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
return self._do_copy_nodes_graph(nodes, index_map, writer,
749
write_index, output_lines, pb)
753
def _do_copy_nodes_graph(self, nodes, index_map, writer, write_index,
755
# for record verification
756
knit_data = _KnitData(None)
757
# for line extraction when requested (inventories only)
759
factory = knit.KnitPlainFactory()
760
# plan a readv on each source pack:
762
nodes = sorted(nodes)
763
# how to map this into knit.py - or knit.py into this?
764
# we don't want the typical knit logic, we want grouping by pack
765
# at this point - perhaps a helper library for the following code
766
# duplication points?
769
pb.update("Copied record", record_index, len(nodes))
770
for index, key, value, references in nodes:
771
if index not in request_groups:
772
request_groups[index] = []
773
request_groups[index].append((key, value, references))
774
for index, items in request_groups.iteritems():
775
pack_readv_requests = []
776
for key, value, references in items:
777
# ---- KnitGraphIndex.get_position
778
bits = value[1:].split(' ')
779
offset, length = int(bits[0]), int(bits[1])
780
pack_readv_requests.append((offset, length, (key, value[0], references)))
781
# linear scan up the pack
782
pack_readv_requests.sort()
784
transport, path = index_map[index]
785
reader = pack.make_readv_reader(transport, path,
786
[offset[0:2] for offset in pack_readv_requests])
787
for (names, read_func), (_1, _2, (key, eol_flag, references)) in \
788
izip(reader.iter_records(), pack_readv_requests):
789
raw_data = read_func(None)
792
# read the entire thing
793
content, _ = knit_data._parse_record(version_id, raw_data)
794
if len(references[-1]) == 0:
795
line_iterator = factory.get_fulltext_content(content)
797
line_iterator = factory.get_linedelta_content(content)
798
for line in line_iterator:
799
yield line, version_id
801
# check the header only
802
df, _ = knit_data._parse_record_header(version_id, raw_data)
804
pos, size = writer.add_bytes_record(raw_data, names)
805
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
806
pb.update("Copied record", record_index)
810
class ReconcilePacker(Packer):
811
"""A packer which regenerates indices etc as it copies.
813
This is used by ``bzr reconcile`` to cause parent text pointers to be
818
class RepositoryPackCollection(object):
819
"""Management of packs within a repository."""
821
def __init__(self, repo, transport, index_transport, upload_transport,
823
"""Create a new RepositoryPackCollection.
825
:param transport: Addresses the repository base directory
826
(typically .bzr/repository/).
827
:param index_transport: Addresses the directory containing indices.
828
:param upload_transport: Addresses the directory into which packs are written
829
while they're being created.
830
:param pack_transport: Addresses the directory of existing complete packs.
833
self.transport = transport
834
self._index_transport = index_transport
835
self._upload_transport = upload_transport
836
self._pack_transport = pack_transport
837
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
840
self._packs_by_name = {}
841
# the previous pack-names content
842
self._packs_at_load = None
843
# when a pack is being created by this object, the state of that pack.
844
self._new_pack = None
845
# aggregated revision index data
846
self.revision_index = AggregateIndex()
847
self.inventory_index = AggregateIndex()
848
self.text_index = AggregateIndex()
849
self.signature_index = AggregateIndex()
851
def add_pack_to_memory(self, pack):
852
"""Make a Pack object available to the repository to satisfy queries.
854
:param pack: A Pack object.
856
assert pack.name not in self._packs_by_name
857
self.packs.append(pack)
858
self._packs_by_name[pack.name] = pack
859
self.revision_index.add_index(pack.revision_index, pack)
860
self.inventory_index.add_index(pack.inventory_index, pack)
861
self.text_index.add_index(pack.text_index, pack)
862
self.signature_index.add_index(pack.signature_index, pack)
864
def _add_text_to_weave(self, file_id, revision_id, new_lines, parents,
865
nostore_sha, random_revid):
866
file_id_index = GraphIndexPrefixAdapter(
867
self.text_index.combined_index,
869
add_nodes_callback=self.text_index.add_callback)
870
self.repo._text_knit._index._graph_index = file_id_index
871
self.repo._text_knit._index._add_callback = file_id_index.add_nodes
872
return self.repo._text_knit.add_lines_with_ghosts(
873
revision_id, parents, new_lines, nostore_sha=nostore_sha,
874
random_id=random_revid, check_content=False)[0:2]
877
"""Return a list of all the Pack objects this repository has.
879
Note that an in-progress pack being created is not returned.
881
:return: A list of Pack objects for all the packs in the repository.
884
for name in self.names():
885
result.append(self.get_pack_by_name(name))
889
"""Pack the pack collection incrementally.
891
This will not attempt global reorganisation or recompression,
892
rather it will just ensure that the total number of packs does
893
not grow without bound. It uses the _max_pack_count method to
894
determine if autopacking is needed, and the pack_distribution
895
method to determine the number of revisions in each pack.
897
If autopacking takes place then the packs name collection will have
898
been flushed to disk - packing requires updating the name collection
899
in synchronisation with certain steps. Otherwise the names collection
902
:return: True if packing took place.
