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
29
from bzrlib.graph import Graph
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
56
from bzrlib.decorators import needs_read_lock, needs_write_lock
57
from bzrlib.repofmt.knitrepo import KnitRepository
58
from bzrlib.repository import (
61
MetaDirRepositoryFormat,
64
import bzrlib.revision as _mod_revision
65
from bzrlib.store.revision.knit import KnitRevisionStore
66
from bzrlib.store.versioned import VersionedFileStore
67
from bzrlib.trace import mutter, note, warning
70
class PackCommitBuilder(CommitBuilder):
71
"""A subclass of CommitBuilder to add texts with pack semantics.
73
Specifically this uses one knit object rather than one knit object per
74
added text, reducing memory and object pressure.
77
def __init__(self, repository, parents, config, timestamp=None,
78
timezone=None, committer=None, revprops=None,
80
CommitBuilder.__init__(self, repository, parents, config,
81
timestamp=timestamp, timezone=timezone, committer=committer,
82
revprops=revprops, revision_id=revision_id)
83
self._file_graph = Graph(
84
repository._pack_collection.text_index.combined_index)
86
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
87
return self.repository._pack_collection._add_text_to_weave(file_id,
88
self._new_revision_id, new_lines, parents, nostore_sha,
91
def _heads(self, file_id, revision_ids):
92
keys = [(file_id, revision_id) for revision_id in revision_ids]
93
return set([key[1] for key in self._file_graph.heads(keys)])
96
class PackRootCommitBuilder(RootCommitBuilder):
97
"""A subclass of RootCommitBuilder to add texts with pack semantics.
99
Specifically this uses one knit object rather than one knit object per
100
added text, reducing memory and object pressure.
103
def __init__(self, repository, parents, config, timestamp=None,
104
timezone=None, committer=None, revprops=None,
106
CommitBuilder.__init__(self, repository, parents, config,
107
timestamp=timestamp, timezone=timezone, committer=committer,
108
revprops=revprops, revision_id=revision_id)
109
self._file_graph = Graph(
110
repository._pack_collection.text_index.combined_index)
112
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
113
return self.repository._pack_collection._add_text_to_weave(file_id,
114
self._new_revision_id, new_lines, parents, nostore_sha,
117
def _heads(self, file_id, revision_ids):
118
keys = [(file_id, revision_id) for revision_id in revision_ids]
119
return set([key[1] for key in self._file_graph.heads(keys)])
123
"""An in memory proxy for a pack and its indices.
125
This is a base class that is not directly used, instead the classes
126
ExistingPack and NewPack are used.
129
def __init__(self, revision_index, inventory_index, text_index,
131
"""Create a pack instance.
133
:param revision_index: A GraphIndex for determining what revisions are
134
present in the Pack and accessing the locations of their texts.
135
:param inventory_index: A GraphIndex for determining what inventories are
136
present in the Pack and accessing the locations of their
138
:param text_index: A GraphIndex for determining what file texts
139
are present in the pack and accessing the locations of their
140
texts/deltas (via (fileid, revisionid) tuples).
141
:param revision_index: A GraphIndex for determining what signatures are
142
present in the Pack and accessing the locations of their texts.
144
self.revision_index = revision_index
145
self.inventory_index = inventory_index
146
self.text_index = text_index
147
self.signature_index = signature_index
149
def access_tuple(self):
150
"""Return a tuple (transport, name) for the pack content."""
151
return self.pack_transport, self.file_name()
154
"""Get the file name for the pack on disk."""
155
return self.name + '.pack'
157
def get_revision_count(self):
158
return self.revision_index.key_count()
160
def inventory_index_name(self, name):
161
"""The inv index is the name + .iix."""
162
return self.index_name('inventory', name)
164
def revision_index_name(self, name):
165
"""The revision index is the name + .rix."""
166
return self.index_name('revision', name)
168
def signature_index_name(self, name):
169
"""The signature index is the name + .six."""
170
return self.index_name('signature', name)
172
def text_index_name(self, name):
173
"""The text index is the name + .tix."""
174
return self.index_name('text', name)
177
class ExistingPack(Pack):
178
"""An in memory proxy for an existing .pack and its disk indices."""
180
def __init__(self, pack_transport, name, revision_index, inventory_index,
181
text_index, signature_index):
182
"""Create an ExistingPack object.
184
:param pack_transport: The transport where the pack file resides.
185
:param name: The name of the pack on disk in the pack_transport.
187
Pack.__init__(self, revision_index, inventory_index, text_index,
190
self.pack_transport = pack_transport
191
assert None not in (revision_index, inventory_index, text_index,
192
signature_index, name, pack_transport)
194
def __eq__(self, other):
195
return self.__dict__ == other.__dict__
197
def __ne__(self, other):
198
return not self.__eq__(other)
201
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
202
id(self), self.transport, self.name)
206
"""An in memory proxy for a pack which is being created."""
208
# A map of index 'type' to the file extension and position in the
210
index_definitions = {
211
'revision': ('.rix', 0),
212
'inventory': ('.iix', 1),
214
'signature': ('.six', 3),
217
def __init__(self, upload_transport, index_transport, pack_transport,
218
upload_suffix='', file_mode=None):
219
"""Create a NewPack instance.
221
:param upload_transport: A writable transport for the pack to be
222
incrementally uploaded to.
223
:param index_transport: A writable transport for the pack's indices to
224
be written to when the pack is finished.
225
:param pack_transport: A writable transport for the pack to be renamed
226
to when the upload is complete. This *must* be the same as
227
upload_transport.clone('../packs').
228
:param upload_suffix: An optional suffix to be given to any temporary
229
files created during the pack creation. e.g '.autopack'
230
:param file_mode: An optional file mode to create the new files with.
232
# The relative locations of the packs are constrained, but all are
233
# passed in because the caller has them, so as to avoid object churn.
235
# Revisions: parents list, no text compression.
236
InMemoryGraphIndex(reference_lists=1),
237
# Inventory: We want to map compression only, but currently the
238
# knit code hasn't been updated enough to understand that, so we
239
# have a regular 2-list index giving parents and compression
241
InMemoryGraphIndex(reference_lists=2),
242
# Texts: compression and per file graph, for all fileids - so two
243
# reference lists and two elements in the key tuple.
244
InMemoryGraphIndex(reference_lists=2, key_elements=2),
245
# Signatures: Just blobs to store, no compression, no parents
247
InMemoryGraphIndex(reference_lists=0),
249
# where should the new pack be opened
250
self.upload_transport = upload_transport
251
# where are indices written out to
252
self.index_transport = index_transport
253
# where is the pack renamed to when it is finished?
254
self.pack_transport = pack_transport
255
# What file mode to upload the pack and indices with.
256
self._file_mode = file_mode
257
# tracks the content written to the .pack file.
258
self._hash = md5.new()
259
# a four-tuple with the length in bytes of the indices, once the pack
260
# is finalised. (rev, inv, text, sigs)
261
self.index_sizes = None
262
# How much data to cache when writing packs. Note that this is not
263
# synchronised with reads, because it's not in the transport layer, so
264
# is not safe unless the client knows it won't be reading from the pack
266
self._cache_limit = 0
267
# the temporary pack file name.
268
self.random_name = rand_chars(20) + upload_suffix
269
# when was this pack started ?
270
self.start_time = time.time()
271
# open an output stream for the data added to the pack.
272
self.write_stream = self.upload_transport.open_write_stream(
273
self.random_name, mode=self._file_mode)
274
if 'pack' in debug.debug_flags:
275
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
276
time.ctime(), self.upload_transport.base, self.random_name,
277
time.time() - self.start_time)
278
# A list of byte sequences to be written to the new pack, and the
279
# aggregate size of them. Stored as a list rather than separate
280
# variables so that the _write_data closure below can update them.
281
self._buffer = [[], 0]
282
# create a callable for adding data
284
# robertc says- this is a closure rather than a method on the object
285
# so that the variables are locals, and faster than accessing object
287
def _write_data(bytes, flush=False, _buffer=self._buffer,
288
_write=self.write_stream.write, _update=self._hash.update):
289
_buffer[0].append(bytes)
290
_buffer[1] += len(bytes)
292
if _buffer[1] > self._cache_limit or flush:
293
bytes = ''.join(_buffer[0])
297
# expose this on self, for the occasion when clients want to add data.
298
self._write_data = _write_data
299
# a pack writer object to serialise pack records.
300
self._writer = pack.ContainerWriter(self._write_data)
302
# what state is the pack in? (open, finished, aborted)
306
"""Cancel creating this pack."""
