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.index import (
34
GraphIndexPrefixAdapter,
36
from bzrlib.knit import KnitGraphIndex, _PackAccess, _KnitData
37
from bzrlib.osutils import rand_chars
38
from bzrlib.pack import ContainerWriter
39
from bzrlib.store import revision
54
from bzrlib.decorators import needs_read_lock, needs_write_lock
55
from bzrlib.repofmt.knitrepo import KnitRepository
56
from bzrlib.repository import (
59
MetaDirRepositoryFormat,
62
import bzrlib.revision as _mod_revision
63
from bzrlib.store.revision.knit import KnitRevisionStore
64
from bzrlib.store.versioned import VersionedFileStore
65
from bzrlib.trace import mutter, note, warning
68
class PackCommitBuilder(CommitBuilder):
69
"""A subclass of CommitBuilder to add texts with pack semantics.
71
Specifically this uses one knit object rather than one knit object per
72
added text, reducing memory and object pressure.
75
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
76
return self.repository._pack_collection._add_text_to_weave(file_id,
77
self._new_revision_id, new_lines, parents, nostore_sha,
81
class PackRootCommitBuilder(RootCommitBuilder):
82
"""A subclass of RootCommitBuilder to add texts with pack semantics.
84
Specifically this uses one knit object rather than one knit object per
85
added text, reducing memory and object pressure.
88
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
89
return self.repository._pack_collection._add_text_to_weave(file_id,
90
self._new_revision_id, new_lines, parents, nostore_sha,
95
"""An in memory proxy for a pack and its indices.
97
This is a base class that is not directly used, instead the classes
98
ExistingPack and NewPack are used.
101
def __init__(self, revision_index, inventory_index, text_index,
103
"""Create a pack instance.
105
:param revision_index: A GraphIndex for determining what revisions are
106
present in the Pack and accessing the locations of their texts.
107
:param inventory_index: A GraphIndex for determining what inventories are
108
present in the Pack and accessing the locations of their
110
:param text_index: A GraphIndex for determining what file texts
111
are present in the pack and accessing the locations of their
112
texts/deltas (via (fileid, revisionid) tuples).
113
:param revision_index: A GraphIndex for determining what signatures are
114
present in the Pack and accessing the locations of their texts.
116
self.revision_index = revision_index
117
self.inventory_index = inventory_index
118
self.text_index = text_index
119
self.signature_index = signature_index
121
def access_tuple(self):
122
"""Return a tuple (transport, name) for the pack content."""
123
return self.pack_transport, self.file_name()
126
"""Get the file name for the pack on disk."""
127
return self.name + '.pack'
129
def get_revision_count(self):
130
return self.revision_index.key_count()
132
def inventory_index_name(self, name):
133
"""The inv index is the name + .iix."""
134
return self.index_name('inventory', name)
136
def revision_index_name(self, name):
137
"""The revision index is the name + .rix."""
138
return self.index_name('revision', name)
140
def signature_index_name(self, name):
141
"""The signature index is the name + .six."""
142
return self.index_name('signature', name)
144
def text_index_name(self, name):
145
"""The text index is the name + .tix."""
146
return self.index_name('text', name)
149
class ExistingPack(Pack):
150
"""An in memory proxy for an existing .pack and its disk indices."""
152
def __init__(self, pack_transport, name, revision_index, inventory_index,
153
text_index, signature_index):
154
"""Create an ExistingPack object.
156
:param pack_transport: The transport where the pack file resides.
157
:param name: The name of the pack on disk in the pack_transport.
159
Pack.__init__(self, revision_index, inventory_index, text_index,
162
self.pack_transport = pack_transport
163
assert None not in (revision_index, inventory_index, text_index,
164
signature_index, name, pack_transport)
166
def __eq__(self, other):
167
return self.__dict__ == other.__dict__
169
def __ne__(self, other):
170
return not self.__eq__(other)
173
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
174
id(self), self.transport, self.name)
178
"""An in memory proxy for a pack which is being created."""
180
# A map of index 'type' to the file extension and position in the
182
index_definitions = {
183
'revision': ('.rix', 0),
184
'inventory': ('.iix', 1),
186
'signature': ('.six', 3),
189
def __init__(self, upload_transport, index_transport, pack_transport,
190
upload_suffix='', file_mode=None):
191
"""Create a NewPack instance.
193
:param upload_transport: A writable transport for the pack to be
194
incrementally uploaded to.
195
:param index_transport: A writable transport for the pack's indices to
196
be written to when the pack is finished.
197
:param pack_transport: A writable transport for the pack to be renamed
198
to when the upload is complete. This *must* be the same as
199
upload_transport.clone('../packs').
200
:param upload_suffix: An optional suffix to be given to any temporary
201
files created during the pack creation. e.g '.autopack'
202
:param file_mode: An optional file mode to create the new files with.
204
# The relative locations of the packs are constrained, but all are
205
# passed in because the caller has them, so as to avoid object churn.
207
# Revisions: parents list, no text compression.
208
InMemoryGraphIndex(reference_lists=1),
209
# Inventory: We want to map compression only, but currently the
210
# knit code hasn't been updated enough to understand that, so we
211
# have a regular 2-list index giving parents and compression
213
InMemoryGraphIndex(reference_lists=2),
214
# Texts: compression and per file graph, for all fileids - so two
215
# reference lists and two elements in the key tuple.
216
InMemoryGraphIndex(reference_lists=2, key_elements=2),
217
# Signatures: Just blobs to store, no compression, no parents
219
InMemoryGraphIndex(reference_lists=0),
221
# where should the new pack be opened
222
self.upload_transport = upload_transport
223
# where are indices written out to
224
self.index_transport = index_transport
225
# where is the pack renamed to when it is finished?
226
self.pack_transport = pack_transport
227
# What file mode to upload the pack and indices with.
228
self._file_mode = file_mode
229
# tracks the content written to the .pack file.
230
self._hash = md5.new()
231
# a four-tuple with the length in bytes of the indices, once the pack
232
# is finalised. (rev, inv, text, sigs)
233
self.index_sizes = None
234
# How much data to cache when writing packs. Note that this is not
235
# synchronised with reads, because it's not in the transport layer, so
236
# is not safe unless the client knows it won't be reading from the pack
238
self._cache_limit = 0
239
# the temporary pack file name.
240
self.random_name = rand_chars(20) + upload_suffix
241
# when was this pack started ?
242
self.start_time = time.time()
243
# open an output stream for the data added to the pack.
244
self.write_stream = self.upload_transport.open_write_stream(
245
self.random_name, mode=self._file_mode)
246
if 'pack' in debug.debug_flags:
247
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
248
time.ctime(), self.upload_transport.base, self.random_name,
249
time.time() - self.start_time)
250
# A list of byte sequences to be written to the new pack, and the
251
# aggregate size of them. Stored as a list rather than separate
252
# variables so that the _write_data closure below can update them.
253
self._buffer = [[], 0]
254
# create a callable for adding data
256
# robertc says- this is a closure rather than a method on the object
257
# so that the variables are locals, and faster than accessing object
259
def _write_data(bytes, flush=False, _buffer=self._buffer,
260
_write=self.write_stream.write, _update=self._hash.update):
261
_buffer[0].append(bytes)
262
_buffer[1] += len(bytes)
264
if _buffer[1] > self._cache_limit or flush:
265
bytes = ''.join(_buffer[0])
269
# expose this on self, for the occasion when clients want to add data.
270
self._write_data = _write_data
271
# a pack writer object to serialise pack records.
272
self._writer = pack.ContainerWriter(self._write_data)
274
# what state is the pack in? (open, finished, aborted)
278
"""Cancel creating this pack."""
