« back to all changes in this revision
Viewing changes to bzrlib/repofmt/pack_repo.py
- browse files at revision 5696
- compare with another revision
- download diff
- download tarball
- view history from revision 5696
- Committer: Canonical.com Patch Queue Manager
- Date: 2011-03-03 07:56:45 UTC
- mfrom: (5695.1.1 merge-2.3-to-trunk)
- Revision ID: pqm@pqm.ubuntu.com-20110303075645-hncdmrbpczkxcvt4
(spiv) Merge lp:bzr/2.3 into lp:bzr,
including fix for #726584. (Andrew Bennetts)
including fix for #726584. (Andrew Bennetts)
- files removed:
- bzrlib/plugins/changelog_merge
- bzrlib/plugins/changelog_merge/tests
- bzrlib/plugins/weave_fmt
- bzrlib/plugins/weave_fmt/tests
- bzrlib/tests/per_repository_vf
- files renamed:
- bzrlib/plugins/weave_fmt/repository.py => bzrlib/repofmt/weaverepo.py
- bzrlib/tests/per_repository_vf/test_check_reconcile.py => bzrlib/tests/per_repository/test_check_reconcile.py
- bzrlib/plugins/weave_fmt/xml4.py => bzrlib/xml4.py
- files modified:
- bzrlib/__init__.py
added
removed
bzrlib/repofmt/pack_repo.py
30
30
osutils,
31
31
pack,
32
32
transactions,
33
tsort,
34
33
ui,
34
xml5,
35
xml6,
36
xml7,
35
37
)
36
38
from bzrlib.index import (
37
39
CombinedGraphIndex,
38
40
GraphIndexPrefixAdapter,
39
41
)
42
from bzrlib.knit import (
43
KnitPlainFactory,
44
KnitVersionedFiles,
45
_KnitGraphIndex,
46
_DirectPackAccess,
47
)
48
from bzrlib import tsort
40
49
""")
41
50
from bzrlib import (
51
bzrdir,
42
52
btree_index,
43
53
errors,
44
54
lockable_files,
45
55
lockdir,
56
revision as _mod_revision,
46
57
)
47
58
48
from bzrlib.decorators import needs_read_lock, needs_write_lock, only_raises
59
from bzrlib.decorators import needs_write_lock, only_raises
60
from bzrlib.index import (
61
GraphIndex,
62
InMemoryGraphIndex,
63
)
49
64
from bzrlib.lock import LogicalLockResult
65
from bzrlib.repofmt.knitrepo import KnitRepository
50
66
from bzrlib.repository import (
51
67
CommitBuilder,
52
MetaDirRepository,
53
68
MetaDirRepositoryFormat,
54
69
RepositoryFormat,
55
70
RepositoryWriteLockResult,
56
71
RootCommitBuilder,
72
StreamSource,
57
73
)
58
74
from bzrlib.trace import (
59
75
mutter,
567
583
self._reload_func = reload_func
568
584
self.index_to_pack = {}
569
585
self.combined_index = CombinedGraphIndex([], reload_func=reload_func)
570
self.data_access = pack._DirectPackAccess(self.index_to_pack,
586
self.data_access = _DirectPackAccess(self.index_to_pack,
571
587
reload_func=reload_func,
572
588
flush_func=flush_func)
573
589
self.add_callback = None
657
673
# What text keys to copy. None for 'all texts'. This is set by
658
674
# _copy_inventory_texts
659
675
self._text_filter = None
676
self._extra_init()
677
678
def _extra_init(self):
679
"""A template hook to allow extending the constructor trivially."""
660
680
661
681
def _pack_map_and_index_list(self, index_attribute):
662
682
"""Convert a list of packs to an index pack map and index list.
776
796
777
797
def _copy_revision_texts(self):
778
798
"""Copy revision data to the new pack."""
