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,
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'
203
# The relative locations of the packs are constrained, but all are
204
# passed in because the caller has them, so as to avoid object churn.
206
# Revisions: parents list, no text compression.
207
InMemoryGraphIndex(reference_lists=1),
208
# Inventory: We want to map compression only, but currently the
209
# knit code hasn't been updated enough to understand that, so we
210
# have a regular 2-list index giving parents and compression
212
InMemoryGraphIndex(reference_lists=2),
213
# Texts: compression and per file graph, for all fileids - so two
214
# reference lists and two elements in the key tuple.
215
InMemoryGraphIndex(reference_lists=2, key_elements=2),
216
# Signatures: Just blobs to store, no compression, no parents
218
InMemoryGraphIndex(reference_lists=0),
220
# where should the new pack be opened
221
self.upload_transport = upload_transport
222
# where are indices written out to
223
self.index_transport = index_transport
224
# where is the pack renamed to when it is finished?
225
self.pack_transport = pack_transport
226
# tracks the content written to the .pack file.
227
self._hash = md5.new()
228
# a four-tuple with the length in bytes of the indices, once the pack
229
# is finalised. (rev, inv, text, sigs)
230
self.index_sizes = None
231
# How much data to cache when writing packs. Note that this is not
232
# synchronised with reads, because it's not in the transport layer, so
233
# is not safe unless the client knows it won't be reading from the pack
235
self._cache_limit = 0
236
# the temporary pack file name.
237
self.random_name = rand_chars(20) + upload_suffix
238
# when was this pack started ?
239
self.start_time = time.time()
240
# open an output stream for the data added to the pack.
241
self.write_stream = self.upload_transport.open_write_stream(
243
if 'pack' in debug.debug_flags:
244
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
245
time.ctime(), self.upload_transport.base, self.random_name,
246
time.time() - self.start_time)
247
# A list of byte sequences to be written to the new pack, and the
248
# aggregate size of them. Stored as a list rather than separate
249
# variables so that the _write_data closure below can update them.
250
self._buffer = [[], 0]
251
# create a callable for adding data
253
# robertc says- this is a closure rather than a method on the object
254
# so that the variables are locals, and faster than accessing object
256
def _write_data(bytes, flush=False, _buffer=self._buffer,
257
_write=self.write_stream.write, _update=self._hash.update):
258
_buffer[0].append(bytes)
259
_buffer[1] += len(bytes)
261
if _buffer[1] > self._cache_limit or flush:
262
bytes = ''.join(_buffer[0])
266
# expose this on self, for the occasion when clients want to add data.
267
self._write_data = _write_data
268
# a pack writer object to serialise pack records.
269
self._writer = pack.ContainerWriter(self._write_data)
271
# what state is the pack in? (open, finished, aborted)
275
"""Cancel creating this pack."""
276
self._state = 'aborted'
277
self.write_stream.close()
278
# Remove the temporary pack file.
279
self.upload_transport.delete(self.random_name)
280
# The indices have no state on disk.
282
def access_tuple(self):
283
"""Return a tuple (transport, name) for the pack content."""
284
assert self._state in ('open', 'finished')
285
if self._state == 'finished':
286
return Pack.access_tuple(self)
288
return self.upload_transport, self.random_name
290
def data_inserted(self):
291
"""True if data has been added to this pack."""
292
return bool(self.get_revision_count() or
293
self.inventory_index.key_count() or
294
self.text_index.key_count() or
295
self.signature_index.key_count())
298
"""Finish the new pack.
301
- finalises the content
302
- assigns a name (the md5 of the content, currently)
303
- writes out the associated indices
304
- renames the pack into place.
305
- stores the index size tuple for the pack in the index_sizes
310
self._write_data('', flush=True)
311
self.name = self._hash.hexdigest()
313
# XXX: It'd be better to write them all to temporary names, then
314
# rename them all into place, so that the window when only some are
315
# visible is smaller. On the other hand none will be seen until
316
# they're in the names list.
317
self.index_sizes = [None, None, None, None]
318
self._write_index('revision', self.revision_index, 'revision')
319
self._write_index('inventory', self.inventory_index, 'inventory')
320
self._write_index('text', self.text_index, 'file texts')
321
self._write_index('signature', self.signature_index,
322
'revision signatures')
323
self.write_stream.close()
324
# Note that this will clobber an existing pack with the same name,
325
# without checking for hash collisions. While this is undesirable this
326
# is something that can be rectified in a subsequent release. One way
327
# to rectify it may be to leave the pack at the original name, writing
328
# its pack-names entry as something like 'HASH: index-sizes
329
# temporary-name'. Allocate that and check for collisions, if it is
330
# collision free then rename it into place. If clients know this scheme
331
# they can handle missing-file errors by:
332
# - try for HASH.pack
333
# - try for temporary-name
334
# - refresh the pack-list to see if the pack is now absent
335
self.upload_transport.rename(self.random_name,
336
'../packs/' + self.name + '.pack')
337
self._state = 'finished'
338
if 'pack' in debug.debug_flags:
339
# XXX: size might be interesting?
340
mutter('%s: create_pack: pack renamed into place: %s%s->%s%s t+%6.3fs',
341
time.ctime(), self.upload_transport.base, self.random_name,
342
self.pack_transport, self.name,
343
time.time() - self.start_time)
345
def index_name(self, index_type, name):
346
"""Get the disk name of an index type for pack name 'name'."""
347
return name + NewPack.index_definitions[index_type][0]
349
def index_offset(self, index_type):
350
"""Get the position in a index_size array for a given index type."""
351
return NewPack.index_definitions[index_type][1]
353
def _replace_index_with_readonly(self, index_type):
354
setattr(self, index_type + '_index',
355
GraphIndex(self.index_transport,
356
self.index_name(index_type, self.name),
357
self.index_sizes[self.index_offset(index_type)]))
359
def set_write_cache_size(self, size):
360
self._cache_limit = size
362
def _write_index(self, index_type, index, label):
363
"""Write out an index.
365
:param index_type: The type of index to write - e.g. 'revision'.
366
:param index: The index object to serialise.
367
:param label: What label to give the index e.g. 'revision'.
369
index_name = self.index_name(index_type, self.name)
370
self.index_sizes[self.index_offset(index_type)] = \
371
self.index_transport.put_file(index_name, index.finish())
372
if 'pack' in debug.debug_flags:
373
# XXX: size might be interesting?
374
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
375
time.ctime(), label, self.upload_transport.base,
376
self.random_name, time.time() - self.start_time)
377
# Replace the writable index on this object with a readonly,
378
# presently unloaded index. We should alter
379
# the index layer to make its finish() error if add_node is
380
# subsequently used. RBC
381
self._replace_index_with_readonly(index_type)
384
class AggregateIndex(object):
385
"""An aggregated index for the RepositoryPackCollection.
387
AggregateIndex is reponsible for managing the PackAccess object,
388
Index-To-Pack mapping, and all indices list for a specific type of index
389
such as 'revision index'.
391
A CombinedIndex provides an index on a single key space built up
392
from several on-disk indices. The AggregateIndex builds on this
393
to provide a knit access layer, and allows having up to one writable
394
index within the collection.