279
self._state = 'aborted'
280
self.write_stream.close()
281
# Remove the temporary pack file.
282
self.upload_transport.delete(self.random_name)
283
# The indices have no state on disk.
285
def access_tuple(self):
286
"""Return a tuple (transport, name) for the pack content."""
287
assert self._state in ('open', 'finished')
288
if self._state == 'finished':
289
return Pack.access_tuple(self)
291
return self.upload_transport, self.random_name
293
def data_inserted(self):
294
"""True if data has been added to this pack."""
295
return bool(self.get_revision_count() or
296
self.inventory_index.key_count() or
297
self.text_index.key_count() or
298
self.signature_index.key_count())
301
"""Finish the new pack.
304
- finalises the content
305
- assigns a name (the md5 of the content, currently)
306
- writes out the associated indices
307
- renames the pack into place.
308
- stores the index size tuple for the pack in the index_sizes
313
self._write_data('', flush=True)
314
self.name = self._hash.hexdigest()
316
# XXX: It'd be better to write them all to temporary names, then
317
# rename them all into place, so that the window when only some are
318
# visible is smaller. On the other hand none will be seen until
319
# they're in the names list.
320
self.index_sizes = [None, None, None, None]
321
self._write_index('revision', self.revision_index, 'revision')
322
self._write_index('inventory', self.inventory_index, 'inventory')
323
self._write_index('text', self.text_index, 'file texts')
324
self._write_index('signature', self.signature_index,
325
'revision signatures')
326
self.write_stream.close()
327
# Note that this will clobber an existing pack with the same name,
328
# without checking for hash collisions. While this is undesirable this
329
# is something that can be rectified in a subsequent release. One way
330
# to rectify it may be to leave the pack at the original name, writing
331
# its pack-names entry as something like 'HASH: index-sizes
332
# temporary-name'. Allocate that and check for collisions, if it is
333
# collision free then rename it into place. If clients know this scheme
334
# they can handle missing-file errors by:
335
# - try for HASH.pack
336
# - try for temporary-name
337
# - refresh the pack-list to see if the pack is now absent
338
self.upload_transport.rename(self.random_name,
339
'../packs/' + self.name + '.pack')
340
self._state = 'finished'
341
if 'pack' in debug.debug_flags:
342
# XXX: size might be interesting?
343
mutter('%s: create_pack: pack renamed into place: %s%s->%s%s t+%6.3fs',
344
time.ctime(), self.upload_transport.base, self.random_name,
345
self.pack_transport, self.name,
346
time.time() - self.start_time)
348
def index_name(self, index_type, name):
349
"""Get the disk name of an index type for pack name 'name'."""
350
return name + NewPack.index_definitions[index_type][0]
352
def index_offset(self, index_type):
353
"""Get the position in a index_size array for a given index type."""
354
return NewPack.index_definitions[index_type][1]
356
def _replace_index_with_readonly(self, index_type):
357
setattr(self, index_type + '_index',
358
GraphIndex(self.index_transport,
359
self.index_name(index_type, self.name),
360
self.index_sizes[self.index_offset(index_type)]))
362
def set_write_cache_size(self, size):
363
self._cache_limit = size
365
def _write_index(self, index_type, index, label):
366
"""Write out an index.
368
:param index_type: The type of index to write - e.g. 'revision'.
369
:param index: The index object to serialise.
370
:param label: What label to give the index e.g. 'revision'.
372
index_name = self.index_name(index_type, self.name)
373
self.index_sizes[self.index_offset(index_type)] = \
374
self.index_transport.put_file(index_name, index.finish(),
375
mode=self._file_mode)
376
if 'pack' in debug.debug_flags:
377
# XXX: size might be interesting?
378
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
379
time.ctime(), label, self.upload_transport.base,
380
self.random_name, time.time() - self.start_time)
381
# Replace the writable index on this object with a readonly,
382
# presently unloaded index. We should alter
383
# the index layer to make its finish() error if add_node is
384
# subsequently used. RBC
385
self._replace_index_with_readonly(index_type)
388
class AggregateIndex(object):
389
"""An aggregated index for the RepositoryPackCollection.
391
AggregateIndex is reponsible for managing the PackAccess object,
392
Index-To-Pack mapping, and all indices list for a specific type of index
393
such as 'revision index'.
395
A CombinedIndex provides an index on a single key space built up
396
from several on-disk indices. The AggregateIndex builds on this
397
to provide a knit access layer, and allows having up to one writable
398
index within the collection.
400
# XXX: Probably 'can be written to' could/should be separated from 'acts
401
# like a knit index' -- mbp 20071024
404
"""Create an AggregateIndex."""
405
self.index_to_pack = {}
406
self.combined_index = CombinedGraphIndex([])
407
self.knit_access = _PackAccess(self.index_to_pack)
409
def replace_indices(self, index_to_pack, indices):
410
"""Replace the current mappings with fresh ones.
412
This should probably not be used eventually, rather incremental add and
413
removal of indices. It has been added during refactoring of existing
416
:param index_to_pack: A mapping from index objects to
417
(transport, name) tuples for the pack file data.
418
:param indices: A list of indices.
420
# refresh the revision pack map dict without replacing the instance.
421
self.index_to_pack.clear()
422
self.index_to_pack.update(index_to_pack)
423
# XXX: API break - clearly a 'replace' method would be good?
424
self.combined_index._indices[:] = indices
425
# the current add nodes callback for the current writable index if
427
self.add_callback = None
429
def add_index(self, index, pack):
430
"""Add index to the aggregate, which is an index for Pack pack.
432
Future searches on the aggregate index will seach this new index
433
before all previously inserted indices.
435
:param index: An Index for the pack.
436
:param pack: A Pack instance.
438
# expose it to the index map
439
self.index_to_pack[index] = pack.access_tuple()
440
# put it at the front of the linear index list
441
self.combined_index.insert_index(0, index)
443
def add_writable_index(self, index, pack):
444
"""Add an index which is able to have data added to it.
446
There can be at most one writable index at any time. Any
447
modifications made to the knit are put into this index.
449
:param index: An index from the pack parameter.
450
:param pack: A Pack instance.
452
assert self.add_callback is None, \
453
"%s already has a writable index through %s" % \
454
(self, self.add_callback)
455
# allow writing: queue writes to a new index
456
self.add_index(index, pack)
457
# Updates the index to packs mapping as a side effect,
458
self.knit_access.set_writer(pack._writer, index, pack.access_tuple())
459
self.add_callback = index.add_nodes
462
"""Reset all the aggregate data to nothing."""
463
self.knit_access.set_writer(None, None, (None, None))
464
self.index_to_pack.clear()
465
del self.combined_index._indices[:]
466
self.add_callback = None
468
def remove_index(self, index, pack):
469
"""Remove index from the indices used to answer queries.
471
:param index: An index from the pack parameter.
472
:param pack: A Pack instance.
474
del self.index_to_pack[index]
475
self.combined_index._indices.remove(index)
476
if (self.add_callback is not None and
477
getattr(index, 'add_nodes', None) == self.add_callback):
478
self.add_callback = None
479
self.knit_access.set_writer(None, None, (None, None))
482
class Packer(object):
483
"""Create a pack from packs."""
485
def __init__(self, pack_collection, packs, suffix, revision_ids=None):
488
:param pack_collection: A RepositoryPackCollection object where the
489
new pack is being written to.
490
:param packs: The packs to combine.
491
:param suffix: The suffix to use on the temporary files for the pack.
492
:param revision_ids: Revision ids to limit the pack to.
496
self.revision_ids = revision_ids
497
self._pack_collection = pack_collection
499
def pack(self, pb=None):
500
"""Create a new pack by reading data from other packs.
