261
206
return not self.__eq__(other)
263
208
def __repr__(self):
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)
209
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
210
id(self), self.transport, self.name)
338
213
class NewPack(Pack):
339
214
"""An in memory proxy for a pack which is being created."""
341
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
216
# A map of index 'type' to the file extension and position in the
218
index_definitions = {
219
'revision': ('.rix', 0),
220
'inventory': ('.iix', 1),
222
'signature': ('.six', 3),
225
def __init__(self, upload_transport, index_transport, pack_transport,
226
upload_suffix='', file_mode=None):
342
227
"""Create a NewPack instance.
344
:param pack_collection: A PackCollection into which this is being inserted.
229
:param upload_transport: A writable transport for the pack to be
230
incrementally uploaded to.
231
:param index_transport: A writable transport for the pack's indices to
232
be written to when the pack is finished.
233
:param pack_transport: A writable transport for the pack to be renamed
234
to when the upload is complete. This *must* be the same as
235
upload_transport.clone('../packs').
345
236
:param upload_suffix: An optional suffix to be given to any temporary
346
237
files created during the pack creation. e.g '.autopack'
347
:param file_mode: Unix permissions for newly created file.
238
:param file_mode: An optional file mode to create the new files with.
349
240
# The relative locations of the packs are constrained, but all are
350
241
# 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)
356
242
Pack.__init__(self,
357
243
# Revisions: parents list, no text compression.
358
index_builder_class(reference_lists=1),
244
InMemoryGraphIndex(reference_lists=1),
359
245
# Inventory: We want to map compression only, but currently the
360
246
# knit code hasn't been updated enough to understand that, so we
361
247
# have a regular 2-list index giving parents and compression
363
index_builder_class(reference_lists=2),
249
InMemoryGraphIndex(reference_lists=2),
364
250
# Texts: compression and per file graph, for all fileids - so two
365
251
# reference lists and two elements in the key tuple.
366
index_builder_class(reference_lists=2, key_elements=2),
252
InMemoryGraphIndex(reference_lists=2, key_elements=2),
367
253
# 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.
255
InMemoryGraphIndex(reference_lists=0),
373
self._pack_collection = pack_collection
374
# When we make readonly indices, we need this.
375
self.index_class = pack_collection._index_class
376
257
# where should the new pack be opened
377
self.upload_transport = pack_collection._upload_transport
258
self.upload_transport = upload_transport
378
259
# where are indices written out to
379
self.index_transport = pack_collection._index_transport
260
self.index_transport = index_transport
380
261
# where is the pack renamed to when it is finished?
381
self.pack_transport = pack_collection._pack_transport
262
self.pack_transport = pack_transport
382
263
# What file mode to upload the pack and indices with.
383
264
self._file_mode = file_mode
384
265
# 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)
266
self._hash = md5.new()
267
# a four-tuple with the length in bytes of the indices, once the pack
268
# is finalised. (rev, inv, text, sigs)
388
269
self.index_sizes = None
389
270
# How much data to cache when writing packs. Note that this is not
390
271
# synchronised with reads, because it's not in the transport layer, so
593
476
Future searches on the aggregate index will seach this new index
594
477
before all previously inserted indices.
596
479
:param index: An Index for the pack.
597
480
:param pack: A Pack instance.
599
482
# expose it to the index map
600
483
self.index_to_pack[index] = pack.access_tuple()
601
484
# put it at the front of the linear index list
602
self.combined_index.insert_index(0, index, pack.name)
485
self.combined_index.insert_index(0, index)
604
487
def add_writable_index(self, index, pack):
605
488
"""Add an index which is able to have data added to it.
607
490
There can be at most one writable index at any time. Any
608
491
modifications made to the knit are put into this index.
610
493
:param index: An index from the pack parameter.
611
494
: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))
496
assert self.add_callback is None, \
497
"%s already has a writable index through %s" % \
498
(self, self.add_callback)
617
499
# allow writing: queue writes to a new index
618
500
self.add_index(index, pack)
619
501
# Updates the index to packs mapping as a side effect,
620
self.data_access.set_writer(pack._writer, index, pack.access_tuple())
502
self.knit_access.set_writer(pack._writer, index, pack.access_tuple())
621
503
self.add_callback = index.add_nodes
624
506
"""Reset all the aggregate data to nothing."""
