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.pack import ContainerWriter
38
from bzrlib.store import revision
53
from bzrlib.decorators import needs_read_lock, needs_write_lock
54
from bzrlib.repofmt.knitrepo import KnitRepository
55
from bzrlib.repository import (
58
MetaDirRepositoryFormat,
61
import bzrlib.revision as _mod_revision
62
from bzrlib.store.revision.knit import KnitRevisionStore
63
from bzrlib.store.versioned import VersionedFileStore
64
from bzrlib.trace import mutter, note, warning
67
class PackCommitBuilder(CommitBuilder):
68
"""A subclass of CommitBuilder to add texts with pack semantics.
70
Specifically this uses one knit object rather than one knit object per
71
added text, reducing memory and object pressure.
74
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
75
return self.repository._pack_collection._add_text_to_weave(file_id,
76
self._new_revision_id, new_lines, parents, nostore_sha,
80
class PackRootCommitBuilder(RootCommitBuilder):
81
"""A subclass of RootCommitBuilder to add texts with pack semantics.
83
Specifically this uses one knit object rather than one knit object per
84
added text, reducing memory and object pressure.
87
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
88
return self.repository._pack_collection._add_text_to_weave(file_id,
89
self._new_revision_id, new_lines, parents, nostore_sha,
94
"""An in memory proxy for a pack and its indices.
96
This is a base class that is not directly used, instead the classes
97
ExistingPack and NewPack are used.
100
def __init__(self, revision_index, inventory_index, text_index,
102
"""Create a pack instance.
104
:param revision_index: A GraphIndex for determining what revisions are
105
present in the Pack and accessing the locations of their texts.
106
:param inventory_index: A GraphIndex for determining what inventories are
107
present in the Pack and accessing the locations of their
109
:param text_index: A GraphIndex for determining what file texts
110
are present in the pack and accessing the locations of their
111
texts/deltas (via (fileid, revisionid) tuples).
112
:param revision_index: A GraphIndex for determining what signatures are
113
present in the Pack and accessing the locations of their texts.
115
self.revision_index = revision_index
116
self.inventory_index = inventory_index
117
self.text_index = text_index
118
self.signature_index = signature_index
120
def access_tuple(self):
121
"""Return a tuple (transport, name) for the pack content."""
122
return self.pack_transport, self.file_name()
125
"""Get the file name for the pack on disk."""
126
return self.name + '.pack'
128
def get_revision_count(self):
129
return self.revision_index.key_count()
131
def inventory_index_name(self, name):
132
"""The inv index is the name + .iix."""
133
return self.index_name('inventory', name)
135
def revision_index_name(self, name):
136
"""The revision index is the name + .rix."""
137
return self.index_name('revision', name)
139
def signature_index_name(self, name):
140
"""The signature index is the name + .six."""
141
return self.index_name('signature', name)
143
def text_index_name(self, name):
144
"""The text index is the name + .tix."""
145
return self.index_name('text', name)
148
class ExistingPack(Pack):
149
"""An in memory proxy for an existing .pack and its disk indices."""
151
def __init__(self, pack_transport, name, revision_index, inventory_index,
152
text_index, signature_index):
153
"""Create an ExistingPack object.
155
:param pack_transport: The transport where the pack file resides.
156
:param name: The name of the pack on disk in the pack_transport.
158
Pack.__init__(self, revision_index, inventory_index, text_index,
161
self.pack_transport = pack_transport
162
assert None not in (revision_index, inventory_index, text_index,
163
signature_index, name, pack_transport)
165
def __eq__(self, other):
166
return self.__dict__ == other.__dict__
168
def __ne__(self, other):
169
return not self.__eq__(other)
172
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
173
id(self), self.transport, self.name)
177
"""An in memory proxy for a pack which is being created."""
179
# A map of index 'type' to the file extension and position in the
181
index_definitions = {
182
'revision': ('.rix', 0),
183
'inventory': ('.iix', 1),
185
'signature': ('.six', 3),
188
def __init__(self, upload_transport, index_transport, pack_transport,
190
"""Create a NewPack instance.
192
:param upload_transport: A writable transport for the pack to be
193
incrementally uploaded to.
194
:param index_transport: A writable transport for the pack's indices to
195
be written to when the pack is finished.
196
:param pack_transport: A writable transport for the pack to be renamed
197
to when the upload is complete. This *must* be the same as
198
upload_transport.clone('../packs').
199
:param upload_suffix: An optional suffix to be given to any temporary
200
files created during the pack creation. e.g '.autopack'
202
# The relative locations of the packs are constrained, but all are
203
# passed in because the caller has them, so as to avoid object churn.
205
# Revisions: parents list, no text compression.
206
InMemoryGraphIndex(reference_lists=1),
207
# Inventory: We want to map compression only, but currently the
208
# knit code hasn't been updated enough to understand that, so we
209
# have a regular 2-list index giving parents and compression
211
InMemoryGraphIndex(reference_lists=2),
212
# Texts: compression and per file graph, for all fileids - so two
213
# reference lists and two elements in the key tuple.
214
InMemoryGraphIndex(reference_lists=2, key_elements=2),
215
# Signatures: Just blobs to store, no compression, no parents
217
InMemoryGraphIndex(reference_lists=0),
219
# where should the new pack be opened
220
self.upload_transport = upload_transport
221
# where are indices written out to
222
self.index_transport = index_transport
223
# where is the pack renamed to when it is finished?
224
self.pack_transport = pack_transport
225
# tracks the content written to the .pack file.
226
self._hash = osutils.md5()
227
# a four-tuple with the length in bytes of the indices, once the pack
228
# is finalised. (rev, inv, text, sigs)
229
self.index_sizes = None
230
# How much data to cache when writing packs. Note that this is not
231
# synchronised with reads, because it's not in the transport layer, so
232
# is not safe unless the client knows it won't be reading from the pack
234
self._cache_limit = 0
235
# the temporary pack file name.
236
self.random_name = osutils.rand_chars(20) + upload_suffix
237
# when was this pack started ?
238
self.start_time = time.time()
239
# open an output stream for the data added to the pack.
240
self.write_stream = self.upload_transport.open_write_stream(
242
if 'pack' in debug.debug_flags:
243
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
244
time.ctime(), self.upload_transport.base, self.random_name,
245
time.time() - self.start_time)
246
# A list of byte sequences to be written to the new pack, and the
247
# aggregate size of them. Stored as a list rather than separate
248
# variables so that the _write_data closure below can update them.
249
self._buffer = [[], 0]
250
# create a callable for adding data
252
# robertc says- this is a closure rather than a method on the object
253
# so that the variables are locals, and faster than accessing object
255
def _write_data(bytes, flush=False, _buffer=self._buffer,
256
_write=self.write_stream.write, _update=self._hash.update):
257
_buffer[0].append(bytes)
258
_buffer[1] += len(bytes)
260
if _buffer[1] > self._cache_limit or flush:
261
bytes = ''.join(_buffer[0])
265
# expose this on self, for the occasion when clients want to add data.
266
self._write_data = _write_data
267
# a pack writer object to serialise pack records.
268
self._writer = pack.ContainerWriter(self._write_data)
270
# what state is the pack in? (open, finished, aborted)
274
"""Cancel creating this pack."""
275
self._state = 'aborted'
276
self.write_stream.close()
277
# Remove the temporary pack file.
278
self.upload_transport.delete(self.random_name)
279
# The indices have no state on disk.
281
def access_tuple(self):
282
"""Return a tuple (transport, name) for the pack content."""
283
assert self._state in ('open', 'finished')
284
if self._state == 'finished':
285
return Pack.access_tuple(self)
287
return self.upload_transport, self.random_name
289
def data_inserted(self):
290
"""True if data has been added to this pack."""