502
This does little more than a bulk copy of data. One key difference
503
is that data with the same item key across multiple packs is elided
504
from the output. The new pack is written into the current pack store
505
along with its indices, and the name added to the pack names. The
506
source packs are not altered and are not required to be in the current
509
:param pb: An optional progress bar to use. A nested bar is created if
511
:return: A Pack object, or None if nothing was copied.
513
# open a pack - using the same name as the last temporary file
514
# - which has already been flushed, so its safe.
515
# XXX: - duplicate code warning with start_write_group; fix before
516
# considering 'done'.
517
if self._pack_collection._new_pack is not None:
518
raise errors.BzrError('call to create_pack_from_packs while '
519
'another pack is being written.')
520
if self.revision_ids is not None:
521
if len(self.revision_ids) == 0:
522
# silly fetch request.
525
self.revision_ids = frozenset(self.revision_ids)
527
self.pb = ui.ui_factory.nested_progress_bar()
531
return self._create_pack_from_packs()
537
"""Open a pack for the pack we are creating."""
538
return NewPack(self._pack_collection._upload_transport,
539
self._pack_collection._index_transport,
540
self._pack_collection._pack_transport, upload_suffix=self.suffix,
541
file_mode=self._pack_collection.repo.control_files._file_mode)
543
def _create_pack_from_packs(self):
544
self.pb.update("Opening pack", 0, 5)
545
new_pack = self.open_pack()
546
# buffer data - we won't be reading-back during the pack creation and
547
# this makes a significant difference on sftp pushes.
548
new_pack.set_write_cache_size(1024*1024)
549
if 'pack' in debug.debug_flags:
550
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
551
for a_pack in self.packs]
552
if self.revision_ids is not None:
553
rev_count = len(self.revision_ids)
556
mutter('%s: create_pack: creating pack from source packs: '
557
'%s%s %s revisions wanted %s t=0',
558
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
559
plain_pack_list, rev_count)
561
if self.revision_ids:
562
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
566
# select revision keys
567
revision_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
568
self.packs, 'revision_index')[0]
569
revision_nodes = self._pack_collection._index_contents(revision_index_map, revision_keys)
570
# copy revision keys and adjust values
571
self.pb.update("Copying revision texts", 1)
572
list(self._copy_nodes_graph(revision_nodes, revision_index_map,
573
new_pack._writer, new_pack.revision_index))
574
if 'pack' in debug.debug_flags:
575
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
576
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
577
new_pack.revision_index.key_count(),
578
time.time() - new_pack.start_time)
579
# select inventory keys
580
inv_keys = revision_keys # currently the same keyspace, and note that
581
# querying for keys here could introduce a bug where an inventory item
582
# is missed, so do not change it to query separately without cross
583
# checking like the text key check below.
584
inventory_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
585
self.packs, 'inventory_index')[0]
586
inv_nodes = self._pack_collection._index_contents(inventory_index_map, inv_keys)
587
# copy inventory keys and adjust values
588
# XXX: Should be a helper function to allow different inv representation
590
self.pb.update("Copying inventory texts", 2)
591
inv_lines = self._copy_nodes_graph(inv_nodes, inventory_index_map,
592
new_pack._writer, new_pack.inventory_index, output_lines=True)
593
if self.revision_ids:
594
fileid_revisions = self._pack_collection.repo._find_file_ids_from_xml_inventory_lines(
595
inv_lines, self.revision_ids)
597
for fileid, file_revids in fileid_revisions.iteritems():
599
[(fileid, file_revid) for file_revid in file_revids])
601
# eat the iterator to cause it to execute.
604
if 'pack' in debug.debug_flags:
605
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
606
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
607
new_pack.inventory_index.key_count(),
608
time.time() - new_pack.start_time)
610
text_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
611
self.packs, 'text_index')[0]
612
text_nodes = self._pack_collection._index_contents(text_index_map, text_filter)
613
if text_filter is not None:
614
# We could return the keys copied as part of the return value from
615
# _copy_nodes_graph but this doesn't work all that well with the
616
# need to get line output too, so we check separately, and as we're
617
# going to buffer everything anyway, we check beforehand, which
618
# saves reading knit data over the wire when we know there are
620
text_nodes = set(text_nodes)
621
present_text_keys = set(_node[1] for _node in text_nodes)
622
missing_text_keys = set(text_filter) - present_text_keys
623
if missing_text_keys:
624
# TODO: raise a specific error that can handle many missing
626
a_missing_key = missing_text_keys.pop()
627
raise errors.RevisionNotPresent(a_missing_key[1],
629
# copy text keys and adjust values
630
self.pb.update("Copying content texts", 3)
631
list(self._copy_nodes_graph(text_nodes, text_index_map,
632
new_pack._writer, new_pack.text_index))
633
if 'pack' in debug.debug_flags:
634
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
635
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
636
new_pack.text_index.key_count(),
637
time.time() - new_pack.start_time)
638
# select signature keys
639
signature_filter = revision_keys # same keyspace
640
signature_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
641
self.packs, 'signature_index')[0]
642
signature_nodes = self._pack_collection._index_contents(signature_index_map,
644
# copy signature keys and adjust values
645
self.pb.update("Copying signature texts", 4)
646
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
647
new_pack.signature_index)
648
if 'pack' in debug.debug_flags:
649
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
650
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
651
new_pack.signature_index.key_count(),
652
time.time() - new_pack.start_time)
653
if not new_pack.data_inserted():
656
self.pb.update("Finishing pack", 5)
658
self._pack_collection.allocate(new_pack)
661
def _copy_nodes(self, nodes, index_map, writer, write_index):
662
"""Copy knit nodes between packs with no graph references."""
663
pb = ui.ui_factory.nested_progress_bar()
665
return self._do_copy_nodes(nodes, index_map, writer,
670
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb):
671
# for record verification
672
knit_data = _KnitData(None)
673
# plan a readv on each source pack:
675
nodes = sorted(nodes)
676
# how to map this into knit.py - or knit.py into this?
677
# we don't want the typical knit logic, we want grouping by pack
678
# at this point - perhaps a helper library for the following code
679
# duplication points?
681
for index, key, value in nodes:
682
if index not in request_groups:
683
request_groups[index] = []
684
request_groups[index].append((key, value))
686
pb.update("Copied record", record_index, len(nodes))
687
for index, items in request_groups.iteritems():
688
pack_readv_requests = []
689
for key, value in items:
690
# ---- KnitGraphIndex.get_position
691
bits = value[1:].split(' ')
692
offset, length = int(bits[0]), int(bits[1])
693
pack_readv_requests.append((offset, length, (key, value[0])))
694
# linear scan up the pack
695
pack_readv_requests.sort()
697
transport, path = index_map[index]
698
reader = pack.make_readv_reader(transport, path,
699
[offset[0:2] for offset in pack_readv_requests])
700
for (names, read_func), (_1, _2, (key, eol_flag)) in \
701
izip(reader.iter_records(), pack_readv_requests):
702
raw_data = read_func(None)
703
# check the header only
704
df, _ = knit_data._parse_record_header(key[-1], raw_data)
706
pos, size = writer.add_bytes_record(raw_data, names)
707
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
708
pb.update("Copied record", record_index)
711
def _copy_nodes_graph(self, nodes, index_map, writer, write_index,
713
"""Copy knit nodes between packs.
715
:param output_lines: Return lines present in the copied data as
716
an iterator of line,version_id.
