263
207
return not self.__eq__(other)
265
209
def __repr__(self):
266
return "<%s.%s object at 0x%x, %s, %s" % (
267
self.__class__.__module__, self.__class__.__name__, id(self),
268
self.pack_transport, self.name)
271
class ResumedPack(ExistingPack):
273
def __init__(self, name, revision_index, inventory_index, text_index,
274
signature_index, upload_transport, pack_transport, index_transport,
275
pack_collection, chk_index=None):
276
"""Create a ResumedPack object."""
277
ExistingPack.__init__(self, pack_transport, name, revision_index,
278
inventory_index, text_index, signature_index,
280
self.upload_transport = upload_transport
281
self.index_transport = index_transport
282
self.index_sizes = [None, None, None, None]
284
('revision', revision_index),
285
('inventory', inventory_index),
286
('text', text_index),
287
('signature', signature_index),
289
if chk_index is not None:
290
indices.append(('chk', chk_index))
291
self.index_sizes.append(None)
292
for index_type, index in indices:
293
offset = self.index_offset(index_type)
294
self.index_sizes[offset] = index._size
295
self.index_class = pack_collection._index_class
296
self._pack_collection = pack_collection
297
self._state = 'resumed'
298
# XXX: perhaps check that the .pack file exists?
300
def access_tuple(self):
301
if self._state == 'finished':
302
return Pack.access_tuple(self)
303
elif self._state == 'resumed':
304
return self.upload_transport, self.file_name()
306
raise AssertionError(self._state)
309
self.upload_transport.delete(self.file_name())
310
indices = [self.revision_index, self.inventory_index, self.text_index,
311
self.signature_index]
312
if self.chk_index is not None:
313
indices.append(self.chk_index)
314
for index in indices:
315
index._transport.delete(index._name)
318
self._check_references()
319
index_types = ['revision', 'inventory', 'text', 'signature']
320
if self.chk_index is not None:
321
index_types.append('chk')
322
for index_type in index_types:
323
old_name = self.index_name(index_type, self.name)
324
new_name = '../indices/' + old_name
325
self.upload_transport.rename(old_name, new_name)
326
self._replace_index_with_readonly(index_type)
327
new_name = '../packs/' + self.file_name()
328
self.upload_transport.rename(self.file_name(), new_name)
329
self._state = 'finished'
331
def _get_external_refs(self, index):
332
"""Return compression parents for this index that are not present.
334
This returns any compression parents that are referenced by this index,
335
which are not contained *in* this index. They may be present elsewhere.
337
return index.external_references(1)
210
return "<bzrlib.repofmt.pack_repo.Pack object at 0x%x, %s, %s" % (
211
id(self), self.transport, self.name)
340
214
class NewPack(Pack):
341
215
"""An in memory proxy for a pack which is being created."""
343
def __init__(self, pack_collection, upload_suffix='', file_mode=None):
217
# A map of index 'type' to the file extension and position in the
219
index_definitions = {
220
'revision': ('.rix', 0),
221
'inventory': ('.iix', 1),
223
'signature': ('.six', 3),
226
def __init__(self, upload_transport, index_transport, pack_transport,
227
upload_suffix='', file_mode=None):
344
228
"""Create a NewPack instance.
346
:param pack_collection: A PackCollection into which this is being inserted.
230
:param upload_transport: A writable transport for the pack to be
231
incrementally uploaded to.
232
:param index_transport: A writable transport for the pack's indices to
233
be written to when the pack is finished.
234
:param pack_transport: A writable transport for the pack to be renamed
235
to when the upload is complete. This *must* be the same as
236
upload_transport.clone('../packs').
347
237
:param upload_suffix: An optional suffix to be given to any temporary
348
238
files created during the pack creation. e.g '.autopack'
349
:param file_mode: Unix permissions for newly created file.
239
:param file_mode: An optional file mode to create the new files with.
351
241
# The relative locations of the packs are constrained, but all are
352
242
# passed in because the caller has them, so as to avoid object churn.
353
index_builder_class = pack_collection._index_builder_class
354
if pack_collection.chk_index is not None:
355
chk_index = index_builder_class(reference_lists=0)
358
243
Pack.__init__(self,
359
244
# Revisions: parents list, no text compression.
360
index_builder_class(reference_lists=1),
245
InMemoryGraphIndex(reference_lists=1),
361
246
# Inventory: We want to map compression only, but currently the
362
247
# knit code hasn't been updated enough to understand that, so we
363
248
# have a regular 2-list index giving parents and compression
365
index_builder_class(reference_lists=2),
250
InMemoryGraphIndex(reference_lists=2),
366
251
# Texts: compression and per file graph, for all fileids - so two
367
252
# reference lists and two elements in the key tuple.
368
index_builder_class(reference_lists=2, key_elements=2),
253
InMemoryGraphIndex(reference_lists=2, key_elements=2),
369
254
# Signatures: Just blobs to store, no compression, no parents
371
index_builder_class(reference_lists=0),
372
# CHK based storage - just blobs, no compression or parents.
256
InMemoryGraphIndex(reference_lists=0),
375
self._pack_collection = pack_collection
376
# When we make readonly indices, we need this.
377
self.index_class = pack_collection._index_class
378
258
# where should the new pack be opened
379
self.upload_transport = pack_collection._upload_transport
259
self.upload_transport = upload_transport
380
260
# where are indices written out to
381
self.index_transport = pack_collection._index_transport
261
self.index_transport = index_transport
382
262
# where is the pack renamed to when it is finished?
383
self.pack_transport = pack_collection._pack_transport
263
self.pack_transport = pack_transport
384
264
# What file mode to upload the pack and indices with.
385
265
self._file_mode = file_mode
386
266
# tracks the content written to the .pack file.
387
self._hash = osutils.md5()
388
# a tuple with the length in bytes of the indices, once the pack
389
# is finalised. (rev, inv, text, sigs, chk_if_in_use)
267
self._hash = md5.new()
268
# a four-tuple with the length in bytes of the indices, once the pack
269
# is finalised. (rev, inv, text, sigs)
390
270
self.index_sizes = None
391
271
# How much data to cache when writing packs. Note that this is not
392
272
# synchronised with reads, because it's not in the transport layer, so
595
477
Future searches on the aggregate index will seach this new index
596
478
before all previously inserted indices.
598
480
:param index: An Index for the pack.
599
481
:param pack: A Pack instance.
601
483
# expose it to the index map
602
484
self.index_to_pack[index] = pack.access_tuple()
603
485
# put it at the front of the linear index list
604
self.combined_index.insert_index(0, index, pack.name)
486
self.combined_index.insert_index(0, index)
606
488
def add_writable_index(self, index, pack):
607
489
"""Add an index which is able to have data added to it.
609
491
There can be at most one writable index at any time. Any
610
492
modifications made to the knit are put into this index.
612
494
:param index: An index from the pack parameter.
613
495
:param pack: A Pack instance.
615
if self.add_callback is not None:
616
raise AssertionError(
617
"%s already has a writable index through %s" % \
618
(self, self.add_callback))
497
assert self.add_callback is None, \
498
"%s already has a writable index through %s" % \
499
(self, self.add_callback)
619
500
# allow writing: queue writes to a new index
620
501
self.add_index(index, pack)
621
502
# Updates the index to packs mapping as a side effect,
622
self.data_access.set_writer(pack._writer, index, pack.access_tuple())
503
self.knit_access.set_writer(pack._writer, index, pack.access_tuple())
623
504
self.add_callback = index.add_nodes
626
507
"""Reset all the aggregate data to nothing."""