625
self.data_access.set_writer(None, None, (None, None))
507
self.knit_access.set_writer(None, None, (None, None))
626
508
self.index_to_pack.clear()
627
509
del self.combined_index._indices[:]
628
del self.combined_index._index_names[:]
629
510
self.add_callback = None
631
def remove_index(self, index):
512
def remove_index(self, index, pack):
632
513
"""Remove index from the indices used to answer queries.
634
515
:param index: An index from the pack parameter.
516
:param pack: A Pack instance.
636
518
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]
519
self.combined_index._indices.remove(index)
640
520
if (self.add_callback is not None and
641
521
getattr(index, 'add_nodes', None) == self.add_callback):
642
522
self.add_callback = None
643
self.data_access.set_writer(None, None, (None, None))
523
self.knit_access.set_writer(None, None, (None, None))
646
526
class Packer(object):
647
527
"""Create a pack from packs."""
649
def __init__(self, pack_collection, packs, suffix, revision_ids=None,
529
def __init__(self, pack_collection, packs, suffix, revision_ids=None):
651
530
"""Create a Packer.
653
532
:param pack_collection: A RepositoryPackCollection object where the
1048
def _do_copy_nodes_graph(self, index_map, writer, write_index,
1049
output_lines, pb, readv_group_iter, total_items):
808
def _do_copy_nodes_graph(self, nodes, index_map, writer, write_index,
1050
810
# for record verification
1051
knit = KnitVersionedFiles(None, None)
811
knit_data = _KnitData(None)
1052
812
# for line extraction when requested (inventories only)
1053
813
if output_lines:
1054
factory = KnitPlainFactory()
814
factory = knit.KnitPlainFactory()
815
# plan a readv on each source pack:
817
nodes = sorted(nodes)
818
# how to map this into knit.py - or knit.py into this?
819
# we don't want the typical knit logic, we want grouping by pack
820
# at this point - perhaps a helper library for the following code
821
# duplication points?
1055
823
record_index = 0
1056
pb.update("Copied record", record_index, total_items)
1057
for index, readv_vector, node_vector in readv_group_iter:
824
pb.update("Copied record", record_index, len(nodes))
825
for index, key, value, references in nodes:
826
if index not in request_groups:
827
request_groups[index] = []
828
request_groups[index].append((key, value, references))
829
for index, items in request_groups.iteritems():
830
pack_readv_requests = []
831
for key, value, references in items:
832
# ---- KnitGraphIndex.get_position
833
bits = value[1:].split(' ')
834
offset, length = int(bits[0]), int(bits[1])
835
pack_readv_requests.append((offset, length, (key, value[0], references)))
836
# linear scan up the pack
837
pack_readv_requests.sort()
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):
839
transport, path = index_map[index]
840
reader = pack.make_readv_reader(transport, path,
841
[offset[0:2] for offset in pack_readv_requests])
842
for (names, read_func), (_1, _2, (key, eol_flag, references)) in \
843
izip(reader.iter_records(), pack_readv_requests):
1069
844
raw_data = read_func(None)
1070
846
if output_lines:
1071
847
# read the entire thing
1072
content, _ = knit._parse_record(key[-1], raw_data)
848
content, _ = knit_data._parse_record(version_id, raw_data)
1073
849
if len(references[-1]) == 0:
1074
850
line_iterator = factory.get_fulltext_content(content)
1076
852
line_iterator = factory.get_linedelta_content(content)
1077
853
for line in line_iterator:
854
yield line, version_id
1080
856
# check the header only
1081
df, _ = knit._parse_record_header(key, raw_data)
857
df, _ = knit_data._parse_record_header(version_id, raw_data)
1083
859
pos, size = writer.add_bytes_record(raw_data, names)
1084
860
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1353
1031
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
1032
"""Management of packs within a repository."""
1362
1034
def __init__(self, repo, transport, index_transport, upload_transport,
1363
pack_transport, index_builder_class, index_class,
1365
1036
"""Create a new RepositoryPackCollection.
1367
:param transport: Addresses the repository base directory
1038
:param transport: Addresses the repository base directory
1368
1039
(typically .bzr/repository/).
