13
13
# You should have received a copy of the GNU General Public License
14
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20
17
from bzrlib.lazy_import import lazy_import
21
18
lazy_import(globals(), """
22
19
from itertools import izip
25
24
from bzrlib import (
38
29
from bzrlib.index import (
39
33
CombinedGraphIndex,
40
34
GraphIndexPrefixAdapter,
42
from bzrlib.knit import (
48
from bzrlib import tsort
36
from bzrlib.knit import KnitGraphIndex, _PackAccess, _KnitData
37
from bzrlib.pack import ContainerWriter
38
from bzrlib.store import revision
50
40
from bzrlib import (
55
revision as _mod_revision,
58
from bzrlib.decorators import needs_write_lock, only_raises
59
from bzrlib.btree_index import (
63
from bzrlib.index import (
53
from bzrlib.decorators import needs_read_lock, needs_write_lock
67
54
from bzrlib.repofmt.knitrepo import KnitRepository
68
55
from bzrlib.repository import (
70
58
MetaDirRepositoryFormat,
75
from bzrlib.trace import (
61
import bzrlib.revision as _mod_revision
62
from bzrlib.store.revision.knit import KnitRevisionStore
63
from bzrlib.store.versioned import VersionedFileStore
64
from bzrlib.trace import mutter, note, warning
82
67
class PackCommitBuilder(CommitBuilder):
83
68
"""A subclass of CommitBuilder to add texts with pack semantics.
85
70
Specifically this uses one knit object rather than one knit object per
86
71
added text, reducing memory and object pressure.
89
def __init__(self, repository, parents, config, timestamp=None,
90
timezone=None, committer=None, revprops=None,
92
CommitBuilder.__init__(self, repository, parents, config,
93
timestamp=timestamp, timezone=timezone, committer=committer,
94
revprops=revprops, revision_id=revision_id)
95
self._file_graph = graph.Graph(
96
repository._pack_collection.text_index.combined_index)
98
def _heads(self, file_id, revision_ids):
99
keys = [(file_id, revision_id) for revision_id in revision_ids]
100
return set([key[1] for key in self._file_graph.heads(keys)])
74
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
75
return self.repository._packs._add_text_to_weave(file_id,
76
self._new_revision_id, new_lines, parents, nostore_sha,
103
80
class PackRootCommitBuilder(RootCommitBuilder):
104
81
"""A subclass of RootCommitBuilder to add texts with pack semantics.
106
83
Specifically this uses one knit object rather than one knit object per
107
84
added text, reducing memory and object pressure.
110
def __init__(self, repository, parents, config, timestamp=None,
111
timezone=None, committer=None, revprops=None,
113
CommitBuilder.__init__(self, repository, parents, config,
114
timestamp=timestamp, timezone=timezone, committer=committer,
115
revprops=revprops, revision_id=revision_id)
116
self._file_graph = graph.Graph(
117
repository._pack_collection.text_index.combined_index)
119
def _heads(self, file_id, revision_ids):
120
keys = [(file_id, revision_id) for revision_id in revision_ids]
121
return set([key[1] for key in self._file_graph.heads(keys)])
87
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
88
return self.repository._packs._add_text_to_weave(file_id,
89
self._new_revision_id, new_lines, parents, nostore_sha,
124
93
class Pack(object):
125
"""An in memory proxy for a pack and its indices.
127
This is a base class that is not directly used, instead the classes
128
ExistingPack and NewPack are used.
131
# A map of index 'type' to the file extension and position in the
133
index_definitions = {
135
'revision': ('.rix', 0),
136
'inventory': ('.iix', 1),
138
'signature': ('.six', 3),
141
def __init__(self, revision_index, inventory_index, text_index,
142
signature_index, chk_index=None):
143
"""Create a pack instance.
145
:param revision_index: A GraphIndex for determining what revisions are
146
present in the Pack and accessing the locations of their texts.
147
:param inventory_index: A GraphIndex for determining what inventories are
148
present in the Pack and accessing the locations of their
150
:param text_index: A GraphIndex for determining what file texts
151
are present in the pack and accessing the locations of their
152
texts/deltas (via (fileid, revisionid) tuples).
153
:param signature_index: A GraphIndex for determining what signatures are
154
present in the Pack and accessing the locations of their texts.
155
:param chk_index: A GraphIndex for accessing content by CHK, if the
158
self.revision_index = revision_index
159
self.inventory_index = inventory_index
160
self.text_index = text_index
161
self.signature_index = signature_index
162
self.chk_index = chk_index
164
def access_tuple(self):
165
"""Return a tuple (transport, name) for the pack content."""
166
return self.pack_transport, self.file_name()
168
def _check_references(self):
169
"""Make sure our external references are present.
171
Packs are allowed to have deltas whose base is not in the pack, but it
172
must be present somewhere in this collection. It is not allowed to
173
have deltas based on a fallback repository.
174
(See <https://bugs.launchpad.net/bzr/+bug/288751>)
177
for (index_name, external_refs, index) in [
179
self._get_external_refs(self.text_index),
180
self._pack_collection.text_index.combined_index),
182
self._get_external_refs(self.inventory_index),
183
self._pack_collection.inventory_index.combined_index),
185
missing = external_refs.difference(
186
k for (idx, k, v, r) in
187
index.iter_entries(external_refs))
189
missing_items[index_name] = sorted(list(missing))
191
from pprint import pformat
192
raise errors.BzrCheckError(
193
"Newly created pack file %r has delta references to "
194
"items not in its repository:\n%s"
195
% (self, pformat(missing_items)))
198
"""Get the file name for the pack on disk."""
199
return self.name + '.pack'
94
"""An in memory proxy for a .pack and its indices."""
97
self.revision_index = None
98
self.inventory_index = None
99
self.text_index = None
100
self.signature_index = None
102
self.transport = None
201
104
def get_revision_count(self):
202
105
return self.revision_index.key_count()
204
def index_name(self, index_type, name):
205
"""Get the disk name of an index type for pack name 'name'."""
206
return name + Pack.index_definitions[index_type][0]
208
def index_offset(self, index_type):
209
"""Get the position in a index_size array for a given index type."""
210
return Pack.index_definitions[index_type][1]
212
def inventory_index_name(self, name):
213
"""The inv index is the name + .iix."""
214
return self.index_name('inventory', name)
216
def revision_index_name(self, name):
217
"""The revision index is the name + .rix."""
218
return self.index_name('revision', name)
220
def signature_index_name(self, name):
221
"""The signature index is the name + .six."""
222
return self.index_name('signature', name)
224
def text_index_name(self, name):
225
"""The text index is the name + .tix."""
226
return self.index_name('text', name)
228
def _replace_index_with_readonly(self, index_type):
229
unlimited_cache = False
230
if index_type == 'chk':
231
unlimited_cache = True
232
setattr(self, index_type + '_index',
233
self.index_class(self.index_transport,
234
self.index_name(index_type, self.name),
235
self.index_sizes[self.index_offset(index_type)],
236
unlimited_cache=unlimited_cache))
239
class ExistingPack(Pack):
240
"""An in memory proxy for an existing .pack and its disk indices."""
242
def __init__(self, pack_transport, name, revision_index, inventory_index,
243
text_index, signature_index, chk_index=None):
244
"""Create an ExistingPack object.
246
:param pack_transport: The transport where the pack file resides.
247
:param name: The name of the pack on disk in the pack_transport.
249
Pack.__init__(self, revision_index, inventory_index, text_index,
250
signature_index, chk_index)
252
self.pack_transport = pack_transport
253
if None in (revision_index, inventory_index, text_index,
254
signature_index, name, pack_transport):
255
raise AssertionError()
257
def __eq__(self, other):
258
return self.__dict__ == other.__dict__
260
def __ne__(self, other):
261
return not self.__eq__(other)
264
return "<%s.%s object at 0x%x, %s, %s" % (
265
self.__class__.__module__, self.__class__.__name__, id(self),
266
self.pack_transport, self.name)
269
class ResumedPack(ExistingPack):
271
def __init__(self, name, revision_index, inventory_index, text_index,
272
signature_index, upload_transport, pack_transport, index_transport,
273
pack_collection, chk_index=None):
274
"""Create a ResumedPack object."""
275
ExistingPack.__init__(self, pack_transport, name, revision_index,
276
inventory_index, text_index, signature_index,
278
self.upload_transport = upload_transport
279
self.index_transport = index_transport
280
self.index_sizes = [None, None, None, None]
282
('revision', revision_index),
283
('inventory', inventory_index),
284
('text', text_index),
285
('signature', signature_index),
287
if chk_index is not None:
288
indices.append(('chk', chk_index))
289
self.index_sizes.append(None)
290
for index_type, index in indices:
291
offset = self.index_offset(index_type)
292
self.index_sizes[offset] = index._size
293
self.index_class = pack_collection._index_class
294
self._pack_collection = pack_collection
295
self._state = 'resumed'
296
# XXX: perhaps check that the .pack file exists?
298
def access_tuple(self):
299
if self._state == 'finished':
300
return Pack.access_tuple(self)
301
elif self._state == 'resumed':
302
return self.upload_transport, self.file_name()
304
raise AssertionError(self._state)
307
self.upload_transport.delete(self.file_name())
308
indices = [self.revision_index, self.inventory_index, self.text_index,
309
self.signature_index]
310
if self.chk_index is not None:
311
indices.append(self.chk_index)
312
for index in indices:
313
index._transport.delete(index._name)
316
self._check_references()
317
index_types = ['revision', 'inventory', 'text', 'signature']
318
if self.chk_index is not None:
319
index_types.append('chk')
320
for index_type in index_types:
321
old_name = self.index_name(index_type, self.name)
322
new_name = '../indices/' + old_name
323
self.upload_transport.rename(old_name, new_name)
324
self._replace_index_with_readonly(index_type)
325
new_name = '../packs/' + self.file_name()
326
self.upload_transport.rename(self.file_name(), new_name)
327
self._state = 'finished'
329
def _get_external_refs(self, index):
330
"""Return compression parents for this index that are not present.
332
This returns any compression parents that are referenced by this index,
333
which are not contained *in* this index. They may be present elsewhere.
335
return index.external_references(1)
339
"""An in memory proxy for a pack which is being created."""
341
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
342
"""Create a NewPack instance.
344
:param pack_collection: A PackCollection into which this is being inserted.
345
:param upload_suffix: An optional suffix to be given to any temporary
346
files created during the pack creation. e.g '.autopack'
347
:param file_mode: Unix permissions for newly created file.
349
# The relative locations of the packs are constrained, but all are
350
# passed in because the caller has them, so as to avoid object churn.
351
index_builder_class = pack_collection._index_builder_class
352
if pack_collection.chk_index is not None:
353
chk_index = index_builder_class(reference_lists=0)
357
# Revisions: parents list, no text compression.
358
index_builder_class(reference_lists=1),
359
# Inventory: We want to map compression only, but currently the
360
# knit code hasn't been updated enough to understand that, so we
361
# have a regular 2-list index giving parents and compression
363
index_builder_class(reference_lists=2),
364
# Texts: compression and per file graph, for all fileids - so two
365
# reference lists and two elements in the key tuple.
366
index_builder_class(reference_lists=2, key_elements=2),
367
# Signatures: Just blobs to store, no compression, no parents
369
index_builder_class(reference_lists=0),
370
# CHK based storage - just blobs, no compression or parents.
373
self._pack_collection = pack_collection
374
# When we make readonly indices, we need this.
375
self.index_class = pack_collection._index_class
376
# where should the new pack be opened
377
self.upload_transport = pack_collection._upload_transport
378
# where are indices written out to
379
self.index_transport = pack_collection._index_transport
380
# where is the pack renamed to when it is finished?
381
self.pack_transport = pack_collection._pack_transport
382
# What file mode to upload the pack and indices with.
383
self._file_mode = file_mode
384
# tracks the content written to the .pack file.
385
self._hash = osutils.md5()
386
# a tuple with the length in bytes of the indices, once the pack
387
# is finalised. (rev, inv, text, sigs, chk_if_in_use)
388
self.index_sizes = None
389
# How much data to cache when writing packs. Note that this is not
390
# synchronised with reads, because it's not in the transport layer, so
391
# is not safe unless the client knows it won't be reading from the pack
393
self._cache_limit = 0
394
# the temporary pack file name.
395
self.random_name = osutils.rand_chars(20) + upload_suffix
396
# when was this pack started ?
397
self.start_time = time.time()
398
# open an output stream for the data added to the pack.
399
self.write_stream = self.upload_transport.open_write_stream(
400
self.random_name, mode=self._file_mode)
401
if 'pack' in debug.debug_flags:
402
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
403
time.ctime(), self.upload_transport.base, self.random_name,
404
time.time() - self.start_time)
405
# A list of byte sequences to be written to the new pack, and the
406
# aggregate size of them. Stored as a list rather than separate
407
# variables so that the _write_data closure below can update them.
408
self._buffer = [[], 0]
409
# create a callable for adding data
411
# robertc says- this is a closure rather than a method on the object
412
# so that the variables are locals, and faster than accessing object
414
def _write_data(bytes, flush=False, _buffer=self._buffer,
415
_write=self.write_stream.write, _update=self._hash.update):
416
_buffer[0].append(bytes)
417
_buffer[1] += len(bytes)
419
if _buffer[1] > self._cache_limit or flush:
420
bytes = ''.join(_buffer[0])
424
# expose this on self, for the occasion when clients want to add data.
425
self._write_data = _write_data
426
# a pack writer object to serialise pack records.
427
self._writer = pack.ContainerWriter(self._write_data)
429
# what state is the pack in? (open, finished, aborted)
431
# no name until we finish writing the content
435
"""Cancel creating this pack."""
436
self._state = 'aborted'
437
self.write_stream.close()
438
# Remove the temporary pack file.
439
self.upload_transport.delete(self.random_name)
440
# The indices have no state on disk.
442
def access_tuple(self):
443
"""Return a tuple (transport, name) for the pack content."""
444
if self._state == 'finished':
445
return Pack.access_tuple(self)
446
elif self._state == 'open':
447
return self.upload_transport, self.random_name
449
raise AssertionError(self._state)
451
def data_inserted(self):
452
"""True if data has been added to this pack."""
453
return bool(self.get_revision_count() or
454
self.inventory_index.key_count() or
455
self.text_index.key_count() or
456
self.signature_index.key_count() or
457
(self.chk_index is not None and self.chk_index.key_count()))
459
def finish_content(self):
460
if self.name is not None:
464
self._write_data('', flush=True)
465
self.name = self._hash.hexdigest()
467
def finish(self, suspend=False):
468
"""Finish the new pack.
471
- finalises the content
472
- assigns a name (the md5 of the content, currently)
473
- writes out the associated indices
474
- renames the pack into place.
475
- stores the index size tuple for the pack in the index_sizes
478
self.finish_content()
480
self._check_references()
482
# XXX: It'd be better to write them all to temporary names, then
483
# rename them all into place, so that the window when only some are
484
# visible is smaller. On the other hand none will be seen until
485
# they're in the names list.
486
self.index_sizes = [None, None, None, None]
487
self._write_index('revision', self.revision_index, 'revision', suspend)
488
self._write_index('inventory', self.inventory_index, 'inventory',
490
self._write_index('text', self.text_index, 'file texts', suspend)
491
self._write_index('signature', self.signature_index,
492
'revision signatures', suspend)
493
if self.chk_index is not None:
494
self.index_sizes.append(None)
495
self._write_index('chk', self.chk_index,
496
'content hash bytes', suspend)
497
self.write_stream.close()
498
# Note that this will clobber an existing pack with the same name,
499
# without checking for hash collisions. While this is undesirable this
500
# is something that can be rectified in a subsequent release. One way
501
# to rectify it may be to leave the pack at the original name, writing
502
# its pack-names entry as something like 'HASH: index-sizes
503
# temporary-name'. Allocate that and check for collisions, if it is
504
# collision free then rename it into place. If clients know this scheme
505
# they can handle missing-file errors by:
506
# - try for HASH.pack
507
# - try for temporary-name
508
# - refresh the pack-list to see if the pack is now absent
509
new_name = self.name + '.pack'
511
new_name = '../packs/' + new_name
512
self.upload_transport.rename(self.random_name, new_name)
513
self._state = 'finished'
514
if 'pack' in debug.debug_flags:
515
# XXX: size might be interesting?
516
mutter('%s: create_pack: pack finished: %s%s->%s t+%6.3fs',
517
time.ctime(), self.upload_transport.base, self.random_name,
518
new_name, time.time() - self.start_time)
521
"""Flush any current data."""