291
return bool(self.get_revision_count() or
292
self.inventory_index.key_count() or
293
self.text_index.key_count() or
294
self.signature_index.key_count())
297
"""Finish the new pack.
300
- finalises the content
301
- assigns a name (the md5 of the content, currently)
302
- writes out the associated indices
303
- renames the pack into place.
304
- stores the index size tuple for the pack in the index_sizes
309
self._write_data('', flush=True)
310
self.name = self._hash.hexdigest()
312
# XXX: It'd be better to write them all to temporary names, then
313
# rename them all into place, so that the window when only some are
314
# visible is smaller. On the other hand none will be seen until
315
# they're in the names list.
316
self.index_sizes = [None, None, None, None]
317
self._write_index('revision', self.revision_index, 'revision')
318
self._write_index('inventory', self.inventory_index, 'inventory')
319
self._write_index('text', self.text_index, 'file texts')
320
self._write_index('signature', self.signature_index,
321
'revision signatures')
322
self.write_stream.close()
323
# Note that this will clobber an existing pack with the same name,
324
# without checking for hash collisions. While this is undesirable this
325
# is something that can be rectified in a subsequent release. One way
326
# to rectify it may be to leave the pack at the original name, writing
327
# its pack-names entry as something like 'HASH: index-sizes
328
# temporary-name'. Allocate that and check for collisions, if it is
329
# collision free then rename it into place. If clients know this scheme
330
# they can handle missing-file errors by:
331
# - try for HASH.pack
332
# - try for temporary-name
333
# - refresh the pack-list to see if the pack is now absent
334
self.upload_transport.rename(self.random_name,
335
'../packs/' + self.name + '.pack')
336
self._state = 'finished'
337
if 'pack' in debug.debug_flags:
338
# XXX: size might be interesting?
339
mutter('%s: create_pack: pack renamed into place: %s%s->%s%s t+%6.3fs',
340
time.ctime(), self.upload_transport.base, self.random_name,
341
self.pack_transport, self.name,
342
time.time() - self.start_time)
344
def index_name(self, index_type, name):
345
"""Get the disk name of an index type for pack name 'name'."""
346
return name + NewPack.index_definitions[index_type][0]
348
def index_offset(self, index_type):
349
"""Get the position in a index_size array for a given index type."""
350
return NewPack.index_definitions[index_type][1]
352
def _replace_index_with_readonly(self, index_type):
353
setattr(self, index_type + '_index',
354
GraphIndex(self.index_transport,
355
self.index_name(index_type, self.name),
356
self.index_sizes[self.index_offset(index_type)]))
358
def set_write_cache_size(self, size):
359
self._cache_limit = size
361
def _write_index(self, index_type, index, label):
362
"""Write out an index.
364
:param index_type: The type of index to write - e.g. 'revision'.
365
:param index: The index object to serialise.
366
:param label: What label to give the index e.g. 'revision'.
368
index_name = self.index_name(index_type, self.name)
369
self.index_sizes[self.index_offset(index_type)] = \
370
self.index_transport.put_file(index_name, index.finish())
371
if 'pack' in debug.debug_flags:
372
# XXX: size might be interesting?
373
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
374
time.ctime(), label, self.upload_transport.base,
375
self.random_name, time.time() - self.start_time)
376
# Replace the writable index on this object with a readonly,
377
# presently unloaded index. We should alter
378
# the index layer to make its finish() error if add_node is
379
# subsequently used. RBC
380
self._replace_index_with_readonly(index_type)
383
class AggregateIndex(object):
384
"""An aggregated index for the RepositoryPackCollection.
386
AggregateIndex is reponsible for managing the PackAccess object,
387
Index-To-Pack mapping, and all indices list for a specific type of index
388
such as 'revision index'.
390
A CombinedIndex provides an index on a single key space built up
391
from several on-disk indices. The AggregateIndex builds on this
392
to provide a knit access layer, and allows having up to one writable
393
index within the collection.
395
# XXX: Probably 'can be written to' could/should be separated from 'acts
396
# like a knit index' -- mbp 20071024
399
"""Create an AggregateIndex."""
400
self.index_to_pack = {}
401
self.combined_index = CombinedGraphIndex([])
402
self.knit_access = _PackAccess(self.index_to_pack)
404
def replace_indices(self, index_to_pack, indices):
405
"""Replace the current mappings with fresh ones.
407
This should probably not be used eventually, rather incremental add and
408
removal of indices. It has been added during refactoring of existing
411
:param index_to_pack: A mapping from index objects to
412
(transport, name) tuples for the pack file data.
413
:param indices: A list of indices.
415
# refresh the revision pack map dict without replacing the instance.
416
self.index_to_pack.clear()
417
self.index_to_pack.update(index_to_pack)
418
# XXX: API break - clearly a 'replace' method would be good?
419
self.combined_index._indices[:] = indices
420
# the current add nodes callback for the current writable index if
422
self.add_callback = None
424
def add_index(self, index, pack):
425
"""Add index to the aggregate, which is an index for Pack pack.
427
Future searches on the aggregate index will seach this new index
428
before all previously inserted indices.
430
:param index: An Index for the pack.
431
:param pack: A Pack instance.
433
# expose it to the index map
434
self.index_to_pack[index] = pack.access_tuple()
435
# put it at the front of the linear index list
436
self.combined_index.insert_index(0, index)
438
def add_writable_index(self, index, pack):
439
"""Add an index which is able to have data added to it.
441
There can be at most one writable index at any time. Any
442
modifications made to the knit are put into this index.
444
:param index: An index from the pack parameter.
445
:param pack: A Pack instance.
447
assert self.add_callback is None, \
448
"%s already has a writable index through %s" % \
449
(self, self.add_callback)
450
# allow writing: queue writes to a new index
451
self.add_index(index, pack)
452
# Updates the index to packs mapping as a side effect,
453
self.knit_access.set_writer(pack._writer, index, pack.access_tuple())
454
self.add_callback = index.add_nodes
457
"""Reset all the aggregate data to nothing."""
458
self.knit_access.set_writer(None, None, (None, None))
459
self.index_to_pack.clear()
460
del self.combined_index._indices[:]
461
self.add_callback = None
463
def remove_index(self, index, pack):
464
"""Remove index from the indices used to answer queries.
466
:param index: An index from the pack parameter.
467
:param pack: A Pack instance.
469
del self.index_to_pack[index]
470
self.combined_index._indices.remove(index)
471
if (self.add_callback is not None and
472
getattr(index, 'add_nodes', None) == self.add_callback):
473
self.add_callback = None
474
self.knit_access.set_writer(None, None, (None, None))
477
class Packer(object):
478
"""Create a pack from packs."""
480
def __init__(self, pack_collection, packs, suffix, revision_ids=None):
483
:param pack_collection: A RepositoryPackCollection object where the
484
new pack is being written to.
485
:param packs: The packs to combine.
486
:param suffix: The suffix to use on the temporary files for the pack.
487
:param revision_ids: Revision ids to limit the pack to.
491
self.revision_ids = revision_ids
492
self._pack_collection = pack_collection
494
def pack(self, pb=None):
495
"""Create a new pack by reading data from other packs.
497
This does little more than a bulk copy of data. One key difference
498
is that data with the same item key across multiple packs is elided
499
from the output. The new pack is written into the current pack store
500
along with its indices, and the name added to the pack names. The
501
source packs are not altered and are not required to be in the current
504
:param pb: An optional progress bar to use. A nested bar is created if
506
:return: A Pack object, or None if nothing was copied.