1369
1040
:param index_transport: Addresses the directory containing indices.
1370
1041
:param upload_transport: Addresses the directory into which packs are written
1371
1042
while they're being created.
1372
1043
: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
1045
self.repo = repo
1379
1046
self.transport = transport
1380
1047
self._index_transport = index_transport
1381
1048
self._upload_transport = upload_transport
1382
1049
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,
1050
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
1387
1051
self.packs = []
1388
1052
# name:Pack mapping
1390
1053
self._packs_by_name = {}
1391
1054
# the previous pack-names content
1392
1055
self._packs_at_load = None
1393
1056
# when a pack is being created by this object, the state of that pack.
1394
1057
self._new_pack = None
1395
1058
# 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)
1059
self.revision_index = AggregateIndex()
1060
self.inventory_index = AggregateIndex()
1061
self.text_index = AggregateIndex()
1062
self.signature_index = AggregateIndex()
1421
1064
def add_pack_to_memory(self, pack):
1422
1065
"""Make a Pack object available to the repository to satisfy queries.
1424
1067
: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,))
1069
assert pack.name not in self._packs_by_name
1429
1070
self.packs.append(pack)
1430
1071
self._packs_by_name[pack.name] = pack
1431
1072
self.revision_index.add_index(pack.revision_index, pack)
1432
1073
self.inventory_index.add_index(pack.inventory_index, pack)
1433
1074
self.text_index.add_index(pack.text_index, pack)
1434
1075
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)
1077
def _add_text_to_weave(self, file_id, revision_id, new_lines, parents,
1078
nostore_sha, random_revid):
1079
file_id_index = GraphIndexPrefixAdapter(
1080
self.text_index.combined_index,
1082
add_nodes_callback=self.text_index.add_callback)
1083
self.repo._text_knit._index._graph_index = file_id_index
1084
self.repo._text_knit._index._add_callback = file_id_index.add_nodes
1085
return self.repo._text_knit.add_lines_with_ghosts(
1086
revision_id, parents, new_lines, nostore_sha=nostore_sha,
1087
random_id=random_revid, check_content=False)[0:2]
1438
1089
def all_packs(self):
1439
1090
"""Return a list of all the Pack objects this repository has.
1490
1137
# group their data with the relevant commit, and that may
1491
1138
# involve rewriting ancient history - which autopack tries to
1492
1139
# avoid. Alternatively we could not group the data but treat
1493
# each of these as having a single revision, and thus add
1140
# each of these as having a single revision, and thus add
1494
1141
# one revision for each to the total revision count, to get
1495
1142
# a matching distribution.
1497
1144
existing_packs.append((revision_count, pack))
1498
1145
pack_operations = self.plan_autopack_combinations(
1499
1146
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)
1147
self._execute_pack_operations(pack_operations)
1512
def _execute_pack_operations(self, pack_operations, _packer_class=Packer,
1150
def _execute_pack_operations(self, pack_operations):
1514
1151
"""Execute a series of pack operations.
1516
1153
: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
1156
for revision_count, packs in pack_operations:
1521
1157
# we may have no-ops from the setup logic
1522
1158
if len(packs) == 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()
1160
Packer(self, packs, '.autopack').pack()
1536
1161
for pack in packs:
1537
1162
self._remove_pack_from_memory(pack)
1538
1163
# 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()
1165
self._save_pack_names(clear_obsolete_packs=True)
1166
# Move the old packs out of the way now they are no longer referenced.
1167
for revision_count, packs in pack_operations:
1168
self._obsolete_packs(packs)
1551
1170
def lock_names(self):
1552
1171
"""Acquire the mutex around the pack-names index.
1554
1173
This cannot be used in the middle of a read-only transaction on the
1557
1176
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):
1564
1179
"""Pack the pack collection totally."""