779
raise NotImplementedError(self._copy_revision_texts)
799
# select revisions
800
if self.revision_ids:
801
revision_keys = [(revision_id,) for revision_id in self.revision_ids]
802
else:
803
revision_keys = None
804
# select revision keys
805
revision_index_map, revision_indices = self._pack_map_and_index_list(
806
'revision_index')
807
revision_nodes = self._index_contents(revision_indices, revision_keys)
808
revision_nodes = list(revision_nodes)
809
self._update_pack_order(revision_nodes, revision_index_map)
810
# copy revision keys and adjust values
811
self.pb.update("Copying revision texts", 1)
812
total_items, readv_group_iter = self._revision_node_readv(revision_nodes)
813
list(self._copy_nodes_graph(revision_index_map, self.new_pack._writer,
814
self.new_pack.revision_index, readv_group_iter, total_items))
815
if 'pack' in debug.debug_flags:
816
mutter('%s: create_pack: revisions copied: %s%s %d items t+%6.3fs',
817
time.ctime(), self._pack_collection._upload_transport.base,
818
self.new_pack.random_name,
819
self.new_pack.revision_index.key_count(),
820
time.time() - self.new_pack.start_time)
821
self._revision_keys = revision_keys
780
822
781
823
def _copy_inventory_texts(self):
782
824
"""Copy the inventory texts to the new pack.
785
827
786
828
Sets self._text_filter appropriately.
787
829
"""
788
raise NotImplementedError(self._copy_inventory_texts)
830
# select inventory keys
831
inv_keys = self._revision_keys # currently the same keyspace, and note that
832
# querying for keys here could introduce a bug where an inventory item
833
# is missed, so do not change it to query separately without cross
834
# checking like the text key check below.
835
inventory_index_map, inventory_indices = self._pack_map_and_index_list(
836
'inventory_index')
837
inv_nodes = self._index_contents(inventory_indices, inv_keys)
838
# copy inventory keys and adjust values
839
# XXX: Should be a helper function to allow different inv representation
840
# at this point.
841
self.pb.update("Copying inventory texts", 2)
842
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
843
# Only grab the output lines if we will be processing them
844
output_lines = bool(self.revision_ids)
845
inv_lines = self._copy_nodes_graph(inventory_index_map,
846
self.new_pack._writer, self.new_pack.inventory_index,
847
readv_group_iter, total_items, output_lines=output_lines)
848
if self.revision_ids:
849
self._process_inventory_lines(inv_lines)
850
else:
851
# eat the iterator to cause it to execute.
852
list(inv_lines)
853
self._text_filter = None
854
if 'pack' in debug.debug_flags:
855
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
856
time.ctime(), self._pack_collection._upload_transport.base,
857
self.new_pack.random_name,
858
self.new_pack.inventory_index.key_count(),
859
time.time() - self.new_pack.start_time)
789
860
790
861
def _copy_text_texts(self):
791
raise NotImplementedError(self._copy_text_texts)
862
# select text keys
863
text_index_map, text_nodes = self._get_text_nodes()
864
if self._text_filter is not None:
865
# We could return the keys copied as part of the return value from
866
# _copy_nodes_graph but this doesn't work all that well with the
867
# need to get line output too, so we check separately, and as we're
868
# going to buffer everything anyway, we check beforehand, which
869
# saves reading knit data over the wire when we know there are
870
# mising records.
871
text_nodes = set(text_nodes)
872
present_text_keys = set(_node[1] for _node in text_nodes)
873
missing_text_keys = set(self._text_filter) - present_text_keys
874
if missing_text_keys:
875
# TODO: raise a specific error that can handle many missing
876
# keys.
877
mutter("missing keys during fetch: %r", missing_text_keys)
878
a_missing_key = missing_text_keys.pop()
879
raise errors.RevisionNotPresent(a_missing_key[1],
880
a_missing_key[0])
881
# copy text keys and adjust values
882
self.pb.update("Copying content texts", 3)
883
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
884
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
885
self.new_pack.text_index, readv_group_iter, total_items))
886
self._log_copied_texts()
792
887
793
888
def _create_pack_from_packs(self):
794
raise NotImplementedError(self._create_pack_from_packs)
889
self.pb.update("Opening pack", 0, 5)
890
self.new_pack = self.open_pack()