307
self._state = 'aborted'
308
self.write_stream.close()
309
# Remove the temporary pack file.
310
self.upload_transport.delete(self.random_name)
311
# The indices have no state on disk.
313
def access_tuple(self):
314
"""Return a tuple (transport, name) for the pack content."""
315
assert self._state in ('open', 'finished')
316
if self._state == 'finished':
317
return Pack.access_tuple(self)
319
return self.upload_transport, self.random_name
321
def data_inserted(self):
322
"""True if data has been added to this pack."""
323
return bool(self.get_revision_count() or
324
self.inventory_index.key_count() or
325
self.text_index.key_count() or
326
self.signature_index.key_count())
329
"""Finish the new pack.
332
- finalises the content
333
- assigns a name (the md5 of the content, currently)
334
- writes out the associated indices
335
- renames the pack into place.
336
- stores the index size tuple for the pack in the index_sizes
341
self._write_data('', flush=True)
342
self.name = self._hash.hexdigest()
344
# XXX: It'd be better to write them all to temporary names, then
345
# rename them all into place, so that the window when only some are
346
# visible is smaller. On the other hand none will be seen until
347
# they're in the names list.
348
self.index_sizes = [None, None, None, None]
349
self._write_index('revision', self.revision_index, 'revision')
350
self._write_index('inventory', self.inventory_index, 'inventory')
351
self._write_index('text', self.text_index, 'file texts')
352
self._write_index('signature', self.signature_index,
353
'revision signatures')
354
self.write_stream.close()
355
# Note that this will clobber an existing pack with the same name,
356
# without checking for hash collisions. While this is undesirable this
357
# is something that can be rectified in a subsequent release. One way
358
# to rectify it may be to leave the pack at the original name, writing
359
# its pack-names entry as something like 'HASH: index-sizes
360
# temporary-name'. Allocate that and check for collisions, if it is
361
# collision free then rename it into place. If clients know this scheme
362
# they can handle missing-file errors by:
363
# - try for HASH.pack
364
# - try for temporary-name
365
# - refresh the pack-list to see if the pack is now absent
366
self.upload_transport.rename(self.random_name,
367
'../packs/' + self.name + '.pack')
368
self._state = 'finished'
369
if 'pack' in debug.debug_flags:
370
# XXX: size might be interesting?
371
mutter('%s: create_pack: pack renamed into place: %s%s->%s%s t+%6.3fs',
372
time.ctime(), self.upload_transport.base, self.random_name,
373
self.pack_transport, self.name,
374
time.time() - self.start_time)
376
def index_name(self, index_type, name):
377
"""Get the disk name of an index type for pack name 'name'."""
378
return name + NewPack.index_definitions[index_type][0]
380
def index_offset(self, index_type):
381
"""Get the position in a index_size array for a given index type."""
382
return NewPack.index_definitions[index_type][1]
384
def _replace_index_with_readonly(self, index_type):
385
setattr(self, index_type + '_index',
386
GraphIndex(self.index_transport,
387
self.index_name(index_type, self.name),
388
self.index_sizes[self.index_offset(index_type)]))
390
def set_write_cache_size(self, size):
391
self._cache_limit = size
393
def _write_index(self, index_type, index, label):
394
"""Write out an index.
396
:param index_type: The type of index to write - e.g. 'revision'.
397
:param index: The index object to serialise.
398
:param label: What label to give the index e.g. 'revision'.
400
index_name = self.index_name(index_type, self.name)
401
self.index_sizes[self.index_offset(index_type)] = \
402
self.index_transport.put_file(index_name, index.finish(),
403
mode=self._file_mode)
404
if 'pack' in debug.debug_flags:
405
# XXX: size might be interesting?
406
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
407
time.ctime(), label, self.upload_transport.base,
408
self.random_name, time.time() - self.start_time)
409
# Replace the writable index on this object with a readonly,
410
# presently unloaded index. We should alter
411
# the index layer to make its finish() error if add_node is
412
# subsequently used. RBC
413
self._replace_index_with_readonly(index_type)
416
class AggregateIndex(object):
417
"""An aggregated index for the RepositoryPackCollection.
419
AggregateIndex is reponsible for managing the PackAccess object,
420
Index-To-Pack mapping, and all indices list for a specific type of index
421
such as 'revision index'.
423
A CombinedIndex provides an index on a single key space built up
424
from several on-disk indices. The AggregateIndex builds on this
425
to provide a knit access layer, and allows having up to one writable
426
index within the collection.
428
# XXX: Probably 'can be written to' could/should be separated from 'acts
429
# like a knit index' -- mbp 20071024
432
"""Create an AggregateIndex."""
433
self.index_to_pack = {}
434
self.combined_index = CombinedGraphIndex([])
435
self.knit_access = _PackAccess(self.index_to_pack)
437
def replace_indices(self, index_to_pack, indices):
438
"""Replace the current mappings with fresh ones.
440
This should probably not be used eventually, rather incremental add and
441
removal of indices. It has been added during refactoring of existing
444
:param index_to_pack: A mapping from index objects to
445
(transport, name) tuples for the pack file data.
446
:param indices: A list of indices.
448
# refresh the revision pack map dict without replacing the instance.
449
self.index_to_pack.clear()
450
self.index_to_pack.update(index_to_pack)
451
# XXX: API break - clearly a 'replace' method would be good?
452
self.combined_index._indices[:] = indices
453
# the current add nodes callback for the current writable index if
455
self.add_callback = None
457
def add_index(self, index, pack):
458
"""Add index to the aggregate, which is an index for Pack pack.
460
Future searches on the aggregate index will seach this new index
461
before all previously inserted indices.
463
:param index: An Index for the pack.
464
:param pack: A Pack instance.
466
# expose it to the index map
467
self.index_to_pack[index] = pack.access_tuple()
468
# put it at the front of the linear index list
469
self.combined_index.insert_index(0, index)
471
def add_writable_index(self, index, pack):
472
"""Add an index which is able to have data added to it.
474
There can be at most one writable index at any time. Any
475
modifications made to the knit are put into this index.
477
:param index: An index from the pack parameter.
478
:param pack: A Pack instance.
480
assert self.add_callback is None, \
481
"%s already has a writable index through %s" % \
482
(self, self.add_callback)
483
# allow writing: queue writes to a new index
484
self.add_index(index, pack)
485
# Updates the index to packs mapping as a side effect,
486
self.knit_access.set_writer(pack._writer, index, pack.access_tuple())
487
self.add_callback = index.add_nodes
490
"""Reset all the aggregate data to nothing."""
491
self.knit_access.set_writer(None, None, (None, None))
492
self.index_to_pack.clear()
493
del self.combined_index._indices[:]
494
self.add_callback = None
496
def remove_index(self, index, pack):
497
"""Remove index from the indices used to answer queries.
499
:param index: An index from the pack parameter.
500
:param pack: A Pack instance.
502
del self.index_to_pack[index]
503
self.combined_index._indices.remove(index)
504
if (self.add_callback is not None and
505
getattr(index, 'add_nodes', None) == self.add_callback):
506
self.add_callback = None
507
self.knit_access.set_writer(None, None, (None, None))
510
class Packer(object):
511
"""Create a pack from packs."""
513
def __init__(self, pack_collection, packs, suffix, revision_ids=None):
516
:param pack_collection: A RepositoryPackCollection object where the
517
new pack is being written to.
518
:param packs: The packs to combine.
519
:param suffix: The suffix to use on the temporary files for the pack.