396
# XXX: Probably 'can be written to' could/should be separated from 'acts
397
# like a knit index' -- mbp 20071024
400
"""Create an AggregateIndex."""
401
self.index_to_pack = {}
402
self.combined_index = CombinedGraphIndex([])
403
self.knit_access = _PackAccess(self.index_to_pack)
405
def replace_indices(self, index_to_pack, indices):
406
"""Replace the current mappings with fresh ones.
408
This should probably not be used eventually, rather incremental add and
409
removal of indices. It has been added during refactoring of existing
412
:param index_to_pack: A mapping from index objects to
413
(transport, name) tuples for the pack file data.
414
:param indices: A list of indices.
416
# refresh the revision pack map dict without replacing the instance.
417
self.index_to_pack.clear()
418
self.index_to_pack.update(index_to_pack)
419
# XXX: API break - clearly a 'replace' method would be good?
420
self.combined_index._indices[:] = indices
421
# the current add nodes callback for the current writable index if
423
self.add_callback = None
425
def add_index(self, index, pack):
426
"""Add index to the aggregate, which is an index for Pack pack.
428
Future searches on the aggregate index will seach this new index
429
before all previously inserted indices.
431
:param index: An Index for the pack.
432
:param pack: A Pack instance.
434
# expose it to the index map
435
self.index_to_pack[index] = pack.access_tuple()
436
# put it at the front of the linear index list
437
self.combined_index.insert_index(0, index)
439
def add_writable_index(self, index, pack):
440
"""Add an index which is able to have data added to it.
442
There can be at most one writable index at any time. Any
443
modifications made to the knit are put into this index.
445
:param index: An index from the pack parameter.
446
:param pack: A Pack instance.
448
assert self.add_callback is None, \
449
"%s already has a writable index through %s" % \
450
(self, self.add_callback)
451
# allow writing: queue writes to a new index
452
self.add_index(index, pack)
453
# Updates the index to packs mapping as a side effect,
454
self.knit_access.set_writer(pack._writer, index, pack.access_tuple())
455
self.add_callback = index.add_nodes
458
"""Reset all the aggregate data to nothing."""
459
self.knit_access.set_writer(None, None, (None, None))
460
self.index_to_pack.clear()
461
del self.combined_index._indices[:]
462
self.add_callback = None
464
def remove_index(self, index, pack):
465
"""Remove index from the indices used to answer queries.
467
:param index: An index from the pack parameter.
468
:param pack: A Pack instance.
470
del self.index_to_pack[index]
471
self.combined_index._indices.remove(index)
472
if (self.add_callback is not None and
473
getattr(index, 'add_nodes', None) == self.add_callback):
474
self.add_callback = None
475
self.knit_access.set_writer(None, None, (None, None))
478
class Packer(object):
479
"""Create a pack from packs."""
481
def __init__(self, pack_collection, packs, suffix, revision_ids=None):
484
:param pack_collection: A RepositoryPackCollection object where the
485
new pack is being written to.
486
:param packs: The packs to combine.
487
:param suffix: The suffix to use on the temporary files for the pack.
488
:param revision_ids: Revision ids to limit the pack to.
492
self.revision_ids = revision_ids
493
self._pack_collection = pack_collection
495
def pack(self, pb=None):
496
"""Create a new pack by reading data from other packs.
498
This does little more than a bulk copy of data. One key difference
499
is that data with the same item key across multiple packs is elided
500
from the output. The new pack is written into the current pack store
501
along with its indices, and the name added to the pack names. The
502
source packs are not altered and are not required to be in the current
505
:param pb: An optional progress bar to use. A nested bar is created if
507
:return: A Pack object, or None if nothing was copied.
509
# open a pack - using the same name as the last temporary file
510
# - which has already been flushed, so its safe.
511
# XXX: - duplicate code warning with start_write_group; fix before
512
# considering 'done'.
513
if self._pack_collection._new_pack is not None:
514
raise errors.BzrError('call to create_pack_from_packs while '
515
'another pack is being written.')
516
if self.revision_ids is not None:
517
if len(self.revision_ids) == 0:
518
# silly fetch request.
521
self.revision_ids = frozenset(self.revision_ids)
523
self.pb = ui.ui_factory.nested_progress_bar()
527
return self._create_pack_from_packs()
533
"""Open a pack for the pack we are creating."""
534
return NewPack(self._pack_collection._upload_transport,
535
self._pack_collection._index_transport,
536
self._pack_collection._pack_transport, upload_suffix=self.suffix)
538
def _create_pack_from_packs(self):
539
self.pb.update("Opening pack", 0, 5)
540
new_pack = self.open_pack()
541
# buffer data - we won't be reading-back during the pack creation and
542
# this makes a significant difference on sftp pushes.
543
new_pack.set_write_cache_size(1024*1024)
544
if 'pack' in debug.debug_flags:
545
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
546
for a_pack in self.packs]
547
if self.revision_ids is not None:
548
rev_count = len(self.revision_ids)
551
mutter('%s: create_pack: creating pack from source packs: '
552
'%s%s %s revisions wanted %s t=0',
553
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
554
plain_pack_list, rev_count)
556
if self.revision_ids:
557
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
561
# select revision keys
562
revision_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
563
self.packs, 'revision_index')[0]
564
revision_nodes = self._pack_collection._index_contents(revision_index_map, revision_keys)
565
# copy revision keys and adjust values
566
self.pb.update("Copying revision texts", 1)
567
list(self._copy_nodes_graph(revision_nodes, revision_index_map,
568
new_pack._writer, new_pack.revision_index))
569
if 'pack' in debug.debug_flags:
570
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
571
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
572
new_pack.revision_index.key_count(),
573
time.time() - new_pack.start_time)
574
# select inventory keys
575
inv_keys = revision_keys # currently the same keyspace, and note that
576
# querying for keys here could introduce a bug where an inventory item
577
# is missed, so do not change it to query separately without cross
578
# checking like the text key check below.
579
inventory_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
580
self.packs, 'inventory_index')[0]
581
inv_nodes = self._pack_collection._index_contents(inventory_index_map, inv_keys)
582
# copy inventory keys and adjust values
583
# XXX: Should be a helper function to allow different inv representation
585
self.pb.update("Copying inventory texts", 2)
586
inv_lines = self._copy_nodes_graph(inv_nodes, inventory_index_map,
587
new_pack._writer, new_pack.inventory_index, output_lines=True)
588
if self.revision_ids:
589
fileid_revisions = self._pack_collection.repo._find_file_ids_from_xml_inventory_lines(
590
inv_lines, self.revision_ids)
592
for fileid, file_revids in fileid_revisions.iteritems():
594
[(fileid, file_revid) for file_revid in file_revids])