718
pb = ui.ui_factory.nested_progress_bar()
720
return self._do_copy_nodes_graph(nodes, index_map, writer,
721
write_index, output_lines, pb)
725
def _do_copy_nodes_graph(self, nodes, index_map, writer, write_index,
727
# for record verification
728
knit_data = _KnitData(None)
729
# for line extraction when requested (inventories only)
731
factory = knit.KnitPlainFactory()
732
# plan a readv on each source pack:
734
nodes = sorted(nodes)
735
# how to map this into knit.py - or knit.py into this?
736
# we don't want the typical knit logic, we want grouping by pack
737
# at this point - perhaps a helper library for the following code
738
# duplication points?
741
pb.update("Copied record", record_index, len(nodes))
742
for index, key, value, references in nodes:
743
if index not in request_groups:
744
request_groups[index] = []
745
request_groups[index].append((key, value, references))
746
for index, items in request_groups.iteritems():
747
pack_readv_requests = []
748
for key, value, references in items:
749
# ---- KnitGraphIndex.get_position
750
bits = value[1:].split(' ')
751
offset, length = int(bits[0]), int(bits[1])
752
pack_readv_requests.append((offset, length, (key, value[0], references)))
753
# linear scan up the pack
754
pack_readv_requests.sort()
756
transport, path = index_map[index]
757
reader = pack.make_readv_reader(transport, path,
758
[offset[0:2] for offset in pack_readv_requests])
759
for (names, read_func), (_1, _2, (key, eol_flag, references)) in \
760
izip(reader.iter_records(), pack_readv_requests):
761
raw_data = read_func(None)
764
# read the entire thing
765
content, _ = knit_data._parse_record(version_id, raw_data)
766
if len(references[-1]) == 0:
767
line_iterator = factory.get_fulltext_content(content)
769
line_iterator = factory.get_linedelta_content(content)
770
for line in line_iterator:
771
yield line, version_id
773
# check the header only
774
df, _ = knit_data._parse_record_header(version_id, raw_data)
776
pos, size = writer.add_bytes_record(raw_data, names)
777
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
778
pb.update("Copied record", record_index)
782
class ReconcilePacker(Packer):
783
"""A packer which regenerates indices etc as it copies.
785
This is used by ``bzr reconcile`` to cause parent text pointers to be
790
class RepositoryPackCollection(object):
791
"""Management of packs within a repository."""
793
def __init__(self, repo, transport, index_transport, upload_transport,
795
"""Create a new RepositoryPackCollection.
797
:param transport: Addresses the repository base directory
798
(typically .bzr/repository/).
799
:param index_transport: Addresses the directory containing indices.
800
:param upload_transport: Addresses the directory into which packs are written
801
while they're being created.
802
:param pack_transport: Addresses the directory of existing complete packs.
805
self.transport = transport
806
self._index_transport = index_transport
807
self._upload_transport = upload_transport
808
self._pack_transport = pack_transport
809
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
812
self._packs_by_name = {}
813
# the previous pack-names content
814
self._packs_at_load = None
815
# when a pack is being created by this object, the state of that pack.
816
self._new_pack = None
817
# aggregated revision index data
818
self.revision_index = AggregateIndex()
819
self.inventory_index = AggregateIndex()
820
self.text_index = AggregateIndex()
821
self.signature_index = AggregateIndex()
823
def add_pack_to_memory(self, pack):
824
"""Make a Pack object available to the repository to satisfy queries.
826
:param pack: A Pack object.
828
assert pack.name not in self._packs_by_name
829
self.packs.append(pack)
830
self._packs_by_name[pack.name] = pack
831
self.revision_index.add_index(pack.revision_index, pack)
832
self.inventory_index.add_index(pack.inventory_index, pack)
833
self.text_index.add_index(pack.text_index, pack)
834
self.signature_index.add_index(pack.signature_index, pack)
836
def _add_text_to_weave(self, file_id, revision_id, new_lines, parents,
837
nostore_sha, random_revid):
838
file_id_index = GraphIndexPrefixAdapter(
839
self.text_index.combined_index,
841
add_nodes_callback=self.text_index.add_callback)
842
self.repo._text_knit._index._graph_index = file_id_index
843
self.repo._text_knit._index._add_callback = file_id_index.add_nodes
844
return self.repo._text_knit.add_lines_with_ghosts(
845
revision_id, parents, new_lines, nostore_sha=nostore_sha,
846
random_id=random_revid, check_content=False)[0:2]
849
"""Return a list of all the Pack objects this repository has.
851
Note that an in-progress pack being created is not returned.
853
:return: A list of Pack objects for all the packs in the repository.
856
for name in self.names():
857
result.append(self.get_pack_by_name(name))
861
"""Pack the pack collection incrementally.
863
This will not attempt global reorganisation or recompression,
864
rather it will just ensure that the total number of packs does
865
not grow without bound. It uses the _max_pack_count method to
866
determine if autopacking is needed, and the pack_distribution
867
method to determine the number of revisions in each pack.
869
If autopacking takes place then the packs name collection will have
870
been flushed to disk - packing requires updating the name collection
871
in synchronisation with certain steps. Otherwise the names collection
874
:return: True if packing took place.
876
# XXX: Should not be needed when the management of indices is sane.
877
total_revisions = self.revision_index.combined_index.key_count()
878
total_packs = len(self._names)
879
if self._max_pack_count(total_revisions) >= total_packs:
881
# XXX: the following may want to be a class, to pack with a given
883
mutter('Auto-packing repository %s, which has %d pack files, '
884
'containing %d revisions into %d packs.', self, total_packs,
885
total_revisions, self._max_pack_count(total_revisions))
886
# determine which packs need changing
887
pack_distribution = self.pack_distribution(total_revisions)
889
for pack in self.all_packs():
890
revision_count = pack.get_revision_count()
891
if revision_count == 0:
892
# revision less packs are not generated by normal operation,
893
# only by operations like sign-my-commits, and thus will not
894
# tend to grow rapdily or without bound like commit containing
895
# packs do - leave them alone as packing them really should
896
# group their data with the relevant commit, and that may
897
# involve rewriting ancient history - which autopack tries to
898
# avoid. Alternatively we could not group the data but treat
899
# each of these as having a single revision, and thus add
900
# one revision for each to the total revision count, to get
901
# a matching distribution.
903
existing_packs.append((revision_count, pack))
904
pack_operations = self.plan_autopack_combinations(
905
existing_packs, pack_distribution)
906
self._execute_pack_operations(pack_operations)
909
def _execute_pack_operations(self, pack_operations):
910
"""Execute a series of pack operations.
912
:param pack_operations: A list of [revision_count, packs_to_combine].
915
for revision_count, packs in pack_operations:
916
# we may have no-ops from the setup logic
919
Packer(self, packs, '.autopack').pack()
921
self._remove_pack_from_memory(pack)
922
# record the newly available packs and stop advertising the old
924
self._save_pack_names(clear_obsolete_packs=True)
925
# Move the old packs out of the way now they are no longer referenced.
926
for revision_count, packs in pack_operations:
927
self._obsolete_packs(packs)
929
def lock_names(self):
930
"""Acquire the mutex around the pack-names index.
932
This cannot be used in the middle of a read-only transaction on the
935
self.repo.control_files.lock_write()
938
"""Pack the pack collection totally."""
940
total_packs = len(self._names)
943
total_revisions = self.revision_index.combined_index.key_count()
944
# XXX: the following may want to be a class, to pack with a given
946
mutter('Packing repository %s, which has %d pack files, '
947
'containing %d revisions into 1 packs.', self, total_packs,
949
# determine which packs need changing
950
pack_distribution = [1]
951
pack_operations = [[0, []]]
952
for pack in self.all_packs():
953
revision_count = pack.get_revision_count()
954
pack_operations[-1][0] += revision_count
955
pack_operations[-1][1].append(pack)
956
self._execute_pack_operations(pack_operations)
958
def plan_autopack_combinations(self, existing_packs, pack_distribution):
959
"""Plan a pack operation.