520
:param revision_ids: Revision ids to limit the pack to.
524
self.revision_ids = revision_ids
525
self._pack_collection = pack_collection
527
def pack(self, pb=None):
528
"""Create a new pack by reading data from other packs.
530
This does little more than a bulk copy of data. One key difference
531
is that data with the same item key across multiple packs is elided
532
from the output. The new pack is written into the current pack store
533
along with its indices, and the name added to the pack names. The
534
source packs are not altered and are not required to be in the current
537
:param pb: An optional progress bar to use. A nested bar is created if
539
:return: A Pack object, or None if nothing was copied.
541
# open a pack - using the same name as the last temporary file
542
# - which has already been flushed, so its safe.
543
# XXX: - duplicate code warning with start_write_group; fix before
544
# considering 'done'.
545
if self._pack_collection._new_pack is not None:
546
raise errors.BzrError('call to create_pack_from_packs while '
547
'another pack is being written.')
548
if self.revision_ids is not None:
549
if len(self.revision_ids) == 0:
550
# silly fetch request.
553
self.revision_ids = frozenset(self.revision_ids)
555
self.pb = ui.ui_factory.nested_progress_bar()
559
return self._create_pack_from_packs()
565
"""Open a pack for the pack we are creating."""
566
return NewPack(self._pack_collection._upload_transport,
567
self._pack_collection._index_transport,
568
self._pack_collection._pack_transport, upload_suffix=self.suffix,
569
file_mode=self._pack_collection.repo.control_files._file_mode)
571
def _create_pack_from_packs(self):
572
self.pb.update("Opening pack", 0, 5)
573
new_pack = self.open_pack()
574
# buffer data - we won't be reading-back during the pack creation and
575
# this makes a significant difference on sftp pushes.
576
new_pack.set_write_cache_size(1024*1024)
577
if 'pack' in debug.debug_flags:
578
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
579
for a_pack in self.packs]
580
if self.revision_ids is not None:
581
rev_count = len(self.revision_ids)
584
mutter('%s: create_pack: creating pack from source packs: '
585
'%s%s %s revisions wanted %s t=0',
586
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
587
plain_pack_list, rev_count)
589
if self.revision_ids:
590
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
594
# select revision keys
595
revision_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
596
self.packs, 'revision_index')[0]
597
revision_nodes = self._pack_collection._index_contents(revision_index_map, revision_keys)
598
# copy revision keys and adjust values
599
self.pb.update("Copying revision texts", 1)
600
list(self._copy_nodes_graph(revision_nodes, revision_index_map,
601
new_pack._writer, new_pack.revision_index))
602
if 'pack' in debug.debug_flags:
603
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
604
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
605
new_pack.revision_index.key_count(),
606
time.time() - new_pack.start_time)
607
# select inventory keys
608
inv_keys = revision_keys # currently the same keyspace, and note that
609
# querying for keys here could introduce a bug where an inventory item
610
# is missed, so do not change it to query separately without cross
611
# checking like the text key check below.
612
inventory_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
613
self.packs, 'inventory_index')[0]
614
inv_nodes = self._pack_collection._index_contents(inventory_index_map, inv_keys)
615
# copy inventory keys and adjust values
616
# XXX: Should be a helper function to allow different inv representation
618
self.pb.update("Copying inventory texts", 2)
619
inv_lines = self._copy_nodes_graph(inv_nodes, inventory_index_map,
620
new_pack._writer, new_pack.inventory_index, output_lines=True)
621
if self.revision_ids:
622
fileid_revisions = self._pack_collection.repo._find_file_ids_from_xml_inventory_lines(
623
inv_lines, self.revision_ids)
625
for fileid, file_revids in fileid_revisions.iteritems():
627
[(fileid, file_revid) for file_revid in file_revids])
629
# eat the iterator to cause it to execute.
632
if 'pack' in debug.debug_flags:
633
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
634
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
635
new_pack.inventory_index.key_count(),
636
time.time() - new_pack.start_time)
638
text_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
639
self.packs, 'text_index')[0]
640
text_nodes = self._pack_collection._index_contents(text_index_map, text_filter)
641
if text_filter is not None:
642
# We could return the keys copied as part of the return value from
643
# _copy_nodes_graph but this doesn't work all that well with the
644
# need to get line output too, so we check separately, and as we're
645
# going to buffer everything anyway, we check beforehand, which
646
# saves reading knit data over the wire when we know there are
648
text_nodes = set(text_nodes)
649
present_text_keys = set(_node[1] for _node in text_nodes)
650
missing_text_keys = set(text_filter) - present_text_keys
651
if missing_text_keys:
652
# TODO: raise a specific error that can handle many missing
654
a_missing_key = missing_text_keys.pop()
655
raise errors.RevisionNotPresent(a_missing_key[1],
657
# copy text keys and adjust values
658
self.pb.update("Copying content texts", 3)
659
list(self._copy_nodes_graph(text_nodes, text_index_map,
660
new_pack._writer, new_pack.text_index))
661
if 'pack' in debug.debug_flags:
662
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
663
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
664
new_pack.text_index.key_count(),
665
time.time() - new_pack.start_time)
666
# select signature keys
667
signature_filter = revision_keys # same keyspace
668
signature_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
669
self.packs, 'signature_index')[0]
670
signature_nodes = self._pack_collection._index_contents(signature_index_map,
672
# copy signature keys and adjust values
673
self.pb.update("Copying signature texts", 4)
674
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
675
new_pack.signature_index)
676
if 'pack' in debug.debug_flags:
677
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
678
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
679
new_pack.signature_index.key_count(),
680
time.time() - new_pack.start_time)
681
if not new_pack.data_inserted():
684
self.pb.update("Finishing pack", 5)
686
self._pack_collection.allocate(new_pack)
689
def _copy_nodes(self, nodes, index_map, writer, write_index):
690
"""Copy knit nodes between packs with no graph references."""
691
pb = ui.ui_factory.nested_progress_bar()
693
return self._do_copy_nodes(nodes, index_map, writer,
698
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb):
699
# for record verification
700
knit_data = _KnitData(None)
701
# plan a readv on each source pack:
703
nodes = sorted(nodes)
704
# how to map this into knit.py - or knit.py into this?
705
# we don't want the typical knit logic, we want grouping by pack
706
# at this point - perhaps a helper library for the following code
707
# duplication points?
709
for index, key, value in nodes:
710
if index not in request_groups:
711
request_groups[index] = []
712
request_groups[index].append((key, value))
714
pb.update("Copied record", record_index, len(nodes))
715
for index, items in request_groups.iteritems():
716
pack_readv_requests = []
717
for key, value in items:
718
# ---- KnitGraphIndex.get_position
719
bits = value[1:].split(' ')
720
offset, length = int(bits[0]), int(bits[1])
721
pack_readv_requests.append((offset, length, (key, value[0])))
722
# linear scan up the pack
723
pack_readv_requests.sort()
725
transport, path = index_map[index]
726
reader = pack.make_readv_reader(transport, path,
727
[offset[0:2] for offset in pack_readv_requests])
728
for (names, read_func), (_1, _2, (key, eol_flag)) in \
729
izip(reader.iter_records(), pack_readv_requests):
730
raw_data = read_func(None)
731
# check the header only
732
df, _ = knit_data._parse_record_header(key[-1], raw_data)
734
pos, size = writer.add_bytes_record(raw_data, names)
735
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
736
pb.update("Copied record", record_index)
739
def _copy_nodes_graph(self, nodes, index_map, writer, write_index,
741
"""Copy knit nodes between packs.
743
:param output_lines: Return lines present in the copied data as
744
an iterator of line,version_id.