904
# XXX: Should not be needed when the management of indices is sane.
905
total_revisions = self.revision_index.combined_index.key_count()
906
total_packs = len(self._names)
907
if self._max_pack_count(total_revisions) >= total_packs:
909
# XXX: the following may want to be a class, to pack with a given
911
mutter('Auto-packing repository %s, which has %d pack files, '
912
'containing %d revisions into %d packs.', self, total_packs,
913
total_revisions, self._max_pack_count(total_revisions))
914
# determine which packs need changing
915
pack_distribution = self.pack_distribution(total_revisions)
917
for pack in self.all_packs():
918
revision_count = pack.get_revision_count()
919
if revision_count == 0:
920
# revision less packs are not generated by normal operation,
921
# only by operations like sign-my-commits, and thus will not
922
# tend to grow rapdily or without bound like commit containing
923
# packs do - leave them alone as packing them really should
924
# group their data with the relevant commit, and that may
925
# involve rewriting ancient history - which autopack tries to
926
# avoid. Alternatively we could not group the data but treat
927
# each of these as having a single revision, and thus add
928
# one revision for each to the total revision count, to get
929
# a matching distribution.
931
existing_packs.append((revision_count, pack))
932
pack_operations = self.plan_autopack_combinations(
933
existing_packs, pack_distribution)
934
self._execute_pack_operations(pack_operations)
937
def _execute_pack_operations(self, pack_operations):
938
"""Execute a series of pack operations.
940
:param pack_operations: A list of [revision_count, packs_to_combine].
943
for revision_count, packs in pack_operations:
944
# we may have no-ops from the setup logic
947
Packer(self, packs, '.autopack').pack()
949
self._remove_pack_from_memory(pack)
950
# record the newly available packs and stop advertising the old
952
self._save_pack_names(clear_obsolete_packs=True)
953
# Move the old packs out of the way now they are no longer referenced.
954
for revision_count, packs in pack_operations:
955
self._obsolete_packs(packs)
957
def lock_names(self):
958
"""Acquire the mutex around the pack-names index.
960
This cannot be used in the middle of a read-only transaction on the
963
self.repo.control_files.lock_write()
966
"""Pack the pack collection totally."""
968
total_packs = len(self._names)
971
total_revisions = self.revision_index.combined_index.key_count()
972
# XXX: the following may want to be a class, to pack with a given
974
mutter('Packing repository %s, which has %d pack files, '
975
'containing %d revisions into 1 packs.', self, total_packs,
977
# determine which packs need changing
978
pack_distribution = [1]
979
pack_operations = [[0, []]]
980
for pack in self.all_packs():
981
revision_count = pack.get_revision_count()
982
pack_operations[-1][0] += revision_count
983
pack_operations[-1][1].append(pack)
984
self._execute_pack_operations(pack_operations)
986
def plan_autopack_combinations(self, existing_packs, pack_distribution):
987
"""Plan a pack operation.
989
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
991
:param pack_distribution: A list with the number of revisions desired
994
if len(existing_packs) <= len(pack_distribution):
996
existing_packs.sort(reverse=True)
997
pack_operations = [[0, []]]
998
# plan out what packs to keep, and what to reorganise
999
while len(existing_packs):
1000
# take the largest pack, and if its less than the head of the
1001
# distribution chart we will include its contents in the new pack for
1002
# that position. If its larger, we remove its size from the
1003
# distribution chart
1004
next_pack_rev_count, next_pack = existing_packs.pop(0)
1005
if next_pack_rev_count >= pack_distribution[0]:
1006
# this is already packed 'better' than this, so we can
1007
# not waste time packing it.
1008
while next_pack_rev_count > 0:
1009
next_pack_rev_count -= pack_distribution[0]
1010
if next_pack_rev_count >= 0:
1012
del pack_distribution[0]
1014
# didn't use that entire bucket up
1015
pack_distribution[0] = -next_pack_rev_count
1017
# add the revisions we're going to add to the next output pack
1018
pack_operations[-1][0] += next_pack_rev_count
1019
# allocate this pack to the next pack sub operation
1020
pack_operations[-1][1].append(next_pack)
1021
if pack_operations[-1][0] >= pack_distribution[0]:
1022
# this pack is used up, shift left.
1023
del pack_distribution[0]
1024
pack_operations.append([0, []])
1026
return pack_operations
1028
def ensure_loaded(self):
1029
# NB: if you see an assertion error here, its probably access against
1030
# an unlocked repo. Naughty.
1031
assert self.repo.is_locked()
1032
if self._names is None:
1034
self._packs_at_load = set()
1035
for index, key, value in self._iter_disk_pack_index():
1037
self._names[name] = self._parse_index_sizes(value)
1038
self._packs_at_load.add((key, value))
1039
# populate all the metadata.
1042
def _parse_index_sizes(self, value):
1043
"""Parse a string of index sizes."""
1044
return tuple([int(digits) for digits in value.split(' ')])
1046
def get_pack_by_name(self, name):
1047
"""Get a Pack object by name.
