
# now lets leave it in. (e.g. reconcile -> one pack. But not
1271
total_revisions = self.revision_index.combined_index.key_count()
1272
# XXX: the following may want to be a class, to pack with a given
1274
mutter('Packing repository %s, which has %d pack files, '
1275
'containing %d revisions into 1 packs.', self, total_packs,
1277
# determine which packs need changing
1278
pack_distribution = [1]
1279
pack_operations = [[0, []]]
1280
for pack in self.all_packs():
1281
pack_operations[-1][0] += pack.get_revision_count()
1282
pack_operations[-1][1].append(pack)
1283
self._execute_pack_operations(pack_operations, OptimisingPacker)
1285
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1286
"""Plan a pack operation.
1288
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1290
:param pack_distribution: A list with the number of revisions desired
1293
if len(existing_packs) <= len(pack_distribution):
1295
existing_packs.sort(reverse=True)
1296
pack_operations = [[0, []]]
1297
# plan out what packs to keep, and what to reorganise
1298
while len(existing_packs):
1299
# take the largest pack, and if its less than the head of the
1300
# distribution chart we will include its contents in the new pack for
1301
# that position. If its larger, we remove its size from the
1302
# distribution chart
1303
next_pack_rev_count, next_pack = existing_packs.pop(0)
1304
if next_pack_rev_count >= pack_distribution[0]:
1305
# this is already packed 'better' than this, so we can
1306
# not waste time packing it.
1307
while next_pack_rev_count > 0:
1308
next_pack_rev_count -= pack_distribution[0]
1309
if next_pack_rev_count >= 0:
1311
del pack_distribution[0]
1313
# didn't use that entire bucket up
1314
pack_distribution[0] = -next_pack_rev_count
1316
# add the revisions we're going to add to the next output pack
1317
pack_operations[-1][0] += next_pack_rev_count
1318
# allocate this pack to the next pack sub operation
1319
pack_operations[-1][1].append(next_pack)
1320
if pack_operations[-1][0] >= pack_distribution[0]:
1321
# this pack is used up, shift left.
1322
del pack_distribution[0]
1323
pack_operations.append([0, []])
1325
return pack_operations
1327
def ensure_loaded(self):
1328
# NB: if you see an assertion error here, its probably access against
1329
# an unlocked repo. Naughty.
1330
if not self.repo.is_locked():
1331
raise errors.ObjectNotLocked(self.repo)
1332
if self._names is None:
1334
self._packs_at_load = set()
1335
for index, key, value in self._iter_disk_pack_index():
1337
self._names[name] = self._parse_index_sizes(value)
1338
self._packs_at_load.add((key, value))
1339
# populate all the metadata.
1342
def _parse_index_sizes(self, value):
1343
"""Parse a string of index sizes."""
1344
return tuple([int(digits) for digits in value.split(' ')])
1346
def get_pack_by_name(self, name):
1347
"""Get a Pack object by name.
1349
:param name: The name of the pack - e.g. '123456'
1350
:return: A Pack object.
1353
return self._packs_by_name[name]
1355
rev_index = self._make_index(name, '.rix')
1356
inv_index = self._make_index(name, '.iix')
1357
txt_index = self._make_index(name, '.tix')
1358
sig_index = self._make_index(name, '.six')
1359
result = ExistingPack(self._pack_transport, name, rev_index,
1360
inv_index, txt_index, sig_index)
1361
self.add_pack_to_memory(result)
1364
def allocate(self, a_new_pack):
1365
"""Allocate name in the list of packs.
1367
:param a_new_pack: A NewPack instance to be added to the collection of
1368
packs for this repository.
1370
self.ensure_loaded()
1371
if a_new_pack.name in self._names:
1372
raise errors.BzrError(
1373
'Pack %r already exists in %s' % (a_new_pack.name, self))
1374
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1375
self.add_pack_to_memory(a_new_pack)
1377
def _iter_disk_pack_index(self):
1378
"""Iterate over the contents of the pack-names index.
1380
This is used when loading the list from disk, and before writing to
1381
detect updates from others during our write operation.
1382
:return: An iterator of the index contents.
1384
return GraphIndex(self.transport, 'pack-names', None
1385
).iter_all_entries()
1387
def _make_index(self, name, suffix):
1388
size_offset = self._suffix_offsets[suffix]
1389
index_name = name + suffix
1390
index_size = self._names[name][size_offset]
1392
self._index_transport, index_name, index_size)
1394
def _max_pack_count(self, total_revisions):
1395
"""Return the maximum number of packs to use for total revisions.
1397
:param total_revisions: The total number of revisions in the
1400
if not total_revisions:
1402
digits = str(total_revisions)
1404
for digit in digits:
1405
result += int(digit)
1409
"""Provide an order to the underlying names."""