523
bytes = ''.join(self._buffer[0])
524
self.write_stream.write(bytes)
525
self._hash.update(bytes)
526
self._buffer[:] = [[], 0]
528
def _get_external_refs(self, index):
529
return index._external_references()
531
def set_write_cache_size(self, size):
532
self._cache_limit = size
534
def _write_index(self, index_type, index, label, suspend=False):
535
"""Write out an index.
537
:param index_type: The type of index to write - e.g. 'revision'.
538
:param index: The index object to serialise.
539
:param label: What label to give the index e.g. 'revision'.
541
index_name = self.index_name(index_type, self.name)
543
transport = self.upload_transport
545
transport = self.index_transport
546
self.index_sizes[self.index_offset(index_type)] = transport.put_file(
547
index_name, index.finish(), mode=self._file_mode)
548
if 'pack' in debug.debug_flags:
549
# XXX: size might be interesting?
550
mutter('%s: create_pack: wrote %s index: %s%s t+%6.3fs',
551
time.ctime(), label, self.upload_transport.base,
552
self.random_name, time.time() - self.start_time)
553
# Replace the writable index on this object with a readonly,
554
# presently unloaded index. We should alter
555
# the index layer to make its finish() error if add_node is
556
# subsequently used. RBC
557
self._replace_index_with_readonly(index_type)
560
class AggregateIndex(object):
561
"""An aggregated index for the RepositoryPackCollection.
563
AggregateIndex is reponsible for managing the PackAccess object,
564
Index-To-Pack mapping, and all indices list for a specific type of index
565
such as 'revision index'.
567
A CombinedIndex provides an index on a single key space built up
568
from several on-disk indices. The AggregateIndex builds on this
569
to provide a knit access layer, and allows having up to one writable
570
index within the collection.
572
# XXX: Probably 'can be written to' could/should be separated from 'acts
573
# like a knit index' -- mbp 20071024
575
def __init__(self, reload_func=None, flush_func=None):
576
"""Create an AggregateIndex.
578
:param reload_func: A function to call if we find we are missing an
579
index. Should have the form reload_func() => True if the list of
580
active pack files has changed.
582
self._reload_func = reload_func
583
self.index_to_pack = {}
584
self.combined_index = CombinedGraphIndex([], reload_func=reload_func)
585
self.data_access = _DirectPackAccess(self.index_to_pack,
586
reload_func=reload_func,
587
flush_func=flush_func)
588
self.add_callback = None
590
def add_index(self, index, pack):
591
"""Add index to the aggregate, which is an index for Pack pack.
593
Future searches on the aggregate index will seach this new index
594
before all previously inserted indices.
596
:param index: An Index for the pack.
597
:param pack: A Pack instance.
599
# expose it to the index map
600
self.index_to_pack[index] = pack.access_tuple()
601
# put it at the front of the linear index list
602
self.combined_index.insert_index(0, index, pack.name)
604
def add_writable_index(self, index, pack):
605
"""Add an index which is able to have data added to it.
607
There can be at most one writable index at any time. Any
608
modifications made to the knit are put into this index.
610
:param index: An index from the pack parameter.
611
:param pack: A Pack instance.
613
if self.add_callback is not None:
614
raise AssertionError(
615
"%s already has a writable index through %s" % \
616
(self, self.add_callback))
617
# allow writing: queue writes to a new index
618
self.add_index(index, pack)
619
# Updates the index to packs mapping as a side effect,
620
self.data_access.set_writer(pack._writer, index, pack.access_tuple())
621
self.add_callback = index.add_nodes
624
"""Reset all the aggregate data to nothing."""
625
self.data_access.set_writer(None, None, (None, None))
626
self.index_to_pack.clear()
627
del self.combined_index._indices[:]
628
del self.combined_index._index_names[:]
629
self.add_callback = None
631
def remove_index(self, index):
632
"""Remove index from the indices used to answer queries.
634
:param index: An index from the pack parameter.
636
del self.index_to_pack[index]
637
pos = self.combined_index._indices.index(index)
638
del self.combined_index._indices[pos]
639
del self.combined_index._index_names[pos]
640
if (self.add_callback is not None and
641
getattr(index, 'add_nodes', None) == self.add_callback):
642
self.add_callback = None
643
self.data_access.set_writer(None, None, (None, None))
646
class Packer(object):
647
"""Create a pack from packs."""
649
def __init__(self, pack_collection, packs, suffix, revision_ids=None,
653
:param pack_collection: A RepositoryPackCollection object where the
654
new pack is being written to.
655
:param packs: The packs to combine.
656
:param suffix: The suffix to use on the temporary files for the pack.
657
:param revision_ids: Revision ids to limit the pack to.
658
:param reload_func: A function to call if a pack file/index goes
659
missing. The side effect of calling this function should be to
660
update self.packs. See also AggregateIndex
664
self.revision_ids = revision_ids
665
# The pack object we are creating.
667
self._pack_collection = pack_collection
668
self._reload_func = reload_func
669
# The index layer keys for the revisions being copied. None for 'all
671
self._revision_keys = None
672
# What text keys to copy. None for 'all texts'. This is set by
673
# _copy_inventory_texts
674
self._text_filter = None
677
def _extra_init(self):
678
"""A template hook to allow extending the constructor trivially."""
680
def _pack_map_and_index_list(self, index_attribute):
681
"""Convert a list of packs to an index pack map and index list.
683
:param index_attribute: The attribute that the desired index is found
685
:return: A tuple (map, list) where map contains the dict from
686
index:pack_tuple, and list contains the indices in the preferred
691
for pack_obj in self.packs:
692
index = getattr(pack_obj, index_attribute)
693
indices.append(index)
694
pack_map[index] = pack_obj
695
return pack_map, indices
697
def _index_contents(self, indices, key_filter=None):
698
"""Get an iterable of the index contents from a pack_map.
700
:param indices: The list of indices to query
701
:param key_filter: An optional filter to limit the keys returned.
703
all_index = CombinedGraphIndex(indices)
704
if key_filter is None:
705
return all_index.iter_all_entries()
707
return all_index.iter_entries(key_filter)
709
def pack(self, pb=None):
710
"""Create a new pack by reading data from other packs.
712
This does little more than a bulk copy of data. One key difference
713
is that data with the same item key across multiple packs is elided
714
from the output. The new pack is written into the current pack store
715
along with its indices, and the name added to the pack names. The
716
source packs are not altered and are not required to be in the current
719
:param pb: An optional progress bar to use. A nested bar is created if
721
:return: A Pack object, or None if nothing was copied.
723
# open a pack - using the same name as the last temporary file
724
# - which has already been flushed, so its safe.
725
# XXX: - duplicate code warning with start_write_group; fix before
726
# considering 'done'.
727
if self._pack_collection._new_pack is not None:
728
raise errors.BzrError('call to %s.pack() while another pack is'
730
% (self.__class__.__name__,))
731
if self.revision_ids is not None:
732
if len(self.revision_ids) == 0:
733
# silly fetch request.
736
self.revision_ids = frozenset(self.revision_ids)
737
self.revision_keys = frozenset((revid,) for revid in
740
self.pb = ui.ui_factory.nested_progress_bar()
744
return self._create_pack_from_packs()
750
"""Open a pack for the pack we are creating."""
751
new_pack = self._pack_collection.pack_factory(self._pack_collection,
752
upload_suffix=self.suffix,
753
file_mode=self._pack_collection.repo.bzrdir._get_file_mode())
754
# We know that we will process all nodes in order, and don't need to
755
# query, so don't combine any indices spilled to disk until we are done
756
new_pack.revision_index.set_optimize(combine_backing_indices=False)
757
new_pack.inventory_index.set_optimize(combine_backing_indices=False)
758
new_pack.text_index.set_optimize(combine_backing_indices=False)
759
new_pack.signature_index.set_optimize(combine_backing_indices=False)
762
def _update_pack_order(self, entries, index_to_pack_map):
763
"""Determine how we want our packs to be ordered.
765
This changes the sort order of the self.packs list so that packs unused
766
by 'entries' will be at the end of the list, so that future requests
767
can avoid probing them. Used packs will be at the front of the
768
self.packs list, in the order of their first use in 'entries'.
770
:param entries: A list of (index, ...) tuples
771
:param index_to_pack_map: A mapping from index objects to pack objects.
775
for entry in entries:
777
if index not in seen_indexes:
778
packs.append(index_to_pack_map[index])
779
seen_indexes.add(index)
780
if len(packs) == len(self.packs):
781
if 'pack' in debug.debug_flags:
782
mutter('Not changing pack list, all packs used.')
784
seen_packs = set(packs)
785
for pack in self.packs:
786
if pack not in seen_packs:
789
if 'pack' in debug.debug_flags:
790
old_names = [p.access_tuple()[1] for p in self.packs]
791
new_names = [p.access_tuple()[1] for p in packs]
792
mutter('Reordering packs\nfrom: %s\n to: %s',
793
old_names, new_names)
796
def _copy_revision_texts(self):
797
"""Copy revision data to the new pack."""
799
if self.revision_ids:
800
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
803
# select revision keys
804
revision_index_map, revision_indices = self._pack_map_and_index_list(
806
revision_nodes = self._index_contents(revision_indices, revision_keys)
807
revision_nodes = list(revision_nodes)
808
self._update_pack_order(revision_nodes, revision_index_map)
809
# copy revision keys and adjust values
810
self.pb.update("Copying revision texts", 1)
811
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
812
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
813
self.new_pack.revision_index, readv_group_iter, total_items))
814
if 'pack' in debug.debug_flags:
815
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
816
time.ctime(), self._pack_collection._upload_transport.base,
817
self.new_pack.random_name,
818
self.new_pack.revision_index.key_count(),
819
time.time() - self.new_pack.start_time)
820
self._revision_keys = revision_keys
822
def _copy_inventory_texts(self):
823
"""Copy the inventory texts to the new pack.
825
self._revision_keys is used to determine what inventories to copy.
827
Sets self._text_filter appropriately.
829
# select inventory keys
830
inv_keys = self._revision_keys # currently the same keyspace, and note that
831
# querying for keys here could introduce a bug where an inventory item
832
# is missed, so do not change it to query separately without cross
833
# checking like the text key check below.
834
inventory_index_map, inventory_indices = self._pack_map_and_index_list(
836
inv_nodes = self._index_contents(inventory_indices, inv_keys)
837
# copy inventory keys and adjust values
838
# XXX: Should be a helper function to allow different inv representation
840
self.pb.update("Copying inventory texts", 2)
841
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
842
# Only grab the output lines if we will be processing them
843
output_lines = bool(self.revision_ids)
844
inv_lines = self._copy_nodes_graph(inventory_index_map,
845
self.new_pack._writer, self.new_pack.inventory_index,
846
readv_group_iter, total_items, output_lines=output_lines)
847
if self.revision_ids:
848
self._process_inventory_lines(inv_lines)
850
# eat the iterator to cause it to execute.
852
self._text_filter = None
853
if 'pack' in debug.debug_flags:
854
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
855
time.ctime(), self._pack_collection._upload_transport.base,
856
self.new_pack.random_name,
857
self.new_pack.inventory_index.key_count(),
858
time.time() - self.new_pack.start_time)
860
def _copy_text_texts(self):
862
text_index_map, text_nodes = self._get_text_nodes()
863
if self._text_filter is not None:
864
# We could return the keys copied as part of the return value from
865
# _copy_nodes_graph but this doesn't work all that well with the
866
# need to get line output too, so we check separately, and as we're
867
# going to buffer everything anyway, we check beforehand, which
868
# saves reading knit data over the wire when we know there are
870
text_nodes = set(text_nodes)
871
present_text_keys = set(_node[1] for _node in text_nodes)
872
missing_text_keys = set(self._text_filter) - present_text_keys
873
if missing_text_keys:
874
# TODO: raise a specific error that can handle many missing
876
mutter("missing keys during fetch: %r", missing_text_keys)
877
a_missing_key = missing_text_keys.pop()
878
raise errors.RevisionNotPresent(a_missing_key[1],
880
# copy text keys and adjust values
881
self.pb.update("Copying content texts", 3)
882
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
883
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
884
self.new_pack.text_index, readv_group_iter, total_items))
885
self._log_copied_texts()
887
def _create_pack_from_packs(self):
888
self.pb.update("Opening pack", 0, 5)
889
self.new_pack = self.open_pack()
890
new_pack = self.new_pack
891
# buffer data - we won't be reading-back during the pack creation and
892
# this makes a significant difference on sftp pushes.
893
new_pack.set_write_cache_size(1024*1024)
894
if 'pack' in debug.debug_flags:
895
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
896
for a_pack in self.packs]
897
if self.revision_ids is not None:
898
rev_count = len(self.revision_ids)
901
mutter('%s: create_pack: creating pack from source packs: '
902
'%s%s %s revisions wanted %s t=0',
903
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
904
plain_pack_list, rev_count)
905
self._copy_revision_texts()
906
self._copy_inventory_texts()
907
self._copy_text_texts()
908
# select signature keys
909
signature_filter = self._revision_keys # same keyspace
910
signature_index_map, signature_indices = self._pack_map_and_index_list(
912
signature_nodes = self._index_contents(signature_indices,
914
# copy signature keys and adjust values
915
self.pb.update("Copying signature texts", 4)
916
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
917
new_pack.signature_index)
918
if 'pack' in debug.debug_flags:
919
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
920
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
921
new_pack.signature_index.key_count(),
922
time.time() - new_pack.start_time)
924
# NB XXX: how to check CHK references are present? perhaps by yielding
925
# the items? How should that interact with stacked repos?
926
if new_pack.chk_index is not None:
928
if 'pack' in debug.debug_flags:
929
mutter('%s: create_pack: chk content copied: %s%s %d items t+%6.3fs',
930
time.ctime(), self._pack_collection._upload_transport.base,
931
new_pack.random_name,
932
new_pack.chk_index.key_count(),
933
time.time() - new_pack.start_time)
934
new_pack._check_references()
935
if not self._use_pack(new_pack):
938
self.pb.update("Finishing pack", 5)
940
self._pack_collection.allocate(new_pack)
943
def _copy_chks(self, refs=None):
944
# XXX: Todo, recursive follow-pointers facility when fetching some
946
chk_index_map, chk_indices = self._pack_map_and_index_list(
948
chk_nodes = self._index_contents(chk_indices, refs)
950
# TODO: This isn't strictly tasteful as we are accessing some private
951
# variables (_serializer). Perhaps a better way would be to have
952
# Repository._deserialise_chk_node()
953
search_key_func = chk_map.search_key_registry.get(
954
self._pack_collection.repo._serializer.search_key_name)
955
def accumlate_refs(lines):
956
# XXX: move to a generic location
958
bytes = ''.join(lines)
959
node = chk_map._deserialise(bytes, ("unknown",), search_key_func)
960
new_refs.update(node.refs())
961
self._copy_nodes(chk_nodes, chk_index_map, self.new_pack._writer,
962
self.new_pack.chk_index, output_lines=accumlate_refs)
965
def _copy_nodes(self, nodes, index_map, writer, write_index,
967
"""Copy knit nodes between packs with no graph references.
969
:param output_lines: Output full texts of copied items.
971
pb = ui.ui_factory.nested_progress_bar()
973
return self._do_copy_nodes(nodes, index_map, writer,
974
write_index, pb, output_lines=output_lines)
978
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb,
980
# for record verification
981
knit = KnitVersionedFiles(None, None)
982
# plan a readv on each source pack:
984
nodes = sorted(nodes)
985
# how to map this into knit.py - or knit.py into this?
986
# we don't want the typical knit logic, we want grouping by pack
987
# at this point - perhaps a helper library for the following code
988
# duplication points?
990
for index, key, value in nodes:
991
if index not in request_groups:
992
request_groups[index] = []
993
request_groups[index].append((key, value))
995
pb.update("Copied record", record_index, len(nodes))
996
for index, items in request_groups.iteritems():
997
pack_readv_requests = []
998
for key, value in items:
999
# ---- KnitGraphIndex.get_position
1000
bits = value[1:].split(' ')
1001
offset, length = int(bits[0]), int(bits[1])
1002
pack_readv_requests.append((offset, length, (key, value[0])))
1003
# linear scan up the pack
1004
pack_readv_requests.sort()
1006
pack_obj = index_map[index]
1007
transport, path = pack_obj.access_tuple()
1009
reader = pack.make_readv_reader(transport, path,
1010
[offset[0:2] for offset in pack_readv_requests])
1011
except errors.NoSuchFile:
1012
if self._reload_func is not None:
1015
for (names, read_func), (_1, _2, (key, eol_flag)) in \
1016
izip(reader.iter_records(), pack_readv_requests):
1017
raw_data = read_func(None)
1018
# check the header only
1019
if output_lines is not None:
1020
output_lines(knit._parse_record(key[-1], raw_data)[0])
1022
df, _ = knit._parse_record_header(key, raw_data)
1024
pos, size = writer.add_bytes_record(raw_data, names)
1025
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
1026
pb.update("Copied record", record_index)
1029
def _copy_nodes_graph(self, index_map, writer, write_index,
1030
readv_group_iter, total_items, output_lines=False):
1031
"""Copy knit nodes between packs.