508
# open a pack - using the same name as the last temporary file
509
# - which has already been flushed, so its safe.
510
# XXX: - duplicate code warning with start_write_group; fix before
511
# considering 'done'.
512
if self._pack_collection._new_pack is not None:
513
raise errors.BzrError('call to create_pack_from_packs while '
514
'another pack is being written.')
515
if self.revision_ids is not None:
516
if len(self.revision_ids) == 0:
517
# silly fetch request.
520
self.revision_ids = frozenset(self.revision_ids)
522
self.pb = ui.ui_factory.nested_progress_bar()
526
return self._create_pack_from_packs()
532
"""Open a pack for the pack we are creating."""
533
return NewPack(self._pack_collection._upload_transport,
534
self._pack_collection._index_transport,
535
self._pack_collection._pack_transport, upload_suffix=self.suffix)
537
def _create_pack_from_packs(self):
538
self.pb.update("Opening pack", 0, 5)
539
new_pack = self.open_pack()
540
# buffer data - we won't be reading-back during the pack creation and
541
# this makes a significant difference on sftp pushes.
542
new_pack.set_write_cache_size(1024*1024)
543
if 'pack' in debug.debug_flags:
544
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
545
for a_pack in self.packs]
546
if self.revision_ids is not None:
547
rev_count = len(self.revision_ids)
550
mutter('%s: create_pack: creating pack from source packs: '
551
'%s%s %s revisions wanted %s t=0',
552
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
553
plain_pack_list, rev_count)
555
if self.revision_ids:
556
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
560
# select revision keys
561
revision_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
562
self.packs, 'revision_index')[0]
563
revision_nodes = self._pack_collection._index_contents(revision_index_map, revision_keys)
564
# copy revision keys and adjust values
565
self.pb.update("Copying revision texts", 1)
566
list(self._copy_nodes_graph(revision_nodes, revision_index_map,
567
new_pack._writer, new_pack.revision_index))
568
if 'pack' in debug.debug_flags:
569
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
570
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
571
new_pack.revision_index.key_count(),
572
time.time() - new_pack.start_time)
573
# select inventory keys
574
inv_keys = revision_keys # currently the same keyspace, and note that
575
# querying for keys here could introduce a bug where an inventory item
576
# is missed, so do not change it to query separately without cross
577
# checking like the text key check below.
578
inventory_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
579
self.packs, 'inventory_index')[0]
580
inv_nodes = self._pack_collection._index_contents(inventory_index_map, inv_keys)
581
# copy inventory keys and adjust values
582
# XXX: Should be a helper function to allow different inv representation
584
self.pb.update("Copying inventory texts", 2)
585
inv_lines = self._copy_nodes_graph(inv_nodes, inventory_index_map,
586
new_pack._writer, new_pack.inventory_index, output_lines=True)
587
if self.revision_ids:
588
fileid_revisions = self._pack_collection.repo._find_file_ids_from_xml_inventory_lines(
589
inv_lines, self.revision_ids)
591
for fileid, file_revids in fileid_revisions.iteritems():
593
[(fileid, file_revid) for file_revid in file_revids])
595
# eat the iterator to cause it to execute.
598
if 'pack' in debug.debug_flags:
599
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
600
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
601
new_pack.inventory_index.key_count(),
602
time.time() - new_pack.start_time)
604
text_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
605
self.packs, 'text_index')[0]
606
text_nodes = self._pack_collection._index_contents(text_index_map, text_filter)
607
if text_filter is not None:
608
# We could return the keys copied as part of the return value from
609
# _copy_nodes_graph but this doesn't work all that well with the
610
# need to get line output too, so we check separately, and as we're
611
# going to buffer everything anyway, we check beforehand, which
612
# saves reading knit data over the wire when we know there are
614
text_nodes = set(text_nodes)
615
present_text_keys = set(_node[1] for _node in text_nodes)
616
missing_text_keys = set(text_filter) - present_text_keys
617
if missing_text_keys:
618
# TODO: raise a specific error that can handle many missing
620
a_missing_key = missing_text_keys.pop()
621
raise errors.RevisionNotPresent(a_missing_key[1],
623
# copy text keys and adjust values
624
self.pb.update("Copying content texts", 3)
625
list(self._copy_nodes_graph(text_nodes, text_index_map,
626
new_pack._writer, new_pack.text_index))
627
if 'pack' in debug.debug_flags:
628
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
629
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
630
new_pack.text_index.key_count(),
631
time.time() - new_pack.start_time)
632
# select signature keys
633
signature_filter = revision_keys # same keyspace
634
signature_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
635
self.packs, 'signature_index')[0]
636
signature_nodes = self._pack_collection._index_contents(signature_index_map,
638
# copy signature keys and adjust values
639
self.pb.update("Copying signature texts", 4)
640
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
641
new_pack.signature_index)
642
if 'pack' in debug.debug_flags:
643
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
644
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
645
new_pack.signature_index.key_count(),
646
time.time() - new_pack.start_time)
647
if not new_pack.data_inserted():
650
self.pb.update("Finishing pack", 5)
652
self._pack_collection.allocate(new_pack)
655
def _copy_nodes(self, nodes, index_map, writer, write_index):
656
"""Copy knit nodes between packs with no graph references."""
657
pb = ui.ui_factory.nested_progress_bar()
659
return self._do_copy_nodes(nodes, index_map, writer,
664
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb):
665
# for record verification
666
knit_data = _KnitData(None)
667
# plan a readv on each source pack:
669
nodes = sorted(nodes)
670
# how to map this into knit.py - or knit.py into this?
671
# we don't want the typical knit logic, we want grouping by pack
672
# at this point - perhaps a helper library for the following code
673
# duplication points?
675
for index, key, value in nodes:
676
if index not in request_groups:
677
request_groups[index] = []
678
request_groups[index].append((key, value))
680
pb.update("Copied record", record_index, len(nodes))
681
for index, items in request_groups.iteritems():
682
pack_readv_requests = []
683
for key, value in items:
684
# ---- KnitGraphIndex.get_position
685
bits = value[1:].split(' ')
686
offset, length = int(bits[0]), int(bits[1])
687
pack_readv_requests.append((offset, length, (key, value[0])))
688
# linear scan up the pack
689
pack_readv_requests.sort()
691
transport, path = index_map[index]
692
reader = pack.make_readv_reader(transport, path,
693
[offset[0:2] for offset in pack_readv_requests])
694
for (names, read_func), (_1, _2, (key, eol_flag)) in \
695
izip(reader.iter_records(), pack_readv_requests):
696
raw_data = read_func(None)
697
# check the header only
698
df, _ = knit_data._parse_record_header(key[-1], raw_data)
700
pos, size = writer.add_bytes_record(raw_data, names)
701
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
702
pb.update("Copied record", record_index)
705
def _copy_nodes_graph(self, nodes, index_map, writer, write_index,
707
"""Copy knit nodes between packs.
709
:param output_lines: Return lines present in the copied data as
710
an iterator of line,version_id.