891
new_pack = self.new_pack
892
# buffer data - we won't be reading-back during the pack creation and
893
# this makes a significant difference on sftp pushes.
894
new_pack.set_write_cache_size(1024*1024)
895
if 'pack' in debug.debug_flags:
896
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
897
for a_pack in self.packs]
898
if self.revision_ids is not None:
899
rev_count = len(self.revision_ids)
900
else:
901
rev_count = 'all'
902
mutter('%s: create_pack: creating pack from source packs: '
903
'%s%s %s revisions wanted %s t=0',
904
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
905
plain_pack_list, rev_count)
906
self._copy_revision_texts()
907
self._copy_inventory_texts()
908
self._copy_text_texts()
909
# select signature keys
910
signature_filter = self._revision_keys # same keyspace
911
signature_index_map, signature_indices = self._pack_map_and_index_list(
912
'signature_index')
913
signature_nodes = self._index_contents(signature_indices,
914
signature_filter)
915
# copy signature keys and adjust values
916
self.pb.update("Copying signature texts", 4)
917
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
918
new_pack.signature_index)
919
if 'pack' in debug.debug_flags:
920
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
921
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
922
new_pack.signature_index.key_count(),
923
time.time() - new_pack.start_time)
924
# copy chk contents
925
# NB XXX: how to check CHK references are present? perhaps by yielding
926
# the items? How should that interact with stacked repos?
927
if new_pack.chk_index is not None:
928
self._copy_chks()
929
if 'pack' in debug.debug_flags:
930
mutter('%s: create_pack: chk content copied: %s%s %d items t+%6.3fs',
931
time.ctime(), self._pack_collection._upload_transport.base,
932
new_pack.random_name,
933
new_pack.chk_index.key_count(),
934
time.time() - new_pack.start_time)
935
new_pack._check_references()
936
if not self._use_pack(new_pack):
937
new_pack.abort()
938
return None
939
self.pb.update("Finishing pack", 5)
940
new_pack.finish()
941
self._pack_collection.allocate(new_pack)
942
return new_pack
943
944
def _copy_chks(self, refs=None):
945
# XXX: Todo, recursive follow-pointers facility when fetching some
946
# revisions only.
947
chk_index_map, chk_indices = self._pack_map_and_index_list(
948
'chk_index')
949
chk_nodes = self._index_contents(chk_indices, refs)
950
new_refs = set()
951
# TODO: This isn't strictly tasteful as we are accessing some private
952
# variables (_serializer). Perhaps a better way would be to have
953
# Repository._deserialise_chk_node()
954
search_key_func = chk_map.search_key_registry.get(
955
self._pack_collection.repo._serializer.search_key_name)
956
def accumlate_refs(lines):
957
# XXX: move to a generic location
958
# Yay mismatch:
959
bytes = ''.join(lines)
960
node = chk_map._deserialise(bytes, ("unknown",), search_key_func)
961
new_refs.update(node.refs())
962
self._copy_nodes(chk_nodes, chk_index_map, self.new_pack._writer,
963
self.new_pack.chk_index, output_lines=accumlate_refs)
964
return new_refs
965
966
def _copy_nodes(self, nodes, index_map, writer, write_index,
967
output_lines=None):
968
"""Copy knit nodes between packs with no graph references.
969
970
:param output_lines: Output full texts of copied items.
971
"""
972
pb = ui.ui_factory.nested_progress_bar()
973
try:
974
return self._do_copy_nodes(nodes, index_map, writer,
975
write_index, pb, output_lines=output_lines)
976
finally:
977
pb.finished()
978
979
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb,
980
output_lines=None):
981
# for record verification
982
knit = KnitVersionedFiles(None, None)
983
# plan a readv on each source pack:
984
# group by pack
985
nodes = sorted(nodes)