1033
:param output_lines: Return lines present in the copied data as
1034
an iterator of line,version_id.
1036
pb = ui.ui_factory.nested_progress_bar()
1038
for result in self._do_copy_nodes_graph(index_map, writer,
1039
write_index, output_lines, pb, readv_group_iter, total_items):
1042
# Python 2.4 does not permit try:finally: in a generator.
1048
def _do_copy_nodes_graph(self, index_map, writer, write_index,
1049
output_lines, pb, readv_group_iter, total_items):
1050
# for record verification
1051
knit = KnitVersionedFiles(None, None)
1052
# for line extraction when requested (inventories only)
1054
factory = KnitPlainFactory()
1056
pb.update("Copied record", record_index, total_items)
1057
for index, readv_vector, node_vector in readv_group_iter:
1059
pack_obj = index_map[index]
1060
transport, path = pack_obj.access_tuple()
1062
reader = pack.make_readv_reader(transport, path, readv_vector)
1063
except errors.NoSuchFile:
1064
if self._reload_func is not None:
1067
for (names, read_func), (key, eol_flag, references) in \
1068
izip(reader.iter_records(), node_vector):
1069
raw_data = read_func(None)
1071
# read the entire thing
1072
content, _ = knit._parse_record(key[-1], raw_data)
1073
if len(references[-1]) == 0:
1074
line_iterator = factory.get_fulltext_content(content)
1076
line_iterator = factory.get_linedelta_content(content)
1077
for line in line_iterator:
1080
# check the header only
1081
df, _ = knit._parse_record_header(key, raw_data)
1083
pos, size = writer.add_bytes_record(raw_data, names)
1084
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1085
pb.update("Copied record", record_index)
1088
def _get_text_nodes(self):
1089
text_index_map, text_indices = self._pack_map_and_index_list(
1091
return text_index_map, self._index_contents(text_indices,
1094
def _least_readv_node_readv(self, nodes):
1095
"""Generate request groups for nodes using the least readv's.
1097
:param nodes: An iterable of graph index nodes.
1098
:return: Total node count and an iterator of the data needed to perform
1099
readvs to obtain the data for nodes. Each item yielded by the
1100
iterator is a tuple with:
1101
index, readv_vector, node_vector. readv_vector is a list ready to
1102
hand to the transport readv method, and node_vector is a list of
1103
(key, eol_flag, references) for the node retrieved by the
1104
matching readv_vector.
1106
# group by pack so we do one readv per pack
1107
nodes = sorted(nodes)
1110
for index, key, value, references in nodes:
1111
if index not in request_groups:
1112
request_groups[index] = []
1113
request_groups[index].append((key, value, references))
1115
for index, items in request_groups.iteritems():
1116
pack_readv_requests = []
1117
for key, value, references in items:
1118
# ---- KnitGraphIndex.get_position
1119
bits = value[1:].split(' ')
1120
offset, length = int(bits[0]), int(bits[1])
1121
pack_readv_requests.append(
1122
((offset, length), (key, value[0], references)))
1123
# linear scan up the pack to maximum range combining.
1124
pack_readv_requests.sort()
1125
# split out the readv and the node data.
1126
pack_readv = [readv for readv, node in pack_readv_requests]
1127
node_vector = [node for readv, node in pack_readv_requests]
1128
result.append((index, pack_readv, node_vector))
1129
return total, result
1131
def _log_copied_texts(self):
1132
if 'pack' in debug.debug_flags:
1133
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
1134
time.ctime(), self._pack_collection._upload_transport.base,
1135
self.new_pack.random_name,
1136
self.new_pack.text_index.key_count(),
1137
time.time() - self.new_pack.start_time)
1139
def _process_inventory_lines(self, inv_lines):
1140
"""Use up the inv_lines generator and setup a text key filter."""
1141
repo = self._pack_collection.repo
1142
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
1143
inv_lines, self.revision_keys)
1145
for fileid, file_revids in fileid_revisions.iteritems():
1146
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
1147
self._text_filter = text_filter
1149
def _revision_node_readv(self, revision_nodes):
1150
"""Return the total revisions and the readv's to issue.
1152
:param revision_nodes: The revision index contents for the packs being
1153
incorporated into the new pack.
1154
:return: As per _least_readv_node_readv.
1156
return self._least_readv_node_readv(revision_nodes)
1158
def _use_pack(self, new_pack):
1159
"""Return True if new_pack should be used.
1161
:param new_pack: The pack that has just been created.
1162
:return: True if the pack should be used.
1164
return new_pack.data_inserted()
1167
class OptimisingPacker(Packer):
1168
"""A packer which spends more time to create better disk layouts."""
1170
def _revision_node_readv(self, revision_nodes):
1171
"""Return the total revisions and the readv's to issue.
1173
This sort places revisions in topological order with the ancestors
1176
:param revision_nodes: The revision index contents for the packs being
1177
incorporated into the new pack.
1178
:return: As per _least_readv_node_readv.
1180
# build an ancestors dict
1183
for index, key, value, references in revision_nodes:
1184
ancestors[key] = references[0]
1185
by_key[key] = (index, value, references)
1186
order = tsort.topo_sort(ancestors)
1188
# Single IO is pathological, but it will work as a starting point.
1190
for key in reversed(order):
1191
index, value, references = by_key[key]
1192
# ---- KnitGraphIndex.get_position
1193
bits = value[1:].split(' ')
1194
offset, length = int(bits[0]), int(bits[1])
1196
(index, [(offset, length)], [(key, value[0], references)]))
1197
# TODO: combine requests in the same index that are in ascending order.
1198
return total, requests
1200
def open_pack(self):
1201
"""Open a pack for the pack we are creating."""
1202
new_pack = super(OptimisingPacker, self).open_pack()
1203
# Turn on the optimization flags for all the index builders.
1204
new_pack.revision_index.set_optimize(for_size=True)
1205
new_pack.inventory_index.set_optimize(for_size=True)
1206
new_pack.text_index.set_optimize(for_size=True)
1207
new_pack.signature_index.set_optimize(for_size=True)
1211
class ReconcilePacker(Packer):
1212
"""A packer which regenerates indices etc as it copies.
1214
This is used by ``bzr reconcile`` to cause parent text pointers to be
1218
def _extra_init(self):
1219
self._data_changed = False
1221
def _process_inventory_lines(self, inv_lines):
1222
"""Generate a text key reference map rather for reconciling with."""
1223
repo = self._pack_collection.repo
1224
refs = repo._find_text_key_references_from_xml_inventory_lines(
1226
self._text_refs = refs
1227
# during reconcile we:
1228
# - convert unreferenced texts to full texts
1229
# - correct texts which reference a text not copied to be full texts
1230
# - copy all others as-is but with corrected parents.
1231
# - so at this point we don't know enough to decide what becomes a full
1233
self._text_filter = None
1235
def _copy_text_texts(self):
1236
"""generate what texts we should have and then copy."""
1237
self.pb.update("Copying content texts", 3)
1238
# we have three major tasks here:
1239
# 1) generate the ideal index
1240
repo = self._pack_collection.repo
1241
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1242
_1, key, _2, refs in
1243
self.new_pack.revision_index.iter_all_entries()])
1244
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1245
# 2) generate a text_nodes list that contains all the deltas that can
1246
# be used as-is, with corrected parents.
1249
discarded_nodes = []
1250
NULL_REVISION = _mod_revision.NULL_REVISION
1251
text_index_map, text_nodes = self._get_text_nodes()
1252
for node in text_nodes:
1258
ideal_parents = tuple(ideal_index[node[1]])
1260
discarded_nodes.append(node)
1261
self._data_changed = True
1263
if ideal_parents == (NULL_REVISION,):
1265
if ideal_parents == node[3][0]:
1267
ok_nodes.append(node)
1268
elif ideal_parents[0:1] == node[3][0][0:1]:
1269
# the left most parent is the same, or there are no parents
1270
# today. Either way, we can preserve the representation as
1271
# long as we change the refs to be inserted.
1272
self._data_changed = True
1273
ok_nodes.append((node[0], node[1], node[2],
1274
(ideal_parents, node[3][1])))
1275
self._data_changed = True
1277
# Reinsert this text completely
1278
bad_texts.append((node[1], ideal_parents))
1279
self._data_changed = True
1280
# we're finished with some data.
1283
# 3) bulk copy the ok data
1284
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1285
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1286
self.new_pack.text_index, readv_group_iter, total_items))
1287
# 4) adhoc copy all the other texts.
1288
# We have to topologically insert all texts otherwise we can fail to
1289
# reconcile when parts of a single delta chain are preserved intact,
1290
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1291
# reinserted, and if d3 has incorrect parents it will also be
1292
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1293
# copied), so we will try to delta, but d2 is not currently able to be
1294
# extracted because it's basis d1 is not present. Topologically sorting
1295
# addresses this. The following generates a sort for all the texts that
1296
# are being inserted without having to reference the entire text key
1297
# space (we only topo sort the revisions, which is smaller).
1298
topo_order = tsort.topo_sort(ancestors)
1299
rev_order = dict(zip(topo_order, range(len(topo_order))))
1300
bad_texts.sort(key=lambda key:rev_order.get(key[0][1], 0))
1301
transaction = repo.get_transaction()
1302
file_id_index = GraphIndexPrefixAdapter(
1303
self.new_pack.text_index,
1305
add_nodes_callback=self.new_pack.text_index.add_nodes)
1306
data_access = _DirectPackAccess(
1307
{self.new_pack.text_index:self.new_pack.access_tuple()})
1308
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1309
self.new_pack.access_tuple())
1310
output_texts = KnitVersionedFiles(
1311
_KnitGraphIndex(self.new_pack.text_index,
1312
add_callback=self.new_pack.text_index.add_nodes,
1313
deltas=True, parents=True, is_locked=repo.is_locked),
1314
data_access=data_access, max_delta_chain=200)
1315
for key, parent_keys in bad_texts:
1316
# We refer to the new pack to delta data being output.
1317
# A possible improvement would be to catch errors on short reads
1318
# and only flush then.
1319
self.new_pack.flush()
1321
for parent_key in parent_keys:
1322
if parent_key[0] != key[0]:
1323
# Graph parents must match the fileid
1324
raise errors.BzrError('Mismatched key parent %r:%r' %
1326
parents.append(parent_key[1])
1327
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1328
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1329
output_texts.add_lines(key, parent_keys, text_lines,
1330
random_id=True, check_content=False)
1331
# 5) check that nothing inserted has a reference outside the keyspace.
1332
missing_text_keys = self.new_pack.text_index._external_references()
1333
if missing_text_keys:
1334
raise errors.BzrCheckError('Reference to missing compression parents %r'
1335
% (missing_text_keys,))
1336
self._log_copied_texts()
1338
def _use_pack(self, new_pack):
1339
"""Override _use_pack to check for reconcile having changed content."""
1340
# XXX: we might be better checking this at the copy time.
1341
original_inventory_keys = set()
1342
inv_index = self._pack_collection.inventory_index.combined_index
1343
for entry in inv_index.iter_all_entries():
1344
original_inventory_keys.add(entry[1])
1345
new_inventory_keys = set()
1346
for entry in new_pack.inventory_index.iter_all_entries():
1347
new_inventory_keys.add(entry[1])
1348
if new_inventory_keys != original_inventory_keys:
1349
self._data_changed = True
1350
return new_pack.data_inserted() and self._data_changed
1353
108
class RepositoryPackCollection(object):
1354
"""Management of packs within a repository.
1356
:ivar _names: map of {pack_name: (index_size,)}
1359
pack_factory = NewPack
1360
resumed_pack_factory = ResumedPack
109
"""Management of packs within a repository."""
1362
111
def __init__(self, repo, transport, index_transport, upload_transport,
1363
pack_transport, index_builder_class, index_class,
1365
113
"""Create a new RepositoryPackCollection.
1367
:param transport: Addresses the repository base directory
115
:param transport: Addresses the repository base directory
1368
116
(typically .bzr/repository/).
1369
:param index_transport: Addresses the directory containing indices.
117
:param index_transport: Addresses the directory containing indexes.
1370
118
:param upload_transport: Addresses the directory into which packs are written
1371
119
while they're being created.
1372
120
:param pack_transport: Addresses the directory of existing complete packs.
1373
:param index_builder_class: The index builder class to use.
1374
:param index_class: The index class to use.
1375
:param use_chk_index: Whether to setup and manage a CHK index.
1377
# XXX: This should call self.reset()
1378
122
self.repo = repo
1379
123
self.transport = transport
1380
124
self._index_transport = index_transport
1381
125
self._upload_transport = upload_transport
1382
126
self._pack_transport = pack_transport
1383
self._index_builder_class = index_builder_class
1384
self._index_class = index_class
1385
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3,
127
self._suffix_offsets = {'.rix':0, '.iix':1, '.tix':2, '.six':3}
1390
self._packs_by_name = {}
1391
# the previous pack-names content
1392
self._packs_at_load = None
1393
# when a pack is being created by this object, the state of that pack.
1394
self._new_pack = None
1395
# aggregated revision index data
1396
flush = self._flush_new_pack
1397
self.revision_index = AggregateIndex(self.reload_pack_names, flush)
1398
self.inventory_index = AggregateIndex(self.reload_pack_names, flush)
1399
self.text_index = AggregateIndex(self.reload_pack_names, flush)
1400
self.signature_index = AggregateIndex(self.reload_pack_names, flush)
1401
all_indices = [self.revision_index, self.inventory_index,
1402
self.text_index, self.signature_index]
1404
self.chk_index = AggregateIndex(self.reload_pack_names, flush)
1405
all_indices.append(self.chk_index)
1407
# used to determine if we're using a chk_index elsewhere.
1408
self.chk_index = None
1409
# Tell all the CombinedGraphIndex objects about each other, so they can
1410
# share hints about which pack names to search first.
1411
all_combined = [agg_idx.combined_index for agg_idx in all_indices]
1412
for combined_idx in all_combined:
1413
combined_idx.set_sibling_indices(
1414
set(all_combined).difference([combined_idx]))
1416
self._resumed_packs = []
1419
return '%s(%r)' % (self.__class__.__name__, self.repo)
1421
130
def add_pack_to_memory(self, pack):
1422
131
"""Make a Pack object available to the repository to satisfy queries.
1424
133
:param pack: A Pack object.
1426
if pack.name in self._packs_by_name:
1427
raise AssertionError(
1428
'pack %s already in _packs_by_name' % (pack.name,))
1429
135
self.packs.append(pack)
1430
self._packs_by_name[pack.name] = pack
1431
self.revision_index.add_index(pack.revision_index, pack)
1432
self.inventory_index.add_index(pack.inventory_index, pack)
1433
self.text_index.add_index(pack.text_index, pack)
1434
self.signature_index.add_index(pack.signature_index, pack)
1435
if self.chk_index is not None:
1436
self.chk_index.add_index(pack.chk_index, pack)
1438
def all_packs(self):
1439
"""Return a list of all the Pack objects this repository has.
1441
Note that an in-progress pack being created is not returned.
1443
:return: A list of Pack objects for all the packs in the repository.