712
pb = ui.ui_factory.nested_progress_bar()
714
return self._do_copy_nodes_graph(nodes, index_map, writer,
715
write_index, output_lines, pb)
719
def _do_copy_nodes_graph(self, nodes, index_map, writer, write_index,
721
# for record verification
722
knit_data = _KnitData(None)
723
# for line extraction when requested (inventories only)
725
factory = knit.KnitPlainFactory()
726
# plan a readv on each source pack:
728
nodes = sorted(nodes)
729
# how to map this into knit.py - or knit.py into this?
730
# we don't want the typical knit logic, we want grouping by pack
731
# at this point - perhaps a helper library for the following code
732
# duplication points?
735
pb.update("Copied record", record_index, len(nodes))
736
for index, key, value, references in nodes:
737
if index not in request_groups:
738
request_groups[index] = []
739
request_groups[index].append((key, value, references))
740
for index, items in request_groups.iteritems():
741
pack_readv_requests = []
742
for key, value, references in items:
743
# ---- KnitGraphIndex.get_position
744
bits = value[1:].split(' ')
745
offset, length = int(bits[0]), int(bits[1])
746
pack_readv_requests.append((offset, length, (key, value[0], references)))
747
# linear scan up the pack
748
pack_readv_requests.sort()
750
transport, path = index_map[index]
751
reader = pack.make_readv_reader(transport, path,
752
[offset[0:2] for offset in pack_readv_requests])
753
for (names, read_func), (_1, _2, (key, eol_flag, references)) in \
754
izip(reader.iter_records(), pack_readv_requests):
755
raw_data = read_func(None)
758
# read the entire thing
759
content, _ = knit_data._parse_record(version_id, raw_data)
760
if len(references[-1]) == 0:
761
line_iterator = factory.get_fulltext_content(content)
763
line_iterator = factory.get_linedelta_content(content)
764
for line in line_iterator:
765
yield line, version_id
767
# check the header only
768
df, _ = knit_data._parse_record_header(version_id, raw_data)
770
pos, size = writer.add_bytes_record(raw_data, names)
771
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
772
pb.update("Copied record", record_index)
776
class ReconcilePacker(Packer):
777
"""A packer which regenerates indices etc as it copies.
779
This is used by ``bzr reconcile`` to cause parent text pointers to be
784
class RepositoryPackCollection(object):
785
"""Management of packs within a repository."""
787
def __init__(self, repo, transport, index_transport, upload_transport,
789
"""Create a new RepositoryPackCollection.
791
:param transport: Addresses the repository base directory
792
(typically .bzr/repository/).
793
:param index_transport: Addresses the directory containing indices.
794
:param upload_transport: Addresses the directory into which packs are written
795
while they're being created.
796
:param pack_transport: Addresses the directory of existing complete packs.
799
self.transport = transport
800
self._index_transport = index_transport
801
self._upload_transport = upload_transport
802
self._pack_transport = pack_transport
803
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
806
self._packs_by_name = {}
807
# the previous pack-names content
808
self._packs_at_load = None
809
# when a pack is being created by this object, the state of that pack.
810
self._new_pack = None
811
# aggregated revision index data
812
self.revision_index = AggregateIndex()
813
self.inventory_index = AggregateIndex()
814
self.text_index = AggregateIndex()
815
self.signature_index = AggregateIndex()
817
def add_pack_to_memory(self, pack):
818
"""Make a Pack object available to the repository to satisfy queries.
820
:param pack: A Pack object.
822
assert pack.name not in self._packs_by_name
823
self.packs.append(pack)
824
self._packs_by_name[pack.name] = pack
825
self.revision_index.add_index(pack.revision_index, pack)
826
self.inventory_index.add_index(pack.inventory_index, pack)
827
self.text_index.add_index(pack.text_index, pack)
828
self.signature_index.add_index(pack.signature_index, pack)
830
def _add_text_to_weave(self, file_id, revision_id, new_lines, parents,
831
nostore_sha, random_revid):
832
file_id_index = GraphIndexPrefixAdapter(
833
self.text_index.combined_index,
835
add_nodes_callback=self.text_index.add_callback)
836
self.repo._text_knit._index._graph_index = file_id_index
837
self.repo._text_knit._index._add_callback = file_id_index.add_nodes
838
return self.repo._text_knit.add_lines_with_ghosts(
839
revision_id, parents, new_lines, nostore_sha=nostore_sha,
840
random_id=random_revid, check_content=False)[0:2]
843
"""Return a list of all the Pack objects this repository has.
845
Note that an in-progress pack being created is not returned.
847
:return: A list of Pack objects for all the packs in the repository.
850
for name in self.names():
851
result.append(self.get_pack_by_name(name))
855
"""Pack the pack collection incrementally.
857
This will not attempt global reorganisation or recompression,
858
rather it will just ensure that the total number of packs does
859
not grow without bound. It uses the _max_pack_count method to
860
determine if autopacking is needed, and the pack_distribution
861
method to determine the number of revisions in each pack.
863
If autopacking takes place then the packs name collection will have
864
been flushed to disk - packing requires updating the name collection
865
in synchronisation with certain steps. Otherwise the names collection
868
:return: True if packing took place.
870
# XXX: Should not be needed when the management of indices is sane.
871
total_revisions = self.revision_index.combined_index.key_count()
872
total_packs = len(self._names)
873
if self._max_pack_count(total_revisions) >= total_packs:
875
# XXX: the following may want to be a class, to pack with a given
877
mutter('Auto-packing repository %s, which has %d pack files, '
878
'containing %d revisions into %d packs.', self, total_packs,
879
total_revisions, self._max_pack_count(total_revisions))
880
# determine which packs need changing
881
pack_distribution = self.pack_distribution(total_revisions)
883
for pack in self.all_packs():
884
revision_count = pack.get_revision_count()
885
if revision_count == 0:
886
# revision less packs are not generated by normal operation,
887
# only by operations like sign-my-commits, and thus will not
888
# tend to grow rapdily or without bound like commit containing
889
# packs do - leave them alone as packing them really should
890
# group their data with the relevant commit, and that may
891
# involve rewriting ancient history - which autopack tries to
892
# avoid. Alternatively we could not group the data but treat
893
# each of these as having a single revision, and thus add
894
# one revision for each to the total revision count, to get
895
# a matching distribution.
897
existing_packs.append((revision_count, pack))
898
pack_operations = self.plan_autopack_combinations(
899
existing_packs, pack_distribution)
900
self._execute_pack_operations(pack_operations)
903
def _execute_pack_operations(self, pack_operations):
904
"""Execute a series of pack operations.
906
:param pack_operations: A list of [revision_count, packs_to_combine].
909
for revision_count, packs in pack_operations:
910
# we may have no-ops from the setup logic
913
Packer(self, packs, '.autopack').pack()
915
self._remove_pack_from_memory(pack)
916
# record the newly available packs and stop advertising the old
918
self._save_pack_names(clear_obsolete_packs=True)
919
# Move the old packs out of the way now they are no longer referenced.
920
for revision_count, packs in pack_operations:
921
self._obsolete_packs(packs)
923
def lock_names(self):
924
"""Acquire the mutex around the pack-names index.
926
This cannot be used in the middle of a read-only transaction on the
929
self.repo.control_files.lock_write()
932
"""Pack the pack collection totally."""
934
total_packs = len(self._names)
937
total_revisions = self.revision_index.combined_index.key_count()
938
# XXX: the following may want to be a class, to pack with a given
940
mutter('Packing repository %s, which has %d pack files, '
941
'containing %d revisions into 1 packs.', self, total_packs,
943
# determine which packs need changing
944
pack_distribution = [1]
945
pack_operations = [[0, []]]
946
for pack in self.all_packs():
947
revision_count = pack.get_revision_count()
948
pack_operations[-1][0] += revision_count
949
pack_operations[-1][1].append(pack)
950
self._execute_pack_operations(pack_operations)
952
def plan_autopack_combinations(self, existing_packs, pack_distribution):
953
"""Plan a pack operation.