961
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
963
:param pack_distribution: A list with the number of revisions desired
966
if len(existing_packs) <= len(pack_distribution):
968
existing_packs.sort(reverse=True)
969
pack_operations = [[0, []]]
970
# plan out what packs to keep, and what to reorganise
971
while len(existing_packs):
972
# take the largest pack, and if its less than the head of the
973
# distribution chart we will include its contents in the new pack for
974
# that position. If its larger, we remove its size from the
976
next_pack_rev_count, next_pack = existing_packs.pop(0)
977
if next_pack_rev_count >= pack_distribution[0]:
978
# this is already packed 'better' than this, so we can
979
# not waste time packing it.
980
while next_pack_rev_count > 0:
981
next_pack_rev_count -= pack_distribution[0]
982
if next_pack_rev_count >= 0:
984
del pack_distribution[0]
986
# didn't use that entire bucket up
987
pack_distribution[0] = -next_pack_rev_count
989
# add the revisions we're going to add to the next output pack
990
pack_operations[-1][0] += next_pack_rev_count
991
# allocate this pack to the next pack sub operation
992
pack_operations[-1][1].append(next_pack)
993
if pack_operations[-1][0] >= pack_distribution[0]:
994
# this pack is used up, shift left.
995
del pack_distribution[0]
996
pack_operations.append([0, []])
998
return pack_operations
1000
def ensure_loaded(self):
1001
# NB: if you see an assertion error here, its probably access against
1002
# an unlocked repo. Naughty.
1003
assert self.repo.is_locked()
1004
if self._names is None:
1006
self._packs_at_load = set()
1007
for index, key, value in self._iter_disk_pack_index():
1009
self._names[name] = self._parse_index_sizes(value)
1010
self._packs_at_load.add((key, value))
1011
# populate all the metadata.
1014
def _parse_index_sizes(self, value):
1015
"""Parse a string of index sizes."""
1016
return tuple([int(digits) for digits in value.split(' ')])
1018
def get_pack_by_name(self, name):
1019
"""Get a Pack object by name.
1021
:param name: The name of the pack - e.g. '123456'
1022
:return: A Pack object.
1025
return self._packs_by_name[name]
1027
rev_index = self._make_index(name, '.rix')
1028
inv_index = self._make_index(name, '.iix')
1029
txt_index = self._make_index(name, '.tix')
1030
sig_index = self._make_index(name, '.six')
1031
result = ExistingPack(self._pack_transport, name, rev_index,
1032
inv_index, txt_index, sig_index)
1033
self.add_pack_to_memory(result)
1036
def allocate(self, a_new_pack):
1037
"""Allocate name in the list of packs.
1039
:param a_new_pack: A NewPack instance to be added to the collection of
1040
packs for this repository.
1042
self.ensure_loaded()
1043
if a_new_pack.name in self._names:
1044
# a collision with the packs we know about (not the only possible
1045
# collision, see NewPack.finish() for some discussion). Remove our
1046
# prior reference to it.
1047
self._remove_pack_from_memory(a_new_pack)
1048
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1049
self.add_pack_to_memory(a_new_pack)
1051
def _iter_disk_pack_index(self):
1052
"""Iterate over the contents of the pack-names index.
1054
This is used when loading the list from disk, and before writing to
1055
detect updates from others during our write operation.
1056
:return: An iterator of the index contents.
1058
return GraphIndex(self.transport, 'pack-names', None
1059
).iter_all_entries()
1061
def _make_index(self, name, suffix):
1062
size_offset = self._suffix_offsets[suffix]
1063
index_name = name + suffix
1064
index_size = self._names[name][size_offset]
1066
self._index_transport, index_name, index_size)
1068
def _max_pack_count(self, total_revisions):
1069
"""Return the maximum number of packs to use for total revisions.
1071
:param total_revisions: The total number of revisions in the
1074
if not total_revisions:
1076
digits = str(total_revisions)
1078
for digit in digits:
1079
result += int(digit)
1083
"""Provide an order to the underlying names."""
1084
return sorted(self._names.keys())
1086
def _obsolete_packs(self, packs):
1087
"""Move a number of packs which have been obsoleted out of the way.
1089
Each pack and its associated indices are moved out of the way.
1091
Note: for correctness this function should only be called after a new
1092
pack names index has been written without these pack names, and with
1093
the names of packs that contain the data previously available via these
1096
:param packs: The packs to obsolete.
1097
:param return: None.
1100
pack.pack_transport.rename(pack.file_name(),
1101
'../obsolete_packs/' + pack.file_name())
1102
# TODO: Probably needs to know all possible indices for this pack
1103
# - or maybe list the directory and move all indices matching this
1104
# name whether we recognize it or not?
1105
for suffix in ('.iix', '.six', '.tix', '.rix'):
1106
self._index_transport.rename(pack.name + suffix,
1107
'../obsolete_packs/' + pack.name + suffix)
1109
def pack_distribution(self, total_revisions):
1110
"""Generate a list of the number of revisions to put in each pack.
1112
:param total_revisions: The total number of revisions in the
1115
if total_revisions == 0:
1117
digits = reversed(str(total_revisions))
1119
for exponent, count in enumerate(digits):
1120
size = 10 ** exponent
1121
for pos in range(int(count)):
1123
return list(reversed(result))
1125
def _pack_tuple(self, name):
1126
"""Return a tuple with the transport and file name for a pack name."""
1127
return self._pack_transport, name + '.pack'
1129
def _remove_pack_from_memory(self, pack):
1130
"""Remove pack from the packs accessed by this repository.
1132
Only affects memory state, until self._save_pack_names() is invoked.
1134
self._names.pop(pack.name)
1135
self._packs_by_name.pop(pack.name)
1136
self._remove_pack_indices(pack)
1138
def _remove_pack_indices(self, pack):
1139
"""Remove the indices for pack from the aggregated indices."""
1140
self.revision_index.remove_index(pack.revision_index, pack)
1141
self.inventory_index.remove_index(pack.inventory_index, pack)
1142
self.text_index.remove_index(pack.text_index, pack)
1143
self.signature_index.remove_index(pack.signature_index, pack)
1146
"""Clear all cached data."""
1147
# cached revision data
1148
self.repo._revision_knit = None
1149
self.revision_index.clear()
1150
# cached signature data
1151
self.repo._signature_knit = None
1152
self.signature_index.clear()
1153
# cached file text data
1154
self.text_index.clear()
1155
self.repo._text_knit = None
1156
# cached inventory data
1157
self.inventory_index.clear()
1158
# remove the open pack
1159
self._new_pack = None
1160
# information about packs.
1163
self._packs_by_name = {}
1164
self._packs_at_load = None
1166
def _make_index_map(self, index_suffix):
1167
"""Return information on existing indices.
1169
:param suffix: Index suffix added to pack name.
1171
:returns: (pack_map, indices) where indices is a list of GraphIndex
1172
objects, and pack_map is a mapping from those objects to the
1173
pack tuple they describe.
1175
# TODO: stop using this; it creates new indices unnecessarily.
1176
self.ensure_loaded()
1177
suffix_map = {'.rix': 'revision_index',
1178
'.six': 'signature_index',
1179
'.iix': 'inventory_index',
1180
'.tix': 'text_index',
1182
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1183
suffix_map[index_suffix])
1185
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1186
"""Convert a list of packs to an index pack map and index list.
1188
:param packs: The packs list to process.
1189
:param index_attribute: The attribute that the desired index is found
1191
:return: A tuple (map, list) where map contains the dict from
1192
index:pack_tuple, and lsit contains the indices in the same order
1198
index = getattr(pack, index_attribute)
1199
indices.append(index)
1200
pack_map[index] = (pack.pack_transport, pack.file_name())
1201
return pack_map, indices
1203
def _index_contents(self, pack_map, key_filter=None):
1204
"""Get an iterable of the index contents from a pack_map.