136
if self.repo._revision_all_indices is None:
137
# to make this function more useful, perhaps we should make an
138
# all_indices object in future?
141
self.repo._revision_pack_map[pack.revision_index] = (
142
pack.transport, pack.name + '.pack')
143
self.repo._revision_all_indices.insert_index(0, pack.revision_index)
144
if self.repo._inv_all_indices is not None:
145
# inv 'knit' has been used : update it.
146
self.repo._inv_all_indices.insert_index(0,
147
pack.inventory_index)
148
self.repo._inv_pack_map[pack.inventory_index] = pack.transport, pack.name + '.pack'
149
if self.repo._text_all_indices is not None:
150
# text 'knits' have been used : update it.
151
self.repo._text_all_indices.insert_index(0,
153
if self.repo._signature_all_indices is not None:
154
# sigatures 'knit' accessed : update it.
155
self.repo._signature_all_indices.insert_index(0,
156
pack.signature_index)
158
def _add_text_to_weave(self, file_id, revision_id, new_lines, parents,
159
nostore_sha, random_revid):
160
file_id_index = GraphIndexPrefixAdapter(
161
self.repo._text_all_indices,
163
add_nodes_callback=self.repo._text_write_index.add_nodes)
164
self.repo._text_knit._index._graph_index = file_id_index
165
self.repo._text_knit._index._add_callback = file_id_index.add_nodes
166
return self.repo._text_knit.add_lines_with_ghosts(
167
revision_id, parents, new_lines, nostore_sha=nostore_sha,
168
random_id=random_revid, check_content=False)[0:2]
170
def all_pack_details(self):
171
"""Return a list of all the packs as transport,name tuples.
173
:return: A list of (transport, name) tuples for all the packs in the
176
# XXX: fix me, should be direct rather than indirect
177
if self.repo._revision_all_indices is None:
178
# trigger creation of the all revision index.
179
self.repo._revision_store.get_revision_file(self.repo.get_transaction())
1446
for name in self.names():
1447
result.append(self.get_pack_by_name(name))
181
for index, transport_and_name in self.repo._revision_pack_map.iteritems():
182
result.append(transport_and_name)
1450
185
def autopack(self):
1451
186
"""Pack the pack collection incrementally.
1453
188
This will not attempt global reorganisation or recompression,
1454
189
rather it will just ensure that the total number of packs does
1455
190
not grow without bound. It uses the _max_pack_count method to
1490
225
# group their data with the relevant commit, and that may
1491
226
# involve rewriting ancient history - which autopack tries to
1492
227
# avoid. Alternatively we could not group the data but treat
1493
# each of these as having a single revision, and thus add
228
# each of these as having a single revision, and thus add
1494
229
# one revision for each to the total revision count, to get
1495
230
# a matching distribution.
1497
existing_packs.append((revision_count, pack))
232
existing_packs.append((revision_count, transport_and_name))
1498
233
pack_operations = self.plan_autopack_combinations(
1499
234
existing_packs, pack_distribution)
1500
num_new_packs = len(pack_operations)
1501
num_old_packs = sum([len(po[1]) for po in pack_operations])
1502
num_revs_affected = sum([po[0] for po in pack_operations])
1503
mutter('Auto-packing repository %s, which has %d pack files, '
1504
'containing %d revisions. Packing %d files into %d affecting %d'
1505
' revisions', self, total_packs, total_revisions, num_old_packs,
1506
num_new_packs, num_revs_affected)
1507
result = self._execute_pack_operations(pack_operations,
1508
reload_func=self._restart_autopack)
1509
mutter('Auto-packing repository %s completed', self)
235
self._execute_pack_operations(pack_operations)
238
def create_pack_from_packs(self, revision_index_map, inventory_index_map,
239
text_index_map, signature_index_map, suffix, revision_ids=None):
240
"""Create a new pack by reading data from other packs.
242
This does little more than a bulk copy of data. One key difference
243
is that data with the same item key across multiple packs is elided
244
from the output. The new pack is written into the current pack store
245
along with its indices, and the name added to the pack names. The
246
source packs are not altered.
248
:param revision_index_map: A revision index map.
249
:param inventory_index_map: A inventory index map.
250
:param text_index_map: A text index map.
251
:param signature_index_map: A signature index map.
252
:param revision_ids: Either None, to copy all data, or a list
253
of revision_ids to limit the copied data to the data they
255
:return: A Pack object, or None if nothing was copied.
257
# open a pack - using the same name as the last temporary file
258
# - which has already been flushed, so its safe.
259
# XXX: - duplicate code warning with start_write_group; fix before
260
# considering 'done'.
261
if getattr(self.repo, '_open_pack_tuple', None) is not None:
262
raise errors.BzrError('call to create_pack_from_packs while '
263
'another pack is being written.')
264
if revision_ids is not None and len(revision_ids) == 0:
265
# silly fetch request.
267
random_name = self.repo.control_files._lock.nonce + suffix
268
if 'fetch' in debug.debug_flags:
269
plain_pack_list = ['%s%s' % (transport.base, name) for
270
transport, name in revision_index_map.itervalues()]
271
if revision_ids is not None:
272
rev_count = len(revision_ids)
275
mutter('%s: create_pack: creating pack from source packs: '
276
'%s%s %s revisions wanted %s t=0',
277
time.ctime(), self._upload_transport.base, random_name,
278
plain_pack_list, rev_count)
279
start_time = time.time()
280
write_stream = self._upload_transport.open_write_stream(random_name)
281
if 'fetch' in debug.debug_flags:
282
mutter('%s: create_pack: pack stream open: %s%s t+%6.3fs',
283
time.ctime(), self._upload_transport.base, random_name,
284
time.time() - start_time)
285
pack_hash = md5.new()
287
def write_data(bytes, update=pack_hash.update, write=write_stream.write):
289
if len(buffer) == 640:
290
bytes = ''.join(buffer)
294
writer = pack.ContainerWriter(write_data)
297
revision_index = InMemoryGraphIndex(reference_lists=1)
298
inv_index = InMemoryGraphIndex(reference_lists=2)
299
text_index = InMemoryGraphIndex(reference_lists=2, key_elements=2)
300
signature_index = InMemoryGraphIndex(reference_lists=0)
303
revision_keys = [(revision_id,) for revision_id in revision_ids]
306
revision_nodes = self._index_contents(revision_index_map, revision_keys)
307
# copy revision keys and adjust values
308
list(self._copy_nodes_graph(revision_nodes, revision_index_map, writer,
310
if 'fetch' in debug.debug_flags:
311
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
312
time.ctime(), self._upload_transport.base, random_name,
313
revision_index.key_count(),
314
time.time() - start_time)
315
# select inventory keys
316
inv_keys = revision_keys # currently the same keyspace, and note that
317
# querying for keys here could introduce a bug where an inventory item
318
# is missed, so do not change it to query separately without cross
319
# checking like the text key check below.
320
inv_nodes = self._index_contents(inventory_index_map, inv_keys)
321
# copy inventory keys and adjust values
322
# XXX: Should be a helper function to allow different inv representation
324
inv_lines = self._copy_nodes_graph(inv_nodes, inventory_index_map,
325
writer, inv_index, output_lines=True)
327
fileid_revisions = self.repo._find_file_ids_from_xml_inventory_lines(
328
inv_lines, revision_ids)
330
for fileid, file_revids in fileid_revisions.iteritems():
332
[(fileid, file_revid) for file_revid in file_revids])
334
# eat the iterator to cause it to execute.
337
if 'fetch' in debug.debug_flags:
338
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
339
time.ctime(), self._upload_transport.base, random_name,
340
inv_index.key_count(),
341
time.time() - start_time)
343
text_nodes = self._index_contents(text_index_map, text_filter)
344
if text_filter is not None:
345
# We could return the keys copied as part of the return value from
346
# _copy_nodes_graph but this doesn't work all that well with the
347
# need to get line output too, so we check separately, and as we're
348
# going to buffer everything anyway, we check beforehand, which
349
# saves reading knit data over the wire when we know there are
351
text_nodes = set(text_nodes)
352
present_text_keys = set(_node[1] for _node in text_nodes)
353
missing_text_keys = set(text_filter) - present_text_keys
354
if missing_text_keys:
355
# TODO: raise a specific error that can handle many missing
357
a_missing_key = missing_text_keys.pop()
358
raise errors.RevisionNotPresent(a_missing_key[1],
360
# copy text keys and adjust values
361
list(self._copy_nodes_graph(text_nodes, text_index_map, writer,
363
if 'fetch' in debug.debug_flags:
364
mutter('%s: create_pack: file texts copied: %s%s %d items t+%6.3fs',
365
time.ctime(), self._upload_transport.base, random_name,
366
text_index.key_count(),
367
time.time() - start_time)
368
# select signature keys
369
signature_filter = revision_keys # same keyspace
370
signature_nodes = self._index_contents(signature_index_map,
372
# copy signature keys and adjust values
373
self._copy_nodes(signature_nodes, signature_index_map, writer, signature_index)
374
if 'fetch' in debug.debug_flags:
375
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
376
time.ctime(), self._upload_transport.base, random_name,
377
signature_index.key_count(),
378
time.time() - start_time)
382
bytes = ''.join(buffer)
383
write_stream.write(bytes)
384
pack_hash.update(bytes)
385
new_name = pack_hash.hexdigest()
386
# if nothing has been written, discard the new pack.
387
if 0 == sum((revision_index.key_count(),
388
inv_index.key_count(),
389
text_index.key_count(),
390
signature_index.key_count(),
392
self._upload_transport.delete(random_name)
395
result.name = new_name
396
result.transport = self._upload_transport.clone('../packs/')
398
index_transport = self._index_transport
399
rev_index_name = self.repo._revision_store.name_to_revision_index_name(new_name)
400
revision_index_length = index_transport.put_file(rev_index_name,
401
revision_index.finish())
402
if 'fetch' in debug.debug_flags:
403
# XXX: size might be interesting?
404
mutter('%s: create_pack: wrote revision index: %s%s t+%6.3fs',
405
time.ctime(), self._upload_transport.base, random_name,
406
time.time() - start_time)
407
inv_index_name = self.repo._inv_thunk.name_to_inv_index_name(new_name)
408
inventory_index_length = index_transport.put_file(inv_index_name,
410
if 'fetch' in debug.debug_flags:
411
# XXX: size might be interesting?
412
mutter('%s: create_pack: wrote inventory index: %s%s t+%6.3fs',
413
time.ctime(), self._upload_transport.base, random_name,
414
time.time() - start_time)
415
text_index_name = self.repo.weave_store.name_to_text_index_name(new_name)
416
text_index_length = index_transport.put_file(text_index_name,
418
if 'fetch' in debug.debug_flags:
419
# XXX: size might be interesting?
420
mutter('%s: create_pack: wrote file texts index: %s%s t+%6.3fs',
421
time.ctime(), self._upload_transport.base, random_name,
422
time.time() - start_time)
423
signature_index_name = self.repo._revision_store.name_to_signature_index_name(new_name)
424
signature_index_length = index_transport.put_file(signature_index_name,
425
signature_index.finish())
426
if 'fetch' in debug.debug_flags:
427
# XXX: size might be interesting?
428
mutter('%s: create_pack: wrote revision signatures index: %s%s t+%6.3fs',
429
time.ctime(), self._upload_transport.base, random_name,
430
time.time() - start_time)
432
self.allocate(new_name, revision_index_length, inventory_index_length,
433
text_index_length, signature_index_length)
434
# rename into place. XXX: should rename each index too rather than just
435
# uploading blind under the chosen name.
437
self._upload_transport.rename(random_name, '../packs/' + new_name + '.pack')
438
if 'fetch' in debug.debug_flags:
439
# XXX: size might be interesting?
440
mutter('%s: create_pack: pack renamed into place: %s%s->%s%s t+%6.3fs',
441
time.ctime(), self._upload_transport.base, random_name,
442
result.transport, result.name,
443
time.time() - start_time)
444
result.revision_index = revision_index
445
result.inventory_index = inv_index
446
result.text_index = text_index
447
result.signature_index = signature_index
448
if 'fetch' in debug.debug_flags:
449
# XXX: size might be interesting?
450
mutter('%s: create_pack: finished: %s%s t+%6.3fs',
451
time.ctime(), self._upload_transport.base, random_name,
452
time.time() - start_time)
1512
def _execute_pack_operations(self, pack_operations, _packer_class=Packer,
455
def _execute_pack_operations(self, pack_operations):
1514
456
"""Execute a series of pack operations.
1516
458
:param pack_operations: A list of [revision_count, packs_to_combine].
1517
:param _packer_class: The class of packer to use (default: Packer).
1518
:return: The new pack names.
1520
for revision_count, packs in pack_operations:
461
for revision_count, pack_details in pack_operations:
1521
462
# we may have no-ops from the setup logic
463
if len(pack_details) == 0:
1524
packer = _packer_class(self, packs, '.autopack',
1525
reload_func=reload_func)
1528
except errors.RetryWithNewPacks:
1529
# An exception is propagating out of this context, make sure
1530
# this packer has cleaned up. Packer() doesn't set its new_pack
1531
# state into the RepositoryPackCollection object, so we only
1532
# have access to it directly here.
1533
if packer.new_pack is not None:
1534
packer.new_pack.abort()
1537
self._remove_pack_from_memory(pack)
465
# have a progress bar?
466
self._combine_packs(pack_details)
467
for pack_detail in pack_details:
468
self._remove_pack_by_name(pack_detail[1])
1538
469
# record the newly available packs and stop advertising the old
1540
to_be_obsoleted = []
1541
for _, packs in pack_operations:
1542
to_be_obsoleted.extend(packs)
1543
result = self._save_pack_names(clear_obsolete_packs=True,
1544
obsolete_packs=to_be_obsoleted)
1547
def _flush_new_pack(self):
1548
if self._new_pack is not None:
1549
self._new_pack.flush()
1551
def lock_names(self):
1552
"""Acquire the mutex around the pack-names index.
1554
This cannot be used in the middle of a read-only transaction on the
1557
self.repo.control_files.lock_write()
1559
def _already_packed(self):
1560
"""Is the collection already packed?"""
1561
return not (self.repo._format.pack_compresses or (len(self._names) > 1))
1563
def pack(self, hint=None, clean_obsolete_packs=False):
471
self._save_pack_names()
472
# move the old packs out of the way
473
for revision_count, pack_details in pack_operations:
474
self._obsolete_packs(pack_details)
1564
477
"""Pack the pack collection totally."""
1565
478
self.ensure_loaded()
1566
total_packs = len(self._names)
1567
if self._already_packed():
1569
total_revisions = self.revision_index.combined_index.key_count()
1570
# XXX: the following may want to be a class, to pack with a given
1572
mutter('Packing repository %s, which has %d pack files, '
1573
'containing %d revisions with hint %r.', self, total_packs,
1574
total_revisions, hint)
1575
# determine which packs need changing
1576
pack_operations = [[0, []]]
1577
for pack in self.all_packs():
1578
if hint is None or pack.name in hint:
1579
# Either no hint was provided (so we are packing everything),
1580
# or this pack was included in the hint.
1581
pack_operations[-1][0] += pack.get_revision_count()
1582
pack_operations[-1][1].append(pack)
1583
self._execute_pack_operations(pack_operations, OptimisingPacker)
1585
if clean_obsolete_packs:
1586
self._clear_obsolete_packs()
480
total_packs = len(self._names)
483
if self.repo._revision_all_indices is None:
484
# trigger creation of the all revision index.
485
self.repo._revision_store.get_revision_file(self.repo.get_transaction())
486
total_revisions = self.repo._revision_all_indices.key_count()
487
# XXX: the following may want to be a class, to pack with a given
489
mutter('Packing repository %s, which has %d pack files, '
490
'containing %d revisions into 1 packs.', self, total_packs,
492
# determine which packs need changing
493
pack_distribution = [1]
494
pack_operations = [[0, []]]
495
for index, transport_and_name in self.repo._revision_pack_map.iteritems():
498
revision_count = index.key_count()
499
pack_operations[-1][0] += revision_count
500
pack_operations[-1][1].append(transport_and_name)
501
self._execute_pack_operations(pack_operations)
503
if not self.repo.is_in_write_group():
1588
506
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1589
"""Plan a pack operation.
1591
:param existing_packs: The packs to pack. (A list of (revcount, Pack)
1593
:param pack_distribution: A list with the number of revisions desired
1596
507
if len(existing_packs) <= len(pack_distribution):
1598
509
existing_packs.sort(reverse=True)
1619
530
# add the revisions we're going to add to the next output pack
1620
531
pack_operations[-1][0] += next_pack_rev_count
1621
532
# allocate this pack to the next pack sub operation
1622
pack_operations[-1][1].append(next_pack)
533
pack_operations[-1][1].append(next_pack_details)
1623
534
if pack_operations[-1][0] >= pack_distribution[0]:
1624
535
# this pack is used up, shift left.
1625
536
del pack_distribution[0]
1626
537
pack_operations.append([0, []])
1627
# Now that we know which pack files we want to move, shove them all
1628
# into a single pack file.
1630
final_pack_list = []
1631
for num_revs, pack_files in pack_operations:
1632
final_rev_count += num_revs
1633
final_pack_list.extend(pack_files)
1634
if len(final_pack_list) == 1:
1635
raise AssertionError('We somehow generated an autopack with a'
1636
' single pack file being moved.')