1410
return sorted(self._names.keys())
1412
def _obsolete_packs(self, packs):
1413
"""Move a number of packs which have been obsoleted out of the way.
1415
Each pack and its associated indices are moved out of the way.
1417
Note: for correctness this function should only be called after a new
1418
pack names index has been written without these pack names, and with
1419
the names of packs that contain the data previously available via these
1422
:param packs: The packs to obsolete.
1423
:param return: None.
1426
pack.pack_transport.rename(pack.file_name(),
1427
'../obsolete_packs/' + pack.file_name())
1428
# TODO: Probably needs to know all possible indices for this pack
1429
# - or maybe list the directory and move all indices matching this
1430
# name whether we recognize it or not?
1431
for suffix in ('.iix', '.six', '.tix', '.rix'):
1432
self._index_transport.rename(pack.name + suffix,
1433
'../obsolete_packs/' + pack.name + suffix)
1435
def pack_distribution(self, total_revisions):
1436
"""Generate a list of the number of revisions to put in each pack.
1438
:param total_revisions: The total number of revisions in the
1441
if total_revisions == 0:
1443
digits = reversed(str(total_revisions))
1445
for exponent, count in enumerate(digits):
1446
size = 10 ** exponent
1447
for pos in range(int(count)):
1449
return list(reversed(result))
1451
def _pack_tuple(self, name):
1452
"""Return a tuple with the transport and file name for a pack name."""
1453
return self._pack_transport, name + '.pack'
1455
def _remove_pack_from_memory(self, pack):
1456
"""Remove pack from the packs accessed by this repository.
1458
Only affects memory state, until self._save_pack_names() is invoked.
1460
self._names.pop(pack.name)
1461
self._packs_by_name.pop(pack.name)
1462
self._remove_pack_indices(pack)
1464
def _remove_pack_indices(self, pack):
1465
"""Remove the indices for pack from the aggregated indices."""
1466
self.revision_index.remove_index(pack.revision_index, pack)
1467
self.inventory_index.remove_index(pack.inventory_index, pack)
1468
self.text_index.remove_index(pack.text_index, pack)
1469
self.signature_index.remove_index(pack.signature_index, pack)
1472
"""Clear all cached data."""
1473
# cached revision data
1474
self.repo._revision_knit = None
1475
self.revision_index.clear()
1476
# cached signature data
1477
self.repo._signature_knit = None
1478
self.signature_index.clear()
1479
# cached file text data
1480
self.text_index.clear()
1481
self.repo._text_knit = None
1482
# cached inventory data
1483
self.inventory_index.clear()
1484
# remove the open pack
1485
self._new_pack = None
1486
# information about packs.
1489
self._packs_by_name = {}
1490
self._packs_at_load = None
1492
def _make_index_map(self, index_suffix):
1493
"""Return information on existing indices.
1495
:param suffix: Index suffix added to pack name.
1497
:returns: (pack_map, indices) where indices is a list of GraphIndex
1498
objects, and pack_map is a mapping from those objects to the
1499
pack tuple they describe.
1501
# TODO: stop using this; it creates new indices unnecessarily.
1502
self.ensure_loaded()
1503
suffix_map = {'.rix': 'revision_index',
1504
'.six': 'signature_index',
1505
'.iix': 'inventory_index',
1506
'.tix': 'text_index',
1508
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1509
suffix_map[index_suffix])
1511
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1512
"""Convert a list of packs to an index pack map and index list.
1514
:param packs: The packs list to process.
1515
:param index_attribute: The attribute that the desired index is found
1517
:return: A tuple (map, list) where map contains the dict from
1518
index:pack_tuple, and lsit contains the indices in the same order
1524
index = getattr(pack, index_attribute)
1525
indices.append(index)
1526
pack_map[index] = (pack.pack_transport, pack.file_name())
1527
return pack_map, indices
1529
def _index_contents(self, pack_map, key_filter=None):
1530
"""Get an iterable of the index contents from a pack_map.
1532
:param pack_map: A map from indices to pack details.
1533
:param key_filter: An optional filter to limit the
1536
indices = [index for index in pack_map.iterkeys()]
1537
all_index = CombinedGraphIndex(indices)
1538
if key_filter is None:
1539
return all_index.iter_all_entries()
1541
return all_index.iter_entries(key_filter)
1543
def _unlock_names(self):
1544
"""Release the mutex around the pack-names index."""
1545
self.repo.control_files.unlock()
1547
def _save_pack_names(self, clear_obsolete_packs=False):
1548
"""Save the list of packs.
1550
This will take out the mutex around the pack names list for the
1551
duration of the method call. If concurrent updates have been made, a
1552
three-way merge between the current list and the current in memory list
1555
:param clear_obsolete_packs: If True, clear out the contents of the
1556
obsolete_packs directory.