1206
:param pack_map: A map from indices to pack details.
1207
:param key_filter: An optional filter to limit the
1210
indices = [index for index in pack_map.iterkeys()]
1211
all_index = CombinedGraphIndex(indices)
1212
if key_filter is None:
1213
return all_index.iter_all_entries()
1215
return all_index.iter_entries(key_filter)
1217
def _unlock_names(self):
1218
"""Release the mutex around the pack-names index."""
1219
self.repo.control_files.unlock()
1221
def _save_pack_names(self, clear_obsolete_packs=False):
1222
"""Save the list of packs.
1224
This will take out the mutex around the pack names list for the
1225
duration of the method call. If concurrent updates have been made, a
1226
three-way merge between the current list and the current in memory list
1229
:param clear_obsolete_packs: If True, clear out the contents of the
1230
obsolete_packs directory.
1234
builder = GraphIndexBuilder()
1235
# load the disk nodes across
1237
for index, key, value in self._iter_disk_pack_index():
1238
disk_nodes.add((key, value))
1239
# do a two-way diff against our original content
1240
current_nodes = set()
1241
for name, sizes in self._names.iteritems():
1243
((name, ), ' '.join(str(size) for size in sizes)))
1244
deleted_nodes = self._packs_at_load - current_nodes
1245
new_nodes = current_nodes - self._packs_at_load
1246
disk_nodes.difference_update(deleted_nodes)
1247
disk_nodes.update(new_nodes)
1248
# TODO: handle same-name, index-size-changes here -
1249
# e.g. use the value from disk, not ours, *unless* we're the one
1251
for key, value in disk_nodes:
1252
builder.add_node(key, value)
1253
self.transport.put_file('pack-names', builder.finish(),
1254
mode=self.repo.control_files._file_mode)
1255
# move the baseline forward
1256
self._packs_at_load = disk_nodes
1257
# now clear out the obsolete packs directory
1258
if clear_obsolete_packs:
1259
self.transport.clone('obsolete_packs').delete_multi(
1260
self.transport.list_dir('obsolete_packs'))
1262
self._unlock_names()
1263
# synchronise the memory packs list with what we just wrote:
1264
new_names = dict(disk_nodes)
1265
# drop no longer present nodes
1266
for pack in self.all_packs():
1267
if (pack.name,) not in new_names:
1268
self._remove_pack_from_memory(pack)
1269
# add new nodes/refresh existing ones
1270
for key, value in disk_nodes:
1272
sizes = self._parse_index_sizes(value)
1273
if name in self._names:
1275
if sizes != self._names[name]:
1276
# the pack for name has had its indices replaced - rare but
1277
# important to handle. XXX: probably can never happen today
1278
# because the three-way merge code above does not handle it
1279
# - you may end up adding the same key twice to the new
1280
# disk index because the set values are the same, unless
1281
# the only index shows up as deleted by the set difference
1282
# - which it may. Until there is a specific test for this,
1283
# assume its broken. RBC 20071017.
1284
self._remove_pack_from_memory(self.get_pack_by_name(name))
1285
self._names[name] = sizes
1286
self.get_pack_by_name(name)
1289
self._names[name] = sizes
1290
self.get_pack_by_name(name)
1292
def _start_write_group(self):
1293
# Do not permit preparation for writing if we're not in a 'write lock'.
1294
if not self.repo.is_write_locked():
1295
raise errors.NotWriteLocked(self)
1296
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1297
self._pack_transport, upload_suffix='.pack',
1298
file_mode=self.repo.control_files._file_mode)
1299
# allow writing: queue writes to a new index
1300
self.revision_index.add_writable_index(self._new_pack.revision_index,
1302
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
1304
self.text_index.add_writable_index(self._new_pack.text_index,
1306
self.signature_index.add_writable_index(self._new_pack.signature_index,
1309
# reused revision and signature knits may need updating
1311
# "Hysterical raisins. client code in bzrlib grabs those knits outside
1312
# of write groups and then mutates it inside the write group."
1313
if self.repo._revision_knit is not None:
1314
self.repo._revision_knit._index._add_callback = \
1315
self.revision_index.add_callback
1316
if self.repo._signature_knit is not None:
1317
self.repo._signature_knit._index._add_callback = \
1318
self.signature_index.add_callback
1319
# create a reused knit object for text addition in commit.
1320
self.repo._text_knit = self.repo.weave_store.get_weave_or_empty(
1323
def _abort_write_group(self):
1324
# FIXME: just drop the transient index.
1325
# forget what names there are
1326
self._new_pack.abort()
1327
self._remove_pack_indices(self._new_pack)
1328
self._new_pack = None
1329
self.repo._text_knit = None
1331
def _commit_write_group(self):
1332
self._remove_pack_indices(self._new_pack)
1333
if self._new_pack.data_inserted():
1334
# get all the data to disk and read to use
1335
self._new_pack.finish()
1336
self.allocate(self._new_pack)
1337
self._new_pack = None
1338
if not self.autopack():
1339
# when autopack takes no steps, the names list is still
1341
self._save_pack_names()
1343
self._new_pack.abort()
1344
self._new_pack = None
1345
self.repo._text_knit = None
1348
class KnitPackRevisionStore(KnitRevisionStore):
1349
"""An object to adapt access from RevisionStore's to use KnitPacks.
1351
This class works by replacing the original RevisionStore.
1352
We need to do this because the KnitPackRevisionStore is less
1353
isolated in its layering - it uses services from the repo.
1356
def __init__(self, repo, transport, revisionstore):
1357
"""Create a KnitPackRevisionStore on repo with revisionstore.
1359
This will store its state in the Repository, use the
1360
indices to provide a KnitGraphIndex,
1361
and at the end of transactions write new indices.
1363
KnitRevisionStore.__init__(self, revisionstore.versioned_file_store)
1365
self._serializer = revisionstore._serializer
1366
self.transport = transport
1368
def get_revision_file(self, transaction):
1369
"""Get the revision versioned file object."""
1370
if getattr(self.repo, '_revision_knit', None) is not None:
1371
return self.repo._revision_knit
1372
self.repo._pack_collection.ensure_loaded()
1373
add_callback = self.repo._pack_collection.revision_index.add_callback
1374
# setup knit specific objects
1375
knit_index = KnitGraphIndex(
1376
self.repo._pack_collection.revision_index.combined_index,
1377
add_callback=add_callback)
1378
self.repo._revision_knit = knit.KnitVersionedFile(
1379
'revisions', self.transport.clone('..'),
1380
self.repo.control_files._file_mode,
1381
create=False, access_mode=self.repo._access_mode(),
1382
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1383
access_method=self.repo._pack_collection.revision_index.knit_access)
1384
return self.repo._revision_knit
1386
def get_signature_file(self, transaction):
1387
"""Get the signature versioned file object."""
1388
if getattr(self.repo, '_signature_knit', None) is not None:
1389
return self.repo._signature_knit
1390
self.repo._pack_collection.ensure_loaded()
1391
add_callback = self.repo._pack_collection.signature_index.add_callback
1392
# setup knit specific objects
1393
knit_index = KnitGraphIndex(
1394
self.repo._pack_collection.signature_index.combined_index,
1395
add_callback=add_callback, parents=False)
1396
self.repo._signature_knit = knit.KnitVersionedFile(
1397
'signatures', self.transport.clone('..'),
1398
self.repo.control_files._file_mode,
1399
create=False, access_mode=self.repo._access_mode(),
1400
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1401
access_method=self.repo._pack_collection.signature_index.knit_access)
1402
return self.repo._signature_knit
1405
class KnitPackTextStore(VersionedFileStore):
1406
"""Presents a TextStore abstraction on top of packs.
1408
This class works by replacing the original VersionedFileStore.