955
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
957
:param pack_distribution: A list with the number of revisions desired
960
if len(existing_packs) <= len(pack_distribution):
962
existing_packs.sort(reverse=True)
963
pack_operations = [[0, []]]
964
# plan out what packs to keep, and what to reorganise
965
while len(existing_packs):
966
# take the largest pack, and if its less than the head of the
967
# distribution chart we will include its contents in the new pack for
968
# that position. If its larger, we remove its size from the
970
next_pack_rev_count, next_pack = existing_packs.pop(0)
971
if next_pack_rev_count >= pack_distribution[0]:
972
# this is already packed 'better' than this, so we can
973
# not waste time packing it.
974
while next_pack_rev_count > 0:
975
next_pack_rev_count -= pack_distribution[0]
976
if next_pack_rev_count >= 0:
978
del pack_distribution[0]
980
# didn't use that entire bucket up
981
pack_distribution[0] = -next_pack_rev_count
983
# add the revisions we're going to add to the next output pack
984
pack_operations[-1][0] += next_pack_rev_count
985
# allocate this pack to the next pack sub operation
986
pack_operations[-1][1].append(next_pack)
987
if pack_operations[-1][0] >= pack_distribution[0]:
988
# this pack is used up, shift left.
989
del pack_distribution[0]
990
pack_operations.append([0, []])
992
return pack_operations
994
def ensure_loaded(self):
995
# NB: if you see an assertion error here, its probably access against
996
# an unlocked repo. Naughty.
997
assert self.repo.is_locked()
998
if self._names is None:
1000
self._packs_at_load = set()
1001
for index, key, value in self._iter_disk_pack_index():
1003
self._names[name] = self._parse_index_sizes(value)
1004
self._packs_at_load.add((key, value))
1005
# populate all the metadata.
1008
def _parse_index_sizes(self, value):
1009
"""Parse a string of index sizes."""
1010
return tuple([int(digits) for digits in value.split(' ')])
1012
def get_pack_by_name(self, name):
1013
"""Get a Pack object by name.
1015
:param name: The name of the pack - e.g. '123456'
1016
:return: A Pack object.
1019
return self._packs_by_name[name]
1021
rev_index = self._make_index(name, '.rix')
1022
inv_index = self._make_index(name, '.iix')
1023
txt_index = self._make_index(name, '.tix')
1024
sig_index = self._make_index(name, '.six')
1025
result = ExistingPack(self._pack_transport, name, rev_index,
1026
inv_index, txt_index, sig_index)
1027
self.add_pack_to_memory(result)
1030
def allocate(self, a_new_pack):
1031
"""Allocate name in the list of packs.
1033
:param a_new_pack: A NewPack instance to be added to the collection of
1034
packs for this repository.
1036
self.ensure_loaded()
1037
if a_new_pack.name in self._names:
1038
# a collision with the packs we know about (not the only possible
1039
# collision, see NewPack.finish() for some discussion). Remove our
1040
# prior reference to it.
1041
self._remove_pack_from_memory(a_new_pack)
1042
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1043
self.add_pack_to_memory(a_new_pack)
1045
def _iter_disk_pack_index(self):
1046
"""Iterate over the contents of the pack-names index.
1048
This is used when loading the list from disk, and before writing to
1049
detect updates from others during our write operation.
1050
:return: An iterator of the index contents.
1052
return GraphIndex(self.transport, 'pack-names', None
1053
).iter_all_entries()
1055
def _make_index(self, name, suffix):
1056
size_offset = self._suffix_offsets[suffix]
1057
index_name = name + suffix
1058
index_size = self._names[name][size_offset]
1060
self._index_transport, index_name, index_size)
1062
def _max_pack_count(self, total_revisions):
1063
"""Return the maximum number of packs to use for total revisions.
1065
:param total_revisions: The total number of revisions in the
1068
if not total_revisions:
1070
digits = str(total_revisions)
1072
for digit in digits:
1073
result += int(digit)
1077
"""Provide an order to the underlying names."""
1078
return sorted(self._names.keys())
1080
def _obsolete_packs(self, packs):
1081
"""Move a number of packs which have been obsoleted out of the way.
1083
Each pack and its associated indices are moved out of the way.
1085
Note: for correctness this function should only be called after a new
1086
pack names index has been written without these pack names, and with
1087
the names of packs that contain the data previously available via these
1090
:param packs: The packs to obsolete.
1091
:param return: None.
1094
pack.pack_transport.rename(pack.file_name(),
1095
'../obsolete_packs/' + pack.file_name())
1096
# TODO: Probably needs to know all possible indices for this pack
1097
# - or maybe list the directory and move all indices matching this
1098
# name whether we recognize it or not?
1099
for suffix in ('.iix', '.six', '.tix', '.rix'):
1100
self._index_transport.rename(pack.name + suffix,
1101
'../obsolete_packs/' + pack.name + suffix)
1103
def pack_distribution(self, total_revisions):
1104
"""Generate a list of the number of revisions to put in each pack.
1106
:param total_revisions: The total number of revisions in the
1109
if total_revisions == 0:
1111
digits = reversed(str(total_revisions))
1113
for exponent, count in enumerate(digits):
1114
size = 10 ** exponent
1115
for pos in range(int(count)):
1117
return list(reversed(result))
1119
def _pack_tuple(self, name):
1120
"""Return a tuple with the transport and file name for a pack name."""
1121
return self._pack_transport, name + '.pack'
1123
def _remove_pack_from_memory(self, pack):
1124
"""Remove pack from the packs accessed by this repository.
1126
Only affects memory state, until self._save_pack_names() is invoked.
1128
self._names.pop(pack.name)
1129
self._packs_by_name.pop(pack.name)
1130
self._remove_pack_indices(pack)
1132
def _remove_pack_indices(self, pack):
1133
"""Remove the indices for pack from the aggregated indices."""
1134
self.revision_index.remove_index(pack.revision_index, pack)
1135
self.inventory_index.remove_index(pack.inventory_index, pack)
1136
self.text_index.remove_index(pack.text_index, pack)
1137
self.signature_index.remove_index(pack.signature_index, pack)
1140
"""Clear all cached data."""
1141
# cached revision data
1142
self.repo._revision_knit = None
1143
self.revision_index.clear()
1144
# cached signature data
1145
self.repo._signature_knit = None
1146
self.signature_index.clear()
1147
# cached file text data
1148
self.text_index.clear()
1149
self.repo._text_knit = None
1150
# cached inventory data
1151
self.inventory_index.clear()
1152
# remove the open pack
1153
self._new_pack = None
1154
# information about packs.
1157
self._packs_by_name = {}
1158
self._packs_at_load = None
1160
def _make_index_map(self, index_suffix):
1161
"""Return information on existing indices.
1163
:param suffix: Index suffix added to pack name.
1165
:returns: (pack_map, indices) where indices is a list of GraphIndex
1166
objects, and pack_map is a mapping from those objects to the
1167
pack tuple they describe.
1169
# TODO: stop using this; it creates new indices unnecessarily.
1170
self.ensure_loaded()
1171
suffix_map = {'.rix': 'revision_index',
1172
'.six': 'signature_index',
1173
'.iix': 'inventory_index',
1174
'.tix': 'text_index',
1176
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1177
suffix_map[index_suffix])
1179
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1180
"""Convert a list of packs to an index pack map and index list.
1182
:param packs: The packs list to process.
1183
:param index_attribute: The attribute that the desired index is found
1185
:return: A tuple (map, list) where map contains the dict from
1186
index:pack_tuple, and lsit contains the indices in the same order
1192
index = getattr(pack, index_attribute)
1193
indices.append(index)
1194
pack_map[index] = (pack.pack_transport, pack.file_name())
1195
return pack_map, indices
1197
def _index_contents(self, pack_map, key_filter=None):
1198
"""Get an iterable of the index contents from a pack_map.
1200
:param pack_map: A map from indices to pack details.