1049
:param name: The name of the pack - e.g. '123456'
1050
:return: A Pack object.
1053
return self._packs_by_name[name]
1055
rev_index = self._make_index(name, '.rix')
1056
inv_index = self._make_index(name, '.iix')
1057
txt_index = self._make_index(name, '.tix')
1058
sig_index = self._make_index(name, '.six')
1059
result = ExistingPack(self._pack_transport, name, rev_index,
1060
inv_index, txt_index, sig_index)
1061
self.add_pack_to_memory(result)
1064
def allocate(self, a_new_pack):
1065
"""Allocate name in the list of packs.
1067
:param a_new_pack: A NewPack instance to be added to the collection of
1068
packs for this repository.
1070
self.ensure_loaded()
1071
if a_new_pack.name in self._names:
1072
# a collision with the packs we know about (not the only possible
1073
# collision, see NewPack.finish() for some discussion). Remove our
1074
# prior reference to it.
1075
self._remove_pack_from_memory(a_new_pack)
1076
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1077
self.add_pack_to_memory(a_new_pack)
1079
def _iter_disk_pack_index(self):
1080
"""Iterate over the contents of the pack-names index.
1082
This is used when loading the list from disk, and before writing to
1083
detect updates from others during our write operation.
1084
:return: An iterator of the index contents.
1086
return GraphIndex(self.transport, 'pack-names', None
1087
).iter_all_entries()
1089
def _make_index(self, name, suffix):
1090
size_offset = self._suffix_offsets[suffix]
1091
index_name = name + suffix
1092
index_size = self._names[name][size_offset]
1094
self._index_transport, index_name, index_size)
1096
def _max_pack_count(self, total_revisions):
1097
"""Return the maximum number of packs to use for total revisions.
1099
:param total_revisions: The total number of revisions in the
1102
if not total_revisions:
1104
digits = str(total_revisions)
1106
for digit in digits:
1107
result += int(digit)
1111
"""Provide an order to the underlying names."""
1112
return sorted(self._names.keys())
1114
def _obsolete_packs(self, packs):
1115
"""Move a number of packs which have been obsoleted out of the way.
1117
Each pack and its associated indices are moved out of the way.
1119
Note: for correctness this function should only be called after a new
1120
pack names index has been written without these pack names, and with
1121
the names of packs that contain the data previously available via these
1124
:param packs: The packs to obsolete.
1125
:param return: None.
1128
pack.pack_transport.rename(pack.file_name(),
1129
'../obsolete_packs/' + pack.file_name())
1130
# TODO: Probably needs to know all possible indices for this pack
1131
# - or maybe list the directory and move all indices matching this
1132
# name whether we recognize it or not?
1133
for suffix in ('.iix', '.six', '.tix', '.rix'):
1134
self._index_transport.rename(pack.name + suffix,
1135
'../obsolete_packs/' + pack.name + suffix)
1137
def pack_distribution(self, total_revisions):
1138
"""Generate a list of the number of revisions to put in each pack.
1140
:param total_revisions: The total number of revisions in the
1143
if total_revisions == 0:
1145
digits = reversed(str(total_revisions))
1147
for exponent, count in enumerate(digits):
1148
size = 10 ** exponent
1149
for pos in range(int(count)):
1151
return list(reversed(result))
1153
def _pack_tuple(self, name):
1154
"""Return a tuple with the transport and file name for a pack name."""
1155
return self._pack_transport, name + '.pack'
1157
def _remove_pack_from_memory(self, pack):
1158
"""Remove pack from the packs accessed by this repository.
1160
Only affects memory state, until self._save_pack_names() is invoked.
1162
self._names.pop(pack.name)
1163
self._packs_by_name.pop(pack.name)
1164
self._remove_pack_indices(pack)
1166
def _remove_pack_indices(self, pack):
1167
"""Remove the indices for pack from the aggregated indices."""
1168
self.revision_index.remove_index(pack.revision_index, pack)
1169
self.inventory_index.remove_index(pack.inventory_index, pack)
1170
self.text_index.remove_index(pack.text_index, pack)
1171
self.signature_index.remove_index(pack.signature_index, pack)
1174
"""Clear all cached data."""
1175
# cached revision data
1176
self.repo._revision_knit = None
1177
self.revision_index.clear()
1178
# cached signature data
1179
self.repo._signature_knit = None
1180
self.signature_index.clear()
1181
# cached file text data
1182
self.text_index.clear()
1183
self.repo._text_knit = None
1184
# cached inventory data
1185
self.inventory_index.clear()
1186
# remove the open pack
1187
self._new_pack = None
1188
# information about packs.
1191
self._packs_by_name = {}
1192
self._packs_at_load = None
1194
def _make_index_map(self, index_suffix):
1195
"""Return information on existing indices.
1197
:param suffix: Index suffix added to pack name.
1199
:returns: (pack_map, indices) where indices is a list of GraphIndex
1200
objects, and pack_map is a mapping from those objects to the
1201
pack tuple they describe.