1565
1180
self.ensure_loaded()
1566
1181
total_packs = len(self._names)
1567
if self._already_packed():
1569
1184
total_revisions = self.revision_index.combined_index.key_count()
1570
1185
# XXX: the following may want to be a class, to pack with a given
1572
1187
mutter('Packing repository %s, which has %d pack files, '
1573
'containing %d revisions with hint %r.', self, total_packs,
1574
total_revisions, hint)
1188
'containing %d revisions into 1 packs.', self, total_packs,
1575
1190
# determine which packs need changing
1191
pack_distribution = [1]
1576
1192
pack_operations = [[0, []]]
1577
1193
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()
1194
revision_count = pack.get_revision_count()
1195
pack_operations[-1][0] += revision_count
1196
pack_operations[-1][1].append(pack)
1197
self._execute_pack_operations(pack_operations)
1588
1199
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1589
1200
"""Plan a pack operation.
1873
1403
self._packs_by_name = {}
1874
1404
self._packs_at_load = None
1406
def _make_index_map(self, index_suffix):
1407
"""Return information on existing indices.
1409
:param suffix: Index suffix added to pack name.
1411
:returns: (pack_map, indices) where indices is a list of GraphIndex
1412
objects, and pack_map is a mapping from those objects to the
1413
pack tuple they describe.
1415
# TODO: stop using this; it creates new indices unnecessarily.
1416
self.ensure_loaded()
1417
suffix_map = {'.rix': 'revision_index',
1418
'.six': 'signature_index',
1419
'.iix': 'inventory_index',
1420
'.tix': 'text_index',
1422
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1423
suffix_map[index_suffix])
1425
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1426
"""Convert a list of packs to an index pack map and index list.
1428
:param packs: The packs list to process.
1429
:param index_attribute: The attribute that the desired index is found
1431
:return: A tuple (map, list) where map contains the dict from
1432
index:pack_tuple, and lsit contains the indices in the same order
1438
index = getattr(pack, index_attribute)
1439
indices.append(index)
1440
pack_map[index] = (pack.pack_transport, pack.file_name())
1441
return pack_map, indices
1443
def _index_contents(self, pack_map, key_filter=None):
1444
"""Get an iterable of the index contents from a pack_map.
1446
:param pack_map: A map from indices to pack details.
1447
:param key_filter: An optional filter to limit the
1450
indices = [index for index in pack_map.iterkeys()]
1451
all_index = CombinedGraphIndex(indices)
1452
if key_filter is None:
1453
return all_index.iter_all_entries()
1455
return all_index.iter_entries(key_filter)
1876
1457
def _unlock_names(self):
1877
1458
"""Release the mutex around the pack-names index."""
1878
1459
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
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)
1894
# do a two-way diff against our original content
1895
current_nodes = set()
1896
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
1461
def _save_pack_names(self, clear_obsolete_packs=False):
1462
"""Save the list of packs.
1464
This will take out the mutex around the pack names list for the
1465
duration of the method call. If concurrent updates have been made, a
1466
three-way merge between the current list and the current in memory list
1469
:param clear_obsolete_packs: If True, clear out the contents of the
1470
obsolete_packs directory.
1474
builder = GraphIndexBuilder()
1475
# load the disk nodes across
1477
for index, key, value in self._iter_disk_pack_index():
1478
disk_nodes.add((key, value))
1479
# do a two-way diff against our original content
1480
current_nodes = set()
1481
for name, sizes in self._names.iteritems():
1483
((name, ), ' '.join(str(size) for size in sizes)))
1484
deleted_nodes = self._packs_at_load - current_nodes
1485
new_nodes = current_nodes - self._packs_at_load
1486
disk_nodes.difference_update(deleted_nodes)
1487
disk_nodes.update(new_nodes)
1488
# TODO: handle same-name, index-size-changes here -
1489
# e.g. use the value from disk, not ours, *unless* we're the one
1491
for key, value in disk_nodes:
1492
builder.add_node(key, value)
1493
self.transport.put_file('pack-names', builder.finish(),
1494
mode=self.repo.control_files._file_mode)
1495
# move the baseline forward
1496
self._packs_at_load = disk_nodes
1497
# now clear out the obsolete packs directory
1498
if clear_obsolete_packs:
1499
self.transport.clone('obsolete_packs').delete_multi(
1500
self.transport.list_dir('obsolete_packs'))
1502
self._unlock_names()
1503
# synchronise the memory packs list with what we just wrote:
1925
1504
new_names = dict(disk_nodes)
1926
1505
# drop no longer present nodes
1927
1506
for pack in self.all_packs():
1928
1507
if (pack.name,) not in new_names:
1929
removed.append(pack.name)
1930
1508
self._remove_pack_from_memory(pack)
1931
1509
# add new nodes/refresh existing ones
1932
1510
for key, value in disk_nodes:
1946
1524
self._remove_pack_from_memory(self.get_pack_by_name(name))
1947
1525
self._names[name] = sizes
1948
1526
self.get_pack_by_name(name)
1949
modified.append(name)
1952
1529
self._names[name] = sizes
1953
1530
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
1532
def _start_write_group(self):
