1
# Copyright (C) 2005, 2006, 2007, 2008 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
30
from bzrlib.index import (
35
GraphIndexPrefixAdapter,
37
from bzrlib.knit import (
43
from bzrlib.osutils import rand_chars, split_lines
44
from bzrlib.pack import ContainerWriter
45
from bzrlib.store import revision
46
from bzrlib import tsort
61
from bzrlib.decorators import needs_read_lock, needs_write_lock
62
from bzrlib.repofmt.knitrepo import KnitRepository
63
from bzrlib.repository import (
66
MetaDirRepositoryFormat,
70
import bzrlib.revision as _mod_revision
71
from bzrlib.store.versioned import VersionedFileStore
72
from bzrlib.trace import (
80
class PackCommitBuilder(CommitBuilder):
81
"""A subclass of CommitBuilder 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 __init__(self, repository, parents, config, timestamp=None,
88
timezone=None, committer=None, revprops=None,
90
CommitBuilder.__init__(self, repository, parents, config,
91
timestamp=timestamp, timezone=timezone, committer=committer,
92
revprops=revprops, revision_id=revision_id)
93
self._file_graph = graph.Graph(
94
repository._pack_collection.text_index.combined_index)
96
def _heads(self, file_id, revision_ids):
97
keys = [(file_id, revision_id) for revision_id in revision_ids]
98
return set([key[1] for key in self._file_graph.heads(keys)])
101
class PackRootCommitBuilder(RootCommitBuilder):
102
"""A subclass of RootCommitBuilder to add texts with pack semantics.
104
Specifically this uses one knit object rather than one knit object per
105
added text, reducing memory and object pressure.
108
def __init__(self, repository, parents, config, timestamp=None,
109
timezone=None, committer=None, revprops=None,
111
CommitBuilder.__init__(self, repository, parents, config,
112
timestamp=timestamp, timezone=timezone, committer=committer,
113
revprops=revprops, revision_id=revision_id)
114
self._file_graph = graph.Graph(
115
repository._pack_collection.text_index.combined_index)
117
def _heads(self, file_id, revision_ids):
118
keys = [(file_id, revision_id) for revision_id in revision_ids]
119
return set([key[1] for key in self._file_graph.heads(keys)])
123
"""An in memory proxy for a pack and its indices.
125
This is a base class that is not directly used, instead the classes
126
ExistingPack and NewPack are used.
129
def __init__(self, revision_index, inventory_index, text_index,
131
"""Create a pack instance.
133
:param revision_index: A GraphIndex for determining what revisions are
134
present in the Pack and accessing the locations of their texts.
135
:param inventory_index: A GraphIndex for determining what inventories are
136
present in the Pack and accessing the locations of their
138
:param text_index: A GraphIndex for determining what file texts
139
are present in the pack and accessing the locations of their
140
texts/deltas (via (fileid, revisionid) tuples).
141
:param revision_index: A GraphIndex for determining what signatures are
142
present in the Pack and accessing the locations of their texts.
144
self.revision_index = revision_index
145
self.inventory_index = inventory_index
146
self.text_index = text_index
147
self.signature_index = signature_index
149
def access_tuple(self):
150
"""Return a tuple (transport, name) for the pack content."""
151
return self.pack_transport, self.file_name()
154
"""Get the file name for the pack on disk."""
155
return self.name + '.pack'
157
def get_revision_count(self):
158
return self.revision_index.key_count()
160
def inventory_index_name(self, name):
161
"""The inv index is the name + .iix."""
162
return self.index_name('inventory', name)
164
def revision_index_name(self, name):
165
"""The revision index is the name + .rix."""
166
return self.index_name('revision', name)
168
def signature_index_name(self, name):
169
"""The signature index is the name + .six."""
170
return self.index_name('signature', name)
172
def text_index_name(self, name):
173
"""The text index is the name + .tix."""
174
return self.index_name('text', name)
176
def _external_compression_parents_of_texts(self):
179
for node in self.text_index.iter_all_entries():
181
refs.update(node[3][1])
185
class ExistingPack(Pack):
186
"""An in memory proxy for an existing .pack and its disk indices."""
188
def __init__(self, pack_transport, name, revision_index, inventory_index,
189
text_index, signature_index):
190
"""Create an ExistingPack object.
192
:param pack_transport: The transport where the pack file resides.
193
:param name: The name of the pack on disk in the pack_transport.
195
Pack.__init__(self, revision_index, inventory_index, text_index,
198
self.pack_transport = pack_transport
199
if None in (revision_index, inventory_index, text_index,
200
signature_index, name, pack_transport):
201
raise AssertionError()
203
def __eq__(self, other):
204
return self.__dict__ == other.__dict__
206
def __ne__(self, other):
207
return not self.__eq__(other)
210
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
211
id(self), self.transport, self.name)
215
"""An in memory proxy for a pack which is being created."""
217
# A map of index 'type' to the file extension and position in the
219
index_definitions = {
220
'revision': ('.rix', 0),
221
'inventory': ('.iix', 1),
223
'signature': ('.six', 3),
226
def __init__(self, upload_transport, index_transport, pack_transport,
227
upload_suffix='', file_mode=None):
228
"""Create a NewPack instance.
230
:param upload_transport: A writable transport for the pack to be
231
incrementally uploaded to.
232
:param index_transport: A writable transport for the pack's indices to
233
be written to when the pack is finished.
234
:param pack_transport: A writable transport for the pack to be renamed
235
to when the upload is complete. This *must* be the same as
236
upload_transport.clone('../packs').
237
:param upload_suffix: An optional suffix to be given to any temporary
238
files created during the pack creation. e.g '.autopack'
239
:param file_mode: An optional file mode to create the new files with.
241
# The relative locations of the packs are constrained, but all are
242
# passed in because the caller has them, so as to avoid object churn.
244
# Revisions: parents list, no text compression.
245
InMemoryGraphIndex(reference_lists=1),
246
# Inventory: We want to map compression only, but currently the
247
# knit code hasn't been updated enough to understand that, so we
248
# have a regular 2-list index giving parents and compression
250
InMemoryGraphIndex(reference_lists=2),
251
# Texts: compression and per file graph, for all fileids - so two
252
# reference lists and two elements in the key tuple.
253
InMemoryGraphIndex(reference_lists=2, key_elements=2),
254
# Signatures: Just blobs to store, no compression, no parents
256
InMemoryGraphIndex(reference_lists=0),
258
# where should the new pack be opened
259
self.upload_transport = upload_transport
260
# where are indices written out to
261
self.index_transport = index_transport
262
# where is the pack renamed to when it is finished?
263
self.pack_transport = pack_transport
264
# What file mode to upload the pack and indices with.
265
self._file_mode = file_mode
266
# tracks the content written to the .pack file.
267
self._hash = md5.new()
268
# a four-tuple with the length in bytes of the indices, once the pack
269
# is finalised. (rev, inv, text, sigs)
270
self.index_sizes = None
271
# How much data to cache when writing packs. Note that this is not
272
# synchronised with reads, because it's not in the transport layer, so
273
# is not safe unless the client knows it won't be reading from the pack
275
self._cache_limit = 0
276
# the temporary pack file name.
277
self.random_name = rand_chars(20) + upload_suffix
278
# when was this pack started ?
279
self.start_time = time.time()
280
# open an output stream for the data added to the pack.
281
self.write_stream = self.upload_transport.open_write_stream(
282
self.random_name, mode=self._file_mode)
283
if 'pack' in debug.debug_flags:
284
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
285
time.ctime(), self.upload_transport.base, self.random_name,
286
time.time() - self.start_time)
287
# A list of byte sequences to be written to the new pack, and the
288
# aggregate size of them. Stored as a list rather than separate
289
# variables so that the _write_data closure below can update them.
290
self._buffer = [[], 0]
291
# create a callable for adding data
293
# robertc says- this is a closure rather than a method on the object
294
# so that the variables are locals, and faster than accessing object
296
def _write_data(bytes, flush=False, _buffer=self._buffer,
297
_write=self.write_stream.write, _update=self._hash.update):
298
_buffer[0].append(bytes)
299
_buffer[1] += len(bytes)
301
if _buffer[1] > self._cache_limit or flush:
302
bytes = ''.join(_buffer[0])
306
# expose this on self, for the occasion when clients want to add data.
307
self._write_data = _write_data
308
# a pack writer object to serialise pack records.
309
self._writer = pack.ContainerWriter(self._write_data)
311
# what state is the pack in? (open, finished, aborted)
315
"""Cancel creating this pack."""
316
self._state = 'aborted'
317
self.write_stream.close()
318
# Remove the temporary pack file.
319
self.upload_transport.delete(self.random_name)
320
# The indices have no state on disk.
322
def access_tuple(self):
323
"""Return a tuple (transport, name) for the pack content."""