1203
# TODO: stop using this; it creates new indices unnecessarily.
1204
self.ensure_loaded()
1205
suffix_map = {'.rix': 'revision_index',
1206
'.six': 'signature_index',
1207
'.iix': 'inventory_index',
1208
'.tix': 'text_index',
1210
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1211
suffix_map[index_suffix])
1213
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1214
"""Convert a list of packs to an index pack map and index list.
1216
:param packs: The packs list to process.
1217
:param index_attribute: The attribute that the desired index is found
1219
:return: A tuple (map, list) where map contains the dict from
1220
index:pack_tuple, and lsit contains the indices in the same order
1226
index = getattr(pack, index_attribute)
1227
indices.append(index)
1228
pack_map[index] = (pack.pack_transport, pack.file_name())
1229
return pack_map, indices
1231
def _index_contents(self, pack_map, key_filter=None):
1232
"""Get an iterable of the index contents from a pack_map.
1234
:param pack_map: A map from indices to pack details.
1235
:param key_filter: An optional filter to limit the
1238
indices = [index for index in pack_map.iterkeys()]
1239
all_index = CombinedGraphIndex(indices)
1240
if key_filter is None:
1241
return all_index.iter_all_entries()
1243
return all_index.iter_entries(key_filter)
1245
def _unlock_names(self):
1246
"""Release the mutex around the pack-names index."""
1247
self.repo.control_files.unlock()
1249
def _save_pack_names(self, clear_obsolete_packs=False):
1250
"""Save the list of packs.
1252
This will take out the mutex around the pack names list for the
1253
duration of the method call. If concurrent updates have been made, a
1254
three-way merge between the current list and the current in memory list
1257
:param clear_obsolete_packs: If True, clear out the contents of the
1258
obsolete_packs directory.
1262
builder = GraphIndexBuilder()
1263
# load the disk nodes across
1265
for index, key, value in self._iter_disk_pack_index():
1266
disk_nodes.add((key, value))
1267
# do a two-way diff against our original content
1268
current_nodes = set()
1269
for name, sizes in self._names.iteritems():
1271
((name, ), ' '.join(str(size) for size in sizes)))
1272
deleted_nodes = self._packs_at_load - current_nodes
1273
new_nodes = current_nodes - self._packs_at_load
1274
disk_nodes.difference_update(deleted_nodes)
1275
disk_nodes.update(new_nodes)
1276
# TODO: handle same-name, index-size-changes here -
1277
# e.g. use the value from disk, not ours, *unless* we're the one
1279
for key, value in disk_nodes:
1280
builder.add_node(key, value)
1281
self.transport.put_file('pack-names', builder.finish(),
1282
mode=self.repo.control_files._file_mode)
1283
# move the baseline forward
1284
self._packs_at_load = disk_nodes
1285
# now clear out the obsolete packs directory
1286
if clear_obsolete_packs:
1287
self.transport.clone('obsolete_packs').delete_multi(
1288
self.transport.list_dir('obsolete_packs'))
1290
self._unlock_names()
1291
# synchronise the memory packs list with what we just wrote:
1292
new_names = dict(disk_nodes)
1293
# drop no longer present nodes
1294
for pack in self.all_packs():
1295
if (pack.name,) not in new_names:
1296
self._remove_pack_from_memory(pack)
1297
# add new nodes/refresh existing ones
1298
for key, value in disk_nodes:
1300
sizes = self._parse_index_sizes(value)
1301
if name in self._names:
1303
if sizes != self._names[name]:
1304
# the pack for name has had its indices replaced - rare but
1305
# important to handle. XXX: probably can never happen today
1306
# because the three-way merge code above does not handle it
1307
# - you may end up adding the same key twice to the new
1308
# disk index because the set values are the same, unless
1309
# the only index shows up as deleted by the set difference
1310
# - which it may. Until there is a specific test for this,
1311
# assume its broken. RBC 20071017.
1312
self._remove_pack_from_memory(self.get_pack_by_name(name))
1313
self._names[name] = sizes
1314
self.get_pack_by_name(name)
1317
self._names[name] = sizes
1318
self.get_pack_by_name(name)
1320
def _start_write_group(self):
1321
# Do not permit preparation for writing if we're not in a 'write lock'.
1322
if not self.repo.is_write_locked():
1323
raise errors.NotWriteLocked(self)
1324
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1325
self._pack_transport, upload_suffix='.pack',
1326
file_mode=self.repo.control_files._file_mode)
1327
# allow writing: queue writes to a new index
1328
self.revision_index.add_writable_index(self._new_pack.revision_index,
1330
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
1332
self.text_index.add_writable_index(self._new_pack.text_index,
1334
self.signature_index.add_writable_index(self._new_pack.signature_index,
1337
# reused revision and signature knits may need updating
1339
# "Hysterical raisins. client code in bzrlib grabs those knits outside
1340
# of write groups and then mutates it inside the write group."
1341
if self.repo._revision_knit is not None:
1342
self.repo._revision_knit._index._add_callback = \
1343
self.revision_index.add_callback
1344
if self.repo._signature_knit is not None:
1345
self.repo._signature_knit._index._add_callback = \
1346
self.signature_index.add_callback
1347
# create a reused knit object for text addition in commit.
1348
self.repo._text_knit = self.repo.weave_store.get_weave_or_empty(
1351
def _abort_write_group(self):
1352
# FIXME: just drop the transient index.
1353
# forget what names there are
1354
self._new_pack.abort()
1355
self._remove_pack_indices(self._new_pack)
1356
self._new_pack = None
1357
self.repo._text_knit = None
1359
def _commit_write_group(self):
1360
self._remove_pack_indices(self._new_pack)
1361
if self._new_pack.data_inserted():
1362
# get all the data to disk and read to use
1363
self._new_pack.finish()
1364
self.allocate(self._new_pack)
1365
self._new_pack = None
1366
if not self.autopack():
1367
# when autopack takes no steps, the names list is still
1369
self._save_pack_names()
1371
self._new_pack.abort()
1372
self._new_pack = None
1373
self.repo._text_knit = None
1376
class KnitPackRevisionStore(KnitRevisionStore):
1377
"""An object to adapt access from RevisionStore's to use KnitPacks.
1379
This class works by replacing the original RevisionStore.
1380
We need to do this because the KnitPackRevisionStore is less
1381
isolated in its layering - it uses services from the repo.
1384
def __init__(self, repo, transport, revisionstore):
1385
"""Create a KnitPackRevisionStore on repo with revisionstore.
1387
This will store its state in the Repository, use the
1388
indices to provide a KnitGraphIndex,
1389
and at the end of transactions write new indices.
1391
KnitRevisionStore.__init__(self, revisionstore.versioned_file_store)
1393
self._serializer = revisionstore._serializer
1394
self.transport = transport
1396
def get_revision_file(self, transaction):
1397
"""Get the revision versioned file object."""
1398
if getattr(self.repo, '_revision_knit', None) is not None:
1399
return self.repo._revision_knit
1400
self.repo._pack_collection.ensure_loaded()
1401
add_callback = self.repo._pack_collection.revision_index.add_callback
1402
# setup knit specific objects
1403
knit_index = KnitGraphIndex(
1404
self.repo._pack_collection.revision_index.combined_index,
1405
add_callback=add_callback)
1406
self.repo._revision_knit = knit.KnitVersionedFile(
1407
'revisions', self.transport.clone('..'),
1408
self.repo.control_files._file_mode,
1409
create=False, access_mode=self.repo._access_mode(),
1410
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1411
access_method=self.repo._pack_collection.revision_index.knit_access)
1412
return self.repo._revision_knit
1414
def get_signature_file(self, transaction):
1415
"""Get the signature versioned file object."""
1416
if getattr(self.repo, '_signature_knit', None) is not None:
1417
return self.repo._signature_knit
1418
self.repo._pack_collection.ensure_loaded()
1419
add_callback = self.repo._pack_collection.signature_index.add_callback
1420
# setup knit specific objects
1421
knit_index = KnitGraphIndex(
1422
self.repo._pack_collection.signature_index.combined_index,
1423
add_callback=add_callback, parents=False)
1424
self.repo._signature_knit = knit.KnitVersionedFile(
1425
'signatures', self.transport.clone('..'),
1426
self.repo.control_files._file_mode,
1427
create=False, access_mode=self.repo._access_mode(),
1428
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1429
access_method=self.repo._pack_collection.signature_index.knit_access)
1430
return self.repo._signature_knit
1433
class KnitPackTextStore(VersionedFileStore):
1434
"""Presents a TextStore abstraction on top of packs.
1436
This class works by replacing the original VersionedFileStore.
1437
We need to do this because the KnitPackRevisionStore is less
1438
isolated in its layering - it uses services from the repo and shares them
1439
with all the data written in a single write group.
1442
def __init__(self, repo, transport, weavestore):
1443
"""Create a KnitPackTextStore on repo with weavestore.
1445
This will store its state in the Repository, use the
1446
indices FileNames to provide a KnitGraphIndex,
1447
and at the end of transactions write new indices.
1449
# don't call base class constructor - it's not suitable.
1450
# no transient data stored in the transaction
1452
self._precious = False
1454
self.transport = transport
1455
self.weavestore = weavestore
1456
# XXX for check() which isn't updated yet
1457
self._transport = weavestore._transport
1459
def get_weave_or_empty(self, file_id, transaction):
1460
"""Get a 'Knit' backed by the .tix indices.
1462
The transaction parameter is ignored.
1464
self.repo._pack_collection.ensure_loaded()
1465
add_callback = self.repo._pack_collection.text_index.add_callback
1466
# setup knit specific objects
1467
file_id_index = GraphIndexPrefixAdapter(
1468
self.repo._pack_collection.text_index.combined_index,
1469
(file_id, ), 1, add_nodes_callback=add_callback)
1470
knit_index = KnitGraphIndex(file_id_index,
1471
add_callback=file_id_index.add_nodes,
1472
deltas=True, parents=True)
1473
return knit.KnitVersionedFile('text:' + file_id,
1474
self.transport.clone('..'),
1477
access_method=self.repo._pack_collection.text_index.knit_access,
1478
factory=knit.KnitPlainFactory())
1480
get_weave = get_weave_or_empty
1483
"""Generate a list of the fileids inserted, for use by check."""