1638
return [[final_rev_count, final_pack_list]]
539
return pack_operations
541
def _combine_packs(self, pack_details):
542
"""Combine the data from the packs listed in pack_details.
544
This does little more than a bulk copy of data. One key difference
545
is that data with the same item key across multiple packs is elided
546
from the output. The new pack is written into the current pack store
547
along with its indices, and the name added to the pack names. The
548
source packs are not altered.
550
:param pack_details: A list of tuples with the transport and pack name
554
# select revision keys
555
revision_index_map = self._revision_index_map(pack_details)
556
# select inventory keys
557
inv_index_map = self._inv_index_map(pack_details)
559
text_index_map = self._text_index_map(pack_details)
560
# select signature keys
561
signature_index_map = self._signature_index_map(pack_details)
562
self.create_pack_from_packs(revision_index_map, inv_index_map,
563
text_index_map, signature_index_map, '.autopack')
565
def _copy_nodes(self, nodes, index_map, writer, write_index):
566
# plan a readv on each source pack:
568
nodes = sorted(nodes)
569
# how to map this into knit.py - or knit.py into this?
570
# we don't want the typical knit logic, we want grouping by pack
571
# at this point - perhaps a helper library for the following code
572
# duplication points?
574
for index, key, value in nodes:
575
if index not in request_groups:
576
request_groups[index] = []
577
request_groups[index].append((key, value))
578
for index, items in request_groups.iteritems():
579
pack_readv_requests = []
580
for key, value in items:
581
# ---- KnitGraphIndex.get_position
582
bits = value[1:].split(' ')
583
offset, length = int(bits[0]), int(bits[1])
584
pack_readv_requests.append((offset, length, (key, value[0])))
585
# linear scan up the pack
586
pack_readv_requests.sort()
588
transport, path = index_map[index]
589
reader = pack.make_readv_reader(transport, path,
590
[offset[0:2] for offset in pack_readv_requests])
591
for (names, read_func), (_1, _2, (key, eol_flag)) in \
592
izip(reader.iter_records(), pack_readv_requests):
593
raw_data = read_func(None)
594
pos, size = writer.add_bytes_record(raw_data, names)
595
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
597
def _copy_nodes_graph(self, nodes, index_map, writer, write_index,
599
"""Copy knit nodes between packs.
601
:param output_lines: Return lines present in the copied data as
604
# for record verification
605
knit_data = _KnitData(None)
606
# for line extraction when requested (inventories only)
608
factory = knit.KnitPlainFactory()
609
# plan a readv on each source pack:
611
nodes = sorted(nodes)
612
# how to map this into knit.py - or knit.py into this?
613
# we don't want the typical knit logic, we want grouping by pack
614
# at this point - perhaps a helper library for the following code
615
# duplication points?
617
for index, key, value, references in nodes:
618
if index not in request_groups:
619
request_groups[index] = []
620
request_groups[index].append((key, value, references))
621
for index, items in request_groups.iteritems():
622
pack_readv_requests = []
623
for key, value, references in items:
624
# ---- KnitGraphIndex.get_position
625
bits = value[1:].split(' ')
626
offset, length = int(bits[0]), int(bits[1])
627
pack_readv_requests.append((offset, length, (key, value[0], references)))
628
# linear scan up the pack
629
pack_readv_requests.sort()
631
transport, path = index_map[index]
632
reader = pack.make_readv_reader(transport, path,
633
[offset[0:2] for offset in pack_readv_requests])
634
for (names, read_func), (_1, _2, (key, eol_flag, references)) in \
635
izip(reader.iter_records(), pack_readv_requests):
636
raw_data = read_func(None)
638
# read the entire thing
639
content, _ = knit_data._parse_record(key[-1], raw_data)
640
if len(references[-1]) == 0:
641
line_iterator = factory.get_fulltext_content(content)
643
line_iterator = factory.get_linedelta_content(content)
644
for line in line_iterator:
647
# check the header only
648
df, _ = knit_data._parse_record_header(key[-1], raw_data)
650
pos, size = writer.add_bytes_record(raw_data, names)
651
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1640
653
def ensure_loaded(self):
1641
"""Ensure we have read names from disk.
1643
:return: True if the disk names had not been previously read.
1645
# NB: if you see an assertion error here, its probably access against
1646
# an unlocked repo. Naughty.
1647
if not self.repo.is_locked():
1648
raise errors.ObjectNotLocked(self.repo)
1649
654
if self._names is None:
1650
655
self._names = {}
1651
self._packs_at_load = set()
1652
for index, key, value in self._iter_disk_pack_index():
656
for index, key, value in \
657
GraphIndex(self.transport, 'pack-names', None
658
).iter_all_entries():
1654
self._names[name] = self._parse_index_sizes(value)
1655
self._packs_at_load.add((key, value))
1659
# populate all the metadata.
1663
def _parse_index_sizes(self, value):
1664
"""Parse a string of index sizes."""
1665
return tuple([int(digits) for digits in value.split(' ')])
1667
def get_pack_by_name(self, name):
1668
"""Get a Pack object by name.
1670
:param name: The name of the pack - e.g. '123456'
1671
:return: A Pack object.
1674
return self._packs_by_name[name]
1676
rev_index = self._make_index(name, '.rix')
1677
inv_index = self._make_index(name, '.iix')
1678
txt_index = self._make_index(name, '.tix')
1679
sig_index = self._make_index(name, '.six')
1680
if self.chk_index is not None:
1681
chk_index = self._make_index(name, '.cix', unlimited_cache=True)
1684
result = ExistingPack(self._pack_transport, name, rev_index,
1685
inv_index, txt_index, sig_index, chk_index)
1686
self.add_pack_to_memory(result)
1689
def _resume_pack(self, name):
1690
"""Get a suspended Pack object by name.
1692
:param name: The name of the pack - e.g. '123456'
1693
:return: A Pack object.
1695
if not re.match('[a-f0-9]{32}', name):
1696
# Tokens should be md5sums of the suspended pack file, i.e. 32 hex
1698
raise errors.UnresumableWriteGroup(
1699
self.repo, [name], 'Malformed write group token')
1701
rev_index = self._make_index(name, '.rix', resume=True)
1702
inv_index = self._make_index(name, '.iix', resume=True)
1703
txt_index = self._make_index(name, '.tix', resume=True)
1704
sig_index = self._make_index(name, '.six', resume=True)
1705
if self.chk_index is not None:
1706
chk_index = self._make_index(name, '.cix', resume=True,
1707
unlimited_cache=True)
1710
result = self.resumed_pack_factory(name, rev_index, inv_index,
1711
txt_index, sig_index, self._upload_transport,
1712
self._pack_transport, self._index_transport, self,
1713
chk_index=chk_index)
1714
except errors.NoSuchFile, e:
1715
raise errors.UnresumableWriteGroup(self.repo, [name], str(e))
1716
self.add_pack_to_memory(result)
1717
self._resumed_packs.append(result)
1720
def allocate(self, a_new_pack):
660
sizes = [int(digits) for digits in value.split(' ')]
661
self._names[name] = sizes
663
def allocate(self, name, revision_index_length, inventory_index_length,
664
text_index_length, signature_index_length):
1721
665
"""Allocate name in the list of packs.
1723
:param a_new_pack: A NewPack instance to be added to the collection of
1724
packs for this repository.
667
:param name: The basename - e.g. the md5 hash hexdigest.
668
:param revision_index_length: The length of the revision index in
670
:param inventory_index_length: The length of the inventory index in
672
:param text_index_length: The length of the text index in bytes.
673
:param signature_index_length: The length of the signature index in
1726
676
self.ensure_loaded()
1727
if a_new_pack.name in self._names:
1728
raise errors.BzrError(
1729
'Pack %r already exists in %s' % (a_new_pack.name, self))
1730
self._names[a_new_pack.name] = tuple(a_new_pack.index_sizes)
1731
self.add_pack_to_memory(a_new_pack)
1733
def _iter_disk_pack_index(self):
1734
"""Iterate over the contents of the pack-names index.
1736
This is used when loading the list from disk, and before writing to
1737
detect updates from others during our write operation.
1738
:return: An iterator of the index contents.
677
if name in self._names:
678
raise errors.DuplicateKey(name)
679
self._names[name] = (revision_index_length, inventory_index_length,
680
text_index_length, signature_index_length)
682
def _make_index_map(self, suffix):
683
"""Return information on existing indexes.
685
:param suffix: Index suffix added to pack name.
687
:returns: (pack_map, indices) where indices is a list of GraphIndex
688
objects, and pack_map is a mapping from those objects to the
689
pack tuple they describe.
1740
return self._index_class(self.transport, 'pack-names', None
1741
).iter_all_entries()
1743
def _make_index(self, name, suffix, resume=False, unlimited_cache=False):
1744
691
size_offset = self._suffix_offsets[suffix]
1745
index_name = name + suffix
1747
transport = self._upload_transport
1748
index_size = transport.stat(index_name).st_size
1750
transport = self._index_transport
695
for name in self.names():
696
# TODO: maybe this should expose size to us to allow
697
# sorting of the indices for better performance ?
698
index_name = name + suffix
1751
699
index_size = self._names[name][size_offset]
1752
return self._index_class(transport, index_name, index_size,
1753
unlimited_cache=unlimited_cache)
700
new_index = GraphIndex(
701
self._index_transport, index_name, index_size)
702
indices.append(new_index)
703
pack_map[new_index] = self._pack_tuple(name)
704
return pack_map, indices
1755
706
def _max_pack_count(self, total_revisions):
1756
707
"""Return the maximum number of packs to use for total revisions.
1758
709
:param total_revisions: The total number of revisions in the
1825
765
"""Return a tuple with the transport and file name for a pack name."""
1826
766
return self._pack_transport, name + '.pack'
1828
def _remove_pack_from_memory(self, pack):
1829
"""Remove pack from the packs accessed by this repository.
1831
Only affects memory state, until self._save_pack_names() is invoked.
1833
self._names.pop(pack.name)
1834
self._packs_by_name.pop(pack.name)
1835
self._remove_pack_indices(pack)
1836
self.packs.remove(pack)
1838
def _remove_pack_indices(self, pack, ignore_missing=False):
1839
"""Remove the indices for pack from the aggregated indices.
1841
:param ignore_missing: Suppress KeyErrors from calling remove_index.
1843
for index_type in Pack.index_definitions.keys():
1844
attr_name = index_type + '_index'
1845
aggregate_index = getattr(self, attr_name)
1846
if aggregate_index is not None:
1847
pack_index = getattr(pack, attr_name)
1849
aggregate_index.remove_index(pack_index)
768
def _remove_pack_by_name(self, name):
770
self._names.pop(name[:-5])
1855
772
def reset(self):
1856
"""Clear all cached data."""
1857
# cached revision data
1858
self.revision_index.clear()
1859
# cached signature data
1860
self.signature_index.clear()
1861
# cached file text data
1862
self.text_index.clear()
1863
# cached inventory data
1864
self.inventory_index.clear()
1866
if self.chk_index is not None:
1867
self.chk_index.clear()
1868
# remove the open pack
1869
self._new_pack = None
1870
# information about packs.
1871
773
self._names = None
1873
self._packs_by_name = {}
1874
self._packs_at_load = None
1876
def _unlock_names(self):
1877
"""Release the mutex around the pack-names index."""
1878
self.repo.control_files.unlock()
1880
def _diff_pack_names(self):
1881
"""Read the pack names from disk, and compare it to the one in memory.
1883
:return: (disk_nodes, deleted_nodes, new_nodes)
1884
disk_nodes The final set of nodes that should be referenced
1885
deleted_nodes Nodes which have been removed from when we started
1886
new_nodes Nodes that are newly introduced
776
def _make_index_to_pack_map(self, pack_details, index_suffix):
777
"""Given a list (transport,name), return a map of (index)->(transport, name)."""
778
# the simplest thing for now is to create new index objects.
779
# this should really reuse the existing index objects for these
780
# packs - this means making the way they are managed in the repo be
782
size_offset = self._suffix_offsets[index_suffix]
784
for transport, name in pack_details:
785
index_name = name[:-5] + index_suffix
786
index_size = self._names[index_name][index_size]
787
indices[GraphIndex(self._index_transport, index_name, index_size)] = \
791
def _inv_index_map(self, pack_details):
792
"""Get a map of inv index -> packs for pack_details."""
793
return self._make_index_to_pack_map(pack_details, '.iix')
795
def _revision_index_map(self, pack_details):
796
"""Get a map of revision index -> packs for pack_details."""
797
return self._make_index_to_pack_map(pack_details, '.rix')
799
def _signature_index_map(self, pack_details):
800
"""Get a map of signature index -> packs for pack_details."""
801
return self._make_index_to_pack_map(pack_details, '.six')
803
def _text_index_map(self, pack_details):
804
"""Get a map of text index -> packs for pack_details."""
805
return self._make_index_to_pack_map(pack_details, '.tix')
807
def _index_contents(self, pack_map, key_filter=None):
808
"""Get an iterable of the index contents from a pack_map.
810
:param pack_map: A map from indices to pack details.
811
:param key_filter: An optional filter to limit the
1888
# load the disk nodes across
1890
for index, key, value in self._iter_disk_pack_index():
1891
disk_nodes.add((key, value))
1892
orig_disk_nodes = set(disk_nodes)
814
indices = [index for index in pack_map.iterkeys()]
815
all_index = CombinedGraphIndex(indices)
816
if key_filter is None:
817
return all_index.iter_all_entries()
819
return all_index.iter_entries(key_filter)
1894
# do a two-way diff against our original content
1895
current_nodes = set()
821
def _save_pack_names(self):
822
builder = GraphIndexBuilder()
1896
823
for name, sizes in self._names.iteritems():
1898
((name, ), ' '.join(str(size) for size in sizes)))
1900
# Packs no longer present in the repository, which were present when we
1901
# locked the repository
1902
deleted_nodes = self._packs_at_load - current_nodes
1903
# Packs which this process is adding
1904
new_nodes = current_nodes - self._packs_at_load
1906
# Update the disk_nodes set to include the ones we are adding, and
1907
# remove the ones which were removed by someone else
1908
disk_nodes.difference_update(deleted_nodes)
1909
disk_nodes.update(new_nodes)
1911
return disk_nodes, deleted_nodes, new_nodes, orig_disk_nodes
1913
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1914
"""Given the correct set of pack files, update our saved info.
1916
:return: (removed, added, modified)
1917
removed pack names removed from self._names
1918
added pack names added to self._names
1919
modified pack names that had changed value
1924
## self._packs_at_load = disk_nodes
1925
new_names = dict(disk_nodes)
1926
# drop no longer present nodes
1927
for pack in self.all_packs():
1928
if (pack.name,) not in new_names:
1929
removed.append(pack.name)
1930
self._remove_pack_from_memory(pack)
1931
# add new nodes/refresh existing ones
1932
for key, value in disk_nodes:
1934
sizes = self._parse_index_sizes(value)
1935
if name in self._names:
1937
if sizes != self._names[name]:
1938
# the pack for name has had its indices replaced - rare but
1939
# important to handle. XXX: probably can never happen today
1940
# because the three-way merge code above does not handle it
1941
# - you may end up adding the same key twice to the new
1942
# disk index because the set values are the same, unless
1943
# the only index shows up as deleted by the set difference
1944
# - which it may. Until there is a specific test for this,
1945
# assume its broken. RBC 20071017.
1946
self._remove_pack_from_memory(self.get_pack_by_name(name))
1947
self._names[name] = sizes
1948
self.get_pack_by_name(name)
1949
modified.append(name)
1952
self._names[name] = sizes
1953
self.get_pack_by_name(name)
1955
return removed, added, modified
1957
def _save_pack_names(self, clear_obsolete_packs=False, obsolete_packs=None):
1958
"""Save the list of packs.
1960
This will take out the mutex around the pack names list for the
1961
duration of the method call. If concurrent updates have been made, a
1962
three-way merge between the current list and the current in memory list
1965
:param clear_obsolete_packs: If True, clear out the contents of the
1966
obsolete_packs directory.
1967
:param obsolete_packs: Packs that are obsolete once the new pack-names
1968
file has been written.