986
# how to map this into knit.py - or knit.py into this?
987
# we don't want the typical knit logic, we want grouping by pack
988
# at this point - perhaps a helper library for the following code
989
# duplication points?
990
request_groups = {}
991
for index, key, value in nodes:
992
if index not in request_groups:
993
request_groups[index] = []
994
request_groups[index].append((key, value))
995
record_index = 0
996
pb.update("Copied record", record_index, len(nodes))
997
for index, items in request_groups.iteritems():
998
pack_readv_requests = []
999
for key, value in items:
1000
# ---- KnitGraphIndex.get_position
1001
bits = value[1:].split(' ')
1002
offset, length = int(bits[0]), int(bits[1])
1003
pack_readv_requests.append((offset, length, (key, value[0])))
1004
# linear scan up the pack
1005
pack_readv_requests.sort()
1006
# copy the data
1007
pack_obj = index_map[index]
1008
transport, path = pack_obj.access_tuple()
1009
try:
1010
reader = pack.make_readv_reader(transport, path,
1011
[offset[0:2] for offset in pack_readv_requests])
1012
except errors.NoSuchFile:
1013
if self._reload_func is not None:
1014
self._reload_func()
1015
raise
1016
for (names, read_func), (_1, _2, (key, eol_flag)) in \
1017
izip(reader.iter_records(), pack_readv_requests):
1018
raw_data = read_func(None)
1019
# check the header only
1020
if output_lines is not None:
1021
output_lines(knit._parse_record(key[-1], raw_data)[0])
1022
else:
1023
df, _ = knit._parse_record_header(key, raw_data)
1024
df.close()
1025
pos, size = writer.add_bytes_record(raw_data, names)
1026
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
1027
pb.update("Copied record", record_index)
1028
record_index += 1
1029
1030
def _copy_nodes_graph(self, index_map, writer, write_index,
1031
readv_group_iter, total_items, output_lines=False):
1032
"""Copy knit nodes between packs.
1033
1034
:param output_lines: Return lines present in the copied data as
1035
an iterator of line,version_id.
1036
"""
1037
pb = ui.ui_factory.nested_progress_bar()
1038
try:
1039
for result in self._do_copy_nodes_graph(index_map, writer,
1040
write_index, output_lines, pb, readv_group_iter, total_items):
1041
yield result
1042
except Exception:
1043
# Python 2.4 does not permit try:finally: in a generator.
1044
pb.finished()
1045
raise
1046
else:
1047
pb.finished()
1048
1049
def _do_copy_nodes_graph(self, index_map, writer, write_index,
1050
output_lines, pb, readv_group_iter, total_items):
1051
# for record verification
1052
knit = KnitVersionedFiles(None, None)
1053
# for line extraction when requested (inventories only)
1054
if output_lines:
1055
factory = KnitPlainFactory()
1056
record_index = 0
1057
pb.update("Copied record", record_index, total_items)
1058
for index, readv_vector, node_vector in readv_group_iter:
1059
# copy the data
1060
pack_obj = index_map[index]
1061
transport, path = pack_obj.access_tuple()
1062
try:
1063
reader = pack.make_readv_reader(transport, path, readv_vector)
1064
except errors.NoSuchFile:
1065
if self._reload_func is not None:
1066
self._reload_func()
1067
raise
1068
for (names, read_func), (key, eol_flag, references) in \
1069
izip(reader.iter_records(), node_vector):
1070
raw_data = read_func(None)
1071
if output_lines:
1072
# read the entire thing
1073
content, _ = knit._parse_record(key[-1], raw_data)
1074
if len(references[-1]) == 0:
1075
line_iterator = factory.get_fulltext_content(content)
1076
else:
1077
line_iterator = factory.get_linedelta_content(content)
1078
for line in line_iterator:
1079
yield line, key
1080
else:
1081
# check the header only
1082
df, _ = knit._parse_record_header(key, raw_data)
1083
df.close()
1084
pos, size = writer.add_bytes_record(raw_data, names)
1085
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1086
pb.update("Copied record", record_index)
1087
record_index += 1
1088
1089
def _get_text_nodes(self):
1090
text_index_map, text_indices = self._pack_map_and_index_list(
1091
'text_index')
1092
return text_index_map, self._index_contents(text_indices,
1093
self._text_filter)
795
1094
796
1095
def _least_readv_node_readv(self, nodes):
797
1096
"""Generate request groups for nodes using the least readv's.