1560
builder = GraphIndexBuilder()
1561
# load the disk nodes across
1563
for index, key, value in self._iter_disk_pack_index():
1564
disk_nodes.add((key, value))
1565
# do a two-way diff against our original content
1566
current_nodes = set()
1567
for name, sizes in self._names.iteritems():
1569
((name, ), ' '.join(str(size) for size in sizes)))
1570
deleted_nodes = self._packs_at_load - current_nodes
1571
new_nodes = current_nodes - self._packs_at_load
1572
disk_nodes.difference_update(deleted_nodes)
1573
disk_nodes.update(new_nodes)
1574
# TODO: handle same-name, index-size-changes here -
1575
# e.g. use the value from disk, not ours, *unless* we're the one
1577
for key, value in disk_nodes:
1578
builder.add_node(key, value)
1579
self.transport.put_file('pack-names', builder.finish(),
1580
mode=self.repo.control_files._file_mode)
1581
# move the baseline forward
1582
self._packs_at_load = disk_nodes
1583
# now clear out the obsolete packs directory
1584
if clear_obsolete_packs:
1585
self.transport.clone('obsolete_packs').delete_multi(
1586
self.transport.list_dir('obsolete_packs'))
1588
self._unlock_names()
1589
# synchronise the memory packs list with what we just wrote:
1590
new_names = dict(disk_nodes)
1591
# drop no longer present nodes
1592
for pack in self.all_packs():
1593
if (pack.name,) not in new_names:
1594
self._remove_pack_from_memory(pack)
1595
# add new nodes/refresh existing ones
1596
for key, value in disk_nodes:
1598
sizes = self._parse_index_sizes(value)
1599
if name in self._names:
1601
if sizes != self._names[name]:
1602
# the pack for name has had its indices replaced - rare but
1603
# important to handle. XXX: probably can never happen today
1604
# because the three-way merge code above does not handle it
1605
# - you may end up adding the same key twice to the new
1606
# disk index because the set values are the same, unless
1607
# the only index shows up as deleted by the set difference
1608
# - which it may. Until there is a specific test for this,
1609
# assume its broken. RBC 20071017.
1610
self._remove_pack_from_memory(self.get_pack_by_name(name))
1611
self._names[name] = sizes
1612
self.get_pack_by_name(name)
1615
self._names[name] = sizes
1616
self.get_pack_by_name(name)
1618
def _start_write_group(self):
1619
# Do not permit preparation for writing if we're not in a 'write lock'.
1620
if not self.repo.is_write_locked():
1621
raise errors.NotWriteLocked(self)
1622
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1623
self._pack_transport, upload_suffix='.pack',
1624
file_mode=self.repo.control_files._file_mode)
1625
# allow writing: queue writes to a new index
1626
self.revision_index.add_writable_index(self._new_pack.revision_index,
1628
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
1630
self.text_index.add_writable_index(self._new_pack.text_index,
1632
self.signature_index.add_writable_index(self._new_pack.signature_index,
1635
# reused revision and signature knits may need updating
1637
# "Hysterical raisins. client code in bzrlib grabs those knits outside
1638
# of write groups and then mutates it inside the write group."
1639
if self.repo._revision_knit is not None:
1640
self.repo._revision_knit._index._add_callback = \
1641
self.revision_index.add_callback
1642
if self.repo._signature_knit is not None:
1643
self.repo._signature_knit._index._add_callback = \
1644
self.signature_index.add_callback
1645
# create a reused knit object for text addition in commit.
1646
self.repo._text_knit = self.repo.weave_store.get_weave_or_empty(
1649
def _abort_write_group(self):
1650
# FIXME: just drop the transient index.
1651
# forget what names there are
1652
if self._new_pack is not None:
1653
self._new_pack.abort()
1654
self._remove_pack_indices(self._new_pack)
1655
self._new_pack = None
1656
self.repo._text_knit = None
1658
def _commit_write_group(self):
1659
self._remove_pack_indices(self._new_pack)
1660
if self._new_pack.data_inserted():
1661
# get all the data to disk and read to use
1662
self._new_pack.finish()
1663
self.allocate(self._new_pack)
1664
self._new_pack = None
1665
if not self.autopack():
1666
# when autopack takes no steps, the names list is still
1668
self._save_pack_names()
1670
self._new_pack.abort()
1671
self._new_pack = None
1672
self.repo._text_knit = None
1675
class KnitPackRevisionStore(KnitRevisionStore):
1676
"""An object to adapt access from RevisionStore's to use KnitPacks.
1678
This class works by replacing the original RevisionStore.