324
if self._state == 'finished':
325
return Pack.access_tuple(self)
326
elif self._state == 'open':
327
return self.upload_transport, self.random_name
329
raise AssertionError(self._state)
331
def data_inserted(self):
332
"""True if data has been added to this pack."""
333
return bool(self.get_revision_count() or
334
self.inventory_index.key_count() or
335
self.text_index.key_count() or
336
self.signature_index.key_count())
339
"""Finish the new pack.
342
- finalises the content
343
- assigns a name (the md5 of the content, currently)
344
- writes out the associated indices
345
- renames the pack into place.
346
- stores the index size tuple for the pack in the index_sizes
351
self._write_data('', flush=True)
352
self.name = self._hash.hexdigest()
354
# XXX: It'd be better to write them all to temporary names, then
355
# rename them all into place, so that the window when only some are
356
# visible is smaller. On the other hand none will be seen until
357
# they're in the names list.
358
self.index_sizes = [None, None, None, None]
359
self._write_index('revision', self.revision_index, 'revision')
360
self._write_index('inventory', self.inventory_index, 'inventory')
361
self._write_index('text', self.text_index, 'file texts')
362
self._write_index('signature', self.signature_index,
363
'revision signatures')
364
self.write_stream.close()
365
# Note that this will clobber an existing pack with the same name,
366
# without checking for hash collisions. While this is undesirable this
367
# is something that can be rectified in a subsequent release. One way
368
# to rectify it may be to leave the pack at the original name, writing
369
# its pack-names entry as something like 'HASH: index-sizes
370
# temporary-name'. Allocate that and check for collisions, if it is
371
# collision free then rename it into place. If clients know this scheme
372
# they can handle missing-file errors by:
373
# - try for HASH.pack
374
# - try for temporary-name
375
# - refresh the pack-list to see if the pack is now absent
376
self.upload_transport.rename(self.random_name,
377
'../packs/' + self.name + '.pack')
378
self._state = 'finished'
379
if 'pack' in debug.debug_flags:
380
# XXX: size might be interesting?
381
mutter('%s: create_pack: pack renamed into place: %s%s->%s%s t+%6.3fs',
382
time.ctime(), self.upload_transport.base, self.random_name,
383
self.pack_transport, self.name,
384
time.time() - self.start_time)
387
"""Flush any current data."""
389
bytes = ''.join(self._buffer[0])
390
self.write_stream.write(bytes)
391
self._hash.update(bytes)
392
self._buffer[:] = [[], 0]
394
def index_name(self, index_type, name):
395
"""Get the disk name of an index type for pack name 'name'."""
396
return name + NewPack.index_definitions[index_type][0]
398
def index_offset(self, index_type):
399
"""Get the position in a index_size array for a given index type."""
400
return NewPack.index_definitions[index_type][1]
402
def _replace_index_with_readonly(self, index_type):
403
setattr(self, index_type + '_index',
404
GraphIndex(self.index_transport,
405
self.index_name(index_type, self.name),
406
self.index_sizes[self.index_offset(index_type)]))
408
def set_write_cache_size(self, size):
409
self._cache_limit = size
411
def _write_index(self, index_type, index, label):
412
"""Write out an index.
414
:param index_type: The type of index to write - e.g. 'revision'.
415
:param index: The index object to serialise.
416
:param label: What label to give the index e.g. 'revision'.
418
index_name = self.index_name(index_type, self.name)
419
self.index_sizes[self.index_offset(index_type)] = \
420
self.index_transport.put_file(index_name, index.finish(),
421
mode=self._file_mode)
422
if 'pack' in debug.debug_flags:
423
# XXX: size might be interesting?
424
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
425
time.ctime(), label, self.upload_transport.base,
426
self.random_name, time.time() - self.start_time)
427
# Replace the writable index on this object with a readonly,
428
# presently unloaded index. We should alter
429
# the index layer to make its finish() error if add_node is
430
# subsequently used. RBC
431
self._replace_index_with_readonly(index_type)
434
class AggregateIndex(object):
435
"""An aggregated index for the RepositoryPackCollection.
437
AggregateIndex is reponsible for managing the PackAccess object,
438
Index-To-Pack mapping, and all indices list for a specific type of index
439
such as 'revision index'.
441
A CombinedIndex provides an index on a single key space built up
442
from several on-disk indices. The AggregateIndex builds on this
443
to provide a knit access layer, and allows having up to one writable
444
index within the collection.
446
# XXX: Probably 'can be written to' could/should be separated from 'acts
447
# like a knit index' -- mbp 20071024
450
"""Create an AggregateIndex."""
451
self.index_to_pack = {}
452
self.combined_index = CombinedGraphIndex([])
453
self.data_access = _DirectPackAccess(self.index_to_pack)
454
self.add_callback = None
456
def replace_indices(self, index_to_pack, indices):
457
"""Replace the current mappings with fresh ones.
459
This should probably not be used eventually, rather incremental add and
460
removal of indices. It has been added during refactoring of existing
463
:param index_to_pack: A mapping from index objects to
464
(transport, name) tuples for the pack file data.
465
:param indices: A list of indices.
467
# refresh the revision pack map dict without replacing the instance.
468
self.index_to_pack.clear()
469
self.index_to_pack.update(index_to_pack)
470
# XXX: API break - clearly a 'replace' method would be good?
471
self.combined_index._indices[:] = indices
472
# the current add nodes callback for the current writable index if
474
self.add_callback = None
476
def add_index(self, index, pack):
477
"""Add index to the aggregate, which is an index for Pack pack.
479
Future searches on the aggregate index will seach this new index
480
before all previously inserted indices.
482
:param index: An Index for the pack.
483
:param pack: A Pack instance.
485
# expose it to the index map
486
self.index_to_pack[index] = pack.access_tuple()
487
# put it at the front of the linear index list
488
self.combined_index.insert_index(0, index)
490
def add_writable_index(self, index, pack):
491
"""Add an index which is able to have data added to it.
493
There can be at most one writable index at any time. Any
494
modifications made to the knit are put into this index.
496
:param index: An index from the pack parameter.
497
:param pack: A Pack instance.
499
if self.add_callback is not None:
500
raise AssertionError(
501
"%s already has a writable index through %s" % \
502
(self, self.add_callback))
503
# allow writing: queue writes to a new index
504
self.add_index(index, pack)
505
# Updates the index to packs mapping as a side effect,
506
self.data_access.set_writer(pack._writer, index, pack.access_tuple())
507
self.add_callback = index.add_nodes
510
"""Reset all the aggregate data to nothing."""
511
self.data_access.set_writer(None, None, (None, None))
512
self.index_to_pack.clear()
513
del self.combined_index._indices[:]
514
self.add_callback = None
516
def remove_index(self, index, pack):
517
"""Remove index from the indices used to answer queries.
519
:param index: An index from the pack parameter.
520
:param pack: A Pack instance.
522
del self.index_to_pack[index]
523
self.combined_index._indices.remove(index)
524
if (self.add_callback is not None and
525
getattr(index, 'add_nodes', None) == self.add_callback):
526
self.add_callback = None
527
self.data_access.set_writer(None, None, (None, None))
530
class Packer(object):
531
"""Create a pack from packs."""
533
def __init__(self, pack_collection, packs, suffix, revision_ids=None):
536
:param pack_collection: A RepositoryPackCollection object where the
537
new pack is being written to.
538
:param packs: The packs to combine.
539
:param suffix: The suffix to use on the temporary files for the pack.
540
:param revision_ids: Revision ids to limit the pack to.
544
self.revision_ids = revision_ids
545
# The pack object we are creating.
547
self._pack_collection = pack_collection
548
# The index layer keys for the revisions being copied. None for 'all
550
self._revision_keys = None
551
# What text keys to copy. None for 'all texts'. This is set by
552
# _copy_inventory_texts
553
self._text_filter = None
556
def _extra_init(self):
557
"""A template hook to allow extending the constructor trivially."""
559
def pack(self, pb=None):
560
"""Create a new pack by reading data from other packs.
562
This does little more than a bulk copy of data. One key difference
563
is that data with the same item key across multiple packs is elided
564
from the output. The new pack is written into the current pack store
565
along with its indices, and the name added to the pack names. The
566
source packs are not altered and are not required to be in the current
569
:param pb: An optional progress bar to use. A nested bar is created if
571
:return: A Pack object, or None if nothing was copied.
573
# open a pack - using the same name as the last temporary file
574
# - which has already been flushed, so its safe.
575
# XXX: - duplicate code warning with start_write_group; fix before
576
# considering 'done'.
577
if self._pack_collection._new_pack is not None:
578
raise errors.BzrError('call to create_pack_from_packs while '
579
'another pack is being written.')