596
# eat the iterator to cause it to execute.
599
if 'pack' in debug.debug_flags:
600
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
601
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
602
new_pack.inventory_index.key_count(),
603
time.time() - new_pack.start_time)
605
text_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
606
self.packs, 'text_index')[0]
607
text_nodes = self._pack_collection._index_contents(text_index_map, text_filter)
608
if text_filter is not None:
609
# We could return the keys copied as part of the return value from
610
# _copy_nodes_graph but this doesn't work all that well with the
611
# need to get line output too, so we check separately, and as we're
612
# going to buffer everything anyway, we check beforehand, which
613
# saves reading knit data over the wire when we know there are
615
text_nodes = set(text_nodes)
616
present_text_keys = set(_node[1] for _node in text_nodes)
617
missing_text_keys = set(text_filter) - present_text_keys
618
if missing_text_keys:
619
# TODO: raise a specific error that can handle many missing
621
a_missing_key = missing_text_keys.pop()
622
raise errors.RevisionNotPresent(a_missing_key[1],
624
# copy text keys and adjust values
625
self.pb.update("Copying content texts", 3)
626
list(self._copy_nodes_graph(text_nodes, text_index_map,
627
new_pack._writer, new_pack.text_index))
628
if 'pack' in debug.debug_flags:
629
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
630
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
631
new_pack.text_index.key_count(),
632
time.time() - new_pack.start_time)
633
# select signature keys
634
signature_filter = revision_keys # same keyspace
635
signature_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
636
self.packs, 'signature_index')[0]
637
signature_nodes = self._pack_collection._index_contents(signature_index_map,
639
# copy signature keys and adjust values
640
self.pb.update("Copying signature texts", 4)
641
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
642
new_pack.signature_index)
643
if 'pack' in debug.debug_flags:
644
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
645
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
646
new_pack.signature_index.key_count(),
647
time.time() - new_pack.start_time)
648
if not new_pack.data_inserted():
651
self.pb.update("Finishing pack", 5)
653
self._pack_collection.allocate(new_pack)
656
def _copy_nodes(self, nodes, index_map, writer, write_index):
657
"""Copy knit nodes between packs with no graph references."""
658
pb = ui.ui_factory.nested_progress_bar()
660
return self._do_copy_nodes(nodes, index_map, writer,
665
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb):
666
# for record verification
667
knit_data = _KnitData(None)
668
# plan a readv on each source pack:
670
nodes = sorted(nodes)
671
# how to map this into knit.py - or knit.py into this?
672
# we don't want the typical knit logic, we want grouping by pack
673
# at this point - perhaps a helper library for the following code
674
# duplication points?
676
for index, key, value in nodes:
677
if index not in request_groups:
678
request_groups[index] = []
679
request_groups[index].append((key, value))
681
pb.update("Copied record", record_index, len(nodes))
682
for index, items in request_groups.iteritems():
683
pack_readv_requests = []
684
for key, value in items:
685
# ---- KnitGraphIndex.get_position
686
bits = value[1:].split(' ')
687
offset, length = int(bits[0]), int(bits[1])
688
pack_readv_requests.append((offset, length, (key, value[0])))
689
# linear scan up the pack
690
pack_readv_requests.sort()
692
transport, path = index_map[index]
693
reader = pack.make_readv_reader(transport, path,
694
[offset[0:2] for offset in pack_readv_requests])
695
for (names, read_func), (_1, _2, (key, eol_flag)) in \
696
izip(reader.iter_records(), pack_readv_requests):
697
raw_data = read_func(None)
698
# check the header only
699
df, _ = knit_data._parse_record_header(key[-1], raw_data)
701
pos, size = writer.add_bytes_record(raw_data, names)
702
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
703
pb.update("Copied record", record_index)
706
def _copy_nodes_graph(self, nodes, index_map, writer, write_index,
708
"""Copy knit nodes between packs.
710
:param output_lines: Return lines present in the copied data as
711
an iterator of line,version_id.
713
pb = ui.ui_factory.nested_progress_bar()
715
return self._do_copy_nodes_graph(nodes, index_map, writer,
716
write_index, output_lines, pb)
720
def _do_copy_nodes_graph(self, nodes, index_map, writer, write_index,
722
# for record verification
723
knit_data = _KnitData(None)
724
# for line extraction when requested (inventories only)
726
factory = knit.KnitPlainFactory()
727
# plan a readv on each source pack:
729
nodes = sorted(nodes)
730
# how to map this into knit.py - or knit.py into this?
731
# we don't want the typical knit logic, we want grouping by pack
732
# at this point - perhaps a helper library for the following code
733
# duplication points?
736
pb.update("Copied record", record_index, len(nodes))
737
for index, key, value, references in nodes:
738
if index not in request_groups:
739
request_groups[index] = []
740
request_groups[index].append((key, value, references))
741
for index, items in request_groups.iteritems():
742
pack_readv_requests = []
743
for key, value, references in items:
744
# ---- KnitGraphIndex.get_position
745
bits = value[1:].split(' ')
746
offset, length = int(bits[0]), int(bits[1])
747
pack_readv_requests.append((offset, length, (key, value[0], references)))
748
# linear scan up the pack
749
pack_readv_requests.sort()
751
transport, path = index_map[index]
752
reader = pack.make_readv_reader(transport, path,
753
[offset[0:2] for offset in pack_readv_requests])
754
for (names, read_func), (_1, _2, (key, eol_flag, references)) in \
755
izip(reader.iter_records(), pack_readv_requests):
756
raw_data = read_func(None)
759
# read the entire thing
760
content, _ = knit_data._parse_record(version_id, raw_data)
761
if len(references[-1]) == 0:
762
line_iterator = factory.get_fulltext_content(content)
764
line_iterator = factory.get_linedelta_content(content)
765
for line in line_iterator:
766
yield line, version_id
768
# check the header only
769
df, _ = knit_data._parse_record_header(version_id, raw_data)
771
pos, size = writer.add_bytes_record(raw_data, names)
772
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
773
pb.update("Copied record", record_index)
777
class ReconcilePacker(Packer):
778
"""A packer which regenerates indices etc as it copies.
780
This is used by ``bzr reconcile`` to cause parent text pointers to be
785
class RepositoryPackCollection(object):
786
"""Management of packs within a repository."""
788
def __init__(self, repo, transport, index_transport, upload_transport,
790
"""Create a new RepositoryPackCollection.
792
:param transport: Addresses the repository base directory
793
(typically .bzr/repository/).
794
:param index_transport: Addresses the directory containing indices.
795
:param upload_transport: Addresses the directory into which packs are written
796
while they're being created.
797
:param pack_transport: Addresses the directory of existing complete packs.
800
self.transport = transport
801
self._index_transport = index_transport
802
self._upload_transport = upload_transport
803
self._pack_transport = pack_transport
804
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
807
self._packs_by_name = {}
808
# the previous pack-names content
809
self._packs_at_load = None
810
# when a pack is being created by this object, the state of that pack.
811
self._new_pack = None
812
# aggregated revision index data
813
self.revision_index = AggregateIndex()
814
self.inventory_index = AggregateIndex()
815
self.text_index = AggregateIndex()
816
self.signature_index = AggregateIndex()
818
def add_pack_to_memory(self, pack):
819
"""Make a Pack object available to the repository to satisfy queries.
821
:param pack: A Pack object.