1484
self.repo._pack_collection.ensure_loaded()
1486
for index, key, value, refs in \
1487
self.repo._pack_collection.text_index.combined_index.iter_all_entries():
1492
class InventoryKnitThunk(object):
1493
"""An object to manage thunking get_inventory_weave to pack based knits."""
1495
def __init__(self, repo, transport):
1496
"""Create an InventoryKnitThunk for repo at transport.
1498
This will store its state in the Repository, use the
1499
indices FileNames to provide a KnitGraphIndex,
1500
and at the end of transactions write a new index..
1503
self.transport = transport
1505
def get_weave(self):
1506
"""Get a 'Knit' that contains inventory data."""
1507
self.repo._pack_collection.ensure_loaded()
1508
add_callback = self.repo._pack_collection.inventory_index.add_callback
1509
# setup knit specific objects
1510
knit_index = KnitGraphIndex(
1511
self.repo._pack_collection.inventory_index.combined_index,
1512
add_callback=add_callback, deltas=True, parents=True)
1513
return knit.KnitVersionedFile(
1514
'inventory', self.transport.clone('..'),
1515
self.repo.control_files._file_mode,
1516
create=False, access_mode=self.repo._access_mode(),
1517
index=knit_index, delta=True, factory=knit.KnitPlainFactory(),
1518
access_method=self.repo._pack_collection.inventory_index.knit_access)
1521
class KnitPackRepository(KnitRepository):
1522
"""Experimental graph-knit using repository."""
1524
def __init__(self, _format, a_bzrdir, control_files, _revision_store,
1525
control_store, text_store, _commit_builder_class, _serializer):
1526
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
1527
_revision_store, control_store, text_store, _commit_builder_class,
1529
index_transport = control_files._transport.clone('indices')
1530
self._pack_collection = RepositoryPackCollection(self, control_files._transport,
1532
control_files._transport.clone('upload'),
1533
control_files._transport.clone('packs'))
1534
self._revision_store = KnitPackRevisionStore(self, index_transport, self._revision_store)
1535
self.weave_store = KnitPackTextStore(self, index_transport, self.weave_store)
1536
self._inv_thunk = InventoryKnitThunk(self, index_transport)
1537
# True when the repository object is 'write locked' (as opposed to the
1538
# physical lock only taken out around changes to the pack-names list.)
1539
# Another way to represent this would be a decorator around the control
1540
# files object that presents logical locks as physical ones - if this
1541
# gets ugly consider that alternative design. RBC 20071011
1542
self._write_lock_count = 0
1543
self._transaction = None
1545
self._reconcile_does_inventory_gc = True
1546
self._reconcile_fixes_text_parents = False
1547
self._reconcile_backsup_inventory = False
1549
def _abort_write_group(self):
1550
self._pack_collection._abort_write_group()
1552
def _access_mode(self):
1553
"""Return 'w' or 'r' for depending on whether a write lock is active.
1555
This method is a helper for the Knit-thunking support objects.
1557
if self.is_write_locked():
1561
def _find_inconsistent_revision_parents(self):
1562
"""Find revisions with incorrectly cached parents.
1564
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1565
parents-in-revision).
1567
assert self.is_locked()
1568
pb = ui.ui_factory.nested_progress_bar()
1571
revision_nodes = self._pack_collection.revision_index \
1572
.combined_index.iter_all_entries()
1573
index_positions = []
1574
# Get the cached index values for all revisions, and also the location
1575
# in each index of the revision text so we can perform linear IO.
1576
for index, key, value, refs in revision_nodes:
1577
pos, length = value[1:].split(' ')
1578
index_positions.append((index, int(pos), key[0],
1579
tuple(parent[0] for parent in refs[0])))
1580
pb.update("Reading revision index.", 0, 0)
1581
index_positions.sort()
1582
batch_count = len(index_positions) / 1000 + 1
1583
pb.update("Checking cached revision graph.", 0, batch_count)
1584
for offset in xrange(batch_count):
1585
pb.update("Checking cached revision graph.", offset)
1586
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1589
rev_ids = [item[2] for item in to_query]
1590
revs = self.get_revisions(rev_ids)
1591
for revision, item in zip(revs, to_query):
1592
index_parents = item[3]
1593
rev_parents = tuple(revision.parent_ids)
1594
if index_parents != rev_parents:
1595
result.append((revision.revision_id, index_parents, rev_parents))
1600
def get_parents(self, revision_ids):
1601
"""See StackedParentsProvider.get_parents.