580
if self.revision_ids is not None:
581
if len(self.revision_ids) == 0:
582
# silly fetch request.
585
self.revision_ids = frozenset(self.revision_ids)
586
self.revision_keys = frozenset((revid,) for revid in
589
self.pb = ui.ui_factory.nested_progress_bar()
593
return self._create_pack_from_packs()
599
"""Open a pack for the pack we are creating."""
600
return NewPack(self._pack_collection._upload_transport,
601
self._pack_collection._index_transport,
602
self._pack_collection._pack_transport, upload_suffix=self.suffix,
603
file_mode=self._pack_collection.repo.control_files._file_mode)
605
def _copy_revision_texts(self):
606
"""Copy revision data to the new pack."""
608
if self.revision_ids:
609
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
612
# select revision keys
613
revision_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
614
self.packs, 'revision_index')[0]
615
revision_nodes = self._pack_collection._index_contents(revision_index_map, revision_keys)
616
# copy revision keys and adjust values
617
self.pb.update("Copying revision texts", 1)
618
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
619
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
620
self.new_pack.revision_index, readv_group_iter, total_items))
621
if 'pack' in debug.debug_flags:
622
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
623
time.ctime(), self._pack_collection._upload_transport.base,
624
self.new_pack.random_name,
625
self.new_pack.revision_index.key_count(),
626
time.time() - self.new_pack.start_time)
627
self._revision_keys = revision_keys
629
def _copy_inventory_texts(self):
630
"""Copy the inventory texts to the new pack.
632
self._revision_keys is used to determine what inventories to copy.
634
Sets self._text_filter appropriately.
636
# select inventory keys
637
inv_keys = self._revision_keys # currently the same keyspace, and note that
638
# querying for keys here could introduce a bug where an inventory item
639
# is missed, so do not change it to query separately without cross
640
# checking like the text key check below.
641
inventory_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
642
self.packs, 'inventory_index')[0]
643
inv_nodes = self._pack_collection._index_contents(inventory_index_map, inv_keys)
644
# copy inventory keys and adjust values
645
# XXX: Should be a helper function to allow different inv representation
647
self.pb.update("Copying inventory texts", 2)
648
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
649
# Only grab the output lines if we will be processing them
650
output_lines = bool(self.revision_ids)
651
inv_lines = self._copy_nodes_graph(inventory_index_map,
652
self.new_pack._writer, self.new_pack.inventory_index,
653
readv_group_iter, total_items, output_lines=output_lines)
654
if self.revision_ids:
655
self._process_inventory_lines(inv_lines)
657
# eat the iterator to cause it to execute.
659
self._text_filter = None
660
if 'pack' in debug.debug_flags:
661
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
662
time.ctime(), self._pack_collection._upload_transport.base,
663
self.new_pack.random_name,
664
self.new_pack.inventory_index.key_count(),
665
time.time() - self.new_pack.start_time)
667
def _copy_text_texts(self):
669
text_index_map, text_nodes = self._get_text_nodes()
670
if self._text_filter is not None:
671
# We could return the keys copied as part of the return value from
672
# _copy_nodes_graph but this doesn't work all that well with the
673
# need to get line output too, so we check separately, and as we're
674
# going to buffer everything anyway, we check beforehand, which
675
# saves reading knit data over the wire when we know there are
677
text_nodes = set(text_nodes)
678
present_text_keys = set(_node[1] for _node in text_nodes)
679
missing_text_keys = set(self._text_filter) - present_text_keys
680
if missing_text_keys:
681
# TODO: raise a specific error that can handle many missing
683
a_missing_key = missing_text_keys.pop()
684
raise errors.RevisionNotPresent(a_missing_key[1],
686
# copy text keys and adjust values
687
self.pb.update("Copying content texts", 3)
688
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
689
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
690
self.new_pack.text_index, readv_group_iter, total_items))
691
self._log_copied_texts()
693
def _check_references(self):
694
"""Make sure our external refereneces are present."""
695
external_refs = self.new_pack._external_compression_parents_of_texts()
697
index = self._pack_collection.text_index.combined_index
698
found_items = list(index.iter_entries(external_refs))
699
if len(found_items) != len(external_refs):
700
found_keys = set(k for idx, k, refs, value in found_items)
701
missing_items = external_refs - found_keys
702
missing_file_id, missing_revision_id = missing_items.pop()
703
raise errors.RevisionNotPresent(missing_revision_id,
706
def _create_pack_from_packs(self):
707
self.pb.update("Opening pack", 0, 5)
708
self.new_pack = self.open_pack()
709
new_pack = self.new_pack
710
# buffer data - we won't be reading-back during the pack creation and
711
# this makes a significant difference on sftp pushes.
712
new_pack.set_write_cache_size(1024*1024)
713
if 'pack' in debug.debug_flags:
714
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
715
for a_pack in self.packs]
716
if self.revision_ids is not None:
717
rev_count = len(self.revision_ids)
720
mutter('%s: create_pack: creating pack from source packs: '
721
'%s%s %s revisions wanted %s t=0',
722
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
723
plain_pack_list, rev_count)
724
self._copy_revision_texts()
725
self._copy_inventory_texts()
726
self._copy_text_texts()
727
# select signature keys
728
signature_filter = self._revision_keys # same keyspace
729
signature_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
730
self.packs, 'signature_index')[0]
731
signature_nodes = self._pack_collection._index_contents(signature_index_map,
733
# copy signature keys and adjust values
734
self.pb.update("Copying signature texts", 4)
735
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
736
new_pack.signature_index)
737
if 'pack' in debug.debug_flags:
738
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
739
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
740
new_pack.signature_index.key_count(),
741
time.time() - new_pack.start_time)
742
self._check_references()
743
if not self._use_pack(new_pack):
746
self.pb.update("Finishing pack", 5)
748
self._pack_collection.allocate(new_pack)
751
def _copy_nodes(self, nodes, index_map, writer, write_index):
752
"""Copy knit nodes between packs with no graph references."""
753
pb = ui.ui_factory.nested_progress_bar()
755
return self._do_copy_nodes(nodes, index_map, writer,
760
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb):
761
# for record verification
762
knit = KnitVersionedFiles(None, None)
763
# plan a readv on each source pack:
765
nodes = sorted(nodes)
766
# how to map this into knit.py - or knit.py into this?
767
# we don't want the typical knit logic, we want grouping by pack
768
# at this point - perhaps a helper library for the following code
769
# duplication points?
771
for index, key, value in nodes:
772
if index not in request_groups:
773
request_groups[index] = []
774
request_groups[index].append((key, value))
776
pb.update("Copied record", record_index, len(nodes))
777
for index, items in request_groups.iteritems():
778
pack_readv_requests = []
779
for key, value in items:
780
# ---- KnitGraphIndex.get_position
781
bits = value[1:].split(' ')
782
offset, length = int(bits[0]), int(bits[1])
783
pack_readv_requests.append((offset, length, (key, value[0])))
784
# linear scan up the pack
785
pack_readv_requests.sort()
787
transport, path = index_map[index]
788
reader = pack.make_readv_reader(transport, path,
789
[offset[0:2] for offset in pack_readv_requests])
790
for (names, read_func), (_1, _2, (key, eol_flag)) in \
791
izip(reader.iter_records(), pack_readv_requests):
792
raw_data = read_func(None)
793
# check the header only
794
df, _ = knit._parse_record_header(key, raw_data)
796
pos, size = writer.add_bytes_record(raw_data, names)
797
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
798
pb.update("Copied record", record_index)
801
def _copy_nodes_graph(self, index_map, writer, write_index,
802
readv_group_iter, total_items, output_lines=False):
803
"""Copy knit nodes between packs.
805
:param output_lines: Return lines present in the copied data as
806
an iterator of line,version_id.
808
pb = ui.ui_factory.nested_progress_bar()
810
for result in self._do_copy_nodes_graph(index_map, writer,
811
write_index, output_lines, pb, readv_group_iter, total_items):
814
# Python 2.4 does not permit try:finally: in a generator.
820
def _do_copy_nodes_graph(self, index_map, writer, write_index,
821
output_lines, pb, readv_group_iter, total_items):
822
# for record verification
823
knit = KnitVersionedFiles(None, None)
824
# for line extraction when requested (inventories only)
826
factory = KnitPlainFactory()
828
pb.update("Copied record", record_index, total_items)
829
for index, readv_vector, node_vector in readv_group_iter:
831
transport, path = index_map[index]
832
reader = pack.make_readv_reader(transport, path, readv_vector)
833
for (names, read_func), (key, eol_flag, references) in \
834
izip(reader.iter_records(), node_vector):
835
raw_data = read_func(None)
837
# read the entire thing
838
content, _ = knit._parse_record(key[-1], raw_data)
839
if len(references[-1]) == 0:
840
line_iterator = factory.get_fulltext_content(content)
842
line_iterator = factory.get_linedelta_content(content)
843
for line in line_iterator:
846
# check the header only
847
df, _ = knit._parse_record_header(key, raw_data)
849
pos, size = writer.add_bytes_record(raw_data, names)
850
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
851
pb.update("Copied record", record_index)
854
def _get_text_nodes(self):
855
text_index_map = self._pack_collection._packs_list_to_pack_map_and_index_list(
856
self.packs, 'text_index')[0]
857
return text_index_map, self._pack_collection._index_contents(text_index_map,
860
def _least_readv_node_readv(self, nodes):
861
"""Generate request groups for nodes using the least readv's.