627
self.data_access.set_writer(None, None, (None, None))
508
self.knit_access.set_writer(None, None, (None, None))
628
509
self.index_to_pack.clear()
629
510
del self.combined_index._indices[:]
630
del self.combined_index._index_names[:]
631
511
self.add_callback = None
633
def remove_index(self, index):
513
def remove_index(self, index, pack):
634
514
"""Remove index from the indices used to answer queries.
636
516
:param index: An index from the pack parameter.
517
:param pack: A Pack instance.
638
519
del self.index_to_pack[index]
639
pos = self.combined_index._indices.index(index)
640
del self.combined_index._indices[pos]
641
del self.combined_index._index_names[pos]
520
self.combined_index._indices.remove(index)
642
521
if (self.add_callback is not None and
643
522
getattr(index, 'add_nodes', None) == self.add_callback):
644
523
self.add_callback = None
645
self.data_access.set_writer(None, None, (None, None))
524
self.knit_access.set_writer(None, None, (None, None))
648
527
class Packer(object):
649
528
"""Create a pack from packs."""
651
def __init__(self, pack_collection, packs, suffix, revision_ids=None,
530
def __init__(self, pack_collection, packs, suffix, revision_ids=None):
653
531
"""Create a Packer.
655
533
:param pack_collection: A RepositoryPackCollection object where the
1050
def _do_copy_nodes_graph(self, index_map, writer, write_index,
1051
output_lines, pb, readv_group_iter, total_items):
809
def _do_copy_nodes_graph(self, nodes, index_map, writer, write_index,
1052
811
# for record verification
1053
knit = KnitVersionedFiles(None, None)
812
knit_data = _KnitData(None)
1054
813
# for line extraction when requested (inventories only)
1055
814
if output_lines:
1056
factory = KnitPlainFactory()
815
factory = knit.KnitPlainFactory()
816
# plan a readv on each source pack:
818
nodes = sorted(nodes)
819
# how to map this into knit.py - or knit.py into this?
820
# we don't want the typical knit logic, we want grouping by pack
821
# at this point - perhaps a helper library for the following code
822
# duplication points?
1057
824
record_index = 0
1058
pb.update("Copied record", record_index, total_items)
1059
for index, readv_vector, node_vector in readv_group_iter:
825
pb.update("Copied record", record_index, len(nodes))
826
for index, key, value, references in nodes:
827
if index not in request_groups:
828
request_groups[index] = []
829
request_groups[index].append((key, value, references))
830
for index, items in request_groups.iteritems():
831
pack_readv_requests = []
832
for key, value, references in items:
833
# ---- KnitGraphIndex.get_position
834
bits = value[1:].split(' ')
835
offset, length = int(bits[0]), int(bits[1])
836
pack_readv_requests.append((offset, length, (key, value[0], references)))
837
# linear scan up the pack
838
pack_readv_requests.sort()
1061
pack_obj = index_map[index]
1062
transport, path = pack_obj.access_tuple()
1064
reader = pack.make_readv_reader(transport, path, readv_vector)
1065
except errors.NoSuchFile:
1066
if self._reload_func is not None:
1069
for (names, read_func), (key, eol_flag, references) in \
1070
izip(reader.iter_records(), node_vector):
840
transport, path = index_map[index]
841
reader = pack.make_readv_reader(transport, path,
842
[offset[0:2] for offset in pack_readv_requests])
843
for (names, read_func), (_1, _2, (key, eol_flag, references)) in \
844
izip(reader.iter_records(), pack_readv_requests):
1071
845
raw_data = read_func(None)
1072
847
if output_lines:
1073
848
# read the entire thing
1074
content, _ = knit._parse_record(key[-1], raw_data)
849
content, _ = knit_data._parse_record(version_id, raw_data)
1075
850
if len(references[-1]) == 0:
1076
851
line_iterator = factory.get_fulltext_content(content)
1078
853
line_iterator = factory.get_linedelta_content(content)
1079
854
for line in line_iterator:
855
yield line, version_id
1082
857
# check the header only
1083
df, _ = knit._parse_record_header(key, raw_data)
858
df, _ = knit_data._parse_record_header(version_id, raw_data)
1085
860
pos, size = writer.add_bytes_record(raw_data, names)
1086
861
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1355
1048
class RepositoryPackCollection(object):
1356
"""Management of packs within a repository.
1358
:ivar _names: map of {pack_name: (index_size,)}
1361
pack_factory = NewPack
1362
resumed_pack_factory = ResumedPack
1049
"""Management of packs within a repository."""
1364
1051
def __init__(self, repo, transport, index_transport, upload_transport,
1365
pack_transport, index_builder_class, index_class,
1367
1053
"""Create a new RepositoryPackCollection.
1369
:param transport: Addresses the repository base directory
1055
:param transport: Addresses the repository base directory
1370
1056
(typically .bzr/repository/).
1371
1057
:param index_transport: Addresses the directory containing indices.
1372
1058
:param upload_transport: Addresses the directory into which packs are written
1373
1059
while they're being created.
1374
1060
:param pack_transport: Addresses the directory of existing complete packs.
1375
:param index_builder_class: The index builder class to use.
1376
:param index_class: The index class to use.
1377
:param use_chk_index: Whether to setup and manage a CHK index.
1379
# XXX: This should call self.reset()
1380
1062
self.repo = repo
1381
1063
self.transport = transport
1382
1064
self._index_transport = index_transport
1383
1065
self._upload_transport = upload_transport
1384
1066
self._pack_transport = pack_transport
1385
self._index_builder_class = index_builder_class
1386
self._index_class = index_class
1387
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3,
1067
self._suffix_offsets = {'.rix': 0, '.iix': 1, '.tix': 2, '.six': 3}
1389
1068
self.packs = []
1390
1069
# name:Pack mapping
1392
1070
self._packs_by_name = {}
1393
1071
# the previous pack-names content
1394
1072
self._packs_at_load = None
1395
1073
# when a pack is being created by this object, the state of that pack.
1396
1074
self._new_pack = None
1397
1075
# aggregated revision index data
1398
flush = self._flush_new_pack
1399
self.revision_index = AggregateIndex(self.reload_pack_names, flush)
1400
self.inventory_index = AggregateIndex(self.reload_pack_names, flush)
1401
self.text_index = AggregateIndex(self.reload_pack_names, flush)
1402
self.signature_index = AggregateIndex(self.reload_pack_names, flush)
1403
all_indices = [self.revision_index, self.inventory_index,
1404
self.text_index, self.signature_index]
1406
self.chk_index = AggregateIndex(self.reload_pack_names, flush)
1407
all_indices.append(self.chk_index)
1409
# used to determine if we're using a chk_index elsewhere.
1410
self.chk_index = None
1411
# Tell all the CombinedGraphIndex objects about each other, so they can
1412
# share hints about which pack names to search first.
1413
all_combined = [agg_idx.combined_index for agg_idx in all_indices]
1414
for combined_idx in all_combined:
1415
combined_idx.set_sibling_indices(
1416
set(all_combined).difference([combined_idx]))
1418
self._resumed_packs = []
1421
return '%s(%r)' % (self.__class__.__name__, self.repo)
1076
self.revision_index = AggregateIndex()
1077
self.inventory_index = AggregateIndex()
1078
self.text_index = AggregateIndex()
1079
self.signature_index = AggregateIndex()
1423
1081
def add_pack_to_memory(self, pack):
1424
1082
"""Make a Pack object available to the repository to satisfy queries.
1426
1084
:param pack: A Pack object.