2067
1533
# Do not permit preparation for writing if we're not in a 'write lock'.
2068
1534
if not self.repo.is_write_locked():
2069
1535
raise errors.NotWriteLocked(self)
2070
self._new_pack = self.pack_factory(self, upload_suffix='.pack',
2071
file_mode=self.repo.bzrdir._get_file_mode())
1536
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1537
self._pack_transport, upload_suffix='.pack',
1538
file_mode=self.repo.control_files._file_mode)
2072
1539
# allow writing: queue writes to a new index
2073
1540
self.revision_index.add_writable_index(self._new_pack.revision_index,
2074
1541
self._new_pack)
2076
1543
self._new_pack)
2077
1544
self.text_index.add_writable_index(self._new_pack.text_index,
2078
1545
self._new_pack)
2079
self._new_pack.text_index.set_optimize(combine_backing_indices=False)
2080
1546
self.signature_index.add_writable_index(self._new_pack.signature_index,
2081
1547
self._new_pack)
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
1549
# reused revision and signature knits may need updating
1551
# "Hysterical raisins. client code in bzrlib grabs those knits outside
1552
# of write groups and then mutates it inside the write group."
1553
if self.repo._revision_knit is not None:
1554
self.repo._revision_knit._index._add_callback = \
1555
self.revision_index.add_callback
1556
if self.repo._signature_knit is not None:
1557
self.repo._signature_knit._index._add_callback = \
1558
self.signature_index.add_callback
1559
# create a reused knit object for text addition in commit.
1560
self.repo._text_knit = self.repo.weave_store.get_weave_or_empty(
2093
1563
def _abort_write_group(self):
2094
1564
# FIXME: just drop the transient index.
2095
1565
# 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
1566
self._new_pack.abort()
1567
self._remove_pack_indices(self._new_pack)
1568
self._new_pack = None
1569
self.repo._text_knit = None
2129
1571
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
1572
self._remove_pack_indices(self._new_pack)
2149
any_new_content = False
2150
1573
if self._new_pack.data_inserted():
2151
1574
# get all the data to disk and read to use
2152
1575
self._new_pack.finish()
2153
1576
self.allocate(self._new_pack)
2154
1577
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()
1578
if not self.autopack():
2171
1579
# 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
1581
self._save_pack_names()
2186
1583
self._new_pack.abort()
2187
1584
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)
1585
self.repo._text_knit = None
1588
class KnitPackRevisionStore(KnitRevisionStore):
1589
"""An object to adapt access from RevisionStore's to use KnitPacks.
1591
This class works by replacing the original RevisionStore.
1592
We need to do this because the KnitPackRevisionStore is less
1593
isolated in its layering - it uses services from the repo.
1596
def __init__(self, repo, transport, revisionstore):
1597
"""Create a KnitPackRevisionStore on repo with revisionstore.
1599
This will store its state in the Repository, use the
1600
indices to provide a KnitGraphIndex,
1601
and at the end of transactions write new indices.
1603
KnitRevisionStore.__init__(self, revisionstore.versioned_file_store)
1605
self._serializer = revisionstore._serializer
1606
self.transport = transport
1608
def get_revision_file(self, transaction):
1609
"""Get the revision versioned file object."""
1610
if getattr(self.repo, '_revision_knit', None) is not None:
1611
return self.repo._revision_knit
1612
self.repo._pack_collection.ensure_loaded()
1613
add_callback = self.repo._pack_collection.revision_index.add_callback
1614
# setup knit specific objects
1615
knit_index = KnitGraphIndex(
1616
self.repo._pack_collection.revision_index.combined_index,
1617
add_callback=add_callback)
1618
self.repo._revision_knit = knit.KnitVersionedFile(
1619
'revisions', self.transport.clone('..'),
1620
self.repo.control_files._file_mode,
1621
create=False, access_mode=self.repo._access_mode(),
1622
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1623
access_method=self.repo._pack_collection.revision_index.knit_access)
1624
return self.repo._revision_knit
1626
def get_signature_file(self, transaction):
1627
"""Get the signature versioned file object."""