863
:param nodes: An iterable of graph index nodes.
864
:return: Total node count and an iterator of the data needed to perform
865
readvs to obtain the data for nodes. Each item yielded by the
866
iterator is a tuple with:
867
index, readv_vector, node_vector. readv_vector is a list ready to
868
hand to the transport readv method, and node_vector is a list of
869
(key, eol_flag, references) for the the node retrieved by the
870
matching readv_vector.
872
# group by pack so we do one readv per pack
873
nodes = sorted(nodes)
876
for index, key, value, references in nodes:
877
if index not in request_groups:
878
request_groups[index] = []
879
request_groups[index].append((key, value, references))
881
for index, items in request_groups.iteritems():
882
pack_readv_requests = []
883
for key, value, references in items:
884
# ---- KnitGraphIndex.get_position
885
bits = value[1:].split(' ')
886
offset, length = int(bits[0]), int(bits[1])
887
pack_readv_requests.append(
888
((offset, length), (key, value[0], references)))
889
# linear scan up the pack to maximum range combining.
890
pack_readv_requests.sort()
891
# split out the readv and the node data.
892
pack_readv = [readv for readv, node in pack_readv_requests]
893
node_vector = [node for readv, node in pack_readv_requests]
894
result.append((index, pack_readv, node_vector))
897
def _log_copied_texts(self):
898
if 'pack' in debug.debug_flags:
899
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
900
time.ctime(), self._pack_collection._upload_transport.base,
901
self.new_pack.random_name,
902
self.new_pack.text_index.key_count(),
903
time.time() - self.new_pack.start_time)
905
def _process_inventory_lines(self, inv_lines):
906
"""Use up the inv_lines generator and setup a text key filter."""
907
repo = self._pack_collection.repo
908
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
909
inv_lines, self.revision_keys)
911
for fileid, file_revids in fileid_revisions.iteritems():
912
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
913
self._text_filter = text_filter
915
def _revision_node_readv(self, revision_nodes):
916
"""Return the total revisions and the readv's to issue.
918
:param revision_nodes: The revision index contents for the packs being
919
incorporated into the new pack.
920
:return: As per _least_readv_node_readv.
922
return self._least_readv_node_readv(revision_nodes)
924
def _use_pack(self, new_pack):
925
"""Return True if new_pack should be used.
927
:param new_pack: The pack that has just been created.
928
:return: True if the pack should be used.
930
return new_pack.data_inserted()
933
class OptimisingPacker(Packer):
934
"""A packer which spends more time to create better disk layouts."""
936
def _revision_node_readv(self, revision_nodes):
937
"""Return the total revisions and the readv's to issue.
939
This sort places revisions in topological order with the ancestors
942
:param revision_nodes: The revision index contents for the packs being
943
incorporated into the new pack.
944
:return: As per _least_readv_node_readv.
946
# build an ancestors dict
949
for index, key, value, references in revision_nodes:
950
ancestors[key] = references[0]
951
by_key[key] = (index, value, references)
952
order = tsort.topo_sort(ancestors)
954
# Single IO is pathological, but it will work as a starting point.
956
for key in reversed(order):
957
index, value, references = by_key[key]
958
# ---- KnitGraphIndex.get_position
959
bits = value[1:].split(' ')
960
offset, length = int(bits[0]), int(bits[1])
962
(index, [(offset, length)], [(key, value[0], references)]))
963
# TODO: combine requests in the same index that are in ascending order.
964
return total, requests
967
class ReconcilePacker(Packer):
968
"""A packer which regenerates indices etc as it copies.
970
This is used by ``bzr reconcile`` to cause parent text pointers to be
974
def _extra_init(self):
975
self._data_changed = False
977
def _process_inventory_lines(self, inv_lines):
978
"""Generate a text key reference map rather for reconciling with."""
979
repo = self._pack_collection.repo
980
refs = repo._find_text_key_references_from_xml_inventory_lines(
982
self._text_refs = refs
983
# during reconcile we:
984
# - convert unreferenced texts to full texts
985
# - correct texts which reference a text not copied to be full texts
986
# - copy all others as-is but with corrected parents.
987
# - so at this point we don't know enough to decide what becomes a full
989
self._text_filter = None
991
def _copy_text_texts(self):
992
"""generate what texts we should have and then copy."""
993
self.pb.update("Copying content texts", 3)
994
# we have three major tasks here:
995
# 1) generate the ideal index
996
repo = self._pack_collection.repo
997
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
999
self.new_pack.revision_index.iter_all_entries()])
1000
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1001
# 2) generate a text_nodes list that contains all the deltas that can
1002
# be used as-is, with corrected parents.
1005
discarded_nodes = []
1006
NULL_REVISION = _mod_revision.NULL_REVISION
1007
text_index_map, text_nodes = self._get_text_nodes()
1008
for node in text_nodes:
1014
ideal_parents = tuple(ideal_index[node[1]])
1016
discarded_nodes.append(node)
1017
self._data_changed = True
1019
if ideal_parents == (NULL_REVISION,):
1021
if ideal_parents == node[3][0]:
1023
ok_nodes.append(node)
1024
elif ideal_parents[0:1] == node[3][0][0:1]:
1025
# the left most parent is the same, or there are no parents
1026
# today. Either way, we can preserve the representation as
1027
# long as we change the refs to be inserted.
1028
self._data_changed = True
1029
ok_nodes.append((node[0], node[1], node[2],
1030
(ideal_parents, node[3][1])))
1031
self._data_changed = True
1033
# Reinsert this text completely
1034
bad_texts.append((node[1], ideal_parents))
1035
self._data_changed = True
1036
# we're finished with some data.
1039
# 3) bulk copy the ok data
1040
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1041
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1042
self.new_pack.text_index, readv_group_iter, total_items))
1043
# 4) adhoc copy all the other texts.
1044
# We have to topologically insert all texts otherwise we can fail to
1045
# reconcile when parts of a single delta chain are preserved intact,
1046
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1047
# reinserted, and if d3 has incorrect parents it will also be
1048
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1049
# copied), so we will try to delta, but d2 is not currently able to be
1050
# extracted because it's basis d1 is not present. Topologically sorting
1051
# addresses this. The following generates a sort for all the texts that
1052
# are being inserted without having to reference the entire text key
1053
# space (we only topo sort the revisions, which is smaller).
1054
topo_order = tsort.topo_sort(ancestors)
1055
rev_order = dict(zip(topo_order, range(len(topo_order))))
1056
bad_texts.sort(key=lambda key:rev_order[key[0][1]])
1057
transaction = repo.get_transaction()
1058
file_id_index = GraphIndexPrefixAdapter(
1059
self.new_pack.text_index,
1061
add_nodes_callback=self.new_pack.text_index.add_nodes)
1062
data_access = _DirectPackAccess(
1063
{self.new_pack.text_index:self.new_pack.access_tuple()})
1064
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1065
self.new_pack.access_tuple())
1066
output_texts = KnitVersionedFiles(
1067
_KnitGraphIndex(self.new_pack.text_index,
1068
add_callback=self.new_pack.text_index.add_nodes,
1069
deltas=True, parents=True, is_locked=repo.is_locked),
1070
data_access=data_access, max_delta_chain=200)
1071
for key, parent_keys in bad_texts:
1072
# We refer to the new pack to delta data being output.
1073
# A possible improvement would be to catch errors on short reads
1074
# and only flush then.
1075
self.new_pack.flush()
1077
for parent_key in parent_keys:
1078
if parent_key[0] != key[0]:
1079
# Graph parents must match the fileid
1080
raise errors.BzrError('Mismatched key parent %r:%r' %
1082
parents.append(parent_key[1])
1083
text_lines = split_lines(repo.texts.get_record_stream(
1084
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1085
output_texts.add_lines(key, parent_keys, text_lines,
1086
random_id=True, check_content=False)
1087
# 5) check that nothing inserted has a reference outside the keyspace.
1088
missing_text_keys = self.new_pack._external_compression_parents_of_texts()
1089
if missing_text_keys:
1090
raise errors.BzrError('Reference to missing compression parents %r'
1091
% (missing_text_keys,))
1092
self._log_copied_texts()
1094
def _use_pack(self, new_pack):
1095
"""Override _use_pack to check for reconcile having changed content."""