1969
:return: A list of the names saved that were not previously on disk.
1971
already_obsolete = []
1974
builder = self._index_builder_class()
1975
(disk_nodes, deleted_nodes, new_nodes,
1976
orig_disk_nodes) = self._diff_pack_names()
1977
# TODO: handle same-name, index-size-changes here -
1978
# e.g. use the value from disk, not ours, *unless* we're the one
1980
for key, value in disk_nodes:
1981
builder.add_node(key, value)
1982
self.transport.put_file('pack-names', builder.finish(),
1983
mode=self.repo.bzrdir._get_file_mode())
1984
self._packs_at_load = disk_nodes
1985
if clear_obsolete_packs:
1988
to_preserve = set([o.name for o in obsolete_packs])
1989
already_obsolete = self._clear_obsolete_packs(to_preserve)
1991
self._unlock_names()
1992
# synchronise the memory packs list with what we just wrote:
1993
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1995
# TODO: We could add one more condition here. "if o.name not in
1996
# orig_disk_nodes and o != the new_pack we haven't written to
1997
# disk yet. However, the new pack object is not easily
1998
# accessible here (it would have to be passed through the
1999
# autopacking code, etc.)
2000
obsolete_packs = [o for o in obsolete_packs
2001
if o.name not in already_obsolete]
2002
self._obsolete_packs(obsolete_packs)
2003
return [new_node[0][0] for new_node in new_nodes]
2005
def reload_pack_names(self):
2006
"""Sync our pack listing with what is present in the repository.
2008
This should be called when we find out that something we thought was
2009
present is now missing. This happens when another process re-packs the
2012
:return: True if the in-memory list of packs has been altered at all.
2014
# The ensure_loaded call is to handle the case where the first call
2015
# made involving the collection was to reload_pack_names, where we
2016
# don't have a view of disk contents. Its a bit of a bandaid, and
2017
# causes two reads of pack-names, but its a rare corner case not struck
2018
# with regular push/pull etc.
2019
first_read = self.ensure_loaded()
2022
# out the new value.
2023
(disk_nodes, deleted_nodes, new_nodes,
2024
orig_disk_nodes) = self._diff_pack_names()
2025
# _packs_at_load is meant to be the explicit list of names in
2026
# 'pack-names' at then start. As such, it should not contain any
2027
# pending names that haven't been written out yet.
2028
self._packs_at_load = orig_disk_nodes
2030
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
2031
if removed or added or modified:
2035
def _restart_autopack(self):
2036
"""Reload the pack names list, and restart the autopack code."""
2037
if not self.reload_pack_names():
2038
# Re-raise the original exception, because something went missing
2039
# and a restart didn't find it
2041
raise errors.RetryAutopack(self.repo, False, sys.exc_info())
2043
def _clear_obsolete_packs(self, preserve=None):
2044
"""Delete everything from the obsolete-packs directory.
2046
:return: A list of pack identifiers (the filename without '.pack') that
2047
were found in obsolete_packs.
2050
obsolete_pack_transport = self.transport.clone('obsolete_packs')
2051
if preserve is None:
2053
for filename in obsolete_pack_transport.list_dir('.'):
2054
name, ext = osutils.splitext(filename)
2057
if name in preserve:
2060
obsolete_pack_transport.delete(filename)
2061
except (errors.PathError, errors.TransportError), e:
2062
warning("couldn't delete obsolete pack, skipping it:\n%s"
2066
def _start_write_group(self):
2067
# Do not permit preparation for writing if we're not in a 'write lock'.
2068
if not self.repo.is_write_locked():
824
builder.add_node((name, ), ' '.join(str(size) for size in sizes))
825
self.transport.put_file('pack-names', builder.finish())
828
# cannot add names if we're not in a 'write lock'.
829
if self.repo.control_files._lock_mode != 'w':
2069
830
raise errors.NotWriteLocked(self)
2070
self._new_pack = self.pack_factory(self, upload_suffix='.pack',
2071
file_mode=self.repo.bzrdir._get_file_mode())
2072
# allow writing: queue writes to a new index
2073
self.revision_index.add_writable_index(self._new_pack.revision_index,
2075
self.inventory_index.add_writable_index(self._new_pack.inventory_index,
2077
self.text_index.add_writable_index(self._new_pack.text_index,
2079
self._new_pack.text_index.set_optimize(combine_backing_indices=False)
2080
self.signature_index.add_writable_index(self._new_pack.signature_index,
2082
if self.chk_index is not None:
2083
self.chk_index.add_writable_index(self._new_pack.chk_index,
2085
self.repo.chk_bytes._index._add_callback = self.chk_index.add_callback
2086
self._new_pack.chk_index.set_optimize(combine_backing_indices=False)
2088
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
2089
self.repo.revisions._index._add_callback = self.revision_index.add_callback
2090
self.repo.signatures._index._add_callback = self.signature_index.add_callback
2091
self.repo.texts._index._add_callback = self.text_index.add_callback
832
def _start_write_group(self):
833
random_name = self.repo.control_files._lock.nonce
834
self.repo._open_pack_tuple = (self._upload_transport, random_name + '.pack')
835
write_stream = self._upload_transport.open_write_stream(random_name + '.pack')
836
self._write_stream = write_stream
837
self._open_pack_hash = md5.new()
838
def write_data(bytes, write=write_stream.write,
839
update=self._open_pack_hash.update):
842
self._open_pack_writer = pack.ContainerWriter(write_data)
843
self._open_pack_writer.begin()
845
self.repo._revision_store.setup()
846
self.repo.weave_store.setup()
847
self.repo._inv_thunk.setup()
2093
849
def _abort_write_group(self):
2094
850
# FIXME: just drop the transient index.
851
self.repo._revision_store.reset()
852
self.repo.weave_store.reset()
853
self.repo._inv_thunk.reset()
2095
854
# forget what names there are
2096
if self._new_pack is not None:
2097
operation = cleanup.OperationWithCleanups(self._new_pack.abort)
2098
operation.add_cleanup(setattr, self, '_new_pack', None)
2099
# If we aborted while in the middle of finishing the write
2100
# group, _remove_pack_indices could fail because the indexes are
2101
# already gone. But they're not there we shouldn't fail in this
2102
# case, so we pass ignore_missing=True.
2103
operation.add_cleanup(self._remove_pack_indices, self._new_pack,
2104
ignore_missing=True)
2105
operation.run_simple()
2106
for resumed_pack in self._resumed_packs:
2107
operation = cleanup.OperationWithCleanups(resumed_pack.abort)
2108
# See comment in previous finally block.
2109
operation.add_cleanup(self._remove_pack_indices, resumed_pack,
2110
ignore_missing=True)
2111
operation.run_simple()
2112
del self._resumed_packs[:]
2114
def _remove_resumed_pack_indices(self):
2115
for resumed_pack in self._resumed_packs:
2116
self._remove_pack_indices(resumed_pack)
2117
del self._resumed_packs[:]
2119
def _check_new_inventories(self):
2120
"""Detect missing inventories in this write group.
2122
:returns: list of strs, summarising any problems found. If the list is
2123
empty no problems were found.
2125
# The base implementation does no checks. GCRepositoryPackCollection
856
self._open_pack_hash = None
2129
858
def _commit_write_group(self):
2131
for prefix, versioned_file in (
2132
('revisions', self.repo.revisions),
2133
('inventories', self.repo.inventories),
2134
('texts', self.repo.texts),
2135
('signatures', self.repo.signatures),
2137
missing = versioned_file.get_missing_compression_parent_keys()
2138
all_missing.update([(prefix,) + key for key in missing])
2140
raise errors.BzrCheckError(
2141
"Repository %s has missing compression parent(s) %r "
2142
% (self.repo, sorted(all_missing)))
2143
problems = self._check_new_inventories()
2145
problems_summary = '\n'.join(problems)
2146
raise errors.BzrCheckError(
2147
"Cannot add revision(s) to repository: " + problems_summary)
2148
self._remove_pack_indices(self._new_pack)
2149
any_new_content = False
2150
if self._new_pack.data_inserted():
2151
# get all the data to disk and read to use
2152
self._new_pack.finish()
2153
self.allocate(self._new_pack)
2154
self._new_pack = None
2155
any_new_content = True
2157
self._new_pack.abort()
2158
self._new_pack = None
2159
for resumed_pack in self._resumed_packs:
2160
# XXX: this is a pretty ugly way to turn the resumed pack into a
2161
# properly committed pack.
2162
self._names[resumed_pack.name] = None
2163
self._remove_pack_from_memory(resumed_pack)
2164
resumed_pack.finish()
2165
self.allocate(resumed_pack)
2166
any_new_content = True
2167
del self._resumed_packs[:]
2169
result = self.autopack()
2171
# when autopack takes no steps, the names list is still
2173
return self._save_pack_names()
2177
def _suspend_write_group(self):
2178
tokens = [pack.name for pack in self._resumed_packs]
2179
self._remove_pack_indices(self._new_pack)
2180
if self._new_pack.data_inserted():
2181
# get all the data to disk and read to use
2182
self._new_pack.finish(suspend=True)
2183
tokens.append(self._new_pack.name)
2184
self._new_pack = None
2186
self._new_pack.abort()
2187
self._new_pack = None
2188
self._remove_resumed_pack_indices()
2191
def _resume_write_group(self, tokens):
2192
for token in tokens:
2193
self._resume_pack(token)
2196
class KnitPackRepository(KnitRepository):
2197
"""Repository with knit objects stored inside pack containers.
2199
The layering for a KnitPackRepository is:
2201
Graph | HPSS | Repository public layer |
2202
===================================================
2203
Tuple based apis below, string based, and key based apis above
2204
---------------------------------------------------
2206
Provides .texts, .revisions etc
2207
This adapts the N-tuple keys to physical knit records which only have a
2208
single string identifier (for historical reasons), which in older formats
2209
was always the revision_id, and in the mapped code for packs is always
2210
the last element of key tuples.
2211
---------------------------------------------------
2213
A separate GraphIndex is used for each of the
2214
texts/inventories/revisions/signatures contained within each individual
2215
pack file. The GraphIndex layer works in N-tuples and is unaware of any
2217
===================================================
2221
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
859
data_inserted = (self.repo._revision_store.data_inserted() or
860
self.repo.weave_store.data_inserted() or
861
self.repo._inv_thunk.data_inserted())
863
self._open_pack_writer.end()
864
new_name = self._open_pack_hash.hexdigest()
866
new_pack.name = new_name
867
new_pack.transport = self._upload_transport.clone('../packs/')
869
# new_pack.revision_index =
870
# new_pack.inventory_index =
871
# new_pack.text_index =
872
# new_pack.signature_index =
873
self.repo.weave_store.flush(new_name, new_pack)
874
self.repo._inv_thunk.flush(new_name, new_pack)
875
self.repo._revision_store.flush(new_name, new_pack)
876
self._write_stream.close()
877
self._upload_transport.rename(self.repo._open_pack_tuple[1],
878
'../packs/' + new_name + '.pack')
879
# If this fails, its a hash collision. We should:
880
# - determine if its a collision or
881
# - the same content or
882
# - the existing name is not the actual hash - e.g.
883
# its a deliberate attack or data corruption has
884
# occuring during the write of that file.
885
self.allocate(new_name, new_pack.revision_index_length,
886
new_pack.inventory_index_length, new_pack.text_index_length,
887
new_pack.signature_index_length)
888
self.repo._open_pack_tuple = None
889
if not self.autopack():
890
self._save_pack_names()
892
# remove the pending upload
893
self._upload_transport.delete(self.repo._open_pack_tuple[1])
894
self.repo._revision_store.reset()
895
self.repo.weave_store.reset()
896
self.repo._inv_thunk.reset()
897
# forget what names there are - should just refresh and deal with the
900
self._open_pack_hash = None
901
self._write_stream = None
904
class GraphKnitRevisionStore(KnitRevisionStore):
905
"""An object to adapt access from RevisionStore's to use GraphKnits.
907
This should not live through to production: by production time we should
908
have fully integrated the new indexing and have new data for the
909
repository classes; also we may choose not to do a Knit1 compatible
910
new repository, just a Knit3 one. If neither of these happen, this
911
should definately be cleaned up before merging.
913
This class works by replacing the original RevisionStore.
914
We need to do this because the GraphKnitRevisionStore is less
915
isolated in its layering - it uses services from the repo.
918
def __init__(self, repo, transport, revisionstore):
919
"""Create a GraphKnitRevisionStore on repo with revisionstore.
921
This will store its state in the Repository, use the
922
indices FileNames to provide a KnitGraphIndex,
923
and at the end of transactions write new indices.
925
KnitRevisionStore.__init__(self, revisionstore.versioned_file_store)
927
self._serializer = revisionstore._serializer
928
self.transport = transport
930
def get_revision_file(self, transaction):
931
"""Get the revision versioned file object."""
932
if getattr(self.repo, '_revision_knit', None) is not None:
933
return self.repo._revision_knit
934
pack_map, indices = self.repo._packs._make_index_map('.rix')
935
if self.repo.is_in_write_group():
936
# allow writing: queue writes to a new index
937
indices.insert(0, self.repo._revision_write_index)
938
pack_map[self.repo._revision_write_index] = self.repo._open_pack_tuple
939
writer = self.repo._packs._open_pack_writer, self.repo._revision_write_index
940
add_callback = self.repo._revision_write_index.add_nodes
943
add_callback = None # no data-adding permitted.
944
self.repo._revision_all_indices = CombinedGraphIndex(indices)
945
knit_index = KnitGraphIndex(self.repo._revision_all_indices,
946
add_callback=add_callback)
947
knit_access = _PackAccess(pack_map, writer)
948
self.repo._revision_pack_map = pack_map
949
self.repo._revision_knit_access = knit_access
950
self.repo._revision_knit = knit.KnitVersionedFile(
951
'revisions', self.transport.clone('..'),
952
self.repo.control_files._file_mode,
953
create=False, access_mode=self.repo.control_files._lock_mode,
954
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
955
access_method=knit_access)
956
return self.repo._revision_knit
958
def get_signature_file(self, transaction):
959
"""Get the signature versioned file object."""
960
if getattr(self.repo, '_signature_knit', None) is not None:
961
return self.repo._signature_knit
962
pack_map, indices = self.repo._packs._make_index_map('.six')
963
if self.repo.is_in_write_group():
964
# allow writing: queue writes to a new index
965
indices.insert(0, self.repo._signature_write_index)
966
pack_map[self.repo._signature_write_index] = self.repo._open_pack_tuple
967
writer = self.repo._packs._open_pack_writer, self.repo._signature_write_index
968
add_callback = self.repo._signature_write_index.add_nodes
971
add_callback = None # no data-adding permitted.
972
self.repo._signature_all_indices = CombinedGraphIndex(indices)
973
knit_index = KnitGraphIndex(self.repo._signature_all_indices,
974
add_callback=add_callback, parents=False)
975
knit_access = _PackAccess(pack_map, writer)
976
self.repo._signature_knit_access = knit_access
977
self.repo._signature_knit = knit.KnitVersionedFile(
978
'signatures', self.transport.clone('..'),
979
self.repo.control_files._file_mode,
980
create=False, access_mode=self.repo.control_files._lock_mode,
981
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
982
access_method=knit_access)
983
return self.repo._signature_knit
985
def data_inserted(self):
986
if (getattr(self.repo, '_revision_write_index', None) and
987
self.repo._revision_write_index.key_count()):
989
if (getattr(self.repo, '_signature_write_index', None) and
990
self.repo._signature_write_index.key_count()):
994
def flush(self, new_name, new_pack):
995
"""Write out pending indices."""