838
1137
self.new_pack.text_index.key_count(),
839
1138
time.time() - self.new_pack.start_time)
840
1139
1140
def _process_inventory_lines(self, inv_lines):
1141
"""Use up the inv_lines generator and setup a text key filter."""
1142
repo = self._pack_collection.repo
1143
fileid_revisions = repo._find_file_ids_from_xml_inventory_lines(
1144
inv_lines, self.revision_keys)
1145
text_filter = []
1146
for fileid, file_revids in fileid_revisions.iteritems():
1147
text_filter.extend([(fileid, file_revid) for file_revid in file_revids])
1148
self._text_filter = text_filter
1149
841
1150
def _revision_node_readv(self, revision_nodes):
842
1151
"""Return the total revisions and the readv's to issue.
843
1152
900
1209
return new_pack
901
1210
902
1211
1212
class ReconcilePacker(Packer):
1213
"""A packer which regenerates indices etc as it copies.
1214
1215
This is used by ``bzr reconcile`` to cause parent text pointers to be
1216
regenerated.
1217
"""
1218
1219
def _extra_init(self):
1220
self._data_changed = False
1221
1222
def _process_inventory_lines(self, inv_lines):
1223
"""Generate a text key reference map rather for reconciling with."""
1224
repo = self._pack_collection.repo
1225
refs = repo._serializer._find_text_key_references(inv_lines)
1226
self._text_refs = refs
1227
# during reconcile we:
1228
# - convert unreferenced texts to full texts
1229
# - correct texts which reference a text not copied to be full texts
1230
# - copy all others as-is but with corrected parents.
1231
# - so at this point we don't know enough to decide what becomes a full
1232
# text.
1233
self._text_filter = None
1234
1235
def _copy_text_texts(self):
1236
"""generate what texts we should have and then copy."""
1237
self.pb.update("Copying content texts", 3)
1238
# we have three major tasks here:
1239
# 1) generate the ideal index
1240
repo = self._pack_collection.repo
1241
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1242
_1, key, _2, refs in
1243
self.new_pack.revision_index.iter_all_entries()])
1244
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1245
# 2) generate a text_nodes list that contains all the deltas that can
1246
# be used as-is, with corrected parents.
1247
ok_nodes = []
1248
bad_texts = []
1249
discarded_nodes = []
1250
NULL_REVISION = _mod_revision.NULL_REVISION
1251
text_index_map, text_nodes = self._get_text_nodes()
1252
for node in text_nodes:
1253
# 0 - index
1254
# 1 - key
1255
# 2 - value
1256
# 3 - refs
1257
try:
1258
ideal_parents = tuple(ideal_index[node[1]])
1259
except KeyError:
1260
discarded_nodes.append(node)
1261
self._data_changed = True
1262
else:
1263
if ideal_parents == (NULL_REVISION,):
1264
ideal_parents = ()
1265
if ideal_parents == node[3][0]:
1266
# no change needed.
1267
ok_nodes.append(node)
1268
elif ideal_parents[0:1] == node[3][0][0:1]:
1269
# the left most parent is the same, or there are no parents
1270
# today. Either way, we can preserve the representation as
1271
# long as we change the refs to be inserted.
1272
self._data_changed = True
1273
ok_nodes.append((node[0], node[1], node[2],
1274
(ideal_parents, node[3][1])))
1275
self._data_changed = True
1276
else:
1277
# Reinsert this text completely
1278
bad_texts.append((node[1], ideal_parents))
1279
self._data_changed = True
1280
# we're finished with some data.
1281
del ideal_index
1282
del text_nodes
1283
# 3) bulk copy the ok data
1284
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1285
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1286
self.new_pack.text_index, readv_group_iter, total_items))