1428
if pack.name in self._packs_by_name:
1429
raise AssertionError(
1430
'pack %s already in _packs_by_name' % (pack.name,))
1086
assert pack.name not in self._packs_by_name
1431
1087
self.packs.append(pack)
1432
1088
self._packs_by_name[pack.name] = pack
1433
1089
self.revision_index.add_index(pack.revision_index, pack)
1434
1090
self.inventory_index.add_index(pack.inventory_index, pack)
1435
1091
self.text_index.add_index(pack.text_index, pack)
1436
1092
self.signature_index.add_index(pack.signature_index, pack)
1437
if self.chk_index is not None:
1438
self.chk_index.add_index(pack.chk_index, pack)
1094
def _add_text_to_weave(self, file_id, revision_id, new_lines, parents,
1095
nostore_sha, random_revid):
1096
file_id_index = GraphIndexPrefixAdapter(
1097
self.text_index.combined_index,
1099
add_nodes_callback=self.text_index.add_callback)
1100
self.repo._text_knit._index._graph_index = file_id_index
1101
self.repo._text_knit._index._add_callback = file_id_index.add_nodes
1102
return self.repo._text_knit.add_lines_with_ghosts(
1103
revision_id, parents, new_lines, nostore_sha=nostore_sha,
1104
random_id=random_revid, check_content=False)[0:2]
1440
1106
def all_packs(self):
1441
1107
"""Return a list of all the Pack objects this repository has.
1492
1154
# group their data with the relevant commit, and that may
1493
1155
# involve rewriting ancient history - which autopack tries to
1494
1156
# avoid. Alternatively we could not group the data but treat
1495
# each of these as having a single revision, and thus add
1157
# each of these as having a single revision, and thus add
1496
1158
# one revision for each to the total revision count, to get
1497
1159
# a matching distribution.
1499
1161
existing_packs.append((revision_count, pack))
1500
1162
pack_operations = self.plan_autopack_combinations(
1501
1163
existing_packs, pack_distribution)
1502
num_new_packs = len(pack_operations)
1503
num_old_packs = sum([len(po[1]) for po in pack_operations])
1504
num_revs_affected = sum([po[0] for po in pack_operations])
1505
mutter('Auto-packing repository %s, which has %d pack files, '
1506
'containing %d revisions. Packing %d files into %d affecting %d'
1507
' revisions', self, total_packs, total_revisions, num_old_packs,
1508
num_new_packs, num_revs_affected)
1509
result = self._execute_pack_operations(pack_operations,
1510
reload_func=self._restart_autopack)
1511
mutter('Auto-packing repository %s completed', self)
1164
self._execute_pack_operations(pack_operations)
1514
def _execute_pack_operations(self, pack_operations, _packer_class=Packer,
1167
def _execute_pack_operations(self, pack_operations):
1516
1168
"""Execute a series of pack operations.
1518
1170
:param pack_operations: A list of [revision_count, packs_to_combine].
1519
:param _packer_class: The class of packer to use (default: Packer).
1520
:return: The new pack names.
1522
1173
for revision_count, packs in pack_operations:
1523
1174
# we may have no-ops from the setup logic
1524
1175
if len(packs) == 0:
1526
packer = _packer_class(self, packs, '.autopack',
1527
reload_func=reload_func)
1530
except errors.RetryWithNewPacks:
1531
# An exception is propagating out of this context, make sure
1532
# this packer has cleaned up. Packer() doesn't set its new_pack
1533
# state into the RepositoryPackCollection object, so we only
1534
# have access to it directly here.
1535
if packer.new_pack is not None:
1536
packer.new_pack.abort()
1177
Packer(self, packs, '.autopack').pack()
1538
1178
for pack in packs:
1539
1179
self._remove_pack_from_memory(pack)
1540
1180
# record the newly available packs and stop advertising the old
1542
to_be_obsoleted = []
1543
for _, packs in pack_operations:
1544
to_be_obsoleted.extend(packs)
1545
result = self._save_pack_names(clear_obsolete_packs=True,
1546
obsolete_packs=to_be_obsoleted)
1549
def _flush_new_pack(self):
1550
if self._new_pack is not None:
1551
self._new_pack.flush()
1182
self._save_pack_names(clear_obsolete_packs=True)
1183
# Move the old packs out of the way now they are no longer referenced.
1184
for revision_count, packs in pack_operations:
1185
self._obsolete_packs(packs)
1553
1187
def lock_names(self):
1554
1188
"""Acquire the mutex around the pack-names index.
1556
1190
This cannot be used in the middle of a read-only transaction on the
1559
1193
self.repo.control_files.lock_write()
1561
def _already_packed(self):
1562
"""Is the collection already packed?"""
1563
return not (self.repo._format.pack_compresses or (len(self._names) > 1))
1565
def pack(self, hint=None, clean_obsolete_packs=False):
1566
1196
"""Pack the pack collection totally."""
1567
1197
self.ensure_loaded()
1568
1198
total_packs = len(self._names)
1569
if self._already_packed():
1571
1201
total_revisions = self.revision_index.combined_index.key_count()
1572
1202
# XXX: the following may want to be a class, to pack with a given
1574
1204
mutter('Packing repository %s, which has %d pack files, '
1575
'containing %d revisions with hint %r.', self, total_packs,
1576
total_revisions, hint)
1205
'containing %d revisions into 1 packs.', self, total_packs,
1577
1207
# determine which packs need changing
1208
pack_distribution = [1]
1578
1209
pack_operations = [[0, []]]
1579
1210
for pack in self.all_packs():
1580
if hint is None or pack.name in hint:
1581
# Either no hint was provided (so we are packing everything),
1582
# or this pack was included in the hint.
1583
pack_operations[-1][0] += pack.get_revision_count()
1584
pack_operations[-1][1].append(pack)
1585
self._execute_pack_operations(pack_operations, OptimisingPacker)
1587
if clean_obsolete_packs:
1588
self._clear_obsolete_packs()
1211
revision_count = pack.get_revision_count()
1212
pack_operations[-1][0] += revision_count
1213
pack_operations[-1][1].append(pack)
1214
self._execute_pack_operations(pack_operations)
1590
1216
def plan_autopack_combinations(self, existing_packs, pack_distribution):
1591
1217
"""Plan a pack operation.
1875
1420
self._packs_by_name = {}
1876
1421
self._packs_at_load = None
1423
def _make_index_map(self, index_suffix):
1424
"""Return information on existing indices.
1426
:param suffix: Index suffix added to pack name.
1428
:returns: (pack_map, indices) where indices is a list of GraphIndex
1429
objects, and pack_map is a mapping from those objects to the
1430
pack tuple they describe.
1432
# TODO: stop using this; it creates new indices unnecessarily.
1433
self.ensure_loaded()
1434
suffix_map = {'.rix': 'revision_index',
1435
'.six': 'signature_index',
1436
'.iix': 'inventory_index',
1437
'.tix': 'text_index',
1439
return self._packs_list_to_pack_map_and_index_list(self.all_packs(),
1440
suffix_map[index_suffix])
1442
def _packs_list_to_pack_map_and_index_list(self, packs, index_attribute):
1443
"""Convert a list of packs to an index pack map and index list.
1445
:param packs: The packs list to process.
1446
:param index_attribute: The attribute that the desired index is found
1448
:return: A tuple (map, list) where map contains the dict from
1449
index:pack_tuple, and lsit contains the indices in the same order
1455
index = getattr(pack, index_attribute)
1456
indices.append(index)
1457
pack_map[index] = (pack.pack_transport, pack.file_name())
1458
return pack_map, indices
1460
def _index_contents(self, pack_map, key_filter=None):
1461
"""Get an iterable of the index contents from a pack_map.
1463
:param pack_map: A map from indices to pack details.
1464
:param key_filter: An optional filter to limit the
1467
indices = [index for index in pack_map.iterkeys()]
1468
all_index = CombinedGraphIndex(indices)
1469
if key_filter is None:
1470
return all_index.iter_all_entries()
1472
return all_index.iter_entries(key_filter)
1878
1474
def _unlock_names(self):
1879
1475
"""Release the mutex around the pack-names index."""