1679
We need to do this because the KnitPackRevisionStore is less
1680
isolated in its layering - it uses services from the repo.
1683
def __init__(self, repo, transport, revisionstore):
1684
"""Create a KnitPackRevisionStore on repo with revisionstore.
1686
This will store its state in the Repository, use the
1687
indices to provide a KnitGraphIndex,
1688
and at the end of transactions write new indices.
1690
KnitRevisionStore.__init__(self, revisionstore.versioned_file_store)
1692
self._serializer = revisionstore._serializer
1693
self.transport = transport
1695
def get_revision_file(self, transaction):
1696
"""Get the revision versioned file object."""
1697
if getattr(self.repo, '_revision_knit', None) is not None:
1698
return self.repo._revision_knit
1699
self.repo._pack_collection.ensure_loaded()
1700
add_callback = self.repo._pack_collection.revision_index.add_callback
1701
# setup knit specific objects
1702
knit_index = KnitGraphIndex(
1703
self.repo._pack_collection.revision_index.combined_index,
1704
add_callback=add_callback)
1705
self.repo._revision_knit = knit.KnitVersionedFile(
1706
'revisions', self.transport.clone('..'),
1707
self.repo.control_files._file_mode,
1709
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1710
access_method=self.repo._pack_collection.revision_index.knit_access)
1711
return self.repo._revision_knit
1713
def get_signature_file(self, transaction):
1714
"""Get the signature versioned file object."""
1715
if getattr(self.repo, '_signature_knit', None) is not None:
1716
return self.repo._signature_knit
1717
self.repo._pack_collection.ensure_loaded()
1718
add_callback = self.repo._pack_collection.signature_index.add_callback
1719
# setup knit specific objects
1720
knit_index = KnitGraphIndex(
1721
self.repo._pack_collection.signature_index.combined_index,
1722
add_callback=add_callback, parents=False)
1723
self.repo._signature_knit = knit.KnitVersionedFile(
1724
'signatures', self.transport.clone('..'),
1725
self.repo.control_files._file_mode,
1727
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1728
access_method=self.repo._pack_collection.signature_index.knit_access)
1729
return self.repo._signature_knit
1732
class KnitPackTextStore(VersionedFileStore):
1733
"""Presents a TextStore abstraction on top of packs.
1735
This class works by replacing the original VersionedFileStore.
1736
We need to do this because the KnitPackRevisionStore is less
1737
isolated in its layering - it uses services from the repo and shares them
1738
with all the data written in a single write group.
1741
def __init__(self, repo, transport, weavestore):
1742
"""Create a KnitPackTextStore on repo with weavestore.
1744
This will store its state in the Repository, use the
1745
indices FileNames to provide a KnitGraphIndex,
1746
and at the end of transactions write new indices.
1748
# don't call base class constructor - it's not suitable.
1749
# no transient data stored in the transaction
1751
self._precious = False
1753
self.transport = transport
1754
self.weavestore = weavestore
1755
# XXX for check() which isn't updated yet
1756
self._transport = weavestore._transport
1758
def get_weave_or_empty(self, file_id, transaction):
1759
"""Get a 'Knit' backed by the .tix indices.
1761
The transaction parameter is ignored.
1763
self.repo._pack_collection.ensure_loaded()
1764
add_callback = self.repo._pack_collection.text_index.add_callback
1765
# setup knit specific objects
1766
file_id_index = GraphIndexPrefixAdapter(
1767
self.repo._pack_collection.text_index.combined_index,
1768
(file_id, ), 1, add_nodes_callback=add_callback)
1769
knit_index = KnitGraphIndex(file_id_index,
1770
add_callback=file_id_index.add_nodes,
1771
deltas=True, parents=True)
1772
return knit.KnitVersionedFile('text:' + file_id,
1773
self.transport.clone('..'),
1776
access_method=self.repo._pack_collection.text_index.knit_access,
1777
factory=knit.KnitPlainFactory())
1779
get_weave = get_weave_or_empty
1782
"""Generate a list of the fileids inserted, for use by check."""
1783
self.repo._pack_collection.ensure_loaded()
1785
for index, key, value, refs in \
1786
self.repo._pack_collection.text_index.combined_index.iter_all_entries():
1791
class InventoryKnitThunk(object):
1792
"""An object to manage thunking get_inventory_weave to pack based knits."""
1794
def __init__(self, repo, transport):
1795
"""Create an InventoryKnitThunk for repo at transport.
1797
This will store its state in the Repository, use the
1798
indices FileNames to provide a KnitGraphIndex,
1799
and at the end of transactions write a new index..
1802
self.transport = transport
1804
def get_weave(self):
1805
"""Get a 'Knit' that contains inventory data."""