1628
if getattr(self.repo, '_signature_knit', None) is not None:
1629
return self.repo._signature_knit
1630
self.repo._pack_collection.ensure_loaded()
1631
add_callback = self.repo._pack_collection.signature_index.add_callback
1632
# setup knit specific objects
1633
knit_index = KnitGraphIndex(
1634
self.repo._pack_collection.signature_index.combined_index,
1635
add_callback=add_callback, parents=False)
1636
self.repo._signature_knit = knit.KnitVersionedFile(
1637
'signatures', self.transport.clone('..'),
1638
self.repo.control_files._file_mode,
1639
create=False, access_mode=self.repo._access_mode(),
1640
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1641
access_method=self.repo._pack_collection.signature_index.knit_access)
1642
return self.repo._signature_knit
1645
class KnitPackTextStore(VersionedFileStore):
1646
"""Presents a TextStore abstraction on top of packs.
1648
This class works by replacing the original VersionedFileStore.
1649
We need to do this because the KnitPackRevisionStore is less
1650
isolated in its layering - it uses services from the repo and shares them
1651
with all the data written in a single write group.
1654
def __init__(self, repo, transport, weavestore):
1655
"""Create a KnitPackTextStore on repo with weavestore.
1657
This will store its state in the Repository, use the
1658
indices FileNames to provide a KnitGraphIndex,
1659
and at the end of transactions write new indices.
1661
# don't call base class constructor - it's not suitable.
1662
# no transient data stored in the transaction
1664
self._precious = False
1666
self.transport = transport
1667
self.weavestore = weavestore
1668
# XXX for check() which isn't updated yet
1669
self._transport = weavestore._transport
1671
def get_weave_or_empty(self, file_id, transaction):
1672
"""Get a 'Knit' backed by the .tix indices.
1674
The transaction parameter is ignored.
1676
self.repo._pack_collection.ensure_loaded()
1677
add_callback = self.repo._pack_collection.text_index.add_callback
1678
# setup knit specific objects
1679
file_id_index = GraphIndexPrefixAdapter(
1680
self.repo._pack_collection.text_index.combined_index,
1681
(file_id, ), 1, add_nodes_callback=add_callback)
1682
knit_index = KnitGraphIndex(file_id_index,
1683
add_callback=file_id_index.add_nodes,
1684
deltas=True, parents=True)
1685
return knit.KnitVersionedFile('text:' + file_id,
1686
self.transport.clone('..'),
1689
access_method=self.repo._pack_collection.text_index.knit_access,
1690
factory=knit.KnitPlainFactory())
1692
get_weave = get_weave_or_empty
1695
"""Generate a list of the fileids inserted, for use by check."""
1696
self.repo._pack_collection.ensure_loaded()
1698
for index, key, value, refs in \
1699
self.repo._pack_collection.text_index.combined_index.iter_all_entries():
1704
class InventoryKnitThunk(object):
1705
"""An object to manage thunking get_inventory_weave to pack based knits."""
1707
def __init__(self, repo, transport):
1708
"""Create an InventoryKnitThunk for repo at transport.
1710
This will store its state in the Repository, use the
1711
indices FileNames to provide a KnitGraphIndex,
1712
and at the end of transactions write a new index..
1715
self.transport = transport
1717
def get_weave(self):
1718
"""Get a 'Knit' that contains inventory data."""
1719
self.repo._pack_collection.ensure_loaded()
1720
add_callback = self.repo._pack_collection.inventory_index.add_callback
1721
# setup knit specific objects
1722
knit_index = KnitGraphIndex(
1723
self.repo._pack_collection.inventory_index.combined_index,
1724
add_callback=add_callback, deltas=True, parents=True)
1725
return knit.KnitVersionedFile(
1726
'inventory', self.transport.clone('..'),
1727
self.repo.control_files._file_mode,
1728
create=False, access_mode=self.repo._access_mode(),
1729
index=knit_index, delta=True, factory=knit.KnitPlainFactory(),
1730
access_method=self.repo._pack_collection.inventory_index.knit_access)
2196
1733
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,
1734
"""Experimental graph-knit using repository."""