1880
1476
self.repo.control_files.unlock()
1882
def _diff_pack_names(self):
1883
"""Read the pack names from disk, and compare it to the one in memory.
1885
:return: (disk_nodes, deleted_nodes, new_nodes)
1886
disk_nodes The final set of nodes that should be referenced
1887
deleted_nodes Nodes which have been removed from when we started
1888
new_nodes Nodes that are newly introduced
1890
# load the disk nodes across
1892
for index, key, value in self._iter_disk_pack_index():
1893
disk_nodes.add((key, value))
1894
orig_disk_nodes = set(disk_nodes)
1896
# do a two-way diff against our original content
1897
current_nodes = set()
1898
for name, sizes in self._names.iteritems():
1900
((name, ), ' '.join(str(size) for size in sizes)))
1902
# Packs no longer present in the repository, which were present when we
1903
# locked the repository
1904
deleted_nodes = self._packs_at_load - current_nodes
1905
# Packs which this process is adding
1906
new_nodes = current_nodes - self._packs_at_load
1908
# Update the disk_nodes set to include the ones we are adding, and
1909
# remove the ones which were removed by someone else
1910
disk_nodes.difference_update(deleted_nodes)
1911
disk_nodes.update(new_nodes)
1913
return disk_nodes, deleted_nodes, new_nodes, orig_disk_nodes
1915
def _syncronize_pack_names_from_disk_nodes(self, disk_nodes):
1916
"""Given the correct set of pack files, update our saved info.
1918
:return: (removed, added, modified)
1919
removed pack names removed from self._names
1920
added pack names added to self._names
1921
modified pack names that had changed value
1926
## self._packs_at_load = disk_nodes
1478
def _save_pack_names(self, clear_obsolete_packs=False):
1479
"""Save the list of packs.
1481
This will take out the mutex around the pack names list for the
1482
duration of the method call. If concurrent updates have been made, a
1483
three-way merge between the current list and the current in memory list
1486
:param clear_obsolete_packs: If True, clear out the contents of the
1487
obsolete_packs directory.
1491
builder = GraphIndexBuilder()
1492
# load the disk nodes across
1494
for index, key, value in self._iter_disk_pack_index():
1495
disk_nodes.add((key, value))
1496
# do a two-way diff against our original content
1497
current_nodes = set()
1498
for name, sizes in self._names.iteritems():
1500
((name, ), ' '.join(str(size) for size in sizes)))
1501
deleted_nodes = self._packs_at_load - current_nodes
1502
new_nodes = current_nodes - self._packs_at_load
1503
disk_nodes.difference_update(deleted_nodes)
1504
disk_nodes.update(new_nodes)
1505
# TODO: handle same-name, index-size-changes here -
1506
# e.g. use the value from disk, not ours, *unless* we're the one
1508
for key, value in disk_nodes:
1509
builder.add_node(key, value)
1510
self.transport.put_file('pack-names', builder.finish(),
1511
mode=self.repo.control_files._file_mode)
1512
# move the baseline forward
1513
self._packs_at_load = disk_nodes
1514
# now clear out the obsolete packs directory
1515
if clear_obsolete_packs:
1516
self.transport.clone('obsolete_packs').delete_multi(
1517
self.transport.list_dir('obsolete_packs'))
1519
self._unlock_names()
1520
# synchronise the memory packs list with what we just wrote:
1927
1521
new_names = dict(disk_nodes)
1928
1522
# drop no longer present nodes
1929
1523
for pack in self.all_packs():
1930
1524
if (pack.name,) not in new_names:
1931
removed.append(pack.name)
1932
1525
self._remove_pack_from_memory(pack)
1933
1526
# add new nodes/refresh existing ones
1934
1527
for key, value in disk_nodes:
1948
1541
self._remove_pack_from_memory(self.get_pack_by_name(name))
1949
1542
self._names[name] = sizes
1950
1543
self.get_pack_by_name(name)
1951
modified.append(name)
1954
1546
self._names[name] = sizes
1955
1547
self.get_pack_by_name(name)
1957
return removed, added, modified
1959
def _save_pack_names(self, clear_obsolete_packs=False, obsolete_packs=None):
1960
"""Save the list of packs.
1962
This will take out the mutex around the pack names list for the
1963
duration of the method call. If concurrent updates have been made, a
1964
three-way merge between the current list and the current in memory list
1967
:param clear_obsolete_packs: If True, clear out the contents of the
1968
obsolete_packs directory.
1969
:param obsolete_packs: Packs that are obsolete once the new pack-names
1970
file has been written.
1971
:return: A list of the names saved that were not previously on disk.
1973
already_obsolete = []
1976
builder = self._index_builder_class()
1977
(disk_nodes, deleted_nodes, new_nodes,
1978
orig_disk_nodes) = self._diff_pack_names()
1979
# TODO: handle same-name, index-size-changes here -
1980
# e.g. use the value from disk, not ours, *unless* we're the one
1982
for key, value in disk_nodes:
1983
builder.add_node(key, value)
1984
self.transport.put_file('pack-names', builder.finish(),
1985
mode=self.repo.bzrdir._get_file_mode())
1986
self._packs_at_load = disk_nodes
1987
if clear_obsolete_packs:
1990
to_preserve = set([o.name for o in obsolete_packs])
1991
already_obsolete = self._clear_obsolete_packs(to_preserve)
1993
self._unlock_names()
1994
# synchronise the memory packs list with what we just wrote:
1995
self._syncronize_pack_names_from_disk_nodes(disk_nodes)
1997
# TODO: We could add one more condition here. "if o.name not in
1998
# orig_disk_nodes and o != the new_pack we haven't written to
1999
# disk yet. However, the new pack object is not easily
2000
# accessible here (it would have to be passed through the
2001
# autopacking code, etc.)
2002
obsolete_packs = [o for o in obsolete_packs
2003
if o.name not in already_obsolete]
2004
self._obsolete_packs(obsolete_packs)
2005
return [new_node[0][0] for new_node in new_nodes]
2007
def reload_pack_names(self):
2008
"""Sync our pack listing with what is present in the repository.
2010
This should be called when we find out that something we thought was
2011
present is now missing. This happens when another process re-packs the
2014
:return: True if the in-memory list of packs has been altered at all.
2016
# The ensure_loaded call is to handle the case where the first call
2017
# made involving the collection was to reload_pack_names, where we
2018
# don't have a view of disk contents. Its a bit of a bandaid, and
2019
# causes two reads of pack-names, but its a rare corner case not struck
2020
# with regular push/pull etc.
2021
first_read = self.ensure_loaded()
2024
# out the new value.
2025
(disk_nodes, deleted_nodes, new_nodes,
2026
orig_disk_nodes) = self._diff_pack_names()
2027
# _packs_at_load is meant to be the explicit list of names in
2028
# 'pack-names' at then start. As such, it should not contain any
2029
# pending names that haven't been written out yet.
2030
self._packs_at_load = orig_disk_nodes
2032
modified) = self._syncronize_pack_names_from_disk_nodes(disk_nodes)
2033
if removed or added or modified:
2037
def _restart_autopack(self):
2038
"""Reload the pack names list, and restart the autopack code."""
2039
if not self.reload_pack_names():
2040
# Re-raise the original exception, because something went missing
2041
# and a restart didn't find it
2043
raise errors.RetryAutopack(self.repo, False, sys.exc_info())
2045
def _clear_obsolete_packs(self, preserve=None):
2046
"""Delete everything from the obsolete-packs directory.