1096
# XXX: we might be better checking this at the copy time.
1097
original_inventory_keys = set()
1098
inv_index = self._pack_collection.inventory_index.combined_index
1099
for entry in inv_index.iter_all_entries():
1100
original_inventory_keys.add(entry[1])
1101
new_inventory_keys = set()
1102
for entry in new_pack.inventory_index.iter_all_entries():
1103
new_inventory_keys.add(entry[1])
1104
if new_inventory_keys != original_inventory_keys:
1105
self._data_changed = True
1106
return new_pack.data_inserted() and self._data_changed
1109
class RepositoryPackCollection(object):
1110
"""Management of packs within a repository."""
1112
def __init__(self, repo, transport, index_transport, upload_transport,
1114
"""Create a new RepositoryPackCollection.
1116
:param transport: Addresses the repository base directory
1117
(typically .bzr/repository/).
1118
:param index_transport: Addresses the directory containing indices.
1119
:param upload_transport: Addresses the directory into which packs are written
1120
while they're being created.
1121
:param pack_transport: Addresses the directory of existing complete packs.
1124
self.transport = transport
1125
self._index_transport = index_transport
1126
self._upload_transport = upload_transport
1127
self._pack_transport = pack_transport
1128
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
1131
self._packs_by_name = {}
1132
# the previous pack-names content
1133
self._packs_at_load = None
1134
# when a pack is being created by this object, the state of that pack.
1135
self._new_pack = None
1136
# aggregated revision index data
1137
self.revision_index = AggregateIndex()
1138
self.inventory_index = AggregateIndex()
1139
self.text_index = AggregateIndex()
1140
self.signature_index = AggregateIndex()
1142
def add_pack_to_memory(self, pack):
1143
"""Make a Pack object available to the repository to satisfy queries.
1145
:param pack: A Pack object.
1147
if pack.name in self._packs_by_name:
1148
raise AssertionError()
1149
self.packs.append(pack)
1150
self._packs_by_name[pack.name] = pack
1151
self.revision_index.add_index(pack.revision_index, pack)
1152
self.inventory_index.add_index(pack.inventory_index, pack)
1153
self.text_index.add_index(pack.text_index, pack)
1154
self.signature_index.add_index(pack.signature_index, pack)
1156
def all_packs(self):
1157
"""Return a list of all the Pack objects this repository has.
1159
Note that an in-progress pack being created is not returned.
1161
:return: A list of Pack objects for all the packs in the repository.
1164
for name in self.names():
1165
result.append(self.get_pack_by_name(name))
1169
"""Pack the pack collection incrementally.
1171
This will not attempt global reorganisation or recompression,
1172
rather it will just ensure that the total number of packs does
1173
not grow without bound. It uses the _max_pack_count method to
1174
determine if autopacking is needed, and the pack_distribution
1175
method to determine the number of revisions in each pack.
1177
If autopacking takes place then the packs name collection will have
1178
been flushed to disk - packing requires updating the name collection
1179
in synchronisation with certain steps. Otherwise the names collection
1182
:return: True if packing took place.
1184
# XXX: Should not be needed when the management of indices is sane.
1185
total_revisions = self.revision_index.combined_index.key_count()
1186
total_packs = len(self._names)
1187
if self._max_pack_count(total_revisions) >= total_packs:
1189
# XXX: the following may want to be a class, to pack with a given
1191
mutter('Auto-packing repository %s, which has %d pack files, '
1192
'containing %d revisions into %d packs.', self, total_packs,
1193
total_revisions, self._max_pack_count(total_revisions))
1194
# determine which packs need changing
1195
pack_distribution = self.pack_distribution(total_revisions)
1197
for pack in self.all_packs():
1198
revision_count = pack.get_revision_count()
1199
if revision_count == 0:
1200
# revision less packs are not generated by normal operation,
1201
# only by operations like sign-my-commits, and thus will not
1202
# tend to grow rapdily or without bound like commit containing
1203
# packs do - leave them alone as packing them really should
1204
# group their data with the relevant commit, and that may
1205
# involve rewriting ancient history - which autopack tries to
1206
# avoid. Alternatively we could not group the data but treat
1207
# each of these as having a single revision, and thus add
1208
# one revision for each to the total revision count, to get
1209
# a matching distribution.
1211
existing_packs.append((revision_count, pack))
1212
pack_operations = self.plan_autopack_combinations(
1213
existing_packs, pack_distribution)
1214
self._execute_pack_operations(pack_operations)
1217
def _execute_pack_operations(self, pack_operations, _packer_class=Packer):
1218
"""Execute a series of pack operations.
1220
:param pack_operations: A list of [revision_count, packs_to_combine].
1221
:param _packer_class: The class of packer to use (default: Packer).
1224
for revision_count, packs in pack_operations:
1225
# we may have no-ops from the setup logic
1228
_packer_class(self, packs, '.autopack').pack()
1230
self._remove_pack_from_memory(pack)
1231
# record the newly available packs and stop advertising the old
1233
self._save_pack_names(clear_obsolete_packs=True)
1234
# Move the old packs out of the way now they are no longer referenced.
1235
for revision_count, packs in pack_operations:
1236
self._obsolete_packs(packs)
1238
def lock_names(self):
1239
"""Acquire the mutex around the pack-names index.
1241
This cannot be used in the middle of a read-only transaction on the
1244
self.repo.control_files.lock_write()
1247
"""Pack the pack collection totally."""
1248
self.ensure_loaded()
1249
total_packs = len(self._names)
1251
# This is arguably wrong because we might not be optimal, but for
1252
# now lets leave it in. (e.g. reconcile -> one pack. But not
1255
total_revisions = self.revision_index.combined_index.key_count()
1256
# XXX: the following may want to be a class, to pack with a given
1258
mutter('Packing repository %s, which has %d pack files, '
1259
'containing %d revisions into 1 packs.', self, total_packs,
1261
# determine which packs need changing
1262
pack_distribution = [1]
1263
pack_operations = [[0, []]]
1264
for pack in self.all_packs():
1265
pack_operations[-1][0] += pack.get_revision_count()
1266
pack_operations[-1][1].append(pack)
1267
self._execute_pack_operations(pack_operations, OptimisingPacker)
1269
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1270
"""Plan a pack operation.
1272
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1274
:param pack_distribution: A list with the number of revisions desired
1277
if len(existing_packs) <= len(pack_distribution):
1279
existing_packs.sort(reverse=True)
1280
pack_operations = [[0, []]]
1281
# plan out what packs to keep, and what to reorganise
1282
while len(existing_packs):
1283
# take the largest pack, and if its less than the head of the
1284
# distribution chart we will include its contents in the new pack for
1285
# that position. If its larger, we remove its size from the
1286
# distribution chart
1287
next_pack_rev_count, next_pack = existing_packs.pop(0)
1288
if next_pack_rev_count >= pack_distribution[0]:
1289
# this is already packed 'better' than this, so we can
1290
# not waste time packing it.
1291
while next_pack_rev_count > 0:
1292
next_pack_rev_count -= pack_distribution[0]
1293
if next_pack_rev_count >= 0:
1295
del pack_distribution[0]
1297
# didn't use that entire bucket up
1298
pack_distribution[0] = -next_pack_rev_count
1300
# add the revisions we're going to add to the next output pack
1301
pack_operations[-1][0] += next_pack_rev_count
1302
# allocate this pack to the next pack sub operation
1303
pack_operations[-1][1].append(next_pack)
1304
if pack_operations[-1][0] >= pack_distribution[0]:
1305
# this pack is used up, shift left.
1306
del pack_distribution[0]
1307
pack_operations.append([0, []])
1309
return pack_operations
1311
def ensure_loaded(self):
1312
# NB: if you see an assertion error here, its probably access against
1313
# an unlocked repo. Naughty.
1314
if not self.repo.is_locked():
1315
raise errors.ObjectNotLocked(self.repo)
1316
if self._names is None:
1318
self._packs_at_load = set()
1319
for index, key, value in self._iter_disk_pack_index():
1321
self._names[name] = self._parse_index_sizes(value)
1322
self._packs_at_load.add((key, value))
1323
# populate all the metadata.
1326
def _parse_index_sizes(self, value):
1327
"""Parse a string of index sizes."""
1328
return tuple([int(digits) for digits in value.split(' ')])
1330
def get_pack_by_name(self, name):
1331
"""Get a Pack object by name.