996
# write a revision index (might be empty)
997
new_index_name = self.name_to_revision_index_name(new_name)
998
new_pack.revision_index_length = self.transport.put_file(
999
new_index_name, self.repo._revision_write_index.finish())
1000
if self.repo._revision_all_indices is None:
1001
# create a pack map for the autopack code - XXX finish
1002
# making a clear managed list of packs, indices and use
1003
# that in these mapping classes
1004
self.repo._revision_pack_map = self.repo._packs._make_index_map('.rix')[0]
1006
del self.repo._revision_pack_map[self.repo._revision_write_index]
1007
self.repo._revision_write_index = None
1008
new_index = GraphIndex(self.transport, new_index_name,
1009
new_pack.revision_index_length)
1010
self.repo._revision_pack_map[new_index] = (self.repo._packs._pack_tuple(new_name))
1011
# revisions 'knit' accessed : update it.
1012
self.repo._revision_all_indices.insert_index(0, new_index)
1013
# remove the write buffering index. XXX: API break
1014
# - clearly we need a remove_index call too.
1015
del self.repo._revision_all_indices._indices[1]
1016
# reset the knit access writer
1017
self.repo._revision_knit_access.set_writer(None, None, (None, None))
1019
# write a signatures index (might be empty)
1020
new_index_name = self.name_to_signature_index_name(new_name)
1021
new_pack.signature_index_length = self.transport.put_file(
1022
new_index_name, self.repo._signature_write_index.finish())
1023
self.repo._signature_write_index = None
1024
if self.repo._signature_all_indices is not None:
1025
# sigatures 'knit' accessed : update it.
1026
self.repo._signature_all_indices.insert_index(0,
1027
GraphIndex(self.transport, new_index_name,
1028
new_pack.signature_index_length))
1029
# remove the write buffering index. XXX: API break
1030
# - clearly we need a remove_index call too.
1031
del self.repo._signature_all_indices._indices[1]
1032
# reset the knit access writer
1033
self.repo._signature_knit_access.set_writer(None, None, (None, None))
1035
def name_to_revision_index_name(self, name):
1036
"""The revision index is the name + .rix."""
1037
return name + '.rix'
1039
def name_to_signature_index_name(self, name):
1040
"""The signature index is the name + .six."""
1041
return name + '.six'
1044
"""Clear all cached data."""
1045
# cached revision data
1046
self.repo._revision_knit = None
1047
self.repo._revision_write_index = None
1048
self.repo._revision_all_indices = None
1049
self.repo._revision_knit_access = None
1050
# cached signature data
1051
self.repo._signature_knit = None
1052
self.repo._signature_write_index = None
1053
self.repo._signature_all_indices = None
1054
self.repo._signature_knit_access = None
1057
# setup in-memory indices to accumulate data.
1058
self.repo._revision_write_index = InMemoryGraphIndex(1)
1059
self.repo._signature_write_index = InMemoryGraphIndex(0)
1060
# if knit indices have been handed out, add a mutable
1062
if self.repo._revision_knit is not None:
1063
self.repo._revision_all_indices.insert_index(0, self.repo._revision_write_index)
1064
self.repo._revision_knit._index._add_callback = self.repo._revision_write_index.add_nodes
1065
self.repo._revision_knit_access.set_writer(
1066
self.repo._packs._open_pack_writer,
1067
self.repo._revision_write_index, self.repo._open_pack_tuple)
1068
if self.repo._signature_knit is not None:
1069
self.repo._signature_all_indices.insert_index(0, self.repo._signature_write_index)
1070
self.repo._signature_knit._index._add_callback = self.repo._signature_write_index.add_nodes
1071
self.repo._signature_knit_access.set_writer(
1072
self.repo._packs._open_pack_writer,
1073
self.repo._signature_write_index, self.repo._open_pack_tuple)
1076
class GraphKnitTextStore(VersionedFileStore):
1077
"""An object to adapt access from VersionedFileStore's to use GraphKnits.
1079
This should not live through to production: by production time we should
1080
have fully integrated the new indexing and have new data for the
1081
repository classes; also we may choose not to do a Knit1 compatible
1082
new repository, just a Knit3 one. If neither of these happen, this
1083
should definately be cleaned up before merging.
1085
This class works by replacing the original VersionedFileStore.
1086
We need to do this because the GraphKnitRevisionStore is less
1087
isolated in its layering - it uses services from the repo and shares them
1088
with all the data written in a single write group.
1091
def __init__(self, repo, transport, weavestore):
1092
"""Create a GraphKnitTextStore on repo with weavestore.
1094
This will store its state in the Repository, use the
1095
indices FileNames to provide a KnitGraphIndex,
1096
and at the end of transactions write new indices.
1098
# don't call base class constructor - its not suitable.
1099
# no transient data stored in the transaction
1101
self._precious = False
1103
self.transport = transport
1104
self.weavestore = weavestore
1105
# XXX for check() which isn't updated yet
1106
self._transport = weavestore._transport
1108
def data_inserted(self):
1109
# XXX: Should we define __len__ for indices?
1110
if (getattr(self.repo, '_text_write_index', None) and
1111
self.repo._text_write_index.key_count()):
1114
def _ensure_all_index(self, for_write=None):
1115
"""Create the combined index for all texts."""
1116
if getattr(self.repo, '_text_all_indices', None) is not None:
1118
pack_map, indices = self.repo._packs._make_index_map('.tix')
1119
self.repo._text_pack_map = pack_map
1120
if for_write or self.repo.is_in_write_group():
1121
# allow writing: queue writes to a new index
1122
indices.insert(0, self.repo._text_write_index)
1123
self._setup_knit(self.repo.is_in_write_group())
1124
self.repo._text_all_indices = CombinedGraphIndex(indices)
1126
def flush(self, new_name, new_pack):
1127
"""Write the index out to new_name."""
1128
# write a revision index (might be empty)
1129
new_index_name = self.name_to_text_index_name(new_name)
1130
new_pack.text_index_length = self.transport.put_file(
1131
new_index_name, self.repo._text_write_index.finish())
1132
self.repo._text_write_index = None
1133
self._setup_knit(False)
1134
if self.repo._text_all_indices is not None:
1135
# text 'knits' have been used, replace the mutated memory index
1136
# with the new on-disk one. XXX: is this really a good idea?
1137
# perhaps just keep using the memory one ?
1138
self.repo._text_all_indices.insert_index(0,
1139
GraphIndex(self.transport, new_index_name,
1140
new_pack.text_index_length))
1141
# remove the write buffering index. XXX: API break
1142
# - clearly we need a remove_index call too.
1143
del self.repo._text_all_indices._indices[1]
1145
def get_weave_or_empty(self, file_id, transaction, force_write=False):
1146
"""Get a 'Knit' backed by the .tix indices.
1148
The transaction parameter is ignored.
1150
self._ensure_all_index()
1151
if force_write or self.repo.is_in_write_group():
1152
add_callback = self.repo._text_write_index.add_nodes
1153
self.repo._text_pack_map[self.repo._text_write_index] = self.repo._open_pack_tuple
1155
add_callback = None # no data-adding permitted.
1157
file_id_index = GraphIndexPrefixAdapter(self.repo._text_all_indices,
1158
(file_id, ), 1, add_nodes_callback=add_callback)
1159
knit_index = KnitGraphIndex(file_id_index,
1160
add_callback=file_id_index.add_nodes,
1161
deltas=True, parents=True)
1162
return knit.KnitVersionedFile('text:' + file_id,
1163
self.transport.clone('..'),
1166
access_method=self.repo._text_knit_access,
1167
factory=knit.KnitPlainFactory())
1169
get_weave = get_weave_or_empty
1172
"""Generate a list of the fileids inserted, for use by check."""
1173
self._ensure_all_index()
1175
for index, key, value, refs in self.repo._text_all_indices.iter_all_entries():
1179
def name_to_text_index_name(self, name):
1180
"""The text index is the name + .tix."""
1181
return name + '.tix'
1184
"""Clear all cached data."""
1185
# remove any accumlating index of text data
1186
self.repo._text_write_index = None
1188
self.repo._text_knit_access = None
1190
self.repo._text_knit = None
1191
# remove all constructed text data indices
1192
self.repo._text_all_indices = None
1194
self.repo._text_pack_map = None
1197
# setup in-memory indices to accumulate data.
1198
self.repo._text_write_index = InMemoryGraphIndex(reference_lists=2,
1200
# we require that text 'knits' be accessed from within the write
1201
# group to be able to be written to, simply because it makes this
1202
# code cleaner - we don't need to track all 'open' knits and
1204
# prepare to do writes.
1205
self._ensure_all_index(True)
1206
self._setup_knit(True)
1208
def _setup_knit(self, for_write):
1210
writer = (self.repo._packs._open_pack_writer, self.repo._text_write_index)
1213
self.repo._text_knit_access = _PackAccess(
1214
self.repo._text_pack_map, writer)
1216
# a reused knit object for commit specifically.
1217
self.repo._text_knit = self.get_weave_or_empty(
1218
'all-texts', self.repo.get_transaction(), for_write)
1220
self.repo._text_knit = None
1223
class InventoryKnitThunk(object):
1224
"""An object to manage thunking get_inventory_weave to pack based knits."""
1226
def __init__(self, repo, transport):
1227
"""Create an InventoryKnitThunk for repo at transport.
1229
This will store its state in the Repository, use the
1230
indices FileNames to provide a KnitGraphIndex,
1231
and at the end of transactions write a new index..
1234
self.transport = transport
1236
def data_inserted(self):
1237
# XXX: Should we define __len__ for indices?
1238
if (getattr(self.repo, '_inv_write_index', None) and
1239
self.repo._inv_write_index.key_count()):
1242
def _ensure_all_index(self):
1243
"""Create the combined index for all inventories."""
1244
if getattr(self.repo, '_inv_all_indices', None) is not None:
1246
pack_map, indices = self.repo._packs._make_index_map('.iix')
1247
if self.repo.is_in_write_group():
1248
# allow writing: queue writes to a new index
1249
indices.append(self.repo._inv_write_index)
1250
self.repo._inv_all_indices = CombinedGraphIndex(indices)
1251
self.repo._inv_pack_map = pack_map
1253
def flush(self, new_name, new_pack):
1254
"""Write the index out to new_name."""
1255
# write an index (might be empty)
1256
new_index_name = self.name_to_inv_index_name(new_name)
1257
new_pack.inventory_index_length = self.transport.put_file(
1258
new_index_name, self.repo._inv_write_index.finish())
1259
self.repo._inv_write_index = None
1260
if self.repo._inv_all_indices is not None:
1261
# inv 'knit' has been used, replace the mutated memory index
1262
# with the new on-disk one. XXX: is this really a good idea?
1263
# perhaps just keep using the memory one ?
1264
self.repo._inv_all_indices.insert_index(0,
1265
GraphIndex(self.transport, new_index_name,
1266
new_pack.inventory_index_length))
1267
# remove the write buffering index. XXX: API break
1268
# - clearly we need a remove_index call too.
1269
del self.repo._inv_all_indices._indices[1]
1270
self.repo._inv_knit_access.set_writer(None, None, (None, None))
1271
self.repo._inv_pack_map = None
1273
def get_weave(self):
1274
"""Get a 'Knit' that contains inventory data."""
1275
self._ensure_all_index()
1276
filename = 'inventory'
1277
if self.repo.is_in_write_group():
1278
add_callback = self.repo._inv_write_index.add_nodes
1279
self.repo._inv_pack_map[self.repo._inv_write_index] = self.repo._open_pack_tuple
1280
writer = self.repo._packs._open_pack_writer, self.repo._inv_write_index
1282
add_callback = None # no data-adding permitted.
1285
knit_index = KnitGraphIndex(self.repo._inv_all_indices,
1286
add_callback=add_callback,
1287
deltas=True, parents=True)
1288
# TODO - mode support. self.weavestore._file_mode,
1289
knit_access = _PackAccess(self.repo._inv_pack_map, writer)
1290
self.repo._inv_knit_access = knit_access
1291
return knit.KnitVersionedFile('inventory', self.transport.clone('..'),
1293
factory=knit.KnitPlainFactory(),
1294
access_method=knit_access)
1296
def name_to_inv_index_name(self, name):
1297
"""The inv index is the name + .iix."""
1298
return name + '.iix'
1301
"""Clear all cached data."""
1302
# remove any accumlating index of inv data
1303
self.repo._inv_write_index = None
1304
# remove all constructed inv data indices
1305
self.repo._inv_all_indices = None
1306
# remove the knit access object
1307
self.repo._inv_knit_access = None
1308
self.repo._inv_pack_map = None
1311
# setup in-memory indices to accumulate data.
1312
# - we want to map compression only, but currently the knit code hasn't
1313
# been updated enough to understand that, so we have a regular 2-list
1314
# index giving parents and compression source.
1315
self.repo._inv_write_index = InMemoryGraphIndex(reference_lists=2)
1316
# if we have created an inventory index, add the new write index to it
1317
if getattr(self.repo, '_inv_all_indices', None) is not None:
1318
self.repo._inv_all_indices.insert_index(0, self.repo._inv_write_index)
1319
# we don't bother updating the knit layer, because there is not
1320
# defined interface for adding inventories that should need the
1321
# existing knit to be changed - its all behind 'repo.add_inventory'.
1324
class GraphKnitRepository(KnitRepository):
1325
"""Experimental graph-knit using repository."""
1327
def __init__(self, _format, a_bzrdir, control_files, _revision_store,
1328
control_store, text_store, _commit_builder_class, _serializer):
2223
1329
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2224
_commit_builder_class, _serializer)
2225
index_transport = self._transport.clone('indices')
2226
self._pack_collection = RepositoryPackCollection(self, self._transport,
1330
_revision_store, control_store, text_store, _commit_builder_class,
1332
index_transport = control_files._transport.clone('indices')
1333
self._packs = RepositoryPackCollection(self, control_files._transport,
2227
1334
index_transport,
2228
self._transport.clone('upload'),
2229
self._transport.clone('packs'),
2230
_format.index_builder_class,
2231
_format.index_class,
2232
use_chk_index=self._format.supports_chks,
2234
self.inventories = KnitVersionedFiles(
2235
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2236
add_callback=self._pack_collection.inventory_index.add_callback,
2237
deltas=True, parents=True, is_locked=self.is_locked),
2238
data_access=self._pack_collection.inventory_index.data_access,
2239
max_delta_chain=200)
2240
self.revisions = KnitVersionedFiles(
2241
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2242
add_callback=self._pack_collection.revision_index.add_callback,
2243
deltas=False, parents=True, is_locked=self.is_locked,
2244
track_external_parent_refs=True),
2245
data_access=self._pack_collection.revision_index.data_access,
2247
self.signatures = KnitVersionedFiles(
2248
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2249
add_callback=self._pack_collection.signature_index.add_callback,
2250
deltas=False, parents=False, is_locked=self.is_locked),
2251
data_access=self._pack_collection.signature_index.data_access,
2253
self.texts = KnitVersionedFiles(
2254
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2255
add_callback=self._pack_collection.text_index.add_callback,
2256
deltas=True, parents=True, is_locked=self.is_locked),
2257
data_access=self._pack_collection.text_index.data_access,
2258
max_delta_chain=200)
2259
if _format.supports_chks:
2260
# No graph, no compression:- references from chks are between
2261
# different objects not temporal versions of the same; and without
2262
# some sort of temporal structure knit compression will just fail.
2263
self.chk_bytes = KnitVersionedFiles(
2264
_KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2265
add_callback=self._pack_collection.chk_index.add_callback,
2266
deltas=False, parents=False, is_locked=self.is_locked),
2267
data_access=self._pack_collection.chk_index.data_access,
2270
self.chk_bytes = None
2271
# True when the repository object is 'write locked' (as opposed to the
2272
# physical lock only taken out around changes to the pack-names list.)
2273
# Another way to represent this would be a decorator around the control
2274
# files object that presents logical locks as physical ones - if this
2275
# gets ugly consider that alternative design. RBC 20071011
2276
self._write_lock_count = 0
2277
self._transaction = None
1335
control_files._transport.clone('upload'),
1336
control_files._transport.clone('packs'))
1337
self._revision_store = GraphKnitRevisionStore(self, index_transport, self._revision_store)
1338
self.weave_store = GraphKnitTextStore(self, index_transport, self.weave_store)
1339
self._inv_thunk = InventoryKnitThunk(self, index_transport)
2279
self._reconcile_does_inventory_gc = True
2280
self._reconcile_fixes_text_parents = True
2281
self._reconcile_backsup_inventory = False
2283
def _warn_if_deprecated(self, branch=None):
2284
# This class isn't deprecated, but one sub-format is
2285
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2286
super(KnitPackRepository, self)._warn_if_deprecated(branch)
1341
self._reconcile_does_inventory_gc = False
2288
1343
def _abort_write_group(self):
2289
self.revisions._index._key_dependencies.clear()
2290
self._pack_collection._abort_write_group()
2292
def _get_source(self, to_format):
2293
if to_format.network_name() == self._format.network_name():
2294
return KnitPackStreamSource(self, to_format)
2295
return super(KnitPackRepository, self)._get_source(to_format)
2297
def _make_parents_provider(self):
2298
return graph.CachingParentsProvider(self)
1344
self._packs._abort_write_group()
2300
1346
def _refresh_data(self):
2301
if not self.is_locked():
2303
self._pack_collection.reload_pack_names()
1347
if self.control_files._lock_count==1:
1348
self._revision_store.reset()
1349
self.weave_store.reset()
1350
self._inv_thunk.reset()
1351
# forget what names there are
2305
1354
def _start_write_group(self):
2306
self._pack_collection._start_write_group()
1355
self._packs._start_write_group()
2308
1357
def _commit_write_group(self):
2309
hint = self._pack_collection._commit_write_group()
2310
self.revisions._index._key_dependencies.clear()
2313
def suspend_write_group(self):