2048
:return: A list of pack identifiers (the filename without '.pack') that
2049
were found in obsolete_packs.
2052
obsolete_pack_transport = self.transport.clone('obsolete_packs')
2053
if preserve is None:
2055
for filename in obsolete_pack_transport.list_dir('.'):
2056
name, ext = osutils.splitext(filename)
2059
if name in preserve:
2062
obsolete_pack_transport.delete(filename)
2063
except (errors.PathError, errors.TransportError), e:
2064
warning("couldn't delete obsolete pack, skipping it:\n%s"
2068
1549
def _start_write_group(self):
2069
1550
# Do not permit preparation for writing if we're not in a 'write lock'.
2070
1551
if not self.repo.is_write_locked():
2071
1552
raise errors.NotWriteLocked(self)
2072
self._new_pack = self.pack_factory(self, upload_suffix='.pack',
2073
file_mode=self.repo.bzrdir._get_file_mode())
1553
self._new_pack = NewPack(self._upload_transport, self._index_transport,
1554
self._pack_transport, upload_suffix='.pack',
1555
file_mode=self.repo.control_files._file_mode)
2074
1556
# allow writing: queue writes to a new index
2075
1557
self.revision_index.add_writable_index(self._new_pack.revision_index,
2076
1558
self._new_pack)
2078
1560
self._new_pack)
2079
1561
self.text_index.add_writable_index(self._new_pack.text_index,
2080
1562
self._new_pack)
2081
self._new_pack.text_index.set_optimize(combine_backing_indices=False)
2082
1563
self.signature_index.add_writable_index(self._new_pack.signature_index,
2083
1564
self._new_pack)
2084
if self.chk_index is not None:
2085
self.chk_index.add_writable_index(self._new_pack.chk_index,
2087
self.repo.chk_bytes._index._add_callback = self.chk_index.add_callback
2088
self._new_pack.chk_index.set_optimize(combine_backing_indices=False)
2090
self.repo.inventories._index._add_callback = self.inventory_index.add_callback
2091
self.repo.revisions._index._add_callback = self.revision_index.add_callback
2092
self.repo.signatures._index._add_callback = self.signature_index.add_callback
2093
self.repo.texts._index._add_callback = self.text_index.add_callback
1566
# reused revision and signature knits may need updating
1568
# "Hysterical raisins. client code in bzrlib grabs those knits outside
1569
# of write groups and then mutates it inside the write group."
1570
if self.repo._revision_knit is not None:
1571
self.repo._revision_knit._index._add_callback = \
1572
self.revision_index.add_callback
1573
if self.repo._signature_knit is not None:
1574
self.repo._signature_knit._index._add_callback = \
1575
self.signature_index.add_callback
1576
# create a reused knit object for text addition in commit.
1577
self.repo._text_knit = self.repo.weave_store.get_weave_or_empty(
2095
1580
def _abort_write_group(self):
2096
1581
# FIXME: just drop the transient index.
2097
1582
# forget what names there are
2098
if self._new_pack is not None:
2099
operation = cleanup.OperationWithCleanups(self._new_pack.abort)
2100
operation.add_cleanup(setattr, self, '_new_pack', None)
2101
# If we aborted while in the middle of finishing the write
2102
# group, _remove_pack_indices could fail because the indexes are
2103
# already gone. But they're not there we shouldn't fail in this
2104
# case, so we pass ignore_missing=True.
2105
operation.add_cleanup(self._remove_pack_indices, self._new_pack,
2106
ignore_missing=True)
2107
operation.run_simple()
2108
for resumed_pack in self._resumed_packs:
2109
operation = cleanup.OperationWithCleanups(resumed_pack.abort)
2110
# See comment in previous finally block.
2111
operation.add_cleanup(self._remove_pack_indices, resumed_pack,
2112
ignore_missing=True)
2113
operation.run_simple()
2114
del self._resumed_packs[:]
2116
def _remove_resumed_pack_indices(self):
2117
for resumed_pack in self._resumed_packs:
2118
self._remove_pack_indices(resumed_pack)
2119
del self._resumed_packs[:]
2121
def _check_new_inventories(self):
2122
"""Detect missing inventories in this write group.
2124
:returns: list of strs, summarising any problems found. If the list is
2125
empty no problems were found.
2127
# The base implementation does no checks. GCRepositoryPackCollection
1583
self._new_pack.abort()
1584
self._remove_pack_indices(self._new_pack)
1585
self._new_pack = None
1586
self.repo._text_knit = None
2131
1588
def _commit_write_group(self):
2133
for prefix, versioned_file in (
2134
('revisions', self.repo.revisions),
2135
('inventories', self.repo.inventories),
2136
('texts', self.repo.texts),
2137
('signatures', self.repo.signatures),
2139
missing = versioned_file.get_missing_compression_parent_keys()
2140
all_missing.update([(prefix,) + key for key in missing])
2142
raise errors.BzrCheckError(
2143
"Repository %s has missing compression parent(s) %r "
2144
% (self.repo, sorted(all_missing)))
2145
problems = self._check_new_inventories()
2147
problems_summary = '\n'.join(problems)
2148
raise errors.BzrCheckError(
2149
"Cannot add revision(s) to repository: " + problems_summary)
2150
1589
self._remove_pack_indices(self._new_pack)
2151
any_new_content = False
2152
1590
if self._new_pack.data_inserted():
2153
1591
# get all the data to disk and read to use
2154
1592
self._new_pack.finish()
2155
1593
self.allocate(self._new_pack)
2156
1594
self._new_pack = None
2157
any_new_content = True
2159
self._new_pack.abort()
2160
self._new_pack = None
2161
for resumed_pack in self._resumed_packs:
2162
# XXX: this is a pretty ugly way to turn the resumed pack into a
2163
# properly committed pack.
2164
self._names[resumed_pack.name] = None
2165
self._remove_pack_from_memory(resumed_pack)
2166
resumed_pack.finish()
2167
self.allocate(resumed_pack)
2168
any_new_content = True
2169
del self._resumed_packs[:]
2171
result = self.autopack()
1595
if not self.autopack():
2173
1596
# when autopack takes no steps, the names list is still
2175
return self._save_pack_names()
2179
def _suspend_write_group(self):
2180
tokens = [pack.name for pack in self._resumed_packs]
2181
self._remove_pack_indices(self._new_pack)
2182
if self._new_pack.data_inserted():
2183
# get all the data to disk and read to use
2184
self._new_pack.finish(suspend=True)
2185
tokens.append(self._new_pack.name)
2186
self._new_pack = None
1598
self._save_pack_names()
2188
1600
self._new_pack.abort()
2189
1601
self._new_pack = None
2190
self._remove_resumed_pack_indices()
2193
def _resume_write_group(self, tokens):
2194
for token in tokens:
2195
self._resume_pack(token)
1602
self.repo._text_knit = None
1605
class KnitPackRevisionStore(KnitRevisionStore):
1606
"""An object to adapt access from RevisionStore's to use KnitPacks.
1608
This class works by replacing the original RevisionStore.
1609
We need to do this because the KnitPackRevisionStore is less
1610
isolated in its layering - it uses services from the repo.
1613
def __init__(self, repo, transport, revisionstore):
1614
"""Create a KnitPackRevisionStore on repo with revisionstore.
1616
This will store its state in the Repository, use the
1617
indices to provide a KnitGraphIndex,
1618
and at the end of transactions write new indices.
1620
KnitRevisionStore.__init__(self, revisionstore.versioned_file_store)
1622
self._serializer = revisionstore._serializer
1623
self.transport = transport
1625
def get_revision_file(self, transaction):
1626
"""Get the revision versioned file object."""