1333
:param name: The name of the pack - e.g. '123456'
1334
:return: A Pack object.
1337
return self._packs_by_name[name]
1339
rev_index = self._make_index(name, '.rix')
1340
inv_index = self._make_index(name, '.iix')
1341
txt_index = self._make_index(name, '.tix')
1342
sig_index = self._make_index(name, '.six')
1343
result = ExistingPack(self._pack_transport, name, rev_index,
1344
inv_index, txt_index, sig_index)
1345
self.add_pack_to_memory(result)
1348
def allocate(self, a_new_pack):
1349
"""Allocate name in the list of packs.
1351
:param a_new_pack: A NewPack instance to be added to the collection of
1352
packs for this repository.
1354
self.ensure_loaded()
1355
if a_new_pack.name in self._names:
1356
raise errors.BzrError(
1357
'Pack %r already exists in %s' % (a_new_pack.name, self))
1358
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1359
self.add_pack_to_memory(a_new_pack)
1361
def _iter_disk_pack_index(self):
1362
"""Iterate over the contents of the pack-names index.
1364
This is used when loading the list from disk, and before writing to
1365
detect updates from others during our write operation.
1366
:return: An iterator of the index contents.
1368
return GraphIndex(self.transport, 'pack-names', None
1369
).iter_all_entries()
1371
def _make_index(self, name, suffix):
1372
size_offset = self._suffix_offsets[suffix]
1373
index_name = name + suffix
1374
index_size = self._names[name][size_offset]
1376
self._index_transport, index_name, index_size)
1378
def _max_pack_count(self, total_revisions):
1379
"""Return the maximum number of packs to use for total revisions.
1381
:param total_revisions: The total number of revisions in the
1384
if not total_revisions:
1386
digits = str(total_revisions)
1388
for digit in digits:
1389
result += int(digit)
1393
"""Provide an order to the underlying names."""
1394
return sorted(self._names.keys())
1396
def _obsolete_packs(self, packs):
1397
"""Move a number of packs which have been obsoleted out of the way.
1399
Each pack and its associated indices are moved out of the way.
1401
Note: for correctness this function should only be called after a new
1402
pack names index has been written without these pack names, and with
1403
the names of packs that contain the data previously available via these
1406
:param packs: The packs to obsolete.
1407
:param return: None.
1410
pack.pack_transport.rename(pack.file_name(),
1411
'../obsolete_packs/' + pack.file_name())
1412
# TODO: Probably needs to know all possible indices for this pack
1413
# - or maybe list the directory and move all indices matching this
1414
# name whether we recognize it or not?
1415
for suffix in ('.iix', '.six', '.tix', '.rix'):
1416
self._index_transport.rename(pack.name + suffix,
1417
'../obsolete_packs/' + pack.name + suffix)
1419
def pack_distribution(self, total_revisions):
1420
"""Generate a list of the number of revisions to put in each pack.
1422
:param total_revisions: The total number of revisions in the
1425
if total_revisions == 0:
1427
digits = reversed(str(total_revisions))
1429
for exponent, count in enumerate(digits):
1430
size = 10 ** exponent
1431
for pos in range(int(count)):
1433
return list(reversed(result))
1435
def _pack_tuple(self, name):
1436
"""Return a tuple with the transport and file name for a pack name."""
1437
return self._pack_transport, name + '.pack'
1439
def _remove_pack_from_memory(self, pack):
1440
"""Remove pack from the packs accessed by this repository.
1442
Only affects memory state, until self._save_pack_names() is invoked.
1444
self._names.pop(pack.name)
1445
self._packs_by_name.pop(pack.name)
1446
self._remove_pack_indices(pack)
1448
def _remove_pack_indices(self, pack):
1449
"""Remove the indices for pack from the aggregated indices."""
1450
self.revision_index.remove_index(pack.revision_index, pack)
1451
self.inventory_index.remove_index(pack.inventory_index, pack)
1452
self.text_index.remove_index(pack.text_index, pack)
1453
self.signature_index.remove_index(pack.signature_index, pack)
1456
"""Clear all cached data."""
1457
# cached revision data
1458
self.repo._revision_knit = None
1459
self.revision_index.clear()
1460
# cached signature data
1461
self.repo._signature_knit = None
1462
self.signature_index.clear()
1463
# cached file text data
1464
self.text_index.clear()
1465
self.repo._text_knit = None
1466
# cached inventory data
1467
self.inventory_index.clear()
1468
# remove the open pack
1469
self._new_pack = None
1470
# information about packs.
1473
self._packs_by_name = {}
1474
self._packs_at_load = None
1476
def _make_index_map(self, index_suffix):
1477
"""Return information on existing indices.
1479
:param suffix: Index suffix added to pack name.
1481
:returns: (pack_map, indices) where indices is a list of GraphIndex
1482
objects, and pack_map is a mapping from those objects to the
1483
pack tuple they describe.
1485
# TODO: stop using this; it creates new indices unnecessarily.
1486
self.ensure_loaded()
1487
suffix_map = {'.rix': 'revision_index',
1488
'.six': 'signature_index',
1489
'.iix': 'inventory_index',
1490
'.tix': 'text_index',
1492
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1493
suffix_map[index_suffix])
1495
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1496
"""Convert a list of packs to an index pack map and index list.
1498
:param packs: The packs list to process.
1499
:param index_attribute: The attribute that the desired index is found
1501
:return: A tuple (map, list) where map contains the dict from
1502
index:pack_tuple, and lsit contains the indices in the same order
1508
index = getattr(pack, index_attribute)
1509
indices.append(index)
1510
pack_map[index] = (pack.pack_transport, pack.file_name())
1511
return pack_map, indices
1513
def _index_contents(self, pack_map, key_filter=None):
1514
"""Get an iterable of the index contents from a pack_map.
1516
:param pack_map: A map from indices to pack details.
1517
:param key_filter: An optional filter to limit the
1520
indices = [index for index in pack_map.iterkeys()]
1521
all_index = CombinedGraphIndex(indices)
1522
if key_filter is None:
1523
return all_index.iter_all_entries()
1525
return all_index.iter_entries(key_filter)
1527
def _unlock_names(self):
1528
"""Release the mutex around the pack-names index."""
1529
self.repo.control_files.unlock()
1531
def _save_pack_names(self, clear_obsolete_packs=False):
1532
"""Save the list of packs.
1534
This will take out the mutex around the pack names list for the
1535
duration of the method call. If concurrent updates have been made, a
1536
three-way merge between the current list and the current in memory list
1539
:param clear_obsolete_packs: If True, clear out the contents of the
1540
obsolete_packs directory.
1544
builder = GraphIndexBuilder()
1545
# load the disk nodes across
1547
for index, key, value in self._iter_disk_pack_index():
1548
disk_nodes.add((key, value))
1549
# do a two-way diff against our original content
1550
current_nodes = set()
1551
for name, sizes in self._names.iteritems():
1553
((name, ), ' '.join(str(size) for size in sizes)))
1554
deleted_nodes = self._packs_at_load - current_nodes
1555
new_nodes = current_nodes - self._packs_at_load
1556
disk_nodes.difference_update(deleted_nodes)
1557
disk_nodes.update(new_nodes)
1558
# TODO: handle same-name, index-size-changes here -
1559
# e.g. use the value from disk, not ours, *unless* we're the one
1561
for key, value in disk_nodes:
1562
builder.add_node(key, value)
1563
self.transport.put_file('pack-names', builder.finish(),
1564
mode=self.repo.control_files._file_mode)
1565
# move the baseline forward
1566
self._packs_at_load = disk_nodes
1567
# now clear out the obsolete packs directory
1568
if clear_obsolete_packs:
1569
self.transport.clone('obsolete_packs').delete_multi(
1570
self.transport.list_dir('obsolete_packs'))
1572
self._unlock_names()
1573
# synchronise the memory packs list with what we just wrote:
1574
new_names = dict(disk_nodes)
1575
# drop no longer present nodes
1576
for pack in self.all_packs():
1577
if (pack.name,) not in new_names:
1578
self._remove_pack_from_memory(pack)
1579
# add new nodes/refresh existing ones
1580
for key, value in disk_nodes:
1582
sizes = self._parse_index_sizes(value)
1583
if name in self._names:
1585
if sizes != self._names[name]:
1586
# the pack for name has had its indices replaced - rare but
1587
# important to handle. XXX: probably can never happen today
1588
# because the three-way merge code above does not handle it
1589
# - you may end up adding the same key twice to the new
1590
# disk index because the set values are the same, unless
1591
# the only index shows up as deleted by the set difference
1592
# - which it may. Until there is a specific test for this,
1593
# assume its broken. RBC 20071017.
1594
self._remove_pack_from_memory(self.get_pack_by_name(name))
1595
self._names[name] = sizes
1596
self.get_pack_by_name(name)
1599
self._names[name] = sizes
1600
self.get_pack_by_name(name)
1602
def _start_write_group(self):
1603
# Do not permit preparation for writing if we're not in a 'write lock'.
1604
if not self.repo.is_write_locked():
1605
raise errors.NotWriteLocked(self)
1606
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1607
self._pack_transport, upload_suffix='.pack',
1608
file_mode=self.repo.control_files._file_mode)
1609
# allow writing: queue writes to a new index
1610
self.revision_index.add_writable_index(self._new_pack.revision_index,
1612
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
1614
self.text_index.add_writable_index(self._new_pack.text_index,
1616
self.signature_index.add_writable_index(self._new_pack.signature_index,
1619
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
1620
self.repo.revisions._index._add_callback = self.revision_index.add_callback
1621
self.repo.signatures._index._add_callback = self.signature_index.add_callback
1622
self.repo.texts._index._add_callback = self.text_index.add_callback
1624
def _abort_write_group(self):
1625
# FIXME: just drop the transient index.
1626
# forget what names there are
1627
if self._new_pack is not None:
1628
self._new_pack.abort()
1629
self._remove_pack_indices(self._new_pack)
1630
self._new_pack = None
1631
self.repo._text_knit = None
1633
def _commit_write_group(self):
1634
self._remove_pack_indices(self._new_pack)
1635
if self._new_pack.data_inserted():
1636
# get all the data to disk and read to use
1637
self._new_pack.finish()
1638
self.allocate(self._new_pack)
1639
self._new_pack = None
1640
if not self.autopack():
1641
# when autopack takes no steps, the names list is still
1643
self._save_pack_names()
1645
self._new_pack.abort()
1646
self._new_pack = None
1647
self.repo._text_knit = None
1650
class KnitPackRepository(KnitRepository):
1651
"""Experimental graph-knit using repository."""
1653
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
1655
index_transport = control_files._transport.clone('indices')
1656
self._pack_collection = RepositoryPackCollection(self, control_files._transport,
1658
control_files._transport.clone('upload'),
1659
control_files._transport.clone('packs'))
1660
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
1661
_commit_builder_class, _serializer)
1662
self.inventories = KnitVersionedFiles(
1663
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
1664
add_callback=self._pack_collection.inventory_index.add_callback,
1665
deltas=True, parents=True, is_locked=self.is_locked),
1666
data_access=self._pack_collection.inventory_index.data_access,
1667
max_delta_chain=200)
1668
self.revisions = KnitVersionedFiles(
1669
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
1670
add_callback=self._pack_collection.revision_index.add_callback,
1671
deltas=False, parents=True, is_locked=self.is_locked),
1672
data_access=self._pack_collection.revision_index.data_access,
1674
self.signatures = KnitVersionedFiles(
1675
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
1676
add_callback=self._pack_collection.signature_index.add_callback,
1677
deltas=False, parents=False, is_locked=self.is_locked),
1678
data_access=self._pack_collection.signature_index.data_access,
1680
self.texts = KnitVersionedFiles(
1681
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
1682
add_callback=self._pack_collection.text_index.add_callback,
1683
deltas=True, parents=True, is_locked=self.is_locked),
1684
data_access=self._pack_collection.text_index.data_access,
1685
max_delta_chain=200)
1686
# True when the repository object is 'write locked' (as opposed to the
1687
# physical lock only taken out around changes to the pack-names list.)
1688
# Another way to represent this would be a decorator around the control
1689
# files object that presents logical locks as physical ones - if this
1690
# gets ugly consider that alternative design. RBC 20071011
1691
self._write_lock_count = 0
1692
self._transaction = None
1694
self._reconcile_does_inventory_gc = True
1695
self._reconcile_fixes_text_parents = True
1696
self._reconcile_backsup_inventory = False
1698
def _abort_write_group(self):
1699
self._pack_collection._abort_write_group()
1701
def _find_inconsistent_revision_parents(self):
1702
"""Find revisions with incorrectly cached parents.
1704
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1705
parents-in-revision).
1707
if not self.is_locked():
1708
raise errors.ObjectNotLocked(self)
1709
pb = ui.ui_factory.nested_progress_bar()
1712
revision_nodes = self._pack_collection.revision_index \
1713
.combined_index.iter_all_entries()
1714
index_positions = []
1715
# Get the cached index values for all revisions, and also the location
1716
# in each index of the revision text so we can perform linear IO.
1717
for index, key, value, refs in revision_nodes:
1718
pos, length = value[1:].split(' ')
1719
index_positions.append((index, int(pos), key[0],
1720
tuple(parent[0] for parent in refs[0])))
1721
pb.update("Reading revision index.", 0, 0)
1722
index_positions.sort()
1723
batch_count = len(index_positions) / 1000 + 1
1724
pb.update("Checking cached revision graph.", 0, batch_count)
1725
for offset in xrange(batch_count):
1726
pb.update("Checking cached revision graph.", offset)
1727
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1730
rev_ids = [item[2] for item in to_query]
1731
revs = self.get_revisions(rev_ids)
1732
for revision, item in zip(revs, to_query):
1733
index_parents = item[3]
1734
rev_parents = tuple(revision.parent_ids)
1735
if index_parents != rev_parents:
1736
result.append((revision.revision_id, index_parents, rev_parents))
1741
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
1742
def get_parents(self, revision_ids):
1743
"""See graph._StackedParentsProvider.get_parents."""
1744
parent_map = self.get_parent_map(revision_ids)
1745
return [parent_map.get(r, None) for r in revision_ids]
1747
def get_parent_map(self, keys):
1748
"""See graph._StackedParentsProvider.get_parent_map
1750
This implementation accesses the combined revision index to provide
1753
self._pack_collection.ensure_loaded()
1754
index = self._pack_collection.revision_index.combined_index
1756
if _mod_revision.NULL_REVISION in keys:
1757
keys.discard(_mod_revision.NULL_REVISION)
1758
found_parents = {_mod_revision.NULL_REVISION:()}
1761
search_keys = set((revision_id,) for revision_id in keys)
1762
for index, key, value, refs in index.iter_entries(search_keys):
1765
parents = (_mod_revision.NULL_REVISION,)
1767
parents = tuple(parent[0] for parent in parents)
1768
found_parents[key[0]] = parents
1769
return found_parents
1771
def has_revisions(self, revision_ids):
1772
"""See Repository.has_revisions()."""
1773
revision_ids = set(revision_ids)
1774
result = revision_ids.intersection(
1775
set([None, _mod_revision.NULL_REVISION]))
1776
revision_ids.difference_update(result)
1777
index = self._pack_collection.revision_index.combined_index
1778
keys = [(revision_id,) for revision_id in revision_ids]
1779
result.update(node[1][0] for node in index.iter_entries(keys))
1782
def _make_parents_provider(self):
1783
return graph.CachingParentsProvider(self)
1785
def _refresh_data(self):
1786
if self._write_lock_count == 1 or (
1787
self.control_files._lock_count == 1 and
1788
self.control_files._lock_mode == 'r'):
1789
# forget what names there are
1790
self._pack_collection.reset()
1791
# XXX: Better to do an in-memory merge when acquiring a new lock -
1792
# factor out code from _save_pack_names.
1793
self._pack_collection.ensure_loaded()
1795
def _start_write_group(self):
1796
self._pack_collection._start_write_group()
1798
def _commit_write_group(self):
1799
return self._pack_collection._commit_write_group()
1801
def get_transaction(self):
1802
if self._write_lock_count:
1803
return self._transaction
1805
return self.control_files.get_transaction()
1807
def is_locked(self):
1808
return self._write_lock_count or self.control_files.is_locked()
1810
def is_write_locked(self):
1811
return self._write_lock_count
1813
def lock_write(self, token=None):
1814
if not self._write_lock_count and self.is_locked():
1815
raise errors.ReadOnlyError(self)
1816
self._write_lock_count += 1
1817
if self._write_lock_count == 1:
1818
from bzrlib import transactions
1819
self._transaction = transactions.WriteTransaction()
1820
self._refresh_data()
1822
def lock_read(self):
1823
if self._write_lock_count:
1824
self._write_lock_count += 1
1826
self.control_files.lock_read()
1827
self._refresh_data()
1829
def leave_lock_in_place(self):
1830
# not supported - raise an error
1831
raise NotImplementedError(self.leave_lock_in_place)
1833
def dont_leave_lock_in_place(self):
1834
# not supported - raise an error
1835
raise NotImplementedError(self.dont_leave_lock_in_place)
1839
"""Compress the data within the repository.
1841
This will pack all the data to a single pack. In future it may
1842
recompress deltas or do other such expensive operations.