1806
self.repo._pack_collection.ensure_loaded()
1807
add_callback = self.repo._pack_collection.inventory_index.add_callback
1808
# setup knit specific objects
1809
knit_index = KnitGraphIndex(
1810
self.repo._pack_collection.inventory_index.combined_index,
1811
add_callback=add_callback, deltas=True, parents=True)
1812
return knit.KnitVersionedFile(
1813
'inventory', self.transport.clone('..'),
1814
self.repo.control_files._file_mode,
1816
index=knit_index, delta=True, factory=knit.KnitPlainFactory(),
1817
access_method=self.repo._pack_collection.inventory_index.knit_access)
1820
class KnitPackRepository(KnitRepository):
1821
"""Experimental graph-knit using repository."""
1823
def __init__(self, _format, a_bzrdir, control_files, _revision_store,
1824
control_store, text_store, _commit_builder_class, _serializer):
1825
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
1826
_revision_store, control_store, text_store, _commit_builder_class,
1828
index_transport = control_files._transport.clone('indices')
1829
self._pack_collection = RepositoryPackCollection(self, control_files._transport,
1831
control_files._transport.clone('upload'),
1832
control_files._transport.clone('packs'))
1833
self._revision_store = KnitPackRevisionStore(self, index_transport, self._revision_store)
1834
self.weave_store = KnitPackTextStore(self, index_transport, self.weave_store)
1835
self._inv_thunk = InventoryKnitThunk(self, index_transport)
1836
# True when the repository object is 'write locked' (as opposed to the
1837
# physical lock only taken out around changes to the pack-names list.)
1838
# Another way to represent this would be a decorator around the control
1839
# files object that presents logical locks as physical ones - if this
1840
# gets ugly consider that alternative design. RBC 20071011
1841
self._write_lock_count = 0
1842
self._transaction = None
1844
self._reconcile_does_inventory_gc = True
1845
self._reconcile_fixes_text_parents = True
1846
self._reconcile_backsup_inventory = False
1848
def _abort_write_group(self):
1849
self._pack_collection._abort_write_group()
1851
def _find_inconsistent_revision_parents(self):
1852
"""Find revisions with incorrectly cached parents.
1854
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1855
parents-in-revision).
1857
if not self.is_locked():
1858
raise errors.ObjectNotLocked(self)
1859
pb = ui.ui_factory.nested_progress_bar()
1862
revision_nodes = self._pack_collection.revision_index \
1863
.combined_index.iter_all_entries()
1864
index_positions = []
1865
# Get the cached index values for all revisions, and also the location
1866
# in each index of the revision text so we can perform linear IO.
1867
for index, key, value, refs in revision_nodes:
1868
pos, length = value[1:].split(' ')
1869
index_positions.append((index, int(pos), key[0],
1870
tuple(parent[0] for parent in refs[0])))
1871
pb.update("Reading revision index.", 0, 0)
1872
index_positions.sort()
1873
batch_count = len(index_positions) / 1000 + 1
1874
pb.update("Checking cached revision graph.", 0, batch_count)
1875
for offset in xrange(batch_count):
1876
pb.update("Checking cached revision graph.", offset)
1877
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1880
rev_ids = [item[2] for item in to_query]
1881
revs = self.get_revisions(rev_ids)
1882
for revision, item in zip(revs, to_query):
1883
index_parents = item[3]
1884
rev_parents = tuple(revision.parent_ids)
1885
if index_parents != rev_parents:
1886
result.append((revision.revision_id, index_parents, rev_parents))
1891
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
1892
def get_parents(self, revision_ids):
1893
"""See graph._StackedParentsProvider.get_parents."""
1894
parent_map = self.get_parent_map(revision_ids)
1895
return [parent_map.get(r, None) for r in revision_ids]
1897
def get_parent_map(self, keys):
1898
"""See graph._StackedParentsProvider.get_parent_map
1900
This implementation accesses the combined revision index to provide
1903
self._pack_collection.ensure_loaded()
1904
index = self._pack_collection.revision_index.combined_index
1906
if _mod_revision.NULL_REVISION in keys:
1907
keys.discard(_mod_revision.NULL_REVISION)
1908
found_parents = {_mod_revision.NULL_REVISION:()}
1911
search_keys = set((revision_id,) for revision_id in keys)
1912
for index, key, value, refs in index.iter_entries(search_keys):
1915
parents = (_mod_revision.NULL_REVISION,)
1917
parents = tuple(parent[0] for parent in parents)
1918
found_parents[key[0]] = parents
1919
return found_parents
1921
@symbol_versioning.deprecated_method(symbol_versioning.one_four)
1923
def get_revision_graph(self, revision_id=None):
1924
"""Return a dictionary containing the revision graph.
1926
:param revision_id: The revision_id to get a graph from. If None, then
1927
the entire revision graph is returned. This is a deprecated mode of
1928
operation and will be removed in the future.