1201
:param key_filter: An optional filter to limit the
1204
indices = [index for index in pack_map.iterkeys()]
1205
all_index = CombinedGraphIndex(indices)
1206
if key_filter is None:
1207
return all_index.iter_all_entries()
1209
return all_index.iter_entries(key_filter)
1211
def _unlock_names(self):
1212
"""Release the mutex around the pack-names index."""
1213
self.repo.control_files.unlock()
1215
def _save_pack_names(self, clear_obsolete_packs=False):
1216
"""Save the list of packs.
1218
This will take out the mutex around the pack names list for the
1219
duration of the method call. If concurrent updates have been made, a
1220
three-way merge between the current list and the current in memory list
1223
:param clear_obsolete_packs: If True, clear out the contents of the
1224
obsolete_packs directory.
1228
builder = GraphIndexBuilder()
1229
# load the disk nodes across
1231
for index, key, value in self._iter_disk_pack_index():
1232
disk_nodes.add((key, value))
1233
# do a two-way diff against our original content
1234
current_nodes = set()
1235
for name, sizes in self._names.iteritems():
1237
((name, ), ' '.join(str(size) for size in sizes)))
1238
deleted_nodes = self._packs_at_load - current_nodes
1239
new_nodes = current_nodes - self._packs_at_load
1240
disk_nodes.difference_update(deleted_nodes)
1241
disk_nodes.update(new_nodes)
1242
# TODO: handle same-name, index-size-changes here -
1243
# e.g. use the value from disk, not ours, *unless* we're the one
1245
for key, value in disk_nodes:
1246
builder.add_node(key, value)
1247
self.transport.put_file('pack-names', builder.finish())
1248
# move the baseline forward
1249
self._packs_at_load = disk_nodes
1250
# now clear out the obsolete packs directory
1251
if clear_obsolete_packs:
1252
self.transport.clone('obsolete_packs').delete_multi(
1253
self.transport.list_dir('obsolete_packs'))
1255
self._unlock_names()
1256
# synchronise the memory packs list with what we just wrote:
1257
new_names = dict(disk_nodes)
1258
# drop no longer present nodes
1259
for pack in self.all_packs():
1260
if (pack.name,) not in new_names:
1261
self._remove_pack_from_memory(pack)
1262
# add new nodes/refresh existing ones
1263
for key, value in disk_nodes:
1265
sizes = self._parse_index_sizes(value)
1266
if name in self._names:
1268
if sizes != self._names[name]:
1269
# the pack for name has had its indices replaced - rare but
1270
# important to handle. XXX: probably can never happen today
1271
# because the three-way merge code above does not handle it
1272
# - you may end up adding the same key twice to the new
1273
# disk index because the set values are the same, unless
1274
# the only index shows up as deleted by the set difference
1275
# - which it may. Until there is a specific test for this,
1276
# assume its broken. RBC 20071017.
1277
self._remove_pack_from_memory(self.get_pack_by_name(name))
1278
self._names[name] = sizes
1279
self.get_pack_by_name(name)
1282
self._names[name] = sizes
1283
self.get_pack_by_name(name)
1285
def _start_write_group(self):
1286
# Do not permit preparation for writing if we're not in a 'write lock'.
1287
if not self.repo.is_write_locked():
1288
raise errors.NotWriteLocked(self)
1289
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1290
self._pack_transport, upload_suffix='.pack')
1291
# allow writing: queue writes to a new index
1292
self.revision_index.add_writable_index(self._new_pack.revision_index,
1294
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
1296
self.text_index.add_writable_index(self._new_pack.text_index,
1298
self.signature_index.add_writable_index(self._new_pack.signature_index,
1301
# reused revision and signature knits may need updating
1303
# "Hysterical raisins. client code in bzrlib grabs those knits outside
1304
# of write groups and then mutates it inside the write group."
1305
if self.repo._revision_knit is not None:
1306
self.repo._revision_knit._index._add_callback = \
1307
self.revision_index.add_callback
1308
if self.repo._signature_knit is not None:
1309
self.repo._signature_knit._index._add_callback = \
1310
self.signature_index.add_callback
1311
# create a reused knit object for text addition in commit.
1312
self.repo._text_knit = self.repo.weave_store.get_weave_or_empty(
1315
def _abort_write_group(self):
1316
# FIXME: just drop the transient index.
1317
# forget what names there are
1318
self._new_pack.abort()
1319
self._remove_pack_indices(self._new_pack)
1320
self._new_pack = None
1321
self.repo._text_knit = None
1323
def _commit_write_group(self):
1324
self._remove_pack_indices(self._new_pack)
1325
if self._new_pack.data_inserted():
1326
# get all the data to disk and read to use
1327
self._new_pack.finish()
1328
self.allocate(self._new_pack)
1329
self._new_pack = None
1330
if not self.autopack():
1331
# when autopack takes no steps, the names list is still
1333
self._save_pack_names()
1335
self._new_pack.abort()
1336
self._new_pack = None
1337
self.repo._text_knit = None
1340
class KnitPackRevisionStore(KnitRevisionStore):
1341
"""An object to adapt access from RevisionStore's to use KnitPacks.
1343
This class works by replacing the original RevisionStore.
1344
We need to do this because the KnitPackRevisionStore is less
1345
isolated in its layering - it uses services from the repo.
1348
def __init__(self, repo, transport, revisionstore):
1349
"""Create a KnitPackRevisionStore on repo with revisionstore.
1351
This will store its state in the Repository, use the
1352
indices to provide a KnitGraphIndex,
1353
and at the end of transactions write new indices.
1355
KnitRevisionStore.__init__(self, revisionstore.versioned_file_store)
1357
self._serializer = revisionstore._serializer
1358
self.transport = transport
1360
def get_revision_file(self, transaction):
1361
"""Get the revision versioned file object."""
1362
if getattr(self.repo, '_revision_knit', None) is not None:
1363
return self.repo._revision_knit
1364
self.repo._pack_collection.ensure_loaded()
1365
add_callback = self.repo._pack_collection.revision_index.add_callback
1366
# setup knit specific objects
1367
knit_index = KnitGraphIndex(
1368
self.repo._pack_collection.revision_index.combined_index,
1369
add_callback=add_callback)
1370
self.repo._revision_knit = knit.KnitVersionedFile(
1371
'revisions', self.transport.clone('..'),
1372
self.repo.control_files._file_mode,
1373
create=False, access_mode=self.repo._access_mode(),
1374
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1375
access_method=self.repo._pack_collection.revision_index.knit_access)
1376
return self.repo._revision_knit
1378
def get_signature_file(self, transaction):
1379
"""Get the signature versioned file object."""
1380
if getattr(self.repo, '_signature_knit', None) is not None:
1381
return self.repo._signature_knit
1382
self.repo._pack_collection.ensure_loaded()
1383
add_callback = self.repo._pack_collection.signature_index.add_callback
1384
# setup knit specific objects
1385
knit_index = KnitGraphIndex(
1386
self.repo._pack_collection.signature_index.combined_index,
1387
add_callback=add_callback, parents=False)
1388
self.repo._signature_knit = knit.KnitVersionedFile(
1389
'signatures', self.transport.clone('..'),
1390
self.repo.control_files._file_mode,
1391
create=False, access_mode=self.repo._access_mode(),
1392
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1393
access_method=self.repo._pack_collection.signature_index.knit_access)
1394
return self.repo._signature_knit
1397
class KnitPackTextStore(VersionedFileStore):
1398
"""Presents a TextStore abstraction on top of packs.
1400
This class works by replacing the original VersionedFileStore.
1401
We need to do this because the KnitPackRevisionStore is less
1402
isolated in its layering - it uses services from the repo and shares them
1403
with all the data written in a single write group.