1627
if getattr(self.repo, '_revision_knit', None) is not None:
1628
return self.repo._revision_knit
1629
self.repo._pack_collection.ensure_loaded()
1630
add_callback = self.repo._pack_collection.revision_index.add_callback
1631
# setup knit specific objects
1632
knit_index = KnitGraphIndex(
1633
self.repo._pack_collection.revision_index.combined_index,
1634
add_callback=add_callback)
1635
self.repo._revision_knit = knit.KnitVersionedFile(
1636
'revisions', self.transport.clone('..'),
1637
self.repo.control_files._file_mode,
1638
create=False, access_mode=self.repo._access_mode(),
1639
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1640
access_method=self.repo._pack_collection.revision_index.knit_access)
1641
return self.repo._revision_knit
1643
def get_signature_file(self, transaction):
1644
"""Get the signature versioned file object."""
1645
if getattr(self.repo, '_signature_knit', None) is not None:
1646
return self.repo._signature_knit
1647
self.repo._pack_collection.ensure_loaded()
1648
add_callback = self.repo._pack_collection.signature_index.add_callback
1649
# setup knit specific objects
1650
knit_index = KnitGraphIndex(
1651
self.repo._pack_collection.signature_index.combined_index,
1652
add_callback=add_callback, parents=False)
1653
self.repo._signature_knit = knit.KnitVersionedFile(
1654
'signatures', self.transport.clone('..'),
1655
self.repo.control_files._file_mode,
1656
create=False, access_mode=self.repo._access_mode(),
1657
index=knit_index, delta=False, factory=knit.KnitPlainFactory(),
1658
access_method=self.repo._pack_collection.signature_index.knit_access)
1659
return self.repo._signature_knit
1662
class KnitPackTextStore(VersionedFileStore):
1663
"""Presents a TextStore abstraction on top of packs.
1665
This class works by replacing the original VersionedFileStore.
1666
We need to do this because the KnitPackRevisionStore is less
1667
isolated in its layering - it uses services from the repo and shares them
1668
with all the data written in a single write group.
1671
def __init__(self, repo, transport, weavestore):
1672
"""Create a KnitPackTextStore on repo with weavestore.
1674
This will store its state in the Repository, use the
1675
indices FileNames to provide a KnitGraphIndex,
1676
and at the end of transactions write new indices.
1678
# don't call base class constructor - it's not suitable.
1679
# no transient data stored in the transaction
1681
self._precious = False
1683
self.transport = transport
1684
self.weavestore = weavestore
1685
# XXX for check() which isn't updated yet
1686
self._transport = weavestore._transport
1688
def get_weave_or_empty(self, file_id, transaction):
1689
"""Get a 'Knit' backed by the .tix indices.
1691
The transaction parameter is ignored.
1693
self.repo._pack_collection.ensure_loaded()
1694
add_callback = self.repo._pack_collection.text_index.add_callback
1695
# setup knit specific objects
1696
file_id_index = GraphIndexPrefixAdapter(
1697
self.repo._pack_collection.text_index.combined_index,
1698
(file_id, ), 1, add_nodes_callback=add_callback)
1699
knit_index = KnitGraphIndex(file_id_index,
1700
add_callback=file_id_index.add_nodes,
1701
deltas=True, parents=True)
1702
return knit.KnitVersionedFile('text:' + file_id,
1703
self.transport.clone('..'),
1706
access_method=self.repo._pack_collection.text_index.knit_access,
1707
factory=knit.KnitPlainFactory())
1709
get_weave = get_weave_or_empty
1712
"""Generate a list of the fileids inserted, for use by check."""
1713
self.repo._pack_collection.ensure_loaded()
1715
for index, key, value, refs in \
1716
self.repo._pack_collection.text_index.combined_index.iter_all_entries():
1721
class InventoryKnitThunk(object):
1722
"""An object to manage thunking get_inventory_weave to pack based knits."""
1724
def __init__(self, repo, transport):
1725
"""Create an InventoryKnitThunk for repo at transport.
1727
This will store its state in the Repository, use the
1728
indices FileNames to provide a KnitGraphIndex,
1729
and at the end of transactions write a new index..
1732
self.transport = transport
1734
def get_weave(self):
1735
"""Get a 'Knit' that contains inventory data."""
1736
self.repo._pack_collection.ensure_loaded()
1737
add_callback = self.repo._pack_collection.inventory_index.add_callback
1738
# setup knit specific objects
1739
knit_index = KnitGraphIndex(
1740
self.repo._pack_collection.inventory_index.combined_index,
1741
add_callback=add_callback, deltas=True, parents=True)
1742
return knit.KnitVersionedFile(
1743
'inventory', self.transport.clone('..'),
1744
self.repo.control_files._file_mode,
1745
create=False, access_mode=self.repo._access_mode(),
1746
index=knit_index, delta=True, factory=knit.KnitPlainFactory(),
1747
access_method=self.repo._pack_collection.inventory_index.knit_access)
2198
1750
class KnitPackRepository(KnitRepository):
2199
"""Repository with knit objects stored inside pack containers.
2201
The layering for a KnitPackRepository is:
2203
Graph | HPSS | Repository public layer |
2204
===================================================
2205
Tuple based apis below, string based, and key based apis above
2206
---------------------------------------------------
2208
Provides .texts, .revisions etc
2209
This adapts the N-tuple keys to physical knit records which only have a
2210
single string identifier (for historical reasons), which in older formats
2211
was always the revision_id, and in the mapped code for packs is always
2212
the last element of key tuples.
2213
---------------------------------------------------
2215
A separate GraphIndex is used for each of the
2216
texts/inventories/revisions/signatures contained within each individual
2217
pack file. The GraphIndex layer works in N-tuples and is unaware of any
2219
===================================================
2223
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
1751
"""Experimental graph-knit using repository."""
1753
def __init__(self, _format, a_bzrdir, control_files, _revision_store,
1754
control_store, text_store, _commit_builder_class, _serializer):
2225
1755
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2226
_commit_builder_class, _serializer)
2227
index_transport = self._transport.clone('indices')
2228
self._pack_collection = RepositoryPackCollection(self, self._transport,
1756
_revision_store, control_store, text_store, _commit_builder_class,
1758
index_transport = control_files._transport.clone('indices')
1759
self._pack_collection = RepositoryPackCollection(self, control_files._transport,
2229
1760
index_transport,
2230
self._transport.clone('upload'),
2231
self._transport.clone('packs'),
2232
_format.index_builder_class,
2233
_format.index_class,
2234
use_chk_index=self._format.supports_chks,
2236
self.inventories = KnitVersionedFiles(
2237
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2238
add_callback=self._pack_collection.inventory_index.add_callback,
2239
deltas=True, parents=True, is_locked=self.is_locked),
2240
data_access=self._pack_collection.inventory_index.data_access,
2241
max_delta_chain=200)
2242
self.revisions = KnitVersionedFiles(
2243
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2244
add_callback=self._pack_collection.revision_index.add_callback,
2245
deltas=False, parents=True, is_locked=self.is_locked,
2246
track_external_parent_refs=True),
2247
data_access=self._pack_collection.revision_index.data_access,
2249
self.signatures = KnitVersionedFiles(
2250
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2251
add_callback=self._pack_collection.signature_index.add_callback,
2252
deltas=False, parents=False, is_locked=self.is_locked),
2253
data_access=self._pack_collection.signature_index.data_access,
2255
self.texts = KnitVersionedFiles(
2256
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2257
add_callback=self._pack_collection.text_index.add_callback,
2258
deltas=True, parents=True, is_locked=self.is_locked),
2259
data_access=self._pack_collection.text_index.data_access,
2260
max_delta_chain=200)
2261
if _format.supports_chks:
2262
# No graph, no compression:- references from chks are between
2263
# different objects not temporal versions of the same; and without
2264
# some sort of temporal structure knit compression will just fail.
2265
self.chk_bytes = KnitVersionedFiles(
2266
_KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2267
add_callback=self._pack_collection.chk_index.add_callback,
2268
deltas=False, parents=False, is_locked=self.is_locked),
2269
data_access=self._pack_collection.chk_index.data_access,
2272
self.chk_bytes = None
1761
control_files._transport.clone('upload'),
1762
control_files._transport.clone('packs'))
1763
self._revision_store = KnitPackRevisionStore(self, index_transport, self._revision_store)
1764
self.weave_store = KnitPackTextStore(self, index_transport, self.weave_store)
1765
self._inv_thunk = InventoryKnitThunk(self, index_transport)
2273
1766
# True when the repository object is 'write locked' (as opposed to the
2274
# physical lock only taken out around changes to the pack-names list.)
1767
# physical lock only taken out around changes to the pack-names list.)
2275
1768
# Another way to represent this would be a decorator around the control
2276
1769
# files object that presents logical locks as physical ones - if this
2277
1770
# gets ugly consider that alternative design. RBC 20071011
2282
1775
self._reconcile_fixes_text_parents = True
2283
1776
self._reconcile_backsup_inventory = False
2285
def _warn_if_deprecated(self, branch=None):
2286
# This class isn't deprecated, but one sub-format is
2287
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2288
super(KnitPackRepository, self)._warn_if_deprecated(branch)
2290
1778
def _abort_write_group(self):
2291
self.revisions._index._key_dependencies.clear()
2292
1779
self._pack_collection._abort_write_group()
2294
def _get_source(self, to_format):
2295
if to_format.network_name() == self._format.network_name():
2296
return KnitPackStreamSource(self, to_format)
2297
return super(KnitPackRepository, self)._get_source(to_format)
1781
def _access_mode(self):
1782
"""Return 'w' or 'r' for depending on whether a write lock is active.
1784
This method is a helper for the Knit-thunking support objects.
1786
if self.is_write_locked():
1790
def _find_inconsistent_revision_parents(self):
1791
"""Find revisions with incorrectly cached parents.
1793
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1794
parents-in-revision).
1796
if not self.is_locked():
1797
raise errors.ObjectNotLocked(self)
1798
pb = ui.ui_factory.nested_progress_bar()
1801
revision_nodes = self._pack_collection.revision_index \
1802
.combined_index.iter_all_entries()
1803
index_positions = []
1804
# Get the cached index values for all revisions, and also the location
1805
# in each index of the revision text so we can perform linear IO.
1806
for index, key, value, refs in revision_nodes:
1807
pos, length = value[1:].split(' ')
1808
index_positions.append((index, int(pos), key[0],
1809
tuple(parent[0] for parent in refs[0])))
1810
pb.update("Reading revision index.", 0, 0)
1811
index_positions.sort()
1812
batch_count = len(index_positions) / 1000 + 1
1813
pb.update("Checking cached revision graph.", 0, batch_count)
1814
for offset in xrange(batch_count):
1815
pb.update("Checking cached revision graph.", offset)
1816
to_query = index_positions[offset * 1000:(offset + 1) * 1000]
1819
rev_ids = [item[2] for item in to_query]
1820
revs = self.get_revisions(rev_ids)
1821
for revision, item in zip(revs, to_query):
1822
index_parents = item[3]
1823
rev_parents = tuple(revision.parent_ids)
1824
if index_parents != rev_parents:
1825
result.append((revision.revision_id, index_parents, rev_parents))
1830
def get_parents(self, revision_ids):
1831
"""See StackedParentsProvider.get_parents.
1833
This implementation accesses the combined revision index to provide
1836
self._pack_collection.ensure_loaded()
1837
index = self._pack_collection.revision_index.combined_index
1839
for revision_id in revision_ids:
1840
if revision_id != _mod_revision.NULL_REVISION:
1841
search_keys.add((revision_id,))
1842
found_parents = {_mod_revision.NULL_REVISION:[]}
1843
for index, key, value, refs in index.iter_entries(search_keys):
1846
parents = (_mod_revision.NULL_REVISION,)
1848
parents = tuple(parent[0] for parent in parents)
1849
found_parents[key[0]] = parents
1851
for revision_id in revision_ids:
1853
result.append(found_parents[revision_id])
2299
1858
def _make_parents_provider(self):
2300
return graph.CachingParentsProvider(self)
2302
1861
def _refresh_data(self):
2303
if not self.is_locked():
2305
self._pack_collection.reload_pack_names()
1862
if self._write_lock_count == 1 or (
1863
self.control_files._lock_count == 1 and
1864
self.control_files._lock_mode == 'r'):
1865
# forget what names there are
1866
self._pack_collection.reset()
1867
# XXX: Better to do an in-memory merge when acquiring a new lock -
1868
# factor out code from _save_pack_names.
1869
self._pack_collection.ensure_loaded()
2307
1871
def _start_write_group(self):
2308
1872
self._pack_collection._start_write_group()
2310
1874
def _commit_write_group(self):
2311
hint = self._pack_collection._commit_write_group()
2312
self.revisions._index._key_dependencies.clear()
2315
def suspend_write_group(self):
2316
# XXX check self._write_group is self.get_transaction()?
2317
tokens = self._pack_collection._suspend_write_group()
2318
self.revisions._index._key_dependencies.clear()
2319
self._write_group = None
2322
def _resume_write_group(self, tokens):
2323
self._start_write_group()
2325
self._pack_collection._resume_write_group(tokens)
2326
except errors.UnresumableWriteGroup:
2327
self._abort_write_group()
2329
for pack in self._pack_collection._resumed_packs:
2330
self.revisions._index.scan_unvalidated_index(pack.revision_index)
1875
return self._pack_collection._commit_write_group()
1877
def get_inventory_weave(self):
1878
return self._inv_thunk.get_weave()
2332
1880
def get_transaction(self):
2333
1881
if self._write_lock_count:
2429
1948
self.control_files.unlock()
2431
if not self.is_locked():
2432
for repo in self._fallback_repositories:
2436
class KnitPackStreamSource(StreamSource):
2437
"""A StreamSource used to transfer data between same-format KnitPack repos.
2439
This source assumes:
2440
1) Same serialization format for all objects
2441
2) Same root information
2442
3) XML format inventories
2443
4) Atomic inserts (so we can stream inventory texts before text
2448
def __init__(self, from_repository, to_format):
2449
super(KnitPackStreamSource, self).__init__(from_repository, to_format)
2450
self._text_keys = None
2451
self._text_fetch_order = 'unordered'
2453
def _get_filtered_inv_stream(self, revision_ids):
2454
from_repo = self.from_repository
2455
parent_ids = from_repo._find_parent_ids_of_revisions(revision_ids)
2456
parent_keys = [(p,) for p in parent_ids]
2457
find_text_keys = from_repo._find_text_key_references_from_xml_inventory_lines
2458
parent_text_keys = set(find_text_keys(
2459
from_repo._inventory_xml_lines_for_keys(parent_keys)))
2460
content_text_keys = set()
2461
knit = KnitVersionedFiles(None, None)
2462
factory = KnitPlainFactory()
2463
def find_text_keys_from_content(record):
2464
if record.storage_kind not in ('knit-delta-gz', 'knit-ft-gz'):
2465
raise ValueError("Unknown content storage kind for"
2466
" inventory text: %s" % (record.storage_kind,))