1929
:return: a dictionary of revision_id->revision_parents_list.
1931
if 'evil' in debug.debug_flags:
1933
"get_revision_graph scales with size of history.")
1934
# special case NULL_REVISION
1935
if revision_id == _mod_revision.NULL_REVISION:
1937
if revision_id is None:
1938
revision_vf = self._get_revision_vf()
1939
return revision_vf.get_graph()
1940
g = self.get_graph()
1941
first = g.get_parent_map([revision_id])
1942
if revision_id not in first:
1943
raise errors.NoSuchRevision(self, revision_id)
1947
NULL_REVISION = _mod_revision.NULL_REVISION
1948
ghosts = set([NULL_REVISION])
1949
for rev_id, parent_ids in g.iter_ancestry([revision_id]):
1950
if parent_ids is None: # This is a ghost
1953
ancestry[rev_id] = parent_ids
1954
for p in parent_ids:
1956
children[p].append(rev_id)
1958
children[p] = [rev_id]
1960
if NULL_REVISION in ancestry:
1961
del ancestry[NULL_REVISION]
1963
# Find all nodes that reference a ghost, and filter the ghosts out
1964
# of their parent lists. To preserve the order of parents, and
1965
# avoid double filtering nodes, we just find all children first,
1967
children_of_ghosts = set()
1968
for ghost in ghosts:
1969
children_of_ghosts.update(children[ghost])
1971
for child in children_of_ghosts:
1972
ancestry[child] = tuple(p for p in ancestry[child]
1976
def has_revisions(self, revision_ids):
1977
"""See Repository.has_revisions()."""
1978
revision_ids = set(revision_ids)
1979
result = revision_ids.intersection(
1980
set([None, _mod_revision.NULL_REVISION]))
1981
revision_ids.difference_update(result)
1982
index = self._pack_collection.revision_index.combined_index
1983
keys = [(revision_id,) for revision_id in revision_ids]
1984
result.update(node[1][0] for node in index.iter_entries(keys))
1987
def _make_parents_provider(self):
1988
return graph.CachingParentsProvider(self)
1990
def _refresh_data(self):
1991
if self._write_lock_count == 1 or (
1992
self.control_files._lock_count == 1 and
1993
self.control_files._lock_mode == 'r'):
1994
# forget what names there are
1995
self._pack_collection.reset()
1996
# XXX: Better to do an in-memory merge when acquiring a new lock -
1997
# factor out code from _save_pack_names.
1998
self._pack_collection.ensure_loaded()
2000
def _start_write_group(self):
2001
self._pack_collection._start_write_group()
2003
def _commit_write_group(self):
2004
return self._pack_collection._commit_write_group()
2006
def get_inventory_weave(self):
2007
return self._inv_thunk.get_weave()
2009
def get_transaction(self):
2010
if self._write_lock_count:
2011
return self._transaction
2013
return self.control_files.get_transaction()
2015
def is_locked(self):
2016
return self._write_lock_count or self.control_files.is_locked()
2018
def is_write_locked(self):
2019
return self._write_lock_count
2021
def lock_write(self, token=None):
2022
if not self._write_lock_count and self.is_locked():
2023
raise errors.ReadOnlyError(self)
2024
self._write_lock_count += 1
2025
if self._write_lock_count == 1:
2026
from bzrlib import transactions
2027
self._transaction = transactions.WriteTransaction()
2028
self._refresh_data()
2030
def lock_read(self):
2031
if self._write_lock_count:
2032
self._write_lock_count += 1
2034
self.control_files.lock_read()
2035
self._refresh_data()
2037
def leave_lock_in_place(self):
2038
# not supported - raise an error
2039
raise NotImplementedError(self.leave_lock_in_place)
2041
def dont_leave_lock_in_place(self):
2042
# not supported - raise an error
2043
raise NotImplementedError(self.dont_leave_lock_in_place)
2047
"""Compress the data within the repository.
2049
This will pack all the data to a single pack. In future it may
2050
recompress deltas or do other such expensive operations.
2052
self._pack_collection.pack()
2055
def reconcile(self, other=None, thorough=False):
2056
"""Reconcile this repository."""
2057
from bzrlib.reconcile import PackReconciler
2058
reconciler = PackReconciler(self, thorough=thorough)
2059
reconciler.reconcile()
2063
if self._write_lock_count == 1 and self._write_group is not None:
2064
self.abort_write_group()
2065
self._transaction = None
2066
self._write_lock_count = 0
2067
raise errors.BzrError(
2068
'Must end write group before releasing write lock on %s'
2070
if self._write_lock_count:
2071
self._write_lock_count -= 1
2072
if not self._write_lock_count:
2073
transaction = self._transaction
2074
self._transaction = None
2075
transaction.finish()
2077
self.control_files.unlock()
2080
class RepositoryFormatPack(MetaDirRepositoryFormat):
2081
"""Format logic for pack structured repositories.