1287
# 4) adhoc copy all the other texts.
1288
# We have to topologically insert all texts otherwise we can fail to
1289
# reconcile when parts of a single delta chain are preserved intact,
1290
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1291
# reinserted, and if d3 has incorrect parents it will also be
1292
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1293
# copied), so we will try to delta, but d2 is not currently able to be
1294
# extracted because its basis d1 is not present. Topologically sorting
1295
# addresses this. The following generates a sort for all the texts that
1296
# are being inserted without having to reference the entire text key
1297
# space (we only topo sort the revisions, which is smaller).
1298
topo_order = tsort.topo_sort(ancestors)
1299
rev_order = dict(zip(topo_order, range(len(topo_order))))
1300
bad_texts.sort(key=lambda key:rev_order.get(key[0][1], 0))
1301
transaction = repo.get_transaction()
1302
file_id_index = GraphIndexPrefixAdapter(
1303
self.new_pack.text_index,
1304
('blank', ), 1,
1305
add_nodes_callback=self.new_pack.text_index.add_nodes)
1306
data_access = _DirectPackAccess(
1307
{self.new_pack.text_index:self.new_pack.access_tuple()})
1308
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1309
self.new_pack.access_tuple())
1310
output_texts = KnitVersionedFiles(
1311
_KnitGraphIndex(self.new_pack.text_index,
1312
add_callback=self.new_pack.text_index.add_nodes,
1313
deltas=True, parents=True, is_locked=repo.is_locked),
1314
data_access=data_access, max_delta_chain=200)
1315
for key, parent_keys in bad_texts:
1316
# We refer to the new pack to delta data being output.
1317
# A possible improvement would be to catch errors on short reads
1318
# and only flush then.
1319
self.new_pack.flush()
1320
parents = []
1321
for parent_key in parent_keys:
1322
if parent_key[0] != key[0]:
1323
# Graph parents must match the fileid
1324
raise errors.BzrError('Mismatched key parent %r:%r' %
1325
(key, parent_keys))
1326
parents.append(parent_key[1])
1327
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1328
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1329
output_texts.add_lines(key, parent_keys, text_lines,
1330
random_id=True, check_content=False)
1331
# 5) check that nothing inserted has a reference outside the keyspace.
1332
missing_text_keys = self.new_pack.text_index._external_references()
1333
if missing_text_keys:
1334
raise errors.BzrCheckError('Reference to missing compression parents %r'
1335
% (missing_text_keys,))
1336
self._log_copied_texts()
1337
1338
def _use_pack(self, new_pack):
1339
"""Override _use_pack to check for reconcile having changed content."""
1340
# XXX: we might be better checking this at the copy time.
1341
original_inventory_keys = set()
1342
inv_index = self._pack_collection.inventory_index.combined_index
1343
for entry in inv_index.iter_all_entries():
1344
original_inventory_keys.add(entry[1])
1345
new_inventory_keys = set()
1346
for entry in new_pack.inventory_index.iter_all_entries():
1347
new_inventory_keys.add(entry[1])
1348
if new_inventory_keys != original_inventory_keys:
1349
self._data_changed = True
1350
return new_pack.data_inserted() and self._data_changed
1351
1352
903
1353
class RepositoryPackCollection(object):
904
1354
"""Management of packs within a repository.
905
1355
1766
2216
self._resume_pack(token)
1767
2217
1768
2218
1769
class PackRepository(MetaDirRepository):
2219
class KnitPackRepository(KnitRepository):
1770
2220
"""Repository with knit objects stored inside pack containers.
1771
2221
1772
2222
The layering for a KnitPackRepository is:
1775
2225
===================================================
1776
2226
Tuple based apis below, string based, and key based apis above
1777
2227
---------------------------------------------------
1778
VersionedFiles
2228
KnitVersionedFiles
1779
2229
Provides .texts, .revisions etc
1780
2230
This adapts the N-tuple keys to physical knit records which only have a
1781
2231
single string identifier (for historical reasons), which in older formats
1791
2241
1792
2242
"""