823
assert pack.name not in self._packs_by_name
824
self.packs.append(pack)
825
self._packs_by_name[pack.name] = pack
826
self.revision_index.add_index(pack.revision_index, pack)
827
self.inventory_index.add_index(pack.inventory_index, pack)
828
self.text_index.add_index(pack.text_index, pack)
829
self.signature_index.add_index(pack.signature_index, pack)
831
def _add_text_to_weave(self, file_id, revision_id, new_lines, parents,
832
nostore_sha, random_revid):
833
file_id_index = GraphIndexPrefixAdapter(
834
self.text_index.combined_index,
836
add_nodes_callback=self.text_index.add_callback)
837
self.repo._text_knit._index._graph_index = file_id_index
838
self.repo._text_knit._index._add_callback = file_id_index.add_nodes
839
return self.repo._text_knit.add_lines_with_ghosts(
840
revision_id, parents, new_lines, nostore_sha=nostore_sha,
841
random_id=random_revid, check_content=False)[0:2]
844
"""Return a list of all the Pack objects this repository has.
846
Note that an in-progress pack being created is not returned.
848
:return: A list of Pack objects for all the packs in the repository.
851
for name in self.names():
852
result.append(self.get_pack_by_name(name))
856
"""Pack the pack collection incrementally.
858
This will not attempt global reorganisation or recompression,
859
rather it will just ensure that the total number of packs does
860
not grow without bound. It uses the _max_pack_count method to
861
determine if autopacking is needed, and the pack_distribution
862
method to determine the number of revisions in each pack.
864
If autopacking takes place then the packs name collection will have
865
been flushed to disk - packing requires updating the name collection
866
in synchronisation with certain steps. Otherwise the names collection
869
:return: True if packing took place.
871
# XXX: Should not be needed when the management of indices is sane.
872
total_revisions = self.revision_index.combined_index.key_count()
873
total_packs = len(self._names)
874
if self._max_pack_count(total_revisions) >= total_packs:
876
# XXX: the following may want to be a class, to pack with a given
878
mutter('Auto-packing repository %s, which has %d pack files, '
879
'containing %d revisions into %d packs.', self, total_packs,
880
total_revisions, self._max_pack_count(total_revisions))
881
# determine which packs need changing
882
pack_distribution = self.pack_distribution(total_revisions)
884
for pack in self.all_packs():
885
revision_count = pack.get_revision_count()
886
if revision_count == 0:
887
# revision less packs are not generated by normal operation,
888
# only by operations like sign-my-commits, and thus will not
889
# tend to grow rapdily or without bound like commit containing
890
# packs do - leave them alone as packing them really should
891
# group their data with the relevant commit, and that may
892
# involve rewriting ancient history - which autopack tries to
893
# avoid. Alternatively we could not group the data but treat
894
# each of these as having a single revision, and thus add
895
# one revision for each to the total revision count, to get
896
# a matching distribution.
898
existing_packs.append((revision_count, pack))
899
pack_operations = self.plan_autopack_combinations(
900
existing_packs, pack_distribution)
901
self._execute_pack_operations(pack_operations)
904
def _execute_pack_operations(self, pack_operations):
905
"""Execute a series of pack operations.
907
:param pack_operations: A list of [revision_count, packs_to_combine].
910
for revision_count, packs in pack_operations:
911
# we may have no-ops from the setup logic
914
Packer(self, packs, '.autopack').pack()
916
self._remove_pack_from_memory(pack)
917
# record the newly available packs and stop advertising the old
919
self._save_pack_names(clear_obsolete_packs=True)
920
# Move the old packs out of the way now they are no longer referenced.
921
for revision_count, packs in pack_operations:
922
self._obsolete_packs(packs)
924
def lock_names(self):
925
"""Acquire the mutex around the pack-names index.
927
This cannot be used in the middle of a read-only transaction on the
930
self.repo.control_files.lock_write()
933
"""Pack the pack collection totally."""
935
total_packs = len(self._names)
938
total_revisions = self.revision_index.combined_index.key_count()
939
# XXX: the following may want to be a class, to pack with a given
941
mutter('Packing repository %s, which has %d pack files, '
942
'containing %d revisions into 1 packs.', self, total_packs,
944
# determine which packs need changing
945
pack_distribution = [1]
946
pack_operations = [[0, []]]
947
for pack in self.all_packs():
948
revision_count = pack.get_revision_count()
949
pack_operations[-1][0] += revision_count
950
pack_operations[-1][1].append(pack)
951
self._execute_pack_operations(pack_operations)
953
def plan_autopack_combinations(self, existing_packs, pack_distribution):
954
"""Plan a pack operation.
956
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
958
:param pack_distribution: A list with the number of revisions desired
961
if len(existing_packs) <= len(pack_distribution):
963
existing_packs.sort(reverse=True)
964
pack_operations = [[0, []]]
965
# plan out what packs to keep, and what to reorganise
966
while len(existing_packs):
967
# take the largest pack, and if its less than the head of the
968
# distribution chart we will include its contents in the new pack for
969
# that position. If its larger, we remove its size from the
971
next_pack_rev_count, next_pack = existing_packs.pop(0)
972
if next_pack_rev_count >= pack_distribution[0]:
973
# this is already packed 'better' than this, so we can
974
# not waste time packing it.
975
while next_pack_rev_count > 0:
976
next_pack_rev_count -= pack_distribution[0]
977
if next_pack_rev_count >= 0:
979
del pack_distribution[0]
981
# didn't use that entire bucket up
982
pack_distribution[0] = -next_pack_rev_count
984
# add the revisions we're going to add to the next output pack
985
pack_operations[-1][0] += next_pack_rev_count
986
# allocate this pack to the next pack sub operation
987
pack_operations[-1][1].append(next_pack)
988
if pack_operations[-1][0] >= pack_distribution[0]:
989
# this pack is used up, shift left.
990
del pack_distribution[0]
991
pack_operations.append([0, []])
993
return pack_operations
995
def ensure_loaded(self):
996
# NB: if you see an assertion error here, its probably access against
997
# an unlocked repo. Naughty.
998
assert self.repo.is_locked()
999
if self._names is None:
1001
self._packs_at_load = set()
1002
for index, key, value in self._iter_disk_pack_index():
1004
self._names[name] = self._parse_index_sizes(value)
1005
self._packs_at_load.add((key, value))
1006
# populate all the metadata.
1009
def _parse_index_sizes(self, value):
1010
"""Parse a string of index sizes."""
1011
return tuple([int(digits) for digits in value.split(' ')])
1013
def get_pack_by_name(self, name):
1014
"""Get a Pack object by name.
1016
:param name: The name of the pack - e.g. '123456'
1017
:return: A Pack object.
1020
return self._packs_by_name[name]
1022
rev_index = self._make_index(name, '.rix')
1023
inv_index = self._make_index(name, '.iix')
1024
txt_index = self._make_index(name, '.tix')
1025
sig_index = self._make_index(name, '.six')
1026
result = ExistingPack(self._pack_transport, name, rev_index,
1027
inv_index, txt_index, sig_index)
1028
self.add_pack_to_memory(result)
1031
def allocate(self, a_new_pack):
1032
"""Allocate name in the list of packs.