2083
This repository format has:
2084
- a list of packs in pack-names
2085
- packs in packs/NAME.pack
2086
- indices in indices/NAME.{iix,six,tix,rix}
2087
- knit deltas in the packs, knit indices mapped to the indices.
2088
- thunk objects to support the knits programming API.
2089
- a format marker of its own
2090
- an optional 'shared-storage' flag
2091
- an optional 'no-working-trees' flag
2095
# Set this attribute in derived classes to control the repository class
2096
# created by open and initialize.
2097
repository_class = None
2098
# Set this attribute in derived classes to control the
2099
# _commit_builder_class that the repository objects will have passed to
2100
# their constructor.
2101
_commit_builder_class = None
2102
# Set this attribute in derived clases to control the _serializer that the
2103
# repository objects will have passed to their constructor.
2105
# External references are not supported in pack repositories yet.
2106
supports_external_lookups = False
2108
def _get_control_store(self, repo_transport, control_files):
2109
"""Return the control store for this repository."""
2110
return VersionedFileStore(
2113
file_mode=control_files._file_mode,
2114
versionedfile_class=knit.make_file_knit,
2115
versionedfile_kwargs={'factory': knit.KnitPlainFactory()},
2118
def _get_revision_store(self, repo_transport, control_files):
2119
"""See RepositoryFormat._get_revision_store()."""
2120
versioned_file_store = VersionedFileStore(
2122
file_mode=control_files._file_mode,
2125
versionedfile_class=knit.make_file_knit,
2126
versionedfile_kwargs={'delta': False,
2127
'factory': knit.KnitPlainFactory(),
2131
return KnitRevisionStore(versioned_file_store)
2133
def _get_text_store(self, transport, control_files):
2134
"""See RepositoryFormat._get_text_store()."""
2135
return self._get_versioned_file_store('knits',
2138
versionedfile_class=knit.make_file_knit,
2139
versionedfile_kwargs={
2140
'create_parent_dir': True,
2141
'delay_create': True,
2142
'dir_mode': control_files._dir_mode,
2146
def initialize(self, a_bzrdir, shared=False):
2147
"""Create a pack based repository.
2149
:param a_bzrdir: bzrdir to contain the new repository; must already
2151
:param shared: If true the repository will be initialized as a shared
2154
mutter('creating repository in %s.', a_bzrdir.transport.base)
2155
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
2156
builder = GraphIndexBuilder()
2157
files = [('pack-names', builder.finish())]
2158
utf8_files = [('format', self.get_format_string())]
2160
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
2161
return self.open(a_bzrdir=a_bzrdir, _found=True)
2163
def open(self, a_bzrdir, _found=False, _override_transport=None):
2164
"""See RepositoryFormat.open().
2166
:param _override_transport: INTERNAL USE ONLY. Allows opening the
2167
repository at a slightly different url
2168
than normal. I.e. during 'upgrade'.
2171
format = RepositoryFormat.find_format(a_bzrdir)
2172
assert format.__class__ == self.__class__
2173
if _override_transport is not None:
2174
repo_transport = _override_transport
2176
repo_transport = a_bzrdir.get_repository_transport(None)
2177
control_files = lockable_files.LockableFiles(repo_transport,
2178
'lock', lockdir.LockDir)
2179
text_store = self._get_text_store(repo_transport, control_files)
2180
control_store = self._get_control_store(repo_transport, control_files)
2181
_revision_store = self._get_revision_store(repo_transport, control_files)
2182
return self.repository_class(_format=self,
2184
control_files=control_files,
2185
_revision_store=_revision_store,
2186
control_store=control_store,
2187
text_store=text_store,
2188
_commit_builder_class=self._commit_builder_class,
2189
_serializer=self._serializer)
2192
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2193
"""A no-subtrees parameterized Pack repository.
2195
This format was introduced in 0.92.
2198
repository_class = KnitPackRepository
2199
_commit_builder_class = PackCommitBuilder
2200
_serializer = xml5.serializer_v5
2202
def _get_matching_bzrdir(self):
2203
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2205
def _ignore_setting_bzrdir(self, format):
2208
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2210
def get_format_string(self):
2211
"""See RepositoryFormat.get_format_string()."""
2212
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2214
def get_format_description(self):
2215
"""See RepositoryFormat.get_format_description()."""
2216
return "Packs containing knits without subtree support"
2218
def check_conversion_target(self, target_format):
2222
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2223
"""A subtrees parameterized Pack repository.