1736
def __init__(self, _format, a_bzrdir, control_files, _revision_store,
1737
control_store, text_store, _commit_builder_class, _serializer):
2223
1738
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,
1739
_revision_store, control_store, text_store, _commit_builder_class,
1741
index_transport = control_files._transport.clone('indices')
1742
self._pack_collection = RepositoryPackCollection(self, control_files._transport,
2227
1743
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
1744
control_files._transport.clone('upload'),
1745
control_files._transport.clone('packs'))
1746
self._revision_store = KnitPackRevisionStore(self, index_transport, self._revision_store)
1747
self.weave_store = KnitPackTextStore(self, index_transport, self.weave_store)
1748
self._inv_thunk = InventoryKnitThunk(self, index_transport)
2271
1749
# 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.)
1750
# physical lock only taken out around changes to the pack-names list.)
2273
1751
# Another way to represent this would be a decorator around the control
2274
1752
# files object that presents logical locks as physical ones - if this
2275
1753
# gets ugly consider that alternative design. RBC 20071011
2280
1758
self._reconcile_fixes_text_parents = True
2281
1759
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)
2288
1761
def _abort_write_group(self):
2289
self.revisions._index._key_dependencies.clear()
2290
1762
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)
1764
def _access_mode(self):
1765
"""Return 'w' or 'r' for depending on whether a write lock is active.
1767
This method is a helper for the Knit-thunking support objects.
1769
if self.is_write_locked():
1773
def _find_inconsistent_revision_parents(self):
1774
"""Find revisions with incorrectly cached parents.
1776
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1777
parents-in-revision).
1779
if not self.is_locked():
1780
raise errors.ObjectNotLocked(self)
1781
pb = ui.ui_factory.nested_progress_bar()
1784
revision_nodes = self._pack_collection.revision_index \
1785
.combined_index.iter_all_entries()
1786
index_positions = []
1787
# Get the cached index values for all revisions, and also the location
1788
# in each index of the revision text so we can perform linear IO.
1789
for index, key, value, refs in revision_nodes:
1790
pos, length = value[1:].split(' ')
1791
index_positions.append((index, int(pos), key[0],
1792
tuple(parent[0] for parent in refs[0])))
1793
pb.update("Reading revision index.", 0, 0)
1794
index_positions.sort()
1795
batch_count = len(index_positions) / 1000 + 1
1796
pb.update("Checking cached revision graph.", 0, batch_count)
1797
for offset in xrange(batch_count):
1798
pb.update("Checking cached revision graph.", offset)
1799
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1802
rev_ids = [item[2] for item in to_query]
1803
revs = self.get_revisions(rev_ids)
1804
for revision, item in zip(revs, to_query):
1805
index_parents = item[3]
1806
rev_parents = tuple(revision.parent_ids)
1807
if index_parents != rev_parents:
1808
result.append((revision.revision_id, index_parents, rev_parents))
1813
def get_parents(self, revision_ids):
1814
"""See StackedParentsProvider.get_parents.
1816
This implementation accesses the combined revision index to provide
1819
self._pack_collection.ensure_loaded()
1820
index = self._pack_collection.revision_index.combined_index
1822
for revision_id in revision_ids:
1823
if revision_id != _mod_revision.NULL_REVISION:
1824
search_keys.add((revision_id,))
1825
found_parents = {_mod_revision.NULL_REVISION:[]}
1826
for index, key, value, refs in index.iter_entries(search_keys):
1829
parents = (_mod_revision.NULL_REVISION,)
1831
parents = tuple(parent[0] for parent in parents)
1832
found_parents[key[0]] = parents
1834
for revision_id in revision_ids:
1836
result.append(found_parents[revision_id])
2297
1841
def _make_parents_provider(self):
2298
return graph.CachingParentsProvider(self)
2300
1844
def _refresh_data(self):
2301
if not self.is_locked():
2303
self._pack_collection.reload_pack_names()
1845
if self._write_lock_count == 1 or (
1846
self.control_files._lock_count == 1 and
1847
self.control_files._lock_mode == 'r'):
1848
# forget what names there are
1849
self._pack_collection.reset()