746
pb = ui.ui_factory.nested_progress_bar()
748
for result in self._do_copy_nodes_graph(nodes, index_map, writer,
749
write_index, output_lines, pb):
752
# Python 2.4 does not permit try:finally: in a generator.
758
def _do_copy_nodes_graph(self, nodes, index_map, writer, write_index,
760
# for record verification
761
knit_data = _KnitData(None)
762
# for line extraction when requested (inventories only)
764
factory = knit.KnitPlainFactory()
765
# plan a readv on each source pack:
767
nodes = sorted(nodes)
768
# how to map this into knit.py - or knit.py into this?
769
# we don't want the typical knit logic, we want grouping by pack
770
# at this point - perhaps a helper library for the following code
771
# duplication points?
774
pb.update("Copied record", record_index, len(nodes))
775
for index, key, value, references in nodes:
776
if index not in request_groups:
777
request_groups[index] = []
778
request_groups[index].append((key, value, references))
779
for index, items in request_groups.iteritems():
780
pack_readv_requests = []
781
for key, value, references in items:
782
# ---- KnitGraphIndex.get_position
783
bits = value[1:].split(' ')
784
offset, length = int(bits[0]), int(bits[1])
785
pack_readv_requests.append((offset, length, (key, value[0], references)))
786
# linear scan up the pack
787
pack_readv_requests.sort()
789
transport, path = index_map[index]
790
reader = pack.make_readv_reader(transport, path,
791
[offset[0:2] for offset in pack_readv_requests])
792
for (names, read_func), (_1, _2, (key, eol_flag, references)) in \
793
izip(reader.iter_records(), pack_readv_requests):
794
raw_data = read_func(None)
797
# read the entire thing
798
content, _ = knit_data._parse_record(version_id, raw_data)
799
if len(references[-1]) == 0:
800
line_iterator = factory.get_fulltext_content(content)
802
line_iterator = factory.get_linedelta_content(content)
803
for line in line_iterator:
804
yield line, version_id
806
# check the header only
807
df, _ = knit_data._parse_record_header(version_id, raw_data)
809
pos, size = writer.add_bytes_record(raw_data, names)
810
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
811
pb.update("Copied record", record_index)
815
class ReconcilePacker(Packer):
816
"""A packer which regenerates indices etc as it copies.
818
This is used by ``bzr reconcile`` to cause parent text pointers to be
823
class RepositoryPackCollection(object):
824
"""Management of packs within a repository."""
826
def __init__(self, repo, transport, index_transport, upload_transport,
828
"""Create a new RepositoryPackCollection.
830
:param transport: Addresses the repository base directory
831
(typically .bzr/repository/).
832
:param index_transport: Addresses the directory containing indices.
833
:param upload_transport: Addresses the directory into which packs are written
834
while they're being created.
835
:param pack_transport: Addresses the directory of existing complete packs.
838
self.transport = transport
839
self._index_transport = index_transport
840
self._upload_transport = upload_transport
841
self._pack_transport = pack_transport
842
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
845
self._packs_by_name = {}
846
# the previous pack-names content
847
self._packs_at_load = None
848
# when a pack is being created by this object, the state of that pack.
849
self._new_pack = None
850
# aggregated revision index data
851
self.revision_index = AggregateIndex()
852
self.inventory_index = AggregateIndex()
853
self.text_index = AggregateIndex()
854
self.signature_index = AggregateIndex()
856
def add_pack_to_memory(self, pack):
857
"""Make a Pack object available to the repository to satisfy queries.
859
:param pack: A Pack object.
861
assert pack.name not in self._packs_by_name
862
self.packs.append(pack)
863
self._packs_by_name[pack.name] = pack
864
self.revision_index.add_index(pack.revision_index, pack)
865
self.inventory_index.add_index(pack.inventory_index, pack)
866
self.text_index.add_index(pack.text_index, pack)
867
self.signature_index.add_index(pack.signature_index, pack)
869
def _add_text_to_weave(self, file_id, revision_id, new_lines, parents,
870
nostore_sha, random_revid):
871
file_id_index = GraphIndexPrefixAdapter(
872
self.text_index.combined_index,
874
add_nodes_callback=self.text_index.add_callback)
875
self.repo._text_knit._index._graph_index = file_id_index
876
self.repo._text_knit._index._add_callback = file_id_index.add_nodes
877
return self.repo._text_knit.add_lines_with_ghosts(
878
revision_id, parents, new_lines, nostore_sha=nostore_sha,
879
random_id=random_revid, check_content=False)[0:2]
882
"""Return a list of all the Pack objects this repository has.
884
Note that an in-progress pack being created is not returned.
886
:return: A list of Pack objects for all the packs in the repository.
889
for name in self.names():
890
result.append(self.get_pack_by_name(name))
894
"""Pack the pack collection incrementally.
896
This will not attempt global reorganisation or recompression,
897
rather it will just ensure that the total number of packs does
898
not grow without bound. It uses the _max_pack_count method to
899
determine if autopacking is needed, and the pack_distribution
900
method to determine the number of revisions in each pack.
902
If autopacking takes place then the packs name collection will have
903
been flushed to disk - packing requires updating the name collection
904
in synchronisation with certain steps. Otherwise the names collection
907
:return: True if packing took place.
909
# XXX: Should not be needed when the management of indices is sane.
910
total_revisions = self.revision_index.combined_index.key_count()
911
total_packs = len(self._names)
912
if self._max_pack_count(total_revisions) >= total_packs:
914
# XXX: the following may want to be a class, to pack with a given
916
mutter('Auto-packing repository %s, which has %d pack files, '
917
'containing %d revisions into %d packs.', self, total_packs,
918
total_revisions, self._max_pack_count(total_revisions))
919
# determine which packs need changing
920
pack_distribution = self.pack_distribution(total_revisions)
922
for pack in self.all_packs():
923
revision_count = pack.get_revision_count()
924
if revision_count == 0:
925
# revision less packs are not generated by normal operation,
926
# only by operations like sign-my-commits, and thus will not
927
# tend to grow rapdily or without bound like commit containing
928
# packs do - leave them alone as packing them really should
929
# group their data with the relevant commit, and that may
930
# involve rewriting ancient history - which autopack tries to
931
# avoid. Alternatively we could not group the data but treat
932
# each of these as having a single revision, and thus add
933
# one revision for each to the total revision count, to get
934
# a matching distribution.
936
existing_packs.append((revision_count, pack))
937
pack_operations = self.plan_autopack_combinations(
938
existing_packs, pack_distribution)
939
self._execute_pack_operations(pack_operations)
942
def _execute_pack_operations(self, pack_operations):
943
"""Execute a series of pack operations.
945
:param pack_operations: A list of [revision_count, packs_to_combine].
948
for revision_count, packs in pack_operations:
949
# we may have no-ops from the setup logic
952
Packer(self, packs, '.autopack').pack()
954
self._remove_pack_from_memory(pack)
955
# record the newly available packs and stop advertising the old
957
self._save_pack_names(clear_obsolete_packs=True)
958
# Move the old packs out of the way now they are no longer referenced.
959
for revision_count, packs in pack_operations:
960
self._obsolete_packs(packs)
962
def lock_names(self):
963
"""Acquire the mutex around the pack-names index.
965
This cannot be used in the middle of a read-only transaction on the
968
self.repo.control_files.lock_write()
971
"""Pack the pack collection totally."""
973
total_packs = len(self._names)
976
total_revisions = self.revision_index.combined_index.key_count()
977
# XXX: the following may want to be a class, to pack with a given
979
mutter('Packing repository %s, which has %d pack files, '
980
'containing %d revisions into 1 packs.', self, total_packs,
982
# determine which packs need changing
983
pack_distribution = [1]
984
pack_operations = [[0, []]]
985
for pack in self.all_packs():
986
revision_count = pack.get_revision_count()
987
pack_operations[-1][0] += revision_count
988
pack_operations[-1][1].append(pack)
989
self._execute_pack_operations(pack_operations)
991
def plan_autopack_combinations(self, existing_packs, pack_distribution):
992
"""Plan a pack operation.