1034
:param a_new_pack: A NewPack instance to be added to the collection of
1035
packs for this repository.
1037
self.ensure_loaded()
1038
if a_new_pack.name in self._names:
1039
# a collision with the packs we know about (not the only possible
1040
# collision, see NewPack.finish() for some discussion). Remove our
1041
# prior reference to it.
1042
self._remove_pack_from_memory(a_new_pack)
1043
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1044
self.add_pack_to_memory(a_new_pack)
1046
def _iter_disk_pack_index(self):
1047
"""Iterate over the contents of the pack-names index.
1049
This is used when loading the list from disk, and before writing to
1050
detect updates from others during our write operation.
1051
:return: An iterator of the index contents.
1053
return GraphIndex(self.transport, 'pack-names', None
1054
).iter_all_entries()
1056
def _make_index(self, name, suffix):
1057
size_offset = self._suffix_offsets[suffix]
1058
index_name = name + suffix
1059
index_size = self._names[name][size_offset]
1061
self._index_transport, index_name, index_size)
1063
def _max_pack_count(self, total_revisions):
1064
"""Return the maximum number of packs to use for total revisions.
1066
:param total_revisions: The total number of revisions in the
1069
if not total_revisions:
1071
digits = str(total_revisions)
1073
for digit in digits:
1074
result += int(digit)
1078
"""Provide an order to the underlying names."""
1079
return sorted(self._names.keys())
1081
def _obsolete_packs(self, packs):
1082
"""Move a number of packs which have been obsoleted out of the way.
1084
Each pack and its associated indices are moved out of the way.
1086
Note: for correctness this function should only be called after a new
1087
pack names index has been written without these pack names, and with
1088
the names of packs that contain the data previously available via these
1091
:param packs: The packs to obsolete.
1092
:param return: None.
1095
pack.pack_transport.rename(pack.file_name(),
1096
'../obsolete_packs/' + pack.file_name())
1097
# TODO: Probably needs to know all possible indices for this pack
1098
# - or maybe list the directory and move all indices matching this
1099
# name whether we recognize it or not?
1100
for suffix in ('.iix', '.six', '.tix', '.rix'):
1101
self._index_transport.rename(pack.name + suffix,
1102
'../obsolete_packs/' + pack.name + suffix)
1104
def pack_distribution(self, total_revisions):
1105
"""Generate a list of the number of revisions to put in each pack.
1107
:param total_revisions: The total number of revisions in the
1110
if total_revisions == 0:
1112
digits = reversed(str(total_revisions))
1114
for exponent, count in enumerate(digits):
1115
size = 10 ** exponent
1116
for pos in range(int(count)):
1118
return list(reversed(result))
1120
def _pack_tuple(self, name):
1121
"""Return a tuple with the transport and file name for a pack name."""
1122
return self._pack_transport, name + '.pack'
1124
def _remove_pack_from_memory(self, pack):
1125
"""Remove pack from the packs accessed by this repository.
1127
Only affects memory state, until self._save_pack_names() is invoked.
1129
self._names.pop(pack.name)
1130
self._packs_by_name.pop(pack.name)
1131
self._remove_pack_indices(pack)
1133
def _remove_pack_indices(self, pack):
1134
"""Remove the indices for pack from the aggregated indices."""
1135
self.revision_index.remove_index(pack.revision_index, pack)
1136
self.inventory_index.remove_index(pack.inventory_index, pack)
1137
self.text_index.remove_index(pack.text_index, pack)
1138
self.signature_index.remove_index(pack.signature_index, pack)
1141
"""Clear all cached data."""
1142
# cached revision data
1143
self.repo._revision_knit = None
1144
self.revision_index.clear()
1145
# cached signature data
1146
self.repo._signature_knit = None
1147
self.signature_index.clear()
1148
# cached file text data
1149
self.text_index.clear()
1150
self.repo._text_knit = None
1151
# cached inventory data
1152
self.inventory_index.clear()
1153
# remove the open pack
1154
self._new_pack = None
1155
# information about packs.
1158
self._packs_by_name = {}
1159
self._packs_at_load = None
1161
def _make_index_map(self, index_suffix):
1162
"""Return information on existing indices.
1164
:param suffix: Index suffix added to pack name.
1166
:returns: (pack_map, indices) where indices is a list of GraphIndex
1167
objects, and pack_map is a mapping from those objects to the
1168
pack tuple they describe.
1170
# TODO: stop using this; it creates new indices unnecessarily.
1171
self.ensure_loaded()
1172
suffix_map = {'.rix': 'revision_index',
1173
'.six': 'signature_index',
1174
'.iix': 'inventory_index',
1175
'.tix': 'text_index',
1177
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1178
suffix_map[index_suffix])
1180
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1181
"""Convert a list of packs to an index pack map and index list.
1183
:param packs: The packs list to process.
1184
:param index_attribute: The attribute that the desired index is found
1186
:return: A tuple (map, list) where map contains the dict from
1187
index:pack_tuple, and lsit contains the indices in the same order
1193
index = getattr(pack, index_attribute)
1194
indices.append(index)
1195
pack_map[index] = (pack.pack_transport, pack.file_name())
1196
return pack_map, indices
1198
def _index_contents(self, pack_map, key_filter=None):
1199
"""Get an iterable of the index contents from a pack_map.
1201
:param pack_map: A map from indices to pack details.
1202
:param key_filter: An optional filter to limit the
1205
indices = [index for index in pack_map.iterkeys()]
1206
all_index = CombinedGraphIndex(indices)
1207
if key_filter is None:
1208
return all_index.iter_all_entries()
1210
return all_index.iter_entries(key_filter)
1212
def _unlock_names(self):
1213
"""Release the mutex around the pack-names index."""
1214
self.repo.control_files.unlock()
1216
def _save_pack_names(self, clear_obsolete_packs=False):
1217
"""Save the list of packs.
1219
This will take out the mutex around the pack names list for the
1220
duration of the method call. If concurrent updates have been made, a
1221
three-way merge between the current list and the current in memory list
1224
:param clear_obsolete_packs: If True, clear out the contents of the
1225
obsolete_packs directory.