1850
# XXX: Better to do an in-memory merge when acquiring a new lock -
1851
# factor out code from _save_pack_names.
1852
self._pack_collection.ensure_loaded()
2305
1854
def _start_write_group(self):
2306
1855
self._pack_collection._start_write_group()
2308
1857
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)
1858
return self._pack_collection._commit_write_group()
1860
def get_inventory_weave(self):
1861
return self._inv_thunk.get_weave()
2330
1863
def get_transaction(self):
2331
1864
if self._write_lock_count:
2417
1931
self.control_files.unlock()
2419
if not self.is_locked():
2420
for repo in self._fallback_repositories:
2424
class KnitPackStreamSource(StreamSource):
2425
"""A StreamSource used to transfer data between same-format KnitPack repos.
2427
This source assumes:
2428
1) Same serialization format for all objects
2429
2) Same root information
2430
3) XML format inventories
2431
4) Atomic inserts (so we can stream inventory texts before text
2436
def __init__(self, from_repository, to_format):
2437
super(KnitPackStreamSource, self).__init__(from_repository, to_format)
2438
self._text_keys = None
2439
self._text_fetch_order = 'unordered'
2441
def _get_filtered_inv_stream(self, revision_ids):
2442
from_repo = self.from_repository
2443
parent_ids = from_repo._find_parent_ids_of_revisions(revision_ids)
2444
parent_keys = [(p,) for p in parent_ids]
2445
find_text_keys = from_repo._find_text_key_references_from_xml_inventory_lines
2446
parent_text_keys = set(find_text_keys(
2447
from_repo._inventory_xml_lines_for_keys(parent_keys)))
2448
content_text_keys = set()
2449
knit = KnitVersionedFiles(None, None)
2450
factory = KnitPlainFactory()
2451
def find_text_keys_from_content(record):
2452
if record.storage_kind not in ('knit-delta-gz', 'knit-ft-gz'):
2453
raise ValueError("Unknown content storage kind for"
2454
" inventory text: %s" % (record.storage_kind,))
2455
# It's a knit record, it has a _raw_record field (even if it was
2456
# reconstituted from a network stream).
2457
raw_data = record._raw_record
2458
# read the entire thing
2459
revision_id = record.key[-1]
2460
content, _ = knit._parse_record(revision_id, raw_data)
2461
if record.storage_kind == 'knit-delta-gz':
2462
line_iterator = factory.get_linedelta_content(content)
2463
elif record.storage_kind == 'knit-ft-gz':
2464
line_iterator = factory.get_fulltext_content(content)
2465
content_text_keys.update(find_text_keys(
2466
[(line, revision_id) for line in line_iterator]))
2467
revision_keys = [(r,) for r in revision_ids]
2468
def _filtered_inv_stream():
2469
source_vf = from_repo.inventories
2470
stream = source_vf.get_record_stream(revision_keys,
2472
for record in stream:
2473
if record.storage_kind == 'absent':
2474
raise errors.NoSuchRevision(from_repo, record.key)
2475
find_text_keys_from_content(record)
2477
self._text_keys = content_text_keys - parent_text_keys
2478
return ('inventories', _filtered_inv_stream())
2480
def _get_text_stream(self):
2481
# Note: We know we don't have to handle adding root keys, because both
2482
# the source and target are the identical network name.
2483
text_stream = self.from_repository.texts.get_record_stream(
2484
self._text_keys, self._text_fetch_order, False)
2485
return ('texts', text_stream)
2487
def get_stream(self, search):
2488
revision_ids = search.get_keys()
2489
for stream_info in self._fetch_revision_texts(revision_ids):
2491
self._revision_keys = [(rev_id,) for rev_id in revision_ids]
2492
yield self._get_filtered_inv_stream(revision_ids)
2493
yield self._get_text_stream()
2497
1934
class RepositoryFormatPack(MetaDirRepositoryFormat):
2498
1935
"""Format logic for pack structured repositories.
2681
2150
def get_format_description(self):
2682
2151
"""See RepositoryFormat.get_format_description()."""
2683
2152
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")
added
removed
bzrlib/repofmt/pack_repo.py