994
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
996
:param pack_distribution: A list with the number of revisions desired
999
if len(existing_packs) <= len(pack_distribution):
1001
existing_packs.sort(reverse=True)
1002
pack_operations = [[0, []]]
1003
# plan out what packs to keep, and what to reorganise
1004
while len(existing_packs):
1005
# take the largest pack, and if its less than the head of the
1006
# distribution chart we will include its contents in the new pack for
1007
# that position. If its larger, we remove its size from the
1008
# distribution chart
1009
next_pack_rev_count, next_pack = existing_packs.pop(0)
1010
if next_pack_rev_count >= pack_distribution[0]:
1011
# this is already packed 'better' than this, so we can
1012
# not waste time packing it.
1013
while next_pack_rev_count > 0:
1014
next_pack_rev_count -= pack_distribution[0]
1015
if next_pack_rev_count >= 0:
1017
del pack_distribution[0]
1019
# didn't use that entire bucket up
1020
pack_distribution[0] = -next_pack_rev_count
1022
# add the revisions we're going to add to the next output pack
1023
pack_operations[-1][0] += next_pack_rev_count
1024
# allocate this pack to the next pack sub operation
1025
pack_operations[-1][1].append(next_pack)
1026
if pack_operations[-1][0] >= pack_distribution[0]:
1027
# this pack is used up, shift left.
1028
del pack_distribution[0]
1029
pack_operations.append([0, []])
1031
return pack_operations
1033
def ensure_loaded(self):
1034
# NB: if you see an assertion error here, its probably access against
1035
# an unlocked repo. Naughty.
1036
assert self.repo.is_locked()
1037
if self._names is None:
1039
self._packs_at_load = set()
1040
for index, key, value in self._iter_disk_pack_index():
1042
self._names[name] = self._parse_index_sizes(value)
1043
self._packs_at_load.add((key, value))
1044
# populate all the metadata.
1047
def _parse_index_sizes(self, value):
1048
"""Parse a string of index sizes."""
1049
return tuple([int(digits) for digits in value.split(' ')])
1051
def get_pack_by_name(self, name):
1052
"""Get a Pack object by name.
1054
:param name: The name of the pack - e.g. '123456'
1055
:return: A Pack object.
1058
return self._packs_by_name[name]
1060
rev_index = self._make_index(name, '.rix')
1061
inv_index = self._make_index(name, '.iix')
1062
txt_index = self._make_index(name, '.tix')
1063
sig_index = self._make_index(name, '.six')
1064
result = ExistingPack(self._pack_transport, name, rev_index,
1065
inv_index, txt_index, sig_index)
1066
self.add_pack_to_memory(result)
1069
def allocate(self, a_new_pack):
1070
"""Allocate name in the list of packs.
1072
:param a_new_pack: A NewPack instance to be added to the collection of
1073
packs for this repository.
1075
self.ensure_loaded()
1076
if a_new_pack.name in self._names:
1077
# a collision with the packs we know about (not the only possible
1078
# collision, see NewPack.finish() for some discussion). Remove our
1079
# prior reference to it.
1080
self._remove_pack_from_memory(a_new_pack)
1081
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1082
self.add_pack_to_memory(a_new_pack)
1084
def _iter_disk_pack_index(self):
1085
"""Iterate over the contents of the pack-names index.
1087
This is used when loading the list from disk, and before writing to
1088
detect updates from others during our write operation.
1089
:return: An iterator of the index contents.
1091
return GraphIndex(self.transport, 'pack-names', None
1092
).iter_all_entries()
1094
def _make_index(self, name, suffix):
1095
size_offset = self._suffix_offsets[suffix]
1096
index_name = name + suffix
1097
index_size = self._names[name][size_offset]
1099
self._index_transport, index_name, index_size)
1101
def _max_pack_count(self, total_revisions):
1102
"""Return the maximum number of packs to use for total revisions.
1104
:param total_revisions: The total number of revisions in the
1107
if not total_revisions:
1109
digits = str(total_revisions)
1111
for digit in digits:
1112
result += int(digit)
1116
"""Provide an order to the underlying names."""
1117
return sorted(self._names.keys())
1119
def _obsolete_packs(self, packs):
1120
"""Move a number of packs which have been obsoleted out of the way.
1122
Each pack and its associated indices are moved out of the way.
1124
Note: for correctness this function should only be called after a new
1125
pack names index has been written without these pack names, and with
1126
the names of packs that contain the data previously available via these
1129
:param packs: The packs to obsolete.
1130
:param return: None.
1133
pack.pack_transport.rename(pack.file_name(),
1134
'../obsolete_packs/' + pack.file_name())
1135
# TODO: Probably needs to know all possible indices for this pack
1136
# - or maybe list the directory and move all indices matching this
1137
# name whether we recognize it or not?
1138
for suffix in ('.iix', '.six', '.tix', '.rix'):
1139
self._index_transport.rename(pack.name + suffix,
1140
'../obsolete_packs/' + pack.name + suffix)
1142
def pack_distribution(self, total_revisions):
1143
"""Generate a list of the number of revisions to put in each pack.
1145
:param total_revisions: The total number of revisions in the
1148
if total_revisions == 0:
1150
digits = reversed(str(total_revisions))
1152
for exponent, count in enumerate(digits):
1153
size = 10 ** exponent
1154
for pos in range(int(count)):
1156
return list(reversed(result))
1158
def _pack_tuple(self, name):
1159
"""Return a tuple with the transport and file name for a pack name."""
1160
return self._pack_transport, name + '.pack'
1162
def _remove_pack_from_memory(self, pack):
1163
"""Remove pack from the packs accessed by this repository.
1165
Only affects memory state, until self._save_pack_names() is invoked.
1167
self._names.pop(pack.name)
1168
self._packs_by_name.pop(pack.name)
1169
self._remove_pack_indices(pack)
1171
def _remove_pack_indices(self, pack):
1172
"""Remove the indices for pack from the aggregated indices."""
1173
self.revision_index.remove_index(pack.revision_index, pack)
1174
self.inventory_index.remove_index(pack.inventory_index, pack)
1175
self.text_index.remove_index(pack.text_index, pack)
1176
self.signature_index.remove_index(pack.signature_index, pack)
1179
"""Clear all cached data."""
1180
# cached revision data
1181
self.repo._revision_knit = None
1182
self.revision_index.clear()
1183
# cached signature data
1184
self.repo._signature_knit = None
1185
self.signature_index.clear()
1186
# cached file text data
1187
self.text_index.clear()
1188
self.repo._text_knit = None
1189
# cached inventory data
1190
self.inventory_index.clear()
1191
# remove the open pack
1192
self._new_pack = None
1193
# information about packs.
1196
self._packs_by_name = {}
1197
self._packs_at_load = None
1199
def _make_index_map(self, index_suffix):
1200
"""Return information on existing indices.
1202
:param suffix: Index suffix added to pack name.
1204
:returns: (pack_map, indices) where indices is a list of GraphIndex
1205
objects, and pack_map is a mapping from those objects to the
1206
pack tuple they describe.
1208
# TODO: stop using this; it creates new indices unnecessarily.
1209
self.ensure_loaded()
1210
suffix_map = {'.rix': 'revision_index',
1211
'.six': 'signature_index',
1212
'.iix': 'inventory_index',
1213
'.tix': 'text_index',
1215
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1216
suffix_map[index_suffix])
1218
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1219
"""Convert a list of packs to an index pack map and index list.
1221
:param packs: The packs list to process.
1222
:param index_attribute: The attribute that the desired index is found
1224
:return: A tuple (map, list) where map contains the dict from
1225
index:pack_tuple, and lsit contains the indices in the same order
1231
index = getattr(pack, index_attribute)
1232
indices.append(index)
1233
pack_map[index] = (pack.pack_transport, pack.file_name())
1234
return pack_map, indices
1236
def _index_contents(self, pack_map, key_filter=None):
1237
"""Get an iterable of the index contents from a pack_map.
1239
:param pack_map: A map from indices to pack details.
1240
:param key_filter: An optional filter to limit the
1243
indices = [index for index in pack_map.iterkeys()]
1244
all_index = CombinedGraphIndex(indices)
1245
if key_filter is None:
1246
return all_index.iter_all_entries()
1248
return all_index.iter_entries(key_filter)
1250
def _unlock_names(self):
1251
"""Release the mutex around the pack-names index."""