1229
builder = GraphIndexBuilder()
1230
# load the disk nodes across
1232
for index, key, value in self._iter_disk_pack_index():
1233
disk_nodes.add((key, value))
1234
# do a two-way diff against our original content
1235
current_nodes = set()
1236
for name, sizes in self._names.iteritems():
1238
((name, ), ' '.join(str(size) for size in sizes)))
1239
deleted_nodes = self._packs_at_load - current_nodes
1240
new_nodes = current_nodes - self._packs_at_load
1241
disk_nodes.difference_update(deleted_nodes)
1242
disk_nodes.update(new_nodes)
1243
# TODO: handle same-name, index-size-changes here -
1244
# e.g. use the value from disk, not ours, *unless* we're the one
1246
for key, value in disk_nodes:
1247
builder.add_node(key, value)
1248
self.transport.put_file('pack-names', builder.finish())
1249
# move the baseline forward
1250
self._packs_at_load = disk_nodes
1251
# now clear out the obsolete packs directory
1252
if clear_obsolete_packs:
1253
self.transport.clone('obsolete_packs').delete_multi(
1254
self.transport.list_dir('obsolete_packs'))
1256
self._unlock_names()
1257
# synchronise the memory packs list with what we just wrote:
1258
new_names = dict(disk_nodes)
1259
# drop no longer present nodes
1260
for pack in self.all_packs():
1261
if (pack.name,) not in new_names:
1262
self._remove_pack_from_memory(pack)
1263
# add new nodes/refresh existing ones
1264
for key, value in disk_nodes:
1266
sizes = self._parse_index_sizes(value)
1267
if name in self._names:
1269
if sizes != self._names[name]:
1270
# the pack for name has had its indices replaced - rare but
1271
# important to handle. XXX: probably can never happen today
1272
# because the three-way merge code above does not handle it
1273
# - you may end up adding the same key twice to the new
1274
# disk index because the set values are the same, unless
1275
# the only index shows up as deleted by the set difference
1276
# - which it may. Until there is a specific test for this,
1277
# assume its broken. RBC 20071017.
1278
self._remove_pack_from_memory(self.get_pack_by_name(name))
1279
self._names[name] = sizes
1280
self.get_pack_by_name(name)
1283
self._names[name] = sizes
1284
self.get_pack_by_name(name)
1286
def _start_write_group(self):
1287
# Do not permit preparation for writing if we're not in a 'write lock'.
1288
if not self.repo.is_write_locked():
1289
raise errors.NotWriteLocked(self)
1290
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1291
self._pack_transport, upload_suffix='.pack')
1292
# allow writing: queue writes to a new index
1293
self.revision_index.add_writable_index(self._new_pack.revision_index,
1295
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
1297
self.text_index.add_writable_index(self._new_pack.text_index,
1299
self.signature_index.add_writable_index(self._new_pack.signature_index,
1302
# reused revision and signature knits may need updating
1304
# "Hysterical raisins. client code in bzrlib grabs those knits outside
1305
# of write groups and then mutates it inside the write group."
1306
if self.repo._revision_knit is not None:
1307
self.repo._revision_knit._index._add_callback = \
1308
self.revision_index.add_callback
1309
if self.repo._signature_knit is not None:
1310
self.repo._signature_knit._index._add_callback = \
1311
self.signature_index.add_callback
1312
# create a reused knit object for text addition in commit.
1313
self.repo._text_knit = self.repo.weave_store.get_weave_or_empty(
1316
def _abort_write_group(self):
1317
# FIXME: just drop the transient index.
1318
# forget what names there are
1319
self._new_pack.abort()
1320
self._remove_pack_indices(self._new_pack)
1321
self._new_pack = None
1322
self.repo._text_knit = None
1324
def _commit_write_group(self):
1325
self._remove_pack_indices(self._new_pack)
1326
if self._new_pack.data_inserted():
1327
# get all the data to disk and read to use
1328
self._new_pack.finish()
1329
self.allocate(self._new_pack)
1330
self._new_pack = None
1331
if not self.autopack():
1332
# when autopack takes no steps, the names list is still
1334
self._save_pack_names()
1336
self._new_pack.abort()
1337
self._new_pack = None
1338
self.repo._text_knit = None
1341
class KnitPackRevisionStore(KnitRevisionStore):
1342
"""An object to adapt access from RevisionStore's to use KnitPacks.
1344
This class works by replacing the original RevisionStore.
1345
We need to do this because the KnitPackRevisionStore is less
1346
isolated in its layering - it uses services from the repo.
1349
def __init__(self, repo, transport, revisionstore):
1350
"""Create a KnitPackRevisionStore on repo with revisionstore.
1352
This will store its state in the Repository, use the
1353
indices to provide a KnitGraphIndex,
1354
and at the end of transactions write new indices.
1356
KnitRevisionStore.__init__(self, revisionstore.versioned_file_store)
1358
self._serializer = revisionstore._serializer
1359
self.transport = transport
1361
def get_revision_file(self, transaction):
1362
"""Get the revision versioned file object."""
1363
if getattr(self.repo, '_revision_knit', None) is not None:
1364
return self.repo._revision_knit
1365
self.repo._pack_collection.ensure_loaded()
1366
add_callback = self.repo._pack_collection.revision_index.add_callback
1367
# setup knit specific objects
1368
knit_index = KnitGraphIndex(
1369
self.repo._pack_collection.revision_index.combined_index,
1370
add_callback=add_callback)
1371
self.repo._revision_knit = knit.KnitVersionedFile(
1372
'revisions', self.transport.clone('..'),
1373
self.repo.control_files._file_mode,
1374
create=False, access_mode=self.repo._access_mode(),
1375
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1376
access_method=self.repo._pack_collection.revision_index.knit_access)
1377
return self.repo._revision_knit
1379
def get_signature_file(self, transaction):
1380
"""Get the signature versioned file object."""
1381
if getattr(self.repo, '_signature_knit', None) is not None:
1382
return self.repo._signature_knit
1383
self.repo._pack_collection.ensure_loaded()
1384
add_callback = self.repo._pack_collection.signature_index.add_callback
1385
# setup knit specific objects
1386
knit_index = KnitGraphIndex(
1387
self.repo._pack_collection.signature_index.combined_index,
1388
add_callback=add_callback, parents=False)
1389
self.repo._signature_knit = knit.KnitVersionedFile(
1390
'signatures', self.transport.clone('..'),
1391
self.repo.control_files._file_mode,
1392
create=False, access_mode=self.repo._access_mode(),
1393
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1394
access_method=self.repo._pack_collection.signature_index.knit_access)
1395
return self.repo._signature_knit
1398
class KnitPackTextStore(VersionedFileStore):
1399
"""Presents a TextStore abstraction on top of packs.
1401
This class works by replacing the original VersionedFileStore.
1402
We need to do this because the KnitPackRevisionStore is less
1403
isolated in its layering - it uses services from the repo and shares them
1404
with all the data written in a single write group.
1407
def __init__(self, repo, transport, weavestore):
1408
"""Create a KnitPackTextStore on repo with weavestore.
1410
This will store its state in the Repository, use the
1411
indices FileNames to provide a KnitGraphIndex,
1412
and at the end of transactions write new indices.
1414
# don't call base class constructor - it's not suitable.
1415
# no transient data stored in the transaction
1417
self._precious = False
1419
self.transport = transport
1420
self.weavestore = weavestore
1421
# XXX for check() which isn't updated yet
1422
self._transport = weavestore._transport
1424
def get_weave_or_empty(self, file_id, transaction):
1425
"""Get a 'Knit' backed by the .tix indices.
1427
The transaction parameter is ignored.
1429
self.repo._pack_collection.ensure_loaded()
1430
add_callback = self.repo._pack_collection.text_index.add_callback
1431
# setup knit specific objects
1432
file_id_index = GraphIndexPrefixAdapter(
1433
self.repo._pack_collection.text_index.combined_index,
1434
(file_id, ), 1, add_nodes_callback=add_callback)
1435
knit_index = KnitGraphIndex(file_id_index,
1436
add_callback=file_id_index.add_nodes,
1437
deltas=True, parents=True)
1438
return knit.KnitVersionedFile('text:' + file_id,
1439
self.transport.clone('..'),
1442
access_method=self.repo._pack_collection.text_index.knit_access,
1443
factory=knit.KnitPlainFactory())
1445
get_weave = get_weave_or_empty
1448
"""Generate a list of the fileids inserted, for use by check."""