1793
2243
1794
# These attributes are inherited from the Repository base class. Setting
1795
# them to None ensures that if the constructor is changed to not initialize
1796
# them, or a subclass fails to call the constructor, that an error will
1797
# occur rather than the system working but generating incorrect data.
1798
_commit_builder_class = None
1799
_serializer = None
1800
1801
2244
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
1802
2245
_serializer):
1803
MetaDirRepository.__init__(self, _format, a_bzrdir, control_files)
1804
self._commit_builder_class = _commit_builder_class
1805
self._serializer = _serializer
2246
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2247
_commit_builder_class, _serializer)
2248
index_transport = self._transport.clone('indices')
2249
self._pack_collection = RepositoryPackCollection(self, self._transport,
2250
index_transport,
2251
self._transport.clone('upload'),
2252
self._transport.clone('packs'),
2253
_format.index_builder_class,
2254
_format.index_class,
2255
use_chk_index=self._format.supports_chks,
2256
)
2257
self.inventories = KnitVersionedFiles(
2258
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2259
add_callback=self._pack_collection.inventory_index.add_callback,
2260
deltas=True, parents=True, is_locked=self.is_locked),
2261
data_access=self._pack_collection.inventory_index.data_access,
2262
max_delta_chain=200)
2263
self.revisions = KnitVersionedFiles(
2264
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2265
add_callback=self._pack_collection.revision_index.add_callback,
2266
deltas=False, parents=True, is_locked=self.is_locked,
2267
track_external_parent_refs=True),
2268
data_access=self._pack_collection.revision_index.data_access,
2269
max_delta_chain=0)
2270
self.signatures = KnitVersionedFiles(
2271
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2272
add_callback=self._pack_collection.signature_index.add_callback,
2273
deltas=False, parents=False, is_locked=self.is_locked),
2274
data_access=self._pack_collection.signature_index.data_access,
2275
max_delta_chain=0)
2276
self.texts = KnitVersionedFiles(
2277
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2278
add_callback=self._pack_collection.text_index.add_callback,
2279
deltas=True, parents=True, is_locked=self.is_locked),
2280
data_access=self._pack_collection.text_index.data_access,
2281
max_delta_chain=200)
2282
if _format.supports_chks:
2283
# No graph, no compression:- references from chks are between
2284
# different objects not temporal versions of the same; and without
2285
# some sort of temporal structure knit compression will just fail.
2286
self.chk_bytes = KnitVersionedFiles(
2287
_KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2288
add_callback=self._pack_collection.chk_index.add_callback,
2289
deltas=False, parents=False, is_locked=self.is_locked),
2290
data_access=self._pack_collection.chk_index.data_access,
2291
max_delta_chain=0)
2292
else:
2293
self.chk_bytes = None
2294
# True when the repository object is 'write locked' (as opposed to the
2295
# physical lock only taken out around changes to the pack-names list.)
2296
# Another way to represent this would be a decorator around the control
2297
# files object that presents logical locks as physical ones - if this
2298
# gets ugly consider that alternative design. RBC 20071011
2299
self._write_lock_count = 0
2300
self._transaction = None
2301
# for tests
2302
self._reconcile_does_inventory_gc = True
1806
2303
self._reconcile_fixes_text_parents = True
1807
1808
@needs_read_lock
1809
def _all_revision_ids(self):
1810
"""See Repository.all_revision_ids()."""
1811
return [key[0] for key in self.revisions.keys()]
2304
self._reconcile_backsup_inventory = False
1812
2305
1813
2306
def _abort_write_group(self):
1814
2307
self.revisions._index._key_dependencies.clear()
1815
2308
self._pack_collection._abort_write_group()
1816
2309
2310
def _get_source(self, to_format):
2311
if to_format.network_name() == self._format.network_name():
2312
return KnitPackStreamSource(self, to_format)
2313
return super(KnitPackRepository, self)._get_source(to_format)
2314
1817
2315
def _make_parents_provider(self):
1818
2316
return graph.CachingParentsProvider(self)
1819
2317
1925
2423
return reconciler
1926
2424
1927
2425
def _reconcile_pack(self, collection, packs, extension, revs, pb):
1928
raise NotImplementedError(self._reconcile_pack)
2426
packer = ReconcilePacker(collection, packs, extension, revs)
2427
return packer.pack(pb)
1929
2428
1930
2429
@only_raises(errors.LockNotHeld, errors.LockBroken)
1931
2430
def unlock(self):
1950
2449
repo.unlock()
1951
2450
1952
2451
2452
class KnitPackStreamSource(StreamSource):
2453
"""A StreamSource used to transfer data between same-format KnitPack repos.