1252
self.repo.control_files.unlock()
1254
def _save_pack_names(self, clear_obsolete_packs=False):
1255
"""Save the list of packs.
1257
This will take out the mutex around the pack names list for the
1258
duration of the method call. If concurrent updates have been made, a
1259
three-way merge between the current list and the current in memory list
1262
:param clear_obsolete_packs: If True, clear out the contents of the
1263
obsolete_packs directory.
1267
builder = GraphIndexBuilder()
1268
# load the disk nodes across
1270
for index, key, value in self._iter_disk_pack_index():
1271
disk_nodes.add((key, value))
1272
# do a two-way diff against our original content
1273
current_nodes = set()
1274
for name, sizes in self._names.iteritems():
1276
((name, ), ' '.join(str(size) for size in sizes)))
1277
deleted_nodes = self._packs_at_load - current_nodes
1278
new_nodes = current_nodes - self._packs_at_load
1279
disk_nodes.difference_update(deleted_nodes)
1280
disk_nodes.update(new_nodes)
1281
# TODO: handle same-name, index-size-changes here -
1282
# e.g. use the value from disk, not ours, *unless* we're the one
1284
for key, value in disk_nodes:
1285
builder.add_node(key, value)
1286
self.transport.put_file('pack-names', builder.finish(),
1287
mode=self.repo.control_files._file_mode)
1288
# move the baseline forward
1289
self._packs_at_load = disk_nodes
1290
# now clear out the obsolete packs directory
1291
if clear_obsolete_packs:
1292
self.transport.clone('obsolete_packs').delete_multi(
1293
self.transport.list_dir('obsolete_packs'))
1295
self._unlock_names()
1296
# synchronise the memory packs list with what we just wrote:
1297
new_names = dict(disk_nodes)
1298
# drop no longer present nodes
1299
for pack in self.all_packs():
1300
if (pack.name,) not in new_names:
1301
self._remove_pack_from_memory(pack)
1302
# add new nodes/refresh existing ones
1303
for key, value in disk_nodes:
1305
sizes = self._parse_index_sizes(value)
1306
if name in self._names:
1308
if sizes != self._names[name]:
1309
# the pack for name has had its indices replaced - rare but
1310
# important to handle. XXX: probably can never happen today
1311
# because the three-way merge code above does not handle it
1312
# - you may end up adding the same key twice to the new
1313
# disk index because the set values are the same, unless
1314
# the only index shows up as deleted by the set difference
1315
# - which it may. Until there is a specific test for this,
1316
# assume its broken. RBC 20071017.
1317
self._remove_pack_from_memory(self.get_pack_by_name(name))
1318
self._names[name] = sizes
1319
self.get_pack_by_name(name)
1322
self._names[name] = sizes
1323
self.get_pack_by_name(name)
1325
def _start_write_group(self):
1326
# Do not permit preparation for writing if we're not in a 'write lock'.
1327
if not self.repo.is_write_locked():
1328
raise errors.NotWriteLocked(self)
1329
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1330
self._pack_transport, upload_suffix='.pack',
1331
file_mode=self.repo.control_files._file_mode)
1332
# allow writing: queue writes to a new index
1333
self.revision_index.add_writable_index(self._new_pack.revision_index,
1335
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
1337
self.text_index.add_writable_index(self._new_pack.text_index,
1339
self.signature_index.add_writable_index(self._new_pack.signature_index,
1342
# reused revision and signature knits may need updating
1344
# "Hysterical raisins. client code in bzrlib grabs those knits outside
1345
# of write groups and then mutates it inside the write group."
1346
if self.repo._revision_knit is not None:
1347
self.repo._revision_knit._index._add_callback = \
1348
self.revision_index.add_callback
1349
if self.repo._signature_knit is not None:
1350
self.repo._signature_knit._index._add_callback = \
1351
self.signature_index.add_callback
1352
# create a reused knit object for text addition in commit.
1353
self.repo._text_knit = self.repo.weave_store.get_weave_or_empty(
1356
def _abort_write_group(self):
1357
# FIXME: just drop the transient index.
1358
# forget what names there are
1359
self._new_pack.abort()
1360
self._remove_pack_indices(self._new_pack)
1361
self._new_pack = None
1362
self.repo._text_knit = None
1364
def _commit_write_group(self):
1365
self._remove_pack_indices(self._new_pack)
1366
if self._new_pack.data_inserted():
1367
# get all the data to disk and read to use
1368
self._new_pack.finish()
1369
self.allocate(self._new_pack)
1370
self._new_pack = None
1371
if not self.autopack():
1372
# when autopack takes no steps, the names list is still
1374
self._save_pack_names()
1376
self._new_pack.abort()
1377
self._new_pack = None
1378
self.repo._text_knit = None
1381
class KnitPackRevisionStore(KnitRevisionStore):
1382
"""An object to adapt access from RevisionStore's to use KnitPacks.
1384
This class works by replacing the original RevisionStore.
1385
We need to do this because the KnitPackRevisionStore is less
1386
isolated in its layering - it uses services from the repo.
1389
def __init__(self, repo, transport, revisionstore):
1390
"""Create a KnitPackRevisionStore on repo with revisionstore.
1392
This will store its state in the Repository, use the
1393
indices to provide a KnitGraphIndex,
1394
and at the end of transactions write new indices.
1396
KnitRevisionStore.__init__(self, revisionstore.versioned_file_store)
1398
self._serializer = revisionstore._serializer
1399
self.transport = transport
1401
def get_revision_file(self, transaction):
1402
"""Get the revision versioned file object."""
1403
if getattr(self.repo, '_revision_knit', None) is not None:
1404
return self.repo._revision_knit
1405
self.repo._pack_collection.ensure_loaded()
1406
add_callback = self.repo._pack_collection.revision_index.add_callback
1407
# setup knit specific objects
1408
knit_index = KnitGraphIndex(
1409
self.repo._pack_collection.revision_index.combined_index,
1410
add_callback=add_callback)
1411
self.repo._revision_knit = knit.KnitVersionedFile(
1412
'revisions', self.transport.clone('..'),
1413
self.repo.control_files._file_mode,
1414
create=False, access_mode=self.repo._access_mode(),
1415
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1416
access_method=self.repo._pack_collection.revision_index.knit_access)
1417
return self.repo._revision_knit
1419
def get_signature_file(self, transaction):
1420
"""Get the signature versioned file object."""
1421
if getattr(self.repo, '_signature_knit', None) is not None:
1422
return self.repo._signature_knit
1423
self.repo._pack_collection.ensure_loaded()
1424
add_callback = self.repo._pack_collection.signature_index.add_callback
1425
# setup knit specific objects
1426
knit_index = KnitGraphIndex(
1427
self.repo._pack_collection.signature_index.combined_index,
1428
add_callback=add_callback, parents=False)
1429
self.repo._signature_knit = knit.KnitVersionedFile(
1430
'signatures', self.transport.clone('..'),
1431
self.repo.control_files._file_mode,
1432
create=False, access_mode=self.repo._access_mode(),
1433
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1434
access_method=self.repo._pack_collection.signature_index.knit_access)
1435
return self.repo._signature_knit
1438
class KnitPackTextStore(VersionedFileStore):
1439
"""Presents a TextStore abstraction on top of packs.
1441
This class works by replacing the original VersionedFileStore.
1442
We need to do this because the KnitPackRevisionStore is less
1443
isolated in its layering - it uses services from the repo and shares them
1444
with all the data written in a single write group.
1447
def __init__(self, repo, transport, weavestore):
1448
"""Create a KnitPackTextStore on repo with weavestore.
1450
This will store its state in the Repository, use the
1451
indices FileNames to provide a KnitGraphIndex,
1452
and at the end of transactions write new indices.