1449
self.repo._pack_collection.ensure_loaded()
1451
for index, key, value, refs in \
1452
self.repo._pack_collection.text_index.combined_index.iter_all_entries():
1457
class InventoryKnitThunk(object):
1458
"""An object to manage thunking get_inventory_weave to pack based knits."""
1460
def __init__(self, repo, transport):
1461
"""Create an InventoryKnitThunk for repo at transport.
1463
This will store its state in the Repository, use the
1464
indices FileNames to provide a KnitGraphIndex,
1465
and at the end of transactions write a new index..
1468
self.transport = transport
1470
def get_weave(self):
1471
"""Get a 'Knit' that contains inventory data."""
1472
self.repo._pack_collection.ensure_loaded()
1473
add_callback = self.repo._pack_collection.inventory_index.add_callback
1474
# setup knit specific objects
1475
knit_index = KnitGraphIndex(
1476
self.repo._pack_collection.inventory_index.combined_index,
1477
add_callback=add_callback, deltas=True, parents=True)
1478
return knit.KnitVersionedFile(
1479
'inventory', self.transport.clone('..'),
1480
self.repo.control_files._file_mode,
1481
create=False, access_mode=self.repo._access_mode(),
1482
index=knit_index, delta=True, factory=knit.KnitPlainFactory(),
1483
access_method=self.repo._pack_collection.inventory_index.knit_access)
1486
class KnitPackRepository(KnitRepository):
1487
"""Experimental graph-knit using repository."""
1489
def __init__(self, _format, a_bzrdir, control_files, _revision_store,
1490
control_store, text_store, _commit_builder_class, _serializer):
1491
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
1492
_revision_store, control_store, text_store, _commit_builder_class,
1494
index_transport = control_files._transport.clone('indices')
1495
self._pack_collection = RepositoryPackCollection(self, control_files._transport,
1497
control_files._transport.clone('upload'),
1498
control_files._transport.clone('packs'))
1499
self._revision_store = KnitPackRevisionStore(self, index_transport, self._revision_store)
1500
self.weave_store = KnitPackTextStore(self, index_transport, self.weave_store)
1501
self._inv_thunk = InventoryKnitThunk(self, index_transport)
1502
# True when the repository object is 'write locked' (as opposed to the
1503
# physical lock only taken out around changes to the pack-names list.)
1504
# Another way to represent this would be a decorator around the control
1505
# files object that presents logical locks as physical ones - if this
1506
# gets ugly consider that alternative design. RBC 20071011
1507
self._write_lock_count = 0
1508
self._transaction = None
1510
self._reconcile_does_inventory_gc = True
1511
self._reconcile_fixes_text_parents = False
1512
self._reconcile_backsup_inventory = False
1514
def _abort_write_group(self):
1515
self._pack_collection._abort_write_group()
1517
def _access_mode(self):
1518
"""Return 'w' or 'r' for depending on whether a write lock is active.
1520
This method is a helper for the Knit-thunking support objects.
1522
if self.is_write_locked():
1526
def _find_inconsistent_revision_parents(self):
1527
"""Find revisions with incorrectly cached parents.
1529
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1530
parents-in-revision).
1532
assert self.is_locked()
1533
pb = ui.ui_factory.nested_progress_bar()
1536
revision_nodes = self._pack_collection.revision_index \
1537
.combined_index.iter_all_entries()
1538
index_positions = []
1539
# Get the cached index values for all revisions, and also the location
1540
# in each index of the revision text so we can perform linear IO.
1541
for index, key, value, refs in revision_nodes:
1542
pos, length = value[1:].split(' ')
1543
index_positions.append((index, int(pos), key[0],
1544
tuple(parent[0] for parent in refs[0])))
1545
pb.update("Reading revision index.", 0, 0)
1546
index_positions.sort()
1547
batch_count = len(index_positions) / 1000 + 1
1548
pb.update("Checking cached revision graph.", 0, batch_count)
1549
for offset in xrange(batch_count):
1550
pb.update("Checking cached revision graph.", offset)
1551
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1554
rev_ids = [item[2] for item in to_query]
1555
revs = self.get_revisions(rev_ids)
1556
for revision, item in zip(revs, to_query):
1557
index_parents = item[3]
1558
rev_parents = tuple(revision.parent_ids)
1559
if index_parents != rev_parents:
1560
result.append((revision.revision_id, index_parents, rev_parents))
1565
def get_parents(self, revision_ids):
1566
"""See StackedParentsProvider.get_parents.
1568
This implementation accesses the combined revision index to provide
1571
self._pack_collection.ensure_loaded()
1572
index = self._pack_collection.revision_index.combined_index
1574
for revision_id in revision_ids:
1575
if revision_id != _mod_revision.NULL_REVISION:
1576
search_keys.add((revision_id,))
1577
found_parents = {_mod_revision.NULL_REVISION:[]}
1578
for index, key, value, refs in index.iter_entries(search_keys):
1581
parents = (_mod_revision.NULL_REVISION,)
1583
parents = tuple(parent[0] for parent in parents)
1584
found_parents[key[0]] = parents
1586
for revision_id in revision_ids:
1588
result.append(found_parents[revision_id])
1593
def _make_parents_provider(self):
1596
def _refresh_data(self):
1597
if self._write_lock_count == 1 or (
1598
self.control_files._lock_count == 1 and
1599
self.control_files._lock_mode == 'r'):
1600
# forget what names there are
1601
self._pack_collection.reset()
1602
# XXX: Better to do an in-memory merge when acquiring a new lock -
1603
# factor out code from _save_pack_names.
1604
self._pack_collection.ensure_loaded()
1606
def _start_write_group(self):
1607
self._pack_collection._start_write_group()
1609
def _commit_write_group(self):
1610
return self._pack_collection._commit_write_group()
1612
def get_inventory_weave(self):
1613
return self._inv_thunk.get_weave()
1615
def get_transaction(self):
1616
if self._write_lock_count:
1617
return self._transaction
1619
return self.control_files.get_transaction()
1621
def is_locked(self):
1622
return self._write_lock_count or self.control_files.is_locked()
1624
def is_write_locked(self):
1625
return self._write_lock_count
1627
def lock_write(self, token=None):
1628
if not self._write_lock_count and self.is_locked():
1629
raise errors.ReadOnlyError(self)
1630
self._write_lock_count += 1
1631
if self._write_lock_count == 1:
1632
from bzrlib import transactions
1633
self._transaction = transactions.WriteTransaction()
1634
self._refresh_data()
1636
def lock_read(self):
1637
if self._write_lock_count:
1638
self._write_lock_count += 1
1640
self.control_files.lock_read()
1641
self._refresh_data()
1643
def leave_lock_in_place(self):
1644
# not supported - raise an error
1645
raise NotImplementedError(self.leave_lock_in_place)
1647
def dont_leave_lock_in_place(self):
1648
# not supported - raise an error
1649
raise NotImplementedError(self.dont_leave_lock_in_place)
1653
"""Compress the data within the repository.