2467
# It's a knit record, it has a _raw_record field (even if it was
2468
# reconstituted from a network stream).
2469
raw_data = record._raw_record
2470
# read the entire thing
2471
revision_id = record.key[-1]
2472
content, _ = knit._parse_record(revision_id, raw_data)
2473
if record.storage_kind == 'knit-delta-gz':
2474
line_iterator = factory.get_linedelta_content(content)
2475
elif record.storage_kind == 'knit-ft-gz':
2476
line_iterator = factory.get_fulltext_content(content)
2477
content_text_keys.update(find_text_keys(
2478
[(line, revision_id) for line in line_iterator]))
2479
revision_keys = [(r,) for r in revision_ids]
2480
def _filtered_inv_stream():
2481
source_vf = from_repo.inventories
2482
stream = source_vf.get_record_stream(revision_keys,
2484
for record in stream:
2485
if record.storage_kind == 'absent':
2486
raise errors.NoSuchRevision(from_repo, record.key)
2487
find_text_keys_from_content(record)
2489
self._text_keys = content_text_keys - parent_text_keys
2490
return ('inventories', _filtered_inv_stream())
2492
def _get_text_stream(self):
2493
# Note: We know we don't have to handle adding root keys, because both
2494
# the source and target are the identical network name.
2495
text_stream = self.from_repository.texts.get_record_stream(
2496
self._text_keys, self._text_fetch_order, False)
2497
return ('texts', text_stream)
2499
def get_stream(self, search):
2500
revision_ids = search.get_keys()
2501
for stream_info in self._fetch_revision_texts(revision_ids):
2503
self._revision_keys = [(rev_id,) for rev_id in revision_ids]
2504
yield self._get_filtered_inv_stream(revision_ids)
2505
yield self._get_text_stream()
2509
1951
class RepositoryFormatPack(MetaDirRepositoryFormat):
2510
1952
"""Format logic for pack structured repositories.
2693
2167
def get_format_description(self):
2694
2168
"""See RepositoryFormat.get_format_description()."""
2695
2169
return "Packs containing knits with rich root support\n"
2698
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2699
"""Repository that supports external references to allow stacking.
2703
Supports external lookups, which results in non-truncated ghosts after
2704
reconcile compared to pack-0.92 formats.
2707
repository_class = KnitPackRepository
2708
_commit_builder_class = PackCommitBuilder
2709
supports_external_lookups = True
2710
# What index classes to use
2711
index_builder_class = InMemoryGraphIndex
2712
index_class = GraphIndex
2715
def _serializer(self):
2716
return xml5.serializer_v5
2718
def _get_matching_bzrdir(self):
2719
return bzrdir.format_registry.make_bzrdir('1.6')
2721
def _ignore_setting_bzrdir(self, format):
2724
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2726
def get_format_string(self):
2727
"""See RepositoryFormat.get_format_string()."""
2728
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2730
def get_format_description(self):
2731
"""See RepositoryFormat.get_format_description()."""
2732
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2735
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2736
"""A repository with rich roots and stacking.
2738
New in release 1.6.1.
2740
Supports stacking on other repositories, allowing data to be accessed
2741
without being stored locally.
2744
repository_class = KnitPackRepository
2745
_commit_builder_class = PackRootCommitBuilder
2746
rich_root_data = True
2747
supports_tree_reference = False # no subtrees
2748
supports_external_lookups = True
2749
# What index classes to use
2750
index_builder_class = InMemoryGraphIndex
2751
index_class = GraphIndex
2754
def _serializer(self):
2755
return xml6.serializer_v6
2757
def _get_matching_bzrdir(self):
2758
return bzrdir.format_registry.make_bzrdir(
2761
def _ignore_setting_bzrdir(self, format):
2764
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2766
def get_format_string(self):
2767
"""See RepositoryFormat.get_format_string()."""
2768
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2770
def get_format_description(self):
2771
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2774
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2775
"""A repository with rich roots and external references.
2779
Supports external lookups, which results in non-truncated ghosts after
2780
reconcile compared to pack-0.92 formats.
2782
This format was deprecated because the serializer it uses accidentally
2783
supported subtrees, when the format was not intended to. This meant that
2784
someone could accidentally fetch from an incorrect repository.
2787
repository_class = KnitPackRepository
2788
_commit_builder_class = PackRootCommitBuilder
2789
rich_root_data = True
2790
supports_tree_reference = False # no subtrees
2792
supports_external_lookups = True
2793
# What index classes to use
2794
index_builder_class = InMemoryGraphIndex
2795
index_class = GraphIndex
2798
def _serializer(self):
2799
return xml7.serializer_v7
2801
def _get_matching_bzrdir(self):
2802
matching = bzrdir.format_registry.make_bzrdir(
2804
matching.repository_format = self
2807
def _ignore_setting_bzrdir(self, format):
2810
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2812
def get_format_string(self):
2813
"""See RepositoryFormat.get_format_string()."""
2814
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2816
def get_format_description(self):
2817
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2821
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2822
"""A repository with stacking and btree indexes,
2823
without rich roots or subtrees.
2825
This is equivalent to pack-1.6 with B+Tree indices.
2828
repository_class = KnitPackRepository
2829
_commit_builder_class = PackCommitBuilder
2830
supports_external_lookups = True
2831
# What index classes to use
2832
index_builder_class = BTreeBuilder
2833
index_class = BTreeGraphIndex
2836
def _serializer(self):
2837
return xml5.serializer_v5
2839
def _get_matching_bzrdir(self):
2840
return bzrdir.format_registry.make_bzrdir('1.9')
2842
def _ignore_setting_bzrdir(self, format):
2845
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2847
def get_format_string(self):
2848
"""See RepositoryFormat.get_format_string()."""
2849
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2851
def get_format_description(self):
2852
"""See RepositoryFormat.get_format_description()."""
2853
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2856
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2857
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2859
1.6-rich-root with B+Tree indices.
2862
repository_class = KnitPackRepository
2863
_commit_builder_class = PackRootCommitBuilder
2864
rich_root_data = True
2865
supports_tree_reference = False # no subtrees
2866
supports_external_lookups = True
2867
# What index classes to use
2868
index_builder_class = BTreeBuilder
2869
index_class = BTreeGraphIndex
2872
def _serializer(self):
2873
return xml6.serializer_v6
2875
def _get_matching_bzrdir(self):
2876
return bzrdir.format_registry.make_bzrdir(
2879
def _ignore_setting_bzrdir(self, format):
2882
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2884
def get_format_string(self):
2885
"""See RepositoryFormat.get_format_string()."""
2886
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2888
def get_format_description(self):
2889
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2892
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2893
"""A subtrees development repository.
2895
This format should be retained until the second release after bzr 1.7.
2897
1.6.1-subtree[as it might have been] with B+Tree indices.
2899
This is [now] retained until we have a CHK based subtree format in
2903
repository_class = KnitPackRepository
2904
_commit_builder_class = PackRootCommitBuilder
2905
rich_root_data = True
2907
supports_tree_reference = True
2908
supports_external_lookups = True
2909
# What index classes to use
2910
index_builder_class = BTreeBuilder
2911
index_class = BTreeGraphIndex
2914
def _serializer(self):
2915
return xml7.serializer_v7
2917
def _get_matching_bzrdir(self):
2918
return bzrdir.format_registry.make_bzrdir(
2919
'development-subtree')
2921
def _ignore_setting_bzrdir(self, format):
2924
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2926
def get_format_string(self):
2927
"""See RepositoryFormat.get_format_string()."""
2928
return ("Bazaar development format 2 with subtree support "
2929
"(needs bzr.dev from before 1.8)\n")
2931
def get_format_description(self):
2932
"""See RepositoryFormat.get_format_description()."""
2933
return ("Development repository format, currently the same as "
2934
"1.6.1-subtree with B+Tree indices.\n")
added
removed
bzrlib/repofmt/pack_repo.py