1406
def __init__(self, repo, transport, weavestore):
1407
"""Create a KnitPackTextStore on repo with weavestore.
1409
This will store its state in the Repository, use the
1410
indices FileNames to provide a KnitGraphIndex,
1411
and at the end of transactions write new indices.
1413
# don't call base class constructor - it's not suitable.
1414
# no transient data stored in the transaction
1416
self._precious = False
1418
self.transport = transport
1419
self.weavestore = weavestore
1420
# XXX for check() which isn't updated yet
1421
self._transport = weavestore._transport
1423
def get_weave_or_empty(self, file_id, transaction):
1424
"""Get a 'Knit' backed by the .tix indices.
1426
The transaction parameter is ignored.
1428
self.repo._pack_collection.ensure_loaded()
1429
add_callback = self.repo._pack_collection.text_index.add_callback
1430
# setup knit specific objects
1431
file_id_index = GraphIndexPrefixAdapter(
1432
self.repo._pack_collection.text_index.combined_index,
1433
(file_id, ), 1, add_nodes_callback=add_callback)
1434
knit_index = KnitGraphIndex(file_id_index,
1435
add_callback=file_id_index.add_nodes,
1436
deltas=True, parents=True)
1437
return knit.KnitVersionedFile('text:' + file_id,
1438
self.transport.clone('..'),
1441
access_method=self.repo._pack_collection.text_index.knit_access,
1442
factory=knit.KnitPlainFactory())
1444
get_weave = get_weave_or_empty
1447
"""Generate a list of the fileids inserted, for use by check."""
1448
self.repo._pack_collection.ensure_loaded()
1450
for index, key, value, refs in \
1451
self.repo._pack_collection.text_index.combined_index.iter_all_entries():
1456
class InventoryKnitThunk(object):
1457
"""An object to manage thunking get_inventory_weave to pack based knits."""
1459
def __init__(self, repo, transport):
1460
"""Create an InventoryKnitThunk for repo at transport.
1462
This will store its state in the Repository, use the
1463
indices FileNames to provide a KnitGraphIndex,
1464
and at the end of transactions write a new index..
1467
self.transport = transport
1469
def get_weave(self):
1470
"""Get a 'Knit' that contains inventory data."""
1471
self.repo._pack_collection.ensure_loaded()
1472
add_callback = self.repo._pack_collection.inventory_index.add_callback
1473
# setup knit specific objects
1474
knit_index = KnitGraphIndex(
1475
self.repo._pack_collection.inventory_index.combined_index,
1476
add_callback=add_callback, deltas=True, parents=True)
1477
return knit.KnitVersionedFile(
1478
'inventory', self.transport.clone('..'),
1479
self.repo.control_files._file_mode,
1480
create=False, access_mode=self.repo._access_mode(),
1481
index=knit_index, delta=True, factory=knit.KnitPlainFactory(),
1482
access_method=self.repo._pack_collection.inventory_index.knit_access)
1485
class KnitPackRepository(KnitRepository):
1486
"""Experimental graph-knit using repository."""
1488
def __init__(self, _format, a_bzrdir, control_files, _revision_store,
1489
control_store, text_store, _commit_builder_class, _serializer):
1490
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
1491
_revision_store, control_store, text_store, _commit_builder_class,
1493
index_transport = control_files._transport.clone('indices')
1494
self._pack_collection = RepositoryPackCollection(self, control_files._transport,
1496
control_files._transport.clone('upload'),
1497
control_files._transport.clone('packs'))
1498
self._revision_store = KnitPackRevisionStore(self, index_transport, self._revision_store)
1499
self.weave_store = KnitPackTextStore(self, index_transport, self.weave_store)
1500
self._inv_thunk = InventoryKnitThunk(self, index_transport)
1501
# True when the repository object is 'write locked' (as opposed to the
1502
# physical lock only taken out around changes to the pack-names list.)
1503
# Another way to represent this would be a decorator around the control
1504
# files object that presents logical locks as physical ones - if this
1505
# gets ugly consider that alternative design. RBC 20071011
1506
self._write_lock_count = 0
1507
self._transaction = None
1509
self._reconcile_does_inventory_gc = True
1510
self._reconcile_fixes_text_parents = False
1511
self._reconcile_backsup_inventory = False
1513
def _abort_write_group(self):
1514
self._pack_collection._abort_write_group()
1516
def _access_mode(self):
1517
"""Return 'w' or 'r' for depending on whether a write lock is active.
1519
This method is a helper for the Knit-thunking support objects.
1521
if self.is_write_locked():
1525
def _find_inconsistent_revision_parents(self):
1526
"""Find revisions with incorrectly cached parents.
1528
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1529
parents-in-revision).
1531
assert self.is_locked()
1532
pb = ui.ui_factory.nested_progress_bar()
1535
revision_nodes = self._pack_collection.revision_index \
1536
.combined_index.iter_all_entries()
1537
index_positions = []
1538
# Get the cached index values for all revisions, and also the location
1539
# in each index of the revision text so we can perform linear IO.
1540
for index, key, value, refs in revision_nodes:
1541
pos, length = value[1:].split(' ')
1542
index_positions.append((index, int(pos), key[0],
1543
tuple(parent[0] for parent in refs[0])))
1544
pb.update("Reading revision index.", 0, 0)
1545
index_positions.sort()
1546
batch_count = len(index_positions) / 1000 + 1
1547
pb.update("Checking cached revision graph.", 0, batch_count)
1548
for offset in xrange(batch_count):
1549
pb.update("Checking cached revision graph.", offset)
1550
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1553
rev_ids = [item[2] for item in to_query]
1554
revs = self.get_revisions(rev_ids)
1555
for revision, item in zip(revs, to_query):
1556
index_parents = item[3]
1557
rev_parents = tuple(revision.parent_ids)
1558
if index_parents != rev_parents:
1559
result.append((revision.revision_id, index_parents, rev_parents))
1564
def get_parents(self, revision_ids):
1565
"""See StackedParentsProvider.get_parents.
1567
This implementation accesses the combined revision index to provide
1570
self._pack_collection.ensure_loaded()
1571
index = self._pack_collection.revision_index.combined_index
1573
for revision_id in revision_ids:
1574
if revision_id != _mod_revision.NULL_REVISION:
1575
search_keys.add((revision_id,))
1576
found_parents = {_mod_revision.NULL_REVISION:[]}
1577
for index, key, value, refs in index.iter_entries(search_keys):
1580
parents = (_mod_revision.NULL_REVISION,)
1582
parents = tuple(parent[0] for parent in parents)
1583
found_parents[key[0]] = parents
1585
for revision_id in revision_ids:
1587
result.append(found_parents[revision_id])
1592
def _make_parents_provider(self):
1595
def _refresh_data(self):
1596
if self._write_lock_count == 1 or (
1597
self.control_files._lock_count == 1 and
1598
self.control_files._lock_mode == 'r'):
1599
# forget what names there are
1600
self._pack_collection.reset()
1601
# XXX: Better to do an in-memory merge when acquiring a new lock -
1602
# factor out code from _save_pack_names.
1603
self._pack_collection.ensure_loaded()
1605
def _start_write_group(self):
1606
self._pack_collection._start_write_group()
1608
def _commit_write_group(self):
1609
return self._pack_collection._commit_write_group()
1611
def get_inventory_weave(self):
1612
return self._inv_thunk.get_weave()
1614
def get_transaction(self):
1615
if self._write_lock_count:
1616
return self._transaction
1618
return self.control_files.get_transaction()
1620
def is_locked(self):
1621
return self._write_lock_count or self.control_files.is_locked()
1623
def is_write_locked(self):
1624
return self._write_lock_count
1626
def lock_write(self, token=None):
1627
if not self._write_lock_count and self.is_locked():
1628
raise errors.ReadOnlyError(self)
1629
self._write_lock_count += 1
1630
if self._write_lock_count == 1:
1631
from bzrlib import transactions
1632
self._transaction = transactions.WriteTransaction()
1633
self._refresh_data()
1635
def lock_read(self):
1636
if self._write_lock_count:
1637
self._write_lock_count += 1
1639
self.control_files.lock_read()
1640
self._refresh_data()
1642
def leave_lock_in_place(self):
1643
# not supported - raise an error
1644
raise NotImplementedError(self.leave_lock_in_place)
1646
def dont_leave_lock_in_place(self):
1647
# not supported - raise an error
1648
raise NotImplementedError(self.dont_leave_lock_in_place)
1652
"""Compress the data within the repository.