1454
# don't call base class constructor - it's not suitable.
1455
# no transient data stored in the transaction
1457
self._precious = False
1459
self.transport = transport
1460
self.weavestore = weavestore
1461
# XXX for check() which isn't updated yet
1462
self._transport = weavestore._transport
1464
def get_weave_or_empty(self, file_id, transaction):
1465
"""Get a 'Knit' backed by the .tix indices.
1467
The transaction parameter is ignored.
1469
self.repo._pack_collection.ensure_loaded()
1470
add_callback = self.repo._pack_collection.text_index.add_callback
1471
# setup knit specific objects
1472
file_id_index = GraphIndexPrefixAdapter(
1473
self.repo._pack_collection.text_index.combined_index,
1474
(file_id, ), 1, add_nodes_callback=add_callback)
1475
knit_index = KnitGraphIndex(file_id_index,
1476
add_callback=file_id_index.add_nodes,
1477
deltas=True, parents=True)
1478
return knit.KnitVersionedFile('text:' + file_id,
1479
self.transport.clone('..'),
1482
access_method=self.repo._pack_collection.text_index.knit_access,
1483
factory=knit.KnitPlainFactory())
1485
get_weave = get_weave_or_empty
1488
"""Generate a list of the fileids inserted, for use by check."""
1489
self.repo._pack_collection.ensure_loaded()
1491
for index, key, value, refs in \
1492
self.repo._pack_collection.text_index.combined_index.iter_all_entries():
1497
class InventoryKnitThunk(object):
1498
"""An object to manage thunking get_inventory_weave to pack based knits."""
1500
def __init__(self, repo, transport):
1501
"""Create an InventoryKnitThunk for repo at transport.
1503
This will store its state in the Repository, use the
1504
indices FileNames to provide a KnitGraphIndex,
1505
and at the end of transactions write a new index..
1508
self.transport = transport
1510
def get_weave(self):
1511
"""Get a 'Knit' that contains inventory data."""
1512
self.repo._pack_collection.ensure_loaded()
1513
add_callback = self.repo._pack_collection.inventory_index.add_callback
1514
# setup knit specific objects
1515
knit_index = KnitGraphIndex(
1516
self.repo._pack_collection.inventory_index.combined_index,
1517
add_callback=add_callback, deltas=True, parents=True)
1518
return knit.KnitVersionedFile(
1519
'inventory', self.transport.clone('..'),
1520
self.repo.control_files._file_mode,
1521
create=False, access_mode=self.repo._access_mode(),
1522
index=knit_index, delta=True, factory=knit.KnitPlainFactory(),
1523
access_method=self.repo._pack_collection.inventory_index.knit_access)
1526
class KnitPackRepository(KnitRepository):
1527
"""Experimental graph-knit using repository."""
1529
def __init__(self, _format, a_bzrdir, control_files, _revision_store,
1530
control_store, text_store, _commit_builder_class, _serializer):
1531
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
1532
_revision_store, control_store, text_store, _commit_builder_class,
1534
index_transport = control_files._transport.clone('indices')
1535
self._pack_collection = RepositoryPackCollection(self, control_files._transport,
1537
control_files._transport.clone('upload'),
1538
control_files._transport.clone('packs'))
1539
self._revision_store = KnitPackRevisionStore(self, index_transport, self._revision_store)
1540
self.weave_store = KnitPackTextStore(self, index_transport, self.weave_store)
1541
self._inv_thunk = InventoryKnitThunk(self, index_transport)
1542
# True when the repository object is 'write locked' (as opposed to the
1543
# physical lock only taken out around changes to the pack-names list.)
1544
# Another way to represent this would be a decorator around the control
1545
# files object that presents logical locks as physical ones - if this
1546
# gets ugly consider that alternative design. RBC 20071011
1547
self._write_lock_count = 0
1548
self._transaction = None
1550
self._reconcile_does_inventory_gc = True
1551
self._reconcile_fixes_text_parents = False
1552
self._reconcile_backsup_inventory = False
1554
def _abort_write_group(self):
1555
self._pack_collection._abort_write_group()
1557
def _access_mode(self):
1558
"""Return 'w' or 'r' for depending on whether a write lock is active.
1560
This method is a helper for the Knit-thunking support objects.
1562
if self.is_write_locked():
1566
def _find_inconsistent_revision_parents(self):
1567
"""Find revisions with incorrectly cached parents.
1569
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1570
parents-in-revision).
1572
assert self.is_locked()
1573
pb = ui.ui_factory.nested_progress_bar()
1576
revision_nodes = self._pack_collection.revision_index \
1577
.combined_index.iter_all_entries()
1578
index_positions = []
1579
# Get the cached index values for all revisions, and also the location
1580
# in each index of the revision text so we can perform linear IO.
1581
for index, key, value, refs in revision_nodes:
1582
pos, length = value[1:].split(' ')
1583
index_positions.append((index, int(pos), key[0],
1584
tuple(parent[0] for parent in refs[0])))
1585
pb.update("Reading revision index.", 0, 0)
1586
index_positions.sort()
1587
batch_count = len(index_positions) / 1000 + 1
1588
pb.update("Checking cached revision graph.", 0, batch_count)
1589
for offset in xrange(batch_count):
1590
pb.update("Checking cached revision graph.", offset)
1591
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1594
rev_ids = [item[2] for item in to_query]
1595
revs = self.get_revisions(rev_ids)
1596
for revision, item in zip(revs, to_query):
1597
index_parents = item[3]
1598
rev_parents = tuple(revision.parent_ids)
1599
if index_parents != rev_parents:
1600
result.append((revision.revision_id, index_parents, rev_parents))
1605
def get_parents(self, revision_ids):
1606
"""See StackedParentsProvider.get_parents.
1608
This implementation accesses the combined revision index to provide
1611
self._pack_collection.ensure_loaded()
1612
index = self._pack_collection.revision_index.combined_index
1614
for revision_id in revision_ids:
1615
if revision_id != _mod_revision.NULL_REVISION:
1616
search_keys.add((revision_id,))
1617
found_parents = {_mod_revision.NULL_REVISION:[]}
1618
for index, key, value, refs in index.iter_entries(search_keys):
1621
parents = (_mod_revision.NULL_REVISION,)
1623
parents = tuple(parent[0] for parent in parents)
1624
found_parents[key[0]] = parents
1626
for revision_id in revision_ids:
1628
result.append(found_parents[revision_id])
1633
def _make_parents_provider(self):
1636
def _refresh_data(self):
1637
if self._write_lock_count == 1 or (
1638
self.control_files._lock_count == 1 and
1639
self.control_files._lock_mode == 'r'):
1640
# forget what names there are
1641
self._pack_collection.reset()
1642
# XXX: Better to do an in-memory merge when acquiring a new lock -
1643
# factor out code from _save_pack_names.
1644
self._pack_collection.ensure_loaded()
1646
def _start_write_group(self):
1647
self._pack_collection._start_write_group()
1649
def _commit_write_group(self):
1650
return self._pack_collection._commit_write_group()
1652
def get_inventory_weave(self):
1653
return self._inv_thunk.get_weave()
1655
def get_transaction(self):
1656
if self._write_lock_count:
1657
return self._transaction
1659
return self.control_files.get_transaction()
1661
def is_locked(self):
1662
return self._write_lock_count or self.control_files.is_locked()
1664
def is_write_locked(self):
1665
return self._write_lock_count
1667
def lock_write(self, token=None):
1668
if not self._write_lock_count and self.is_locked():
1669
raise errors.ReadOnlyError(self)
1670
self._write_lock_count += 1
1671
if self._write_lock_count == 1:
1672
from bzrlib import transactions
1673
self._transaction = transactions.WriteTransaction()
1674
self._refresh_data()
1676
def lock_read(self):
1677
if self._write_lock_count:
1678
self._write_lock_count += 1
1680
self.control_files.lock_read()
1681
self._refresh_data()
1683
def leave_lock_in_place(self):
1684
# not supported - raise an error
1685
raise NotImplementedError(self.leave_lock_in_place)
1687
def dont_leave_lock_in_place(self):
1688
# not supported - raise an error
1689
raise NotImplementedError(self.dont_leave_lock_in_place)
1693
"""Compress the data within the repository.