1409
We need to do this because the KnitPackRevisionStore is less
1410
isolated in its layering - it uses services from the repo and shares them
1411
with all the data written in a single write group.
1414
def __init__(self, repo, transport, weavestore):
1415
"""Create a KnitPackTextStore on repo with weavestore.
1417
This will store its state in the Repository, use the
1418
indices FileNames to provide a KnitGraphIndex,
1419
and at the end of transactions write new indices.
1421
# don't call base class constructor - it's not suitable.
1422
# no transient data stored in the transaction
1424
self._precious = False
1426
self.transport = transport
1427
self.weavestore = weavestore
1428
# XXX for check() which isn't updated yet
1429
self._transport = weavestore._transport
1431
def get_weave_or_empty(self, file_id, transaction):
1432
"""Get a 'Knit' backed by the .tix indices.
1434
The transaction parameter is ignored.
1436
self.repo._pack_collection.ensure_loaded()
1437
add_callback = self.repo._pack_collection.text_index.add_callback
1438
# setup knit specific objects
1439
file_id_index = GraphIndexPrefixAdapter(
1440
self.repo._pack_collection.text_index.combined_index,
1441
(file_id, ), 1, add_nodes_callback=add_callback)
1442
knit_index = KnitGraphIndex(file_id_index,
1443
add_callback=file_id_index.add_nodes,
1444
deltas=True, parents=True)
1445
return knit.KnitVersionedFile('text:' + file_id,
1446
self.transport.clone('..'),
1449
access_method=self.repo._pack_collection.text_index.knit_access,
1450
factory=knit.KnitPlainFactory())
1452
get_weave = get_weave_or_empty
1455
"""Generate a list of the fileids inserted, for use by check."""
1456
self.repo._pack_collection.ensure_loaded()
1458
for index, key, value, refs in \
1459
self.repo._pack_collection.text_index.combined_index.iter_all_entries():
1464
class InventoryKnitThunk(object):
1465
"""An object to manage thunking get_inventory_weave to pack based knits."""
1467
def __init__(self, repo, transport):
1468
"""Create an InventoryKnitThunk for repo at transport.
1470
This will store its state in the Repository, use the
1471
indices FileNames to provide a KnitGraphIndex,
1472
and at the end of transactions write a new index..
1475
self.transport = transport
1477
def get_weave(self):
1478
"""Get a 'Knit' that contains inventory data."""
1479
self.repo._pack_collection.ensure_loaded()
1480
add_callback = self.repo._pack_collection.inventory_index.add_callback
1481
# setup knit specific objects
1482
knit_index = KnitGraphIndex(
1483
self.repo._pack_collection.inventory_index.combined_index,
1484
add_callback=add_callback, deltas=True, parents=True)
1485
return knit.KnitVersionedFile(
1486
'inventory', self.transport.clone('..'),
1487
self.repo.control_files._file_mode,
1488
create=False, access_mode=self.repo._access_mode(),
1489
index=knit_index, delta=True, factory=knit.KnitPlainFactory(),
1490
access_method=self.repo._pack_collection.inventory_index.knit_access)
1493
class KnitPackRepository(KnitRepository):
1494
"""Experimental graph-knit using repository."""
1496
def __init__(self, _format, a_bzrdir, control_files, _revision_store,
1497
control_store, text_store, _commit_builder_class, _serializer):
1498
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
1499
_revision_store, control_store, text_store, _commit_builder_class,
1501
index_transport = control_files._transport.clone('indices')
1502
self._pack_collection = RepositoryPackCollection(self, control_files._transport,
1504
control_files._transport.clone('upload'),
1505
control_files._transport.clone('packs'))
1506
self._revision_store = KnitPackRevisionStore(self, index_transport, self._revision_store)
1507
self.weave_store = KnitPackTextStore(self, index_transport, self.weave_store)
1508
self._inv_thunk = InventoryKnitThunk(self, index_transport)
1509
# True when the repository object is 'write locked' (as opposed to the
1510
# physical lock only taken out around changes to the pack-names list.)
1511
# Another way to represent this would be a decorator around the control
1512
# files object that presents logical locks as physical ones - if this
1513
# gets ugly consider that alternative design. RBC 20071011
1514
self._write_lock_count = 0
1515
self._transaction = None
1517
self._reconcile_does_inventory_gc = True
1518
self._reconcile_fixes_text_parents = False
1519
self._reconcile_backsup_inventory = False
1521
def _abort_write_group(self):
1522
self._pack_collection._abort_write_group()
1524
def _access_mode(self):
1525
"""Return 'w' or 'r' for depending on whether a write lock is active.
1527
This method is a helper for the Knit-thunking support objects.
1529
if self.is_write_locked():
1533
def _find_inconsistent_revision_parents(self):
1534
"""Find revisions with incorrectly cached parents.
1536
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1537
parents-in-revision).
1539
assert self.is_locked()
1540
pb = ui.ui_factory.nested_progress_bar()
1543
revision_nodes = self._pack_collection.revision_index \
1544
.combined_index.iter_all_entries()
1545
index_positions = []
1546
# Get the cached index values for all revisions, and also the location
1547
# in each index of the revision text so we can perform linear IO.
1548
for index, key, value, refs in revision_nodes:
1549
pos, length = value[1:].split(' ')
1550
index_positions.append((index, int(pos), key[0],
1551
tuple(parent[0] for parent in refs[0])))
1552
pb.update("Reading revision index.", 0, 0)
1553
index_positions.sort()
1554
batch_count = len(index_positions) / 1000 + 1
1555
pb.update("Checking cached revision graph.", 0, batch_count)
1556
for offset in xrange(batch_count):
1557
pb.update("Checking cached revision graph.", offset)
1558
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1561
rev_ids = [item[2] for item in to_query]
1562
revs = self.get_revisions(rev_ids)
1563
for revision, item in zip(revs, to_query):
1564
index_parents = item[3]
1565
rev_parents = tuple(revision.parent_ids)
1566
if index_parents != rev_parents:
1567
result.append((revision.revision_id, index_parents, rev_parents))
1572
def get_parents(self, revision_ids):
1573
"""See StackedParentsProvider.get_parents.
1575
This implementation accesses the combined revision index to provide
1578
self._pack_collection.ensure_loaded()
1579
index = self._pack_collection.revision_index.combined_index
1581
for revision_id in revision_ids:
1582
if revision_id != _mod_revision.NULL_REVISION:
1583
search_keys.add((revision_id,))
1584
found_parents = {_mod_revision.NULL_REVISION:[]}
1585
for index, key, value, refs in index.iter_entries(search_keys):
1588
parents = (_mod_revision.NULL_REVISION,)
1590
parents = tuple(parent[0] for parent in parents)
1591
found_parents[key[0]] = parents
1593
for revision_id in revision_ids:
1595
result.append(found_parents[revision_id])
1600
def _make_parents_provider(self):
1603
def _refresh_data(self):
1604
if self._write_lock_count == 1 or (
1605
self.control_files._lock_count == 1 and
1606
self.control_files._lock_mode == 'r'):
1607
# forget what names there are
1608
self._pack_collection.reset()
1609
# XXX: Better to do an in-memory merge when acquiring a new lock -
1610
# factor out code from _save_pack_names.
1611
self._pack_collection.ensure_loaded()
1613
def _start_write_group(self):
1614
self._pack_collection._start_write_group()
1616
def _commit_write_group(self):
1617
return self._pack_collection._commit_write_group()
1619
def get_inventory_weave(self):
1620
return self._inv_thunk.get_weave()
1622
def get_transaction(self):
1623
if self._write_lock_count:
1624
return self._transaction
1626
return self.control_files.get_transaction()
1628
def is_locked(self):
1629
return self._write_lock_count or self.control_files.is_locked()
1631
def is_write_locked(self):
1632
return self._write_lock_count
1634
def lock_write(self, token=None):
1635
if not self._write_lock_count and self.is_locked():
1636
raise errors.ReadOnlyError(self)
1637
self._write_lock_count += 1
1638
if self._write_lock_count == 1:
1639
from bzrlib import transactions
1640
self._transaction = transactions.WriteTransaction()
1641
self._refresh_data()
1643
def lock_read(self):
1644
if self._write_lock_count:
1645
self._write_lock_count += 1
1647
self.control_files.lock_read()
1648
self._refresh_data()
1650
def leave_lock_in_place(self):
1651
# not supported - raise an error
1652
raise NotImplementedError(self.leave_lock_in_place)
1654
def dont_leave_lock_in_place(self):
1655
# not supported - raise an error
1656
raise NotImplementedError(self.dont_leave_lock_in_place)
1660
"""Compress the data within the repository.