1655
This will pack all the data to a single pack. In future it may
1656
recompress deltas or do other such expensive operations.
1658
self._pack_collection.pack()
1661
def reconcile(self, other=None, thorough=False):
1662
"""Reconcile this repository."""
1663
from bzrlib.reconcile import PackReconciler
1664
reconciler = PackReconciler(self, thorough=thorough)
1665
reconciler.reconcile()
1669
if self._write_lock_count == 1 and self._write_group is not None:
1670
self.abort_write_group()
1671
self._transaction = None
1672
self._write_lock_count = 0
1673
raise errors.BzrError(
1674
'Must end write group before releasing write lock on %s'
1676
if self._write_lock_count:
1677
self._write_lock_count -= 1
1678
if not self._write_lock_count:
1679
transaction = self._transaction
1680
self._transaction = None
1681
transaction.finish()
1683
self.control_files.unlock()
1686
class RepositoryFormatPack(MetaDirRepositoryFormat):
1687
"""Format logic for pack structured repositories.
1689
This repository format has:
1690
- a list of packs in pack-names
1691
- packs in packs/NAME.pack
1692
- indices in indices/NAME.{iix,six,tix,rix}
1693
- knit deltas in the packs, knit indices mapped to the indices.
1694
- thunk objects to support the knits programming API.
1695
- a format marker of its own
1696
- an optional 'shared-storage' flag
1697
- an optional 'no-working-trees' flag
1701
# Set this attribute in derived classes to control the repository class
1702
# created by open and initialize.
1703
repository_class = None
1704
# Set this attribute in derived classes to control the
1705
# _commit_builder_class that the repository objects will have passed to
1706
# their constructor.
1707
_commit_builder_class = None
1708
# Set this attribute in derived clases to control the _serializer that the
1709
# repository objects will have passed to their constructor.
1712
def _get_control_store(self, repo_transport, control_files):
1713
"""Return the control store for this repository."""
1714
return VersionedFileStore(
1717
file_mode=control_files._file_mode,
1718
versionedfile_class=knit.KnitVersionedFile,
1719
versionedfile_kwargs={'factory': knit.KnitPlainFactory()},
1722
def _get_revision_store(self, repo_transport, control_files):
1723
"""See RepositoryFormat._get_revision_store()."""
1724
versioned_file_store = VersionedFileStore(
1726
file_mode=control_files._file_mode,
1729
versionedfile_class=knit.KnitVersionedFile,
1730
versionedfile_kwargs={'delta': False,
1731
'factory': knit.KnitPlainFactory(),
1735
return KnitRevisionStore(versioned_file_store)
1737
def _get_text_store(self, transport, control_files):
1738
"""See RepositoryFormat._get_text_store()."""
1739
return self._get_versioned_file_store('knits',
1742
versionedfile_class=knit.KnitVersionedFile,
1743
versionedfile_kwargs={
1744
'create_parent_dir': True,
1745
'delay_create': True,
1746
'dir_mode': control_files._dir_mode,
1750
def initialize(self, a_bzrdir, shared=False):
1751
"""Create a pack based repository.
1753
:param a_bzrdir: bzrdir to contain the new repository; must already
1755
:param shared: If true the repository will be initialized as a shared
1758
mutter('creating repository in %s.', a_bzrdir.transport.base)
1759
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
1760
builder = GraphIndexBuilder()
1761
files = [('pack-names', builder.finish())]
1762
utf8_files = [('format', self.get_format_string())]
1764
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
1765
return self.open(a_bzrdir=a_bzrdir, _found=True)
1767
def open(self, a_bzrdir, _found=False, _override_transport=None):
1768
"""See RepositoryFormat.open().
1770
:param _override_transport: INTERNAL USE ONLY. Allows opening the
1771
repository at a slightly different url
1772
than normal. I.e. during 'upgrade'.
1775
format = RepositoryFormat.find_format(a_bzrdir)
1776
assert format.__class__ == self.__class__
1777
if _override_transport is not None:
1778
repo_transport = _override_transport
1780
repo_transport = a_bzrdir.get_repository_transport(None)
1781
control_files = lockable_files.LockableFiles(repo_transport,
1782
'lock', lockdir.LockDir)
1783
text_store = self._get_text_store(repo_transport, control_files)
1784
control_store = self._get_control_store(repo_transport, control_files)
1785
_revision_store = self._get_revision_store(repo_transport, control_files)
1786
return self.repository_class(_format=self,
1788
control_files=control_files,
1789
_revision_store=_revision_store,
1790
control_store=control_store,
1791
text_store=text_store,
1792
_commit_builder_class=self._commit_builder_class,
1793
_serializer=self._serializer)
1796
class RepositoryFormatKnitPack1(RepositoryFormatPack):
1797
"""A no-subtrees parameterised Pack repository.
1799
This format was introduced in 0.92.
1802
repository_class = KnitPackRepository
1803
_commit_builder_class = PackCommitBuilder
1804
_serializer = xml5.serializer_v5
1806
def _get_matching_bzrdir(self):
1807
return bzrdir.format_registry.make_bzrdir('knitpack-experimental')
1809
def _ignore_setting_bzrdir(self, format):
1812
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1814
def get_format_string(self):
1815
"""See RepositoryFormat.get_format_string()."""
1816
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
1818
def get_format_description(self):
1819
"""See RepositoryFormat.get_format_description()."""
1820
return "Packs containing knits without subtree support"
1822
def check_conversion_target(self, target_format):
1826
class RepositoryFormatKnitPack3(RepositoryFormatPack):
1827
"""A subtrees parameterised Pack repository.
1829
This repository format uses the xml7 serializer to get:
1830
- support for recording full info about the tree root
1831
- support for recording tree-references
1833
This format was introduced in 0.92.
1836
repository_class = KnitPackRepository
1837
_commit_builder_class = PackRootCommitBuilder
1838
rich_root_data = True
1839
supports_tree_reference = True
1840
_serializer = xml7.serializer_v7
1842
def _get_matching_bzrdir(self):
1843
return bzrdir.format_registry.make_bzrdir(
1844
'knitpack-subtree-experimental')
1846
def _ignore_setting_bzrdir(self, format):
1849
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1851
def check_conversion_target(self, target_format):
1852
if not target_format.rich_root_data:
1853
raise errors.BadConversionTarget(
1854
'Does not support rich root data.', target_format)
1855
if not getattr(target_format, 'supports_tree_reference', False):
1856
raise errors.BadConversionTarget(
1857
'Does not support nested trees', target_format)
1859
def get_format_string(self):
1860
"""See RepositoryFormat.get_format_string()."""
1861
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
1863
def get_format_description(self):
1864
"""See RepositoryFormat.get_format_description()."""
1865
return "Packs containing knits with subtree support\n"
added
removed
bzrlib/repofmt/pack_repo.py