1695
This will pack all the data to a single pack. In future it may
1696
recompress deltas or do other such expensive operations.
1698
self._pack_collection.pack()
1701
def reconcile(self, other=None, thorough=False):
1702
"""Reconcile this repository."""
1703
from bzrlib.reconcile import PackReconciler
1704
reconciler = PackReconciler(self, thorough=thorough)
1705
reconciler.reconcile()
1709
if self._write_lock_count == 1 and self._write_group is not None:
1710
self.abort_write_group()
1711
self._transaction = None
1712
self._write_lock_count = 0
1713
raise errors.BzrError(
1714
'Must end write group before releasing write lock on %s'
1716
if self._write_lock_count:
1717
self._write_lock_count -= 1
1718
if not self._write_lock_count:
1719
transaction = self._transaction
1720
self._transaction = None
1721
transaction.finish()
1723
self.control_files.unlock()
1726
class RepositoryFormatPack(MetaDirRepositoryFormat):
1727
"""Format logic for pack structured repositories.
1729
This repository format has:
1730
- a list of packs in pack-names
1731
- packs in packs/NAME.pack
1732
- indices in indices/NAME.{iix,six,tix,rix}
1733
- knit deltas in the packs, knit indices mapped to the indices.
1734
- thunk objects to support the knits programming API.
1735
- a format marker of its own
1736
- an optional 'shared-storage' flag
1737
- an optional 'no-working-trees' flag
1741
# Set this attribute in derived classes to control the repository class
1742
# created by open and initialize.
1743
repository_class = None
1744
# Set this attribute in derived classes to control the
1745
# _commit_builder_class that the repository objects will have passed to
1746
# their constructor.
1747
_commit_builder_class = None
1748
# Set this attribute in derived clases to control the _serializer that the
1749
# repository objects will have passed to their constructor.
1752
def _get_control_store(self, repo_transport, control_files):
1753
"""Return the control store for this repository."""
1754
return VersionedFileStore(
1757
file_mode=control_files._file_mode,
1758
versionedfile_class=knit.KnitVersionedFile,
1759
versionedfile_kwargs={'factory': knit.KnitPlainFactory()},
1762
def _get_revision_store(self, repo_transport, control_files):
1763
"""See RepositoryFormat._get_revision_store()."""
1764
versioned_file_store = VersionedFileStore(
1766
file_mode=control_files._file_mode,
1769
versionedfile_class=knit.KnitVersionedFile,
1770
versionedfile_kwargs={'delta': False,
1771
'factory': knit.KnitPlainFactory(),
1775
return KnitRevisionStore(versioned_file_store)
1777
def _get_text_store(self, transport, control_files):
1778
"""See RepositoryFormat._get_text_store()."""
1779
return self._get_versioned_file_store('knits',
1782
versionedfile_class=knit.KnitVersionedFile,
1783
versionedfile_kwargs={
1784
'create_parent_dir': True,
1785
'delay_create': True,
1786
'dir_mode': control_files._dir_mode,
1790
def initialize(self, a_bzrdir, shared=False):
1791
"""Create a pack based repository.
1793
:param a_bzrdir: bzrdir to contain the new repository; must already
1795
:param shared: If true the repository will be initialized as a shared
1798
mutter('creating repository in %s.', a_bzrdir.transport.base)
1799
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
1800
builder = GraphIndexBuilder()
1801
files = [('pack-names', builder.finish())]
1802
utf8_files = [('format', self.get_format_string())]
1804
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
1805
return self.open(a_bzrdir=a_bzrdir, _found=True)
1807
def open(self, a_bzrdir, _found=False, _override_transport=None):
1808
"""See RepositoryFormat.open().
1810
:param _override_transport: INTERNAL USE ONLY. Allows opening the
1811
repository at a slightly different url
1812
than normal. I.e. during 'upgrade'.
1815
format = RepositoryFormat.find_format(a_bzrdir)
1816
assert format.__class__ == self.__class__
1817
if _override_transport is not None:
1818
repo_transport = _override_transport
1820
repo_transport = a_bzrdir.get_repository_transport(None)
1821
control_files = lockable_files.LockableFiles(repo_transport,
1822
'lock', lockdir.LockDir)
1823
text_store = self._get_text_store(repo_transport, control_files)
1824
control_store = self._get_control_store(repo_transport, control_files)
1825
_revision_store = self._get_revision_store(repo_transport, control_files)
1826
return self.repository_class(_format=self,
1828
control_files=control_files,
1829
_revision_store=_revision_store,
1830
control_store=control_store,
1831
text_store=text_store,
1832
_commit_builder_class=self._commit_builder_class,
1833
_serializer=self._serializer)
1836
class RepositoryFormatKnitPack1(RepositoryFormatPack):
1837
"""A no-subtrees parameterised Pack repository.
1839
This format was introduced in 0.92.
1842
repository_class = KnitPackRepository
1843
_commit_builder_class = PackCommitBuilder
1844
_serializer = xml5.serializer_v5
1846
def _get_matching_bzrdir(self):
1847
return bzrdir.format_registry.make_bzrdir('pack-0.92')
1849
def _ignore_setting_bzrdir(self, format):
1852
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1854
def get_format_string(self):
1855
"""See RepositoryFormat.get_format_string()."""
1856
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
1858
def get_format_description(self):
1859
"""See RepositoryFormat.get_format_description()."""
1860
return "Packs containing knits without subtree support"
1862
def check_conversion_target(self, target_format):
1866
class RepositoryFormatKnitPack3(RepositoryFormatPack):
1867
"""A subtrees parameterised Pack repository.
1869
This repository format uses the xml7 serializer to get:
1870
- support for recording full info about the tree root
1871
- support for recording tree-references
1873
This format was introduced in 0.92.
1876
repository_class = KnitPackRepository
1877
_commit_builder_class = PackRootCommitBuilder
1878
rich_root_data = True
1879
supports_tree_reference = True
1880
_serializer = xml7.serializer_v7
1882
def _get_matching_bzrdir(self):
1883
return bzrdir.format_registry.make_bzrdir(
1884
'pack-0.92-subtree')
1886
def _ignore_setting_bzrdir(self, format):
1889
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1891
def check_conversion_target(self, target_format):
1892
if not target_format.rich_root_data:
1893
raise errors.BadConversionTarget(
1894
'Does not support rich root data.', target_format)
1895
if not getattr(target_format, 'supports_tree_reference', False):
1896
raise errors.BadConversionTarget(
1897
'Does not support nested trees', target_format)
1899
def get_format_string(self):
1900
"""See RepositoryFormat.get_format_string()."""
1901
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
1903
def get_format_description(self):
1904
"""See RepositoryFormat.get_format_description()."""
1905
return "Packs containing knits with subtree support\n"
1908
class RepositoryFormatKnitPack4(RepositoryFormatPack):
1909
"""A rich-root, no subtrees parameterised Pack repository.
1911
This repository format uses the xml6 serializer to get:
1912
- support for recording full info about the tree root
1914
This format was introduced in 1.0.
1917
repository_class = KnitPackRepository
1918
_commit_builder_class = PackRootCommitBuilder
1919
rich_root_data = True
1920
supports_tree_reference = False
1921
_serializer = xml6.serializer_v6
1923
def _get_matching_bzrdir(self):
1924
return bzrdir.format_registry.make_bzrdir(
1927
def _ignore_setting_bzrdir(self, format):
1930
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1932
def check_conversion_target(self, target_format):
1933
if not target_format.rich_root_data:
1934
raise errors.BadConversionTarget(
1935
'Does not support rich root data.', target_format)
1937
def get_format_string(self):
1938
"""See RepositoryFormat.get_format_string()."""
1939
return ("Bazaar pack repository format 1 with rich root"
1940
" (needs bzr 1.0)\n")
1942
def get_format_description(self):
1943
"""See RepositoryFormat.get_format_description()."""
1944
return "Packs containing knits with rich root support\n"
added
removed
bzrlib/repofmt/pack_repo.py