1844
self._pack_collection.pack()
1847
def reconcile(self, other=None, thorough=False):
1848
"""Reconcile this repository."""
1849
from bzrlib.reconcile import PackReconciler
1850
reconciler = PackReconciler(self, thorough=thorough)
1851
reconciler.reconcile()
1855
if self._write_lock_count == 1 and self._write_group is not None:
1856
self.abort_write_group()
1857
self._transaction = None
1858
self._write_lock_count = 0
1859
raise errors.BzrError(
1860
'Must end write group before releasing write lock on %s'
1862
if self._write_lock_count:
1863
self._write_lock_count -= 1
1864
if not self._write_lock_count:
1865
transaction = self._transaction
1866
self._transaction = None
1867
transaction.finish()
1869
self.control_files.unlock()
1872
class RepositoryFormatPack(MetaDirRepositoryFormat):
1873
"""Format logic for pack structured repositories.
1875
This repository format has:
1876
- a list of packs in pack-names
1877
- packs in packs/NAME.pack
1878
- indices in indices/NAME.{iix,six,tix,rix}
1879
- knit deltas in the packs, knit indices mapped to the indices.
1880
- thunk objects to support the knits programming API.
1881
- a format marker of its own
1882
- an optional 'shared-storage' flag
1883
- an optional 'no-working-trees' flag
1887
# Set this attribute in derived classes to control the repository class
1888
# created by open and initialize.
1889
repository_class = None
1890
# Set this attribute in derived classes to control the
1891
# _commit_builder_class that the repository objects will have passed to
1892
# their constructor.
1893
_commit_builder_class = None
1894
# Set this attribute in derived clases to control the _serializer that the
1895
# repository objects will have passed to their constructor.
1897
# External references are not supported in pack repositories yet.
1898
supports_external_lookups = False
1900
def initialize(self, a_bzrdir, shared=False):
1901
"""Create a pack based repository.
1903
:param a_bzrdir: bzrdir to contain the new repository; must already
1905
:param shared: If true the repository will be initialized as a shared
1908
mutter('creating repository in %s.', a_bzrdir.transport.base)
1909
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
1910
builder = GraphIndexBuilder()
1911
files = [('pack-names', builder.finish())]
1912
utf8_files = [('format', self.get_format_string())]
1914
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
1915
return self.open(a_bzrdir=a_bzrdir, _found=True)
1917
def open(self, a_bzrdir, _found=False, _override_transport=None):
1918
"""See RepositoryFormat.open().
1920
:param _override_transport: INTERNAL USE ONLY. Allows opening the
1921
repository at a slightly different url
1922
than normal. I.e. during 'upgrade'.
1925
format = RepositoryFormat.find_format(a_bzrdir)
1926
if _override_transport is not None:
1927
repo_transport = _override_transport
1929
repo_transport = a_bzrdir.get_repository_transport(None)
1930
control_files = lockable_files.LockableFiles(repo_transport,
1931
'lock', lockdir.LockDir)
1932
return self.repository_class(_format=self,
1934
control_files=control_files,
1935
_commit_builder_class=self._commit_builder_class,
1936
_serializer=self._serializer)
1939
class RepositoryFormatKnitPack1(RepositoryFormatPack):
1940
"""A no-subtrees parameterized Pack repository.
1942
This format was introduced in 0.92.
1945
repository_class = KnitPackRepository
1946
_commit_builder_class = PackCommitBuilder
1947
_serializer = xml5.serializer_v5
1949
def _get_matching_bzrdir(self):
1950
return bzrdir.format_registry.make_bzrdir('pack-0.92')
1952
def _ignore_setting_bzrdir(self, format):
1955
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1957
def get_format_string(self):
1958
"""See RepositoryFormat.get_format_string()."""
1959
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
1961
def get_format_description(self):
1962
"""See RepositoryFormat.get_format_description()."""
1963
return "Packs containing knits without subtree support"
1965
def check_conversion_target(self, target_format):
1969
class RepositoryFormatKnitPack3(RepositoryFormatPack):
1970
"""A subtrees parameterized Pack repository.
1972
This repository format uses the xml7 serializer to get:
1973
- support for recording full info about the tree root
1974
- support for recording tree-references
1976
This format was introduced in 0.92.
1979
repository_class = KnitPackRepository
1980
_commit_builder_class = PackRootCommitBuilder
1981
rich_root_data = True
1982
supports_tree_reference = True
1983
_serializer = xml7.serializer_v7
1985
def _get_matching_bzrdir(self):
1986
return bzrdir.format_registry.make_bzrdir(
1987
'pack-0.92-subtree')
1989
def _ignore_setting_bzrdir(self, format):
1992
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1994
def check_conversion_target(self, target_format):
1995
if not target_format.rich_root_data:
1996
raise errors.BadConversionTarget(
1997
'Does not support rich root data.', target_format)
1998
if not getattr(target_format, 'supports_tree_reference', False):
1999
raise errors.BadConversionTarget(
2000
'Does not support nested trees', target_format)
2002
def get_format_string(self):
2003
"""See RepositoryFormat.get_format_string()."""
2004
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2006
def get_format_description(self):
2007
"""See RepositoryFormat.get_format_description()."""
2008
return "Packs containing knits with subtree support\n"
2011
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2012
"""A rich-root, no subtrees parameterized Pack repository.
2014
This repository format uses the xml6 serializer to get:
2015
- support for recording full info about the tree root
2017
This format was introduced in 1.0.
2020
repository_class = KnitPackRepository
2021
_commit_builder_class = PackRootCommitBuilder
2022
rich_root_data = True
2023
supports_tree_reference = False
2024
_serializer = xml6.serializer_v6
2026
def _get_matching_bzrdir(self):
2027
return bzrdir.format_registry.make_bzrdir(
2030
def _ignore_setting_bzrdir(self, format):
2033
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2035
def check_conversion_target(self, target_format):
2036
if not target_format.rich_root_data:
2037
raise errors.BadConversionTarget(
2038
'Does not support rich root data.', target_format)
2040
def get_format_string(self):
2041
"""See RepositoryFormat.get_format_string()."""
2042
return ("Bazaar pack repository format 1 with rich root"
2043
" (needs bzr 1.0)\n")
2045
def get_format_description(self):
2046
"""See RepositoryFormat.get_format_description()."""
2047
return "Packs containing knits with rich root support\n"
2050
class RepositoryFormatPackDevelopment0(RepositoryFormatPack):
2051
"""A no-subtrees development repository.
2053
This format should be retained until the second release after bzr 1.0.
2055
No changes to the disk behaviour from pack-0.92.
2058
repository_class = KnitPackRepository
2059
_commit_builder_class = PackCommitBuilder
2060
_serializer = xml5.serializer_v5
2062
def _get_matching_bzrdir(self):
2063
return bzrdir.format_registry.make_bzrdir('development0')
2065
def _ignore_setting_bzrdir(self, format):
2068
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2070
def get_format_string(self):
2071
"""See RepositoryFormat.get_format_string()."""
2072
return "Bazaar development format 0 (needs bzr.dev from before 1.3)\n"
2074
def get_format_description(self):
2075
"""See RepositoryFormat.get_format_description()."""
2076
return ("Development repository format, currently the same as "
2079
def check_conversion_target(self, target_format):
2083
class RepositoryFormatPackDevelopment0Subtree(RepositoryFormatPack):
2084
"""A subtrees development repository.
2086
This format should be retained until the second release after bzr 1.0.
2088
No changes to the disk behaviour from pack-0.92-subtree.
2091
repository_class = KnitPackRepository
2092
_commit_builder_class = PackRootCommitBuilder
2093
rich_root_data = True
2094
supports_tree_reference = True
2095
_serializer = xml7.serializer_v7
2097
def _get_matching_bzrdir(self):
2098
return bzrdir.format_registry.make_bzrdir(
2099
'development0-subtree')
2101
def _ignore_setting_bzrdir(self, format):
2104
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2106
def check_conversion_target(self, target_format):
2107
if not target_format.rich_root_data:
2108
raise errors.BadConversionTarget(
2109
'Does not support rich root data.', target_format)
2110
if not getattr(target_format, 'supports_tree_reference', False):
2111
raise errors.BadConversionTarget(
2112
'Does not support nested trees', target_format)
2114
def get_format_string(self):
2115
"""See RepositoryFormat.get_format_string()."""
2116
return ("Bazaar development format 0 with subtree support "
2117
"(needs bzr.dev from before 1.3)\n")
2119
def get_format_description(self):
2120
"""See RepositoryFormat.get_format_description()."""
2121
return ("Development repository format, currently the same as "
2122
"pack-0.92-subtree\n")
added
removed
bzrlib/repofmt/pack_repo.py