1654
This will pack all the data to a single pack. In future it may
1655
recompress deltas or do other such expensive operations.
1657
self._pack_collection.pack()
1660
def reconcile(self, other=None, thorough=False):
1661
"""Reconcile this repository."""
1662
from bzrlib.reconcile import PackReconciler
1663
reconciler = PackReconciler(self, thorough=thorough)
1664
reconciler.reconcile()
1668
if self._write_lock_count == 1 and self._write_group is not None:
1669
self.abort_write_group()
1670
self._transaction = None
1671
self._write_lock_count = 0
1672
raise errors.BzrError(
1673
'Must end write group before releasing write lock on %s'
1675
if self._write_lock_count:
1676
self._write_lock_count -= 1
1677
if not self._write_lock_count:
1678
transaction = self._transaction
1679
self._transaction = None
1680
transaction.finish()
1682
self.control_files.unlock()
1685
class RepositoryFormatPack(MetaDirRepositoryFormat):
1686
"""Format logic for pack structured repositories.
1688
This repository format has:
1689
- a list of packs in pack-names
1690
- packs in packs/NAME.pack
1691
- indices in indices/NAME.{iix,six,tix,rix}
1692
- knit deltas in the packs, knit indices mapped to the indices.
1693
- thunk objects to support the knits programming API.
1694
- a format marker of its own
1695
- an optional 'shared-storage' flag
1696
- an optional 'no-working-trees' flag
1700
# Set this attribute in derived classes to control the repository class
1701
# created by open and initialize.
1702
repository_class = None
1703
# Set this attribute in derived classes to control the
1704
# _commit_builder_class that the repository objects will have passed to
1705
# their constructor.
1706
_commit_builder_class = None
1707
# Set this attribute in derived clases to control the _serializer that the
1708
# repository objects will have passed to their constructor.
1711
def _get_control_store(self, repo_transport, control_files):
1712
"""Return the control store for this repository."""
1713
return VersionedFileStore(
1716
file_mode=control_files._file_mode,
1717
versionedfile_class=knit.KnitVersionedFile,
1718
versionedfile_kwargs={'factory': knit.KnitPlainFactory()},
1721
def _get_revision_store(self, repo_transport, control_files):
1722
"""See RepositoryFormat._get_revision_store()."""
1723
versioned_file_store = VersionedFileStore(
1725
file_mode=control_files._file_mode,
1728
versionedfile_class=knit.KnitVersionedFile,
1729
versionedfile_kwargs={'delta': False,
1730
'factory': knit.KnitPlainFactory(),
1734
return KnitRevisionStore(versioned_file_store)
1736
def _get_text_store(self, transport, control_files):
1737
"""See RepositoryFormat._get_text_store()."""
1738
return self._get_versioned_file_store('knits',
1741
versionedfile_class=knit.KnitVersionedFile,
1742
versionedfile_kwargs={
1743
'create_parent_dir': True,
1744
'delay_create': True,
1745
'dir_mode': control_files._dir_mode,
1749
def initialize(self, a_bzrdir, shared=False):
1750
"""Create a pack based repository.
1752
:param a_bzrdir: bzrdir to contain the new repository; must already
1754
:param shared: If true the repository will be initialized as a shared
1757
mutter('creating repository in %s.', a_bzrdir.transport.base)
1758
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
1759
builder = GraphIndexBuilder()
1760
files = [('pack-names', builder.finish())]
1761
utf8_files = [('format', self.get_format_string())]
1763
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
1764
return self.open(a_bzrdir=a_bzrdir, _found=True)
1766
def open(self, a_bzrdir, _found=False, _override_transport=None):
1767
"""See RepositoryFormat.open().
1769
:param _override_transport: INTERNAL USE ONLY. Allows opening the
1770
repository at a slightly different url
1771
than normal. I.e. during 'upgrade'.
1774
format = RepositoryFormat.find_format(a_bzrdir)
1775
assert format.__class__ == self.__class__
1776
if _override_transport is not None:
1777
repo_transport = _override_transport
1779
repo_transport = a_bzrdir.get_repository_transport(None)
1780
control_files = lockable_files.LockableFiles(repo_transport,
1781
'lock', lockdir.LockDir)
1782
text_store = self._get_text_store(repo_transport, control_files)
1783
control_store = self._get_control_store(repo_transport, control_files)
1784
_revision_store = self._get_revision_store(repo_transport, control_files)
1785
return self.repository_class(_format=self,
1787
control_files=control_files,
1788
_revision_store=_revision_store,
1789
control_store=control_store,
1790
text_store=text_store,
1791
_commit_builder_class=self._commit_builder_class,
1792
_serializer=self._serializer)
1795
class RepositoryFormatKnitPack1(RepositoryFormatPack):
1796
"""A no-subtrees parameterised Pack repository.
1798
This format was introduced in 0.92.
1801
repository_class = KnitPackRepository
1802
_commit_builder_class = PackCommitBuilder
1803
_serializer = xml5.serializer_v5
1805
def _get_matching_bzrdir(self):
1806
return bzrdir.format_registry.make_bzrdir('knitpack-experimental')
1808
def _ignore_setting_bzrdir(self, format):
1811
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1813
def get_format_string(self):
1814
"""See RepositoryFormat.get_format_string()."""
1815
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
1817
def get_format_description(self):
1818
"""See RepositoryFormat.get_format_description()."""
1819
return "Packs containing knits without subtree support"
1821
def check_conversion_target(self, target_format):
1825
class RepositoryFormatKnitPack3(RepositoryFormatPack):
1826
"""A subtrees parameterised Pack repository.
1828
This repository format uses the xml7 serializer to get:
1829
- support for recording full info about the tree root
1830
- support for recording tree-references
1832
This format was introduced in 0.92.
1835
repository_class = KnitPackRepository
1836
_commit_builder_class = PackRootCommitBuilder
1837
rich_root_data = True
1838
supports_tree_reference = True
1839
_serializer = xml7.serializer_v7
1841
def _get_matching_bzrdir(self):
1842
return bzrdir.format_registry.make_bzrdir(
1843
'knitpack-subtree-experimental')
1845
def _ignore_setting_bzrdir(self, format):
1848
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1850
def check_conversion_target(self, target_format):
1851
if not target_format.rich_root_data:
1852
raise errors.BadConversionTarget(
1853
'Does not support rich root data.', target_format)
1854
if not getattr(target_format, 'supports_tree_reference', False):
1855
raise errors.BadConversionTarget(
1856
'Does not support nested trees', target_format)
1858
def get_format_string(self):
1859
"""See RepositoryFormat.get_format_string()."""
1860
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
1862
def get_format_description(self):
1863
"""See RepositoryFormat.get_format_description()."""
1864
return "Packs containing knits with subtree support\n"
added
removed
bzrlib/repofmt/pack_repo.py