2314
# XXX check self._write_group is self.get_transaction()?
2315
tokens = self._pack_collection._suspend_write_group()
2316
self.revisions._index._key_dependencies.clear()
2317
self._write_group = None
2320
def _resume_write_group(self, tokens):
2321
self._start_write_group()
2323
self._pack_collection._resume_write_group(tokens)
2324
except errors.UnresumableWriteGroup:
2325
self._abort_write_group()
2327
for pack in self._pack_collection._resumed_packs:
2328
self.revisions._index.scan_unvalidated_index(pack.revision_index)
2330
def get_transaction(self):
2331
if self._write_lock_count:
2332
return self._transaction
2334
return self.control_files.get_transaction()
2336
def is_locked(self):
2337
return self._write_lock_count or self.control_files.is_locked()
2339
def is_write_locked(self):
2340
return self._write_lock_count
2342
def lock_write(self, token=None):
2343
locked = self.is_locked()
2344
if not self._write_lock_count and locked:
2345
raise errors.ReadOnlyError(self)
2346
self._write_lock_count += 1
2347
if self._write_lock_count == 1:
2348
self._transaction = transactions.WriteTransaction()
2350
if 'relock' in debug.debug_flags and self._prev_lock == 'w':
2351
note('%r was write locked again', self)
2352
self._prev_lock = 'w'
2353
for repo in self._fallback_repositories:
2354
# Writes don't affect fallback repos
2356
self._refresh_data()
2358
def lock_read(self):
2359
locked = self.is_locked()
2360
if self._write_lock_count:
2361
self._write_lock_count += 1
2363
self.control_files.lock_read()
2365
if 'relock' in debug.debug_flags and self._prev_lock == 'r':
2366
note('%r was read locked again', self)
2367
self._prev_lock = 'r'
2368
for repo in self._fallback_repositories:
2370
self._refresh_data()
2372
def leave_lock_in_place(self):
2373
# not supported - raise an error
2374
raise NotImplementedError(self.leave_lock_in_place)
2376
def dont_leave_lock_in_place(self):
2377
# not supported - raise an error
2378
raise NotImplementedError(self.dont_leave_lock_in_place)
1358
return self._packs._commit_write_group()
1360
def get_inventory_weave(self):
1361
return self._inv_thunk.get_weave()
2380
1363
@needs_write_lock
2381
def pack(self, hint=None, clean_obsolete_packs=False):
2382
1365
"""Compress the data within the repository.
2384
1367
This will pack all the data to a single pack. In future it may
2385
1368
recompress deltas or do other such expensive operations.
2387
self._pack_collection.pack(hint=hint, clean_obsolete_packs=clean_obsolete_packs)
2389
1372
@needs_write_lock
2390
1373
def reconcile(self, other=None, thorough=False):
2542
1453
mutter('creating repository in %s.', a_bzrdir.transport.base)
2543
1454
dirs = ['indices', 'obsolete_packs', 'packs', 'upload']
2544
builder = self.index_builder_class()
1455
builder = GraphIndexBuilder()
2545
1456
files = [('pack-names', builder.finish())]
2546
1457
utf8_files = [('format', self.get_format_string())]
2548
1459
self._upload_blank_content(a_bzrdir, dirs, files, utf8_files, shared)
2549
repository = self.open(a_bzrdir=a_bzrdir, _found=True)
2550
self._run_post_repo_init_hooks(repository, a_bzrdir, shared)
1460
return self.open(a_bzrdir=a_bzrdir, _found=True)
2553
1462
def open(self, a_bzrdir, _found=False, _override_transport=None):
2554
1463
"""See RepositoryFormat.open().
2556
1465
:param _override_transport: INTERNAL USE ONLY. Allows opening the
2557
1466
repository at a slightly different url
2558
1467
than normal. I.e. during 'upgrade'.
2561
1470
format = RepositoryFormat.find_format(a_bzrdir)
1471
assert format.__class__ == self.__class__
2562
1472
if _override_transport is not None:
2563
1473
repo_transport = _override_transport
2565
1475
repo_transport = a_bzrdir.get_repository_transport(None)
2566
1476
control_files = lockable_files.LockableFiles(repo_transport,
2567
1477
'lock', lockdir.LockDir)
1478
text_store = self._get_text_store(repo_transport, control_files)
1479
control_store = self._get_control_store(repo_transport, control_files)
1480
_revision_store = self._get_revision_store(repo_transport, control_files)
2568
1481
return self.repository_class(_format=self,
2569
1482
a_bzrdir=a_bzrdir,
2570
1483
control_files=control_files,
1484
_revision_store=_revision_store,
1485
control_store=control_store,
1486
text_store=text_store,
2571
1487
_commit_builder_class=self._commit_builder_class,
2572
1488
_serializer=self._serializer)
2575
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2576
"""A no-subtrees parameterized Pack repository.
2578
This format was introduced in 0.92.
1491
class RepositoryFormatGraphKnit1(RepositoryFormatPack):
1492
"""Experimental pack based repository with knit1 style data.
1494
This repository format has:
1495
- knits for file texts and inventory
1496
- hash subdirectory based stores.
1497
- knits for revisions and signatures
1498
- uses a GraphKnitIndex for revisions.knit.
1499
- TextStores for revisions and signatures.
1500
- a format marker of its own
1501
- an optional 'shared-storage' flag
1502
- an optional 'no-working-trees' flag
1505
This format was introduced in bzr.dev.
2581
repository_class = KnitPackRepository
1508
repository_class = GraphKnitRepository
2582
1509
_commit_builder_class = PackCommitBuilder
2584
def _serializer(self):
2585
return xml5.serializer_v5
2586
# What index classes to use
2587
index_builder_class = InMemoryGraphIndex
2588
index_class = GraphIndex
1510
_serializer = xml5.serializer_v5
2590
1512
def _get_matching_bzrdir(self):
2591
return bzrdir.format_registry.make_bzrdir('pack-0.92')
1513
return bzrdir.format_registry.make_bzrdir('experimental')
2593
1515
def _ignore_setting_bzrdir(self, format):
2596
1518
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1520
def __ne__(self, other):
1521
return self.__class__ is not other.__class__
2598
1523
def get_format_string(self):
2599
1524
"""See RepositoryFormat.get_format_string()."""
2600
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
1525
return "Bazaar Experimental no-subtrees\n"
2602
1527
def get_format_description(self):
2603
1528
"""See RepositoryFormat.get_format_description()."""
2604
return "Packs containing knits without subtree support"
2607
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2608
"""A subtrees parameterized Pack repository.
2610
This repository format uses the xml7 serializer to get:
1529
return "Experimental no-subtrees"
1531
def check_conversion_target(self, target_format):
1535
class RepositoryFormatGraphKnit3(RepositoryFormatPack):
1536
"""Experimental repository with knit3 style data.
1538
This repository format has:
1539
- knits for file texts and inventory
1540
- hash subdirectory based stores.
1541
- knits for revisions and signatures
1542
- uses a GraphKnitIndex for revisions.knit.
1543
- TextStores for revisions and signatures.
1544
- a format marker of its own
1545
- an optional 'shared-storage' flag
1546
- an optional 'no-working-trees' flag
2611
1548
- support for recording full info about the tree root
2612
1549
- support for recording tree-references
2614
This format was introduced in 0.92.
2617
repository_class = KnitPackRepository
2618
_commit_builder_class = PackRootCommitBuilder
2619
rich_root_data = True
2621
supports_tree_reference = True
2623
def _serializer(self):
2624
return xml7.serializer_v7
2625
# What index classes to use
2626
index_builder_class = InMemoryGraphIndex
2627
index_class = GraphIndex
2629
def _get_matching_bzrdir(self):
2630
return bzrdir.format_registry.make_bzrdir(
2631
'pack-0.92-subtree')
2633
def _ignore_setting_bzrdir(self, format):
2636
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2638
def get_format_string(self):
2639
"""See RepositoryFormat.get_format_string()."""
2640
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2642
def get_format_description(self):
2643
"""See RepositoryFormat.get_format_description()."""
2644
return "Packs containing knits with subtree support\n"
2647
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2648
"""A rich-root, no subtrees parameterized Pack repository.
2650
This repository format uses the xml6 serializer to get:
2651
- support for recording full info about the tree root
2653
This format was introduced in 1.0.
2656
repository_class = KnitPackRepository
2657
_commit_builder_class = PackRootCommitBuilder
2658
rich_root_data = True
2659
supports_tree_reference = False
2661
def _serializer(self):
2662
return xml6.serializer_v6
2663
# What index classes to use
2664
index_builder_class = InMemoryGraphIndex
2665
index_class = GraphIndex
2667
def _get_matching_bzrdir(self):
2668
return bzrdir.format_registry.make_bzrdir(
2671
def _ignore_setting_bzrdir(self, format):
2674
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2676
def get_format_string(self):
2677
"""See RepositoryFormat.get_format_string()."""
2678
return ("Bazaar pack repository format 1 with rich root"
2679
" (needs bzr 1.0)\n")
2681
def get_format_description(self):
2682
"""See RepositoryFormat.get_format_description()."""
2683
return "Packs containing knits with rich root support\n"
2686
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2687
"""Repository that supports external references to allow stacking.
2691
Supports external lookups, which results in non-truncated ghosts after
2692
reconcile compared to pack-0.92 formats.
2695
repository_class = KnitPackRepository
2696
_commit_builder_class = PackCommitBuilder
2697
supports_external_lookups = True
2698
# What index classes to use
2699
index_builder_class = InMemoryGraphIndex
2700
index_class = GraphIndex
2703
def _serializer(self):
2704
return xml5.serializer_v5
2706
def _get_matching_bzrdir(self):
2707
return bzrdir.format_registry.make_bzrdir('1.6')
2709
def _ignore_setting_bzrdir(self, format):
2712
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2714
def get_format_string(self):
2715
"""See RepositoryFormat.get_format_string()."""
2716
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2718
def get_format_description(self):
2719
"""See RepositoryFormat.get_format_description()."""
2720
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2723
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2724
"""A repository with rich roots and stacking.
2726
New in release 1.6.1.
2728
Supports stacking on other repositories, allowing data to be accessed
2729
without being stored locally.
2732
repository_class = KnitPackRepository
2733
_commit_builder_class = PackRootCommitBuilder
2734
rich_root_data = True
2735
supports_tree_reference = False # no subtrees
2736
supports_external_lookups = True
2737
# What index classes to use
2738
index_builder_class = InMemoryGraphIndex
2739
index_class = GraphIndex
2742
def _serializer(self):
2743
return xml6.serializer_v6
2745
def _get_matching_bzrdir(self):
2746
return bzrdir.format_registry.make_bzrdir(
2749
def _ignore_setting_bzrdir(self, format):
2752
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2754
def get_format_string(self):
2755
"""See RepositoryFormat.get_format_string()."""
2756
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2758
def get_format_description(self):
2759
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2762
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2763
"""A repository with rich roots and external references.
2767
Supports external lookups, which results in non-truncated ghosts after
2768
reconcile compared to pack-0.92 formats.
2770
This format was deprecated because the serializer it uses accidentally
2771
supported subtrees, when the format was not intended to. This meant that
2772
someone could accidentally fetch from an incorrect repository.
2775
repository_class = KnitPackRepository
2776
_commit_builder_class = PackRootCommitBuilder
2777
rich_root_data = True
2778
supports_tree_reference = False # no subtrees
2780
supports_external_lookups = True
2781
# What index classes to use
2782
index_builder_class = InMemoryGraphIndex
2783
index_class = GraphIndex
2786
def _serializer(self):
2787
return xml7.serializer_v7
2789
def _get_matching_bzrdir(self):
2790
matching = bzrdir.format_registry.make_bzrdir(
2792
matching.repository_format = self
2795
def _ignore_setting_bzrdir(self, format):
2798
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2800
def get_format_string(self):
2801
"""See RepositoryFormat.get_format_string()."""
2802
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2804
def get_format_description(self):
2805
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2809
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2810
"""A repository with stacking and btree indexes,
2811
without rich roots or subtrees.
2813
This is equivalent to pack-1.6 with B+Tree indices.
2816
repository_class = KnitPackRepository
2817
_commit_builder_class = PackCommitBuilder
2818
supports_external_lookups = True
2819
# What index classes to use
2820
index_builder_class = BTreeBuilder
2821
index_class = BTreeGraphIndex
2824
def _serializer(self):
2825
return xml5.serializer_v5
2827
def _get_matching_bzrdir(self):
2828
return bzrdir.format_registry.make_bzrdir('1.9')
2830
def _ignore_setting_bzrdir(self, format):
2833
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2835
def get_format_string(self):
2836
"""See RepositoryFormat.get_format_string()."""
2837
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2839
def get_format_description(self):
2840
"""See RepositoryFormat.get_format_description()."""
2841
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2844
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2845
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2847
1.6-rich-root with B+Tree indices.
2850
repository_class = KnitPackRepository
2851
_commit_builder_class = PackRootCommitBuilder
2852
rich_root_data = True
2853
supports_tree_reference = False # no subtrees
2854
supports_external_lookups = True
2855
# What index classes to use
2856
index_builder_class = BTreeBuilder
2857
index_class = BTreeGraphIndex
2860
def _serializer(self):
2861
return xml6.serializer_v6
2863
def _get_matching_bzrdir(self):
2864
return bzrdir.format_registry.make_bzrdir(
2867
def _ignore_setting_bzrdir(self, format):
2870
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2872
def get_format_string(self):
2873
"""See RepositoryFormat.get_format_string()."""
2874
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2876
def get_format_description(self):
2877
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2880
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2881
"""A subtrees development repository.
2883
This format should be retained until the second release after bzr 1.7.
2885
1.6.1-subtree[as it might have been] with B+Tree indices.
2887
This is [now] retained until we have a CHK based subtree format in
2891
repository_class = KnitPackRepository
2892
_commit_builder_class = PackRootCommitBuilder
2893
rich_root_data = True
2895
supports_tree_reference = True
2896
supports_external_lookups = True
2897
# What index classes to use
2898
index_builder_class = BTreeBuilder
2899
index_class = BTreeGraphIndex
2902
def _serializer(self):
2903
return xml7.serializer_v7
2905
def _get_matching_bzrdir(self):
2906
return bzrdir.format_registry.make_bzrdir(
2907
'development-subtree')
2909
def _ignore_setting_bzrdir(self, format):
2912
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2914
def get_format_string(self):
2915
"""See RepositoryFormat.get_format_string()."""
2916
return ("Bazaar development format 2 with subtree support "
2917
"(needs bzr.dev from before 1.8)\n")
2919
def get_format_description(self):
2920
"""See RepositoryFormat.get_format_description()."""
2921
return ("Development repository format, currently the same as "
2922
"1.6.1-subtree with B+Tree indices.\n")
1552
repository_class = GraphKnitRepository
1553
_commit_builder_class = PackRootCommitBuilder
1554
rich_root_data = True
1555
supports_tree_reference = True
1556
_serializer = xml7.serializer_v7
1558
def _get_matching_bzrdir(self):
1559
return bzrdir.format_registry.make_bzrdir('experimental-subtree')
1561
def _ignore_setting_bzrdir(self, format):
1564
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
1566
def check_conversion_target(self, target_format):
1567
if not target_format.rich_root_data:
1568
raise errors.BadConversionTarget(
1569
'Does not support rich root data.', target_format)
1570
if not getattr(target_format, 'supports_tree_reference', False):
1571
raise errors.BadConversionTarget(
1572
'Does not support nested trees', target_format)
1574
def get_format_string(self):
1575
"""See RepositoryFormat.get_format_string()."""
1576
return "Bazaar Experimental subtrees\n"
1578
def get_format_description(self):
1579
"""See RepositoryFormat.get_format_description()."""
1580
return "Experimental no-subtrees\n"
added
removed
bzrlib/repofmt/pack_repo.py