2225
This repository format uses the xml7 serializer to get:
2226
- support for recording full info about the tree root
2227
- support for recording tree-references
2229
This format was introduced in 0.92.
2232
repository_class = KnitPackRepository
2233
_commit_builder_class = PackRootCommitBuilder
2234
rich_root_data = True
2235
supports_tree_reference = True
2236
_serializer = xml7.serializer_v7
2238
def _get_matching_bzrdir(self):
2239
return bzrdir.format_registry.make_bzrdir(
2240
'pack-0.92-subtree')
2242
def _ignore_setting_bzrdir(self, format):
2245
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2247
def check_conversion_target(self, target_format):
2248
if not target_format.rich_root_data:
2249
raise errors.BadConversionTarget(
2250
'Does not support rich root data.', target_format)
2251
if not getattr(target_format, 'supports_tree_reference', False):
2252
raise errors.BadConversionTarget(
2253
'Does not support nested trees', target_format)
2255
def get_format_string(self):
2256
"""See RepositoryFormat.get_format_string()."""
2257
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2259
def get_format_description(self):
2260
"""See RepositoryFormat.get_format_description()."""
2261
return "Packs containing knits with subtree support\n"
2264
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2265
"""A rich-root, no subtrees parameterized Pack repository.
2267
This repository format uses the xml6 serializer to get:
2268
- support for recording full info about the tree root
2270
This format was introduced in 1.0.
2273
repository_class = KnitPackRepository
2274
_commit_builder_class = PackRootCommitBuilder
2275
rich_root_data = True
2276
supports_tree_reference = False
2277
_serializer = xml6.serializer_v6
2279
def _get_matching_bzrdir(self):
2280
return bzrdir.format_registry.make_bzrdir(
2283
def _ignore_setting_bzrdir(self, format):
2286
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2288
def check_conversion_target(self, target_format):
2289
if not target_format.rich_root_data:
2290
raise errors.BadConversionTarget(
2291
'Does not support rich root data.', target_format)
2293
def get_format_string(self):
2294
"""See RepositoryFormat.get_format_string()."""
2295
return ("Bazaar pack repository format 1 with rich root"
2296
" (needs bzr 1.0)\n")
2298
def get_format_description(self):
2299
"""See RepositoryFormat.get_format_description()."""
2300
return "Packs containing knits with rich root support\n"
2303
class RepositoryFormatPackDevelopment0(RepositoryFormatPack):
2304
"""A no-subtrees development repository.
2306
This format should be retained until the second release after bzr 1.0.
2308
No changes to the disk behaviour from pack-0.92.
2311
repository_class = KnitPackRepository
2312
_commit_builder_class = PackCommitBuilder
2313
_serializer = xml5.serializer_v5
2315
def _get_matching_bzrdir(self):
2316
return bzrdir.format_registry.make_bzrdir('development0')
2318
def _ignore_setting_bzrdir(self, format):
2321
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2323
def get_format_string(self):
2324
"""See RepositoryFormat.get_format_string()."""
2325
return "Bazaar development format 0 (needs bzr.dev from before 1.3)\n"
2327
def get_format_description(self):
2328
"""See RepositoryFormat.get_format_description()."""
2329
return ("Development repository format, currently the same as "
2332
def check_conversion_target(self, target_format):
2336
class RepositoryFormatPackDevelopment0Subtree(RepositoryFormatPack):
2337
"""A subtrees development repository.
2339
This format should be retained until the second release after bzr 1.0.
2341
No changes to the disk behaviour from pack-0.92-subtree.
2344
repository_class = KnitPackRepository
2345
_commit_builder_class = PackRootCommitBuilder
2346
rich_root_data = True
2347
supports_tree_reference = True
2348
_serializer = xml7.serializer_v7
2350
def _get_matching_bzrdir(self):
2351
return bzrdir.format_registry.make_bzrdir(
2352
'development0-subtree')
2354
def _ignore_setting_bzrdir(self, format):
2357
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2359
def check_conversion_target(self, target_format):
2360
if not target_format.rich_root_data:
2361
raise errors.BadConversionTarget(
2362
'Does not support rich root data.', target_format)
2363
if not getattr(target_format, 'supports_tree_reference', False):
2364
raise errors.BadConversionTarget(
2365
'Does not support nested trees', target_format)
2367
def get_format_string(self):
2368
"""See RepositoryFormat.get_format_string()."""
2369
return ("Bazaar development format 0 with subtree support "
2370
"(needs bzr.dev from before 1.3)\n")
2372
def get_format_description(self):
2373
"""See RepositoryFormat.get_format_description()."""
2374
return ("Development repository format, currently the same as "
2375
"pack-0.92-subtree\n")
added
removed
bzrlib/repofmt/pack_repo.py