1603
This implementation accesses the combined revision index to provide
1606
self._pack_collection.ensure_loaded()
1607
index = self._pack_collection.revision_index.combined_index
1609
for revision_id in revision_ids:
1610
if revision_id != _mod_revision.NULL_REVISION:
1611
search_keys.add((revision_id,))
1612
found_parents = {_mod_revision.NULL_REVISION:[]}
1613
for index, key, value, refs in index.iter_entries(search_keys):
1616
parents = (_mod_revision.NULL_REVISION,)
1618
parents = tuple(parent[0] for parent in parents)
1619
found_parents[key[0]] = parents
1621
for revision_id in revision_ids:
1623
result.append(found_parents[revision_id])
1628
def _make_parents_provider(self):
1631
def _refresh_data(self):
1632
if self._write_lock_count == 1 or (
1633
self.control_files._lock_count == 1 and
1634
self.control_files._lock_mode == 'r'):
1635
# forget what names there are
1636
self._pack_collection.reset()
1637
# XXX: Better to do an in-memory merge when acquiring a new lock -
1638
# factor out code from _save_pack_names.
1639
self._pack_collection.ensure_loaded()
1641
def _start_write_group(self):
1642
self._pack_collection._start_write_group()
1644
def _commit_write_group(self):
1645
return self._pack_collection._commit_write_group()
1647
def get_inventory_weave(self):
1648
return self._inv_thunk.get_weave()
1650
def get_transaction(self):
1651
if self._write_lock_count:
1652
return self._transaction
1654
return self.control_files.get_transaction()
1656
def is_locked(self):
1657
return self._write_lock_count or self.control_files.is_locked()
1659
def is_write_locked(self):
1660
return self._write_lock_count
1662
def lock_write(self, token=None):
1663
if not self._write_lock_count and self.is_locked():
1664
raise errors.ReadOnlyError(self)
1665
self._write_lock_count += 1
1666
if self._write_lock_count == 1:
1667
from bzrlib import transactions
1668
self._transaction = transactions.WriteTransaction()
1669
self._refresh_data()
1671
def lock_read(self):
1672
if self._write_lock_count:
1673
self._write_lock_count += 1
1675
self.control_files.lock_read()
1676
self._refresh_data()
1678
def leave_lock_in_place(self):
1679
# not supported - raise an error
1680
raise NotImplementedError(self.leave_lock_in_place)
1682
def dont_leave_lock_in_place(self):
1683
# not supported - raise an error
1684
raise NotImplementedError(self.dont_leave_lock_in_place)
1688
"""Compress the data within the repository.
1690
This will pack all the data to a single pack. In future it may
1691
recompress deltas or do other such expensive operations.
1693
self._pack_collection.pack()
1696
def reconcile(self, other=None, thorough=False):
1697
"""Reconcile this repository."""
1698
from bzrlib.reconcile import PackReconciler
1699
reconciler = PackReconciler(self, thorough=thorough)
1700
reconciler.reconcile()
1704
if self._write_lock_count == 1 and self._write_group is not None:
1705
self.abort_write_group()
1706
self._transaction = None
1707
self._write_lock_count = 0
1708
raise errors.BzrError(
1709
'Must end write group before releasing write lock on %s'
1711
if self._write_lock_count:
1712
self._write_lock_count -= 1
1713
if not self._write_lock_count:
1714
transaction = self._transaction
1715
self._transaction = None
1716
transaction.finish()
1718
self.control_files.unlock()
1721
class RepositoryFormatPack(MetaDirRepositoryFormat):
1722
"""Format logic for pack structured repositories.
1724
This repository format has:
1725
- a list of packs in pack-names
1726
- packs in packs/NAME.pack
1727
- indices in indices/NAME.{iix,six,tix,rix}
1728
- knit deltas in the packs, knit indices mapped to the indices.
1729
- thunk objects to support the knits programming API.
1730
- a format marker of its own
1731
- an optional 'shared-storage' flag
1732
- an optional 'no-working-trees' flag
1736
# Set this attribute in derived classes to control the repository class
1737
# created by open and initialize.
1738
repository_class = None
1739
# Set this attribute in derived classes to control the
1740
# _commit_builder_class that the repository objects will have passed to
1741
# their constructor.
1742
_commit_builder_class = None
1743
# Set this attribute in derived clases to control the _serializer that the
1744
# repository objects will have passed to their constructor.
1747
def _get_control_store(self, repo_transport, control_files):
1748
"""Return the control store for this repository."""
1749
return VersionedFileStore(
1752
file_mode=control_files._file_mode,
1753
versionedfile_class=knit.KnitVersionedFile,
1754
versionedfile_kwargs={'factory': knit.KnitPlainFactory()},
1757
def _get_revision_store(self, repo_transport, control_files):
1758
"""See RepositoryFormat._get_revision_store()."""
1759
versioned_file_store = VersionedFileStore(
1761
file_mode=control_files._file_mode,
1764
versionedfile_class=knit.KnitVersionedFile,
1765
versionedfile_kwargs={'delta': False,
1766
'factory': knit.KnitPlainFactory(),
1770
return KnitRevisionStore(versioned_file_store)
1772
def _get_text_store(self, transport, control_files):
1773
"""See RepositoryFormat._get_text_store()."""
1774
return self._get_versioned_file_store('knits',
1777
versionedfile_class=knit.KnitVersionedFile,
1778
versionedfile_kwargs={
1779
'create_parent_dir': True,
1780
'delay_create': True,
1781
'dir_mode': control_files._dir_mode,
1785
def initialize(self, a_bzrdir, shared=False):
1786
"""Create a pack based repository.