1662
This will pack all the data to a single pack. In future it may
1663
recompress deltas or do other such expensive operations.
1665
self._pack_collection.pack()
1668
def reconcile(self, other=None, thorough=False):
1669
"""Reconcile this repository."""
1670
from bzrlib.reconcile import PackReconciler
1671
reconciler = PackReconciler(self, thorough=thorough)
1672
reconciler.reconcile()
1676
if self._write_lock_count == 1 and self._write_group is not None:
1677
self.abort_write_group()
1678
self._transaction = None
1679
self._write_lock_count = 0
1680
raise errors.BzrError(
1681
'Must end write group before releasing write lock on %s'
1683
if self._write_lock_count:
1684
self._write_lock_count -= 1
1685
if not self._write_lock_count:
1686
transaction = self._transaction
1687
self._transaction = None
1688
transaction.finish()
1690
self.control_files.unlock()
1693
class RepositoryFormatPack(MetaDirRepositoryFormat):
1694
"""Format logic for pack structured repositories.
1696
This repository format has:
1697
- a list of packs in pack-names
1698
- packs in packs/NAME.pack
1699
- indices in indices/NAME.{iix,six,tix,rix}
1700
- knit deltas in the packs, knit indices mapped to the indices.
1701
- thunk objects to support the knits programming API.
1702
- a format marker of its own
1703
- an optional 'shared-storage' flag
1704
- an optional 'no-working-trees' flag
1708
# Set this attribute in derived classes to control the repository class
1709
# created by open and initialize.
1710
repository_class = None
1711
# Set this attribute in derived classes to control the
1712
# _commit_builder_class that the repository objects will have passed to
1713
# their constructor.
1714
_commit_builder_class = None
1715
# Set this attribute in derived clases to control the _serializer that the
1716
# repository objects will have passed to their constructor.
1719
def _get_control_store(self, repo_transport, control_files):
1720
"""Return the control store for this repository."""
1721
return VersionedFileStore(
1724
file_mode=control_files._file_mode,
1725
versionedfile_class=knit.KnitVersionedFile,
1726
versionedfile_kwargs={'factory': knit.KnitPlainFactory()},
1729
def _get_revision_store(self, repo_transport, control_files):
1730
"""See RepositoryFormat._get_revision_store()."""
1731
versioned_file_store = VersionedFileStore(
1733
file_mode=control_files._file_mode,
1736
versionedfile_class=knit.KnitVersionedFile,
1737
versionedfile_kwargs={'delta': False,
1738
'factory': knit.KnitPlainFactory(),
1742
return KnitRevisionStore(versioned_file_store)
1744
def _get_text_store(self, transport, control_files):
1745
"""See RepositoryFormat._get_text_store()."""
1746
return self._get_versioned_file_store('knits',
1749
versionedfile_class=knit.KnitVersionedFile,
1750
versionedfile_kwargs={
1751
'create_parent_dir': True,
1752
'delay_create': True,
1753
'dir_mode': control_files._dir_mode,
1757
def initialize(self, a_bzrdir, shared=False):
1758
"""Create a pack based repository.
1760
:param a_bzrdir: bzrdir to contain the new repository; must already
1762
:param shared: If true the repository will be initialized as a shared
1765
mutter('creating repository in %s.', a_bzrdir.transport.base)
1766
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
1767
builder = GraphIndexBuilder()
1768
files = [('pack-names', builder.finish())]
1769
utf8_files = [('format', self.get_format_string())]
1771
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
1772
return self.open(a_bzrdir=a_bzrdir, _found=True)
1774
def open(self, a_bzrdir, _found=False, _override_transport=None):
1775
"""See RepositoryFormat.open().
1777
:param _override_transport: INTERNAL USE ONLY. Allows opening the
1778
repository at a slightly different url
1779
than normal. I.e. during 'upgrade'.
1782
format = RepositoryFormat.find_format(a_bzrdir)
1783
assert format.__class__ == self.__class__
1784
if _override_transport is not None:
1785
repo_transport = _override_transport
1787
repo_transport = a_bzrdir.get_repository_transport(None)
1788
control_files = lockable_files.LockableFiles(repo_transport,
1789
'lock', lockdir.LockDir)
1790
text_store = self._get_text_store(repo_transport, control_files)
1791
control_store = self._get_control_store(repo_transport, control_files)
1792
_revision_store = self._get_revision_store(repo_transport, control_files)
1793
return self.repository_class(_format=self,
1795
control_files=control_files,
1796
_revision_store=_revision_store,
1797
control_store=control_store,
1798
text_store=text_store,
1799
_commit_builder_class=self._commit_builder_class,
1800
_serializer=self._serializer)
1803
class RepositoryFormatKnitPack1(RepositoryFormatPack):
1804
"""A no-subtrees parameterised Pack repository.
1806
This format was introduced in 0.92.
1809
repository_class = KnitPackRepository
1810
_commit_builder_class = PackCommitBuilder
1811
_serializer = xml5.serializer_v5
1813
def _get_matching_bzrdir(self):
1814
return bzrdir.format_registry.make_bzrdir('knitpack-experimental')
1816
def _ignore_setting_bzrdir(self, format):
1819
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1821
def get_format_string(self):
1822
"""See RepositoryFormat.get_format_string()."""
1823
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
1825
def get_format_description(self):
1826
"""See RepositoryFormat.get_format_description()."""
1827
return "Packs containing knits without subtree support"
1829
def check_conversion_target(self, target_format):
1833
class RepositoryFormatKnitPack3(RepositoryFormatPack):
1834
"""A subtrees parameterised Pack repository.
1836
This repository format uses the xml7 serializer to get:
1837
- support for recording full info about the tree root
1838
- support for recording tree-references
1840
This format was introduced in 0.92.
1843
repository_class = KnitPackRepository
1844
_commit_builder_class = PackRootCommitBuilder
1845
rich_root_data = True
1846
supports_tree_reference = True
1847
_serializer = xml7.serializer_v7
1849
def _get_matching_bzrdir(self):
1850
return bzrdir.format_registry.make_bzrdir(
1851
'knitpack-subtree-experimental')
1853
def _ignore_setting_bzrdir(self, format):
1856
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1858
def check_conversion_target(self, target_format):
1859
if not target_format.rich_root_data:
1860
raise errors.BadConversionTarget(
1861
'Does not support rich root data.', target_format)
1862
if not getattr(target_format, 'supports_tree_reference', False):
1863
raise errors.BadConversionTarget(
1864
'Does not support nested trees', target_format)
1866
def get_format_string(self):
1867
"""See RepositoryFormat.get_format_string()."""
1868
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
1870
def get_format_description(self):
1871
"""See RepositoryFormat.get_format_description()."""
1872
return "Packs containing knits with subtree support\n"
added
removed
bzrlib/repofmt/pack_repo.py