1788
:param a_bzrdir: bzrdir to contain the new repository; must already
1790
:param shared: If true the repository will be initialized as a shared
1793
mutter('creating repository in %s.', a_bzrdir.transport.base)
1794
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
1795
builder = GraphIndexBuilder()
1796
files = [('pack-names', builder.finish())]
1797
utf8_files = [('format', self.get_format_string())]
1799
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
1800
return self.open(a_bzrdir=a_bzrdir, _found=True)
1802
def open(self, a_bzrdir, _found=False, _override_transport=None):
1803
"""See RepositoryFormat.open().
1805
:param _override_transport: INTERNAL USE ONLY. Allows opening the
1806
repository at a slightly different url
1807
than normal. I.e. during 'upgrade'.
1810
format = RepositoryFormat.find_format(a_bzrdir)
1811
assert format.__class__ == self.__class__
1812
if _override_transport is not None:
1813
repo_transport = _override_transport
1815
repo_transport = a_bzrdir.get_repository_transport(None)
1816
control_files = lockable_files.LockableFiles(repo_transport,
1817
'lock', lockdir.LockDir)
1818
text_store = self._get_text_store(repo_transport, control_files)
1819
control_store = self._get_control_store(repo_transport, control_files)
1820
_revision_store = self._get_revision_store(repo_transport, control_files)
1821
return self.repository_class(_format=self,
1823
control_files=control_files,
1824
_revision_store=_revision_store,
1825
control_store=control_store,
1826
text_store=text_store,
1827
_commit_builder_class=self._commit_builder_class,
1828
_serializer=self._serializer)
1831
class RepositoryFormatKnitPack1(RepositoryFormatPack):
1832
"""A no-subtrees parameterised Pack repository.
1834
This format was introduced in 0.92.
1837
repository_class = KnitPackRepository
1838
_commit_builder_class = PackCommitBuilder
1839
_serializer = xml5.serializer_v5
1841
def _get_matching_bzrdir(self):
1842
return bzrdir.format_registry.make_bzrdir('pack-0.92')
1844
def _ignore_setting_bzrdir(self, format):
1847
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1849
def get_format_string(self):
1850
"""See RepositoryFormat.get_format_string()."""
1851
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
1853
def get_format_description(self):
1854
"""See RepositoryFormat.get_format_description()."""
1855
return "Packs containing knits without subtree support"
1857
def check_conversion_target(self, target_format):
1861
class RepositoryFormatKnitPack3(RepositoryFormatPack):
1862
"""A subtrees parameterised Pack repository.
1864
This repository format uses the xml7 serializer to get:
1865
- support for recording full info about the tree root
1866
- support for recording tree-references
1868
This format was introduced in 0.92.
1871
repository_class = KnitPackRepository
1872
_commit_builder_class = PackRootCommitBuilder
1873
rich_root_data = True
1874
supports_tree_reference = True
1875
_serializer = xml7.serializer_v7
1877
def _get_matching_bzrdir(self):
1878
return bzrdir.format_registry.make_bzrdir(
1879
'pack-0.92-subtree')
1881
def _ignore_setting_bzrdir(self, format):
1884
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1886
def check_conversion_target(self, target_format):
1887
if not target_format.rich_root_data:
1888
raise errors.BadConversionTarget(
1889
'Does not support rich root data.', target_format)
1890
if not getattr(target_format, 'supports_tree_reference', False):
1891
raise errors.BadConversionTarget(
1892
'Does not support nested trees', target_format)
1894
def get_format_string(self):
1895
"""See RepositoryFormat.get_format_string()."""
1896
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
1898
def get_format_description(self):
1899
"""See RepositoryFormat.get_format_description()."""
1900
return "Packs containing knits with subtree support\n"
1903
class RepositoryFormatKnitPack4(RepositoryFormatPack):
1904
"""A rich-root, no subtrees parameterised Pack repository.
1906
This repository format uses the xml6 serializer to get:
1907
- support for recording full info about the tree root
1909
This format was introduced in 1.0.
1912
repository_class = KnitPackRepository
1913
_commit_builder_class = PackRootCommitBuilder
1914
rich_root_data = True
1915
supports_tree_reference = False
1916
_serializer = xml6.serializer_v6
1918
def _get_matching_bzrdir(self):
1919
return bzrdir.format_registry.make_bzrdir(
1922
def _ignore_setting_bzrdir(self, format):
1925
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1927
def check_conversion_target(self, target_format):
1928
if not target_format.rich_root_data:
1929
raise errors.BadConversionTarget(
1930
'Does not support rich root data.', target_format)
1932
def get_format_string(self):
1933
"""See RepositoryFormat.get_format_string()."""
1934
return ("Bazaar pack repository format 1 with rich root"
1935
" (needs bzr 1.0)\n")
1937
def get_format_description(self):
1938
"""See RepositoryFormat.get_format_description()."""
1939
return "Packs containing knits with rich root support\n"
added
removed
bzrlib/repofmt/pack_repo.py