2454
2455
This source assumes:
2456
1) Same serialization format for all objects
2457
2) Same root information
2458
3) XML format inventories
2459
4) Atomic inserts (so we can stream inventory texts before text
2460
content)
2461
5) No chk_bytes
2462
"""
2463
2464
def __init__(self, from_repository, to_format):
2465
super(KnitPackStreamSource, self).__init__(from_repository, to_format)
2466
self._text_keys = None
2467
self._text_fetch_order = 'unordered'
2468
2469
def _get_filtered_inv_stream(self, revision_ids):
2470
from_repo = self.from_repository
2471
parent_ids = from_repo._find_parent_ids_of_revisions(revision_ids)
2472
parent_keys = [(p,) for p in parent_ids]
2473
find_text_keys = from_repo._serializer._find_text_key_references
2474
parent_text_keys = set(find_text_keys(
2475
from_repo._inventory_xml_lines_for_keys(parent_keys)))
2476
content_text_keys = set()
2477
knit = KnitVersionedFiles(None, None)
2478
factory = KnitPlainFactory()
2479
def find_text_keys_from_content(record):
2480
if record.storage_kind not in ('knit-delta-gz', 'knit-ft-gz'):
2481
raise ValueError("Unknown content storage kind for"
2482
" inventory text: %s" % (record.storage_kind,))
2483
# It's a knit record, it has a _raw_record field (even if it was
2484
# reconstituted from a network stream).
2485
raw_data = record._raw_record
2486
# read the entire thing
2487
revision_id = record.key[-1]
2488
content, _ = knit._parse_record(revision_id, raw_data)
2489
if record.storage_kind == 'knit-delta-gz':
2490
line_iterator = factory.get_linedelta_content(content)
2491
elif record.storage_kind == 'knit-ft-gz':
2492
line_iterator = factory.get_fulltext_content(content)
2493
content_text_keys.update(find_text_keys(
2494
[(line, revision_id) for line in line_iterator]))
2495
revision_keys = [(r,) for r in revision_ids]
2496
def _filtered_inv_stream():
2497
source_vf = from_repo.inventories
2498
stream = source_vf.get_record_stream(revision_keys,
2499
'unordered', False)
2500
for record in stream:
2501
if record.storage_kind == 'absent':
2502
raise errors.NoSuchRevision(from_repo, record.key)
2503
find_text_keys_from_content(record)
2504
yield record
2505
self._text_keys = content_text_keys - parent_text_keys
2506
return ('inventories', _filtered_inv_stream())
2507
2508
def _get_text_stream(self):
2509
# Note: We know we don't have to handle adding root keys, because both
2510
# the source and target are the identical network name.
2511
text_stream = self.from_repository.texts.get_record_stream(
2512
self._text_keys, self._text_fetch_order, False)
2513
return ('texts', text_stream)
2514
2515
def get_stream(self, search):
2516
revision_ids = search.get_keys()
2517
for stream_info in self._fetch_revision_texts(revision_ids):
2518
yield stream_info
2519
self._revision_keys = [(rev_id,) for rev_id in revision_ids]
2520
yield self._get_filtered_inv_stream(revision_ids)
2521
yield self._get_text_stream()
2522
2523
2524
1953
2525
class RepositoryFormatPack(MetaDirRepositoryFormat):
1954
2526
"""Format logic for pack structured repositories.
1955
2527
1986
2558
index_class = None
1987
2559
_fetch_uses_deltas = True
1988
2560
fast_deltas = False
1989
supports_full_versioned_files = True
1990
supports_funky_characters = True
1991
2561
1992
2562
def initialize(self, a_bzrdir, shared=False):
1993
2563
"""Create a pack based repository.
2030
2600
_serializer=self._serializer)
2031
2601
2032
2602
2603
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2604
"""A no-subtrees parameterized Pack repository.
2605
2606
This format was introduced in 0.92.
2607
"""
2608
2609
repository_class = KnitPackRepository
2610
_commit_builder_class = PackCommitBuilder
2611
@property
2612
def _serializer(self):
2613
return xml5.serializer_v5
2614
# What index classes to use
2615
index_builder_class = InMemoryGraphIndex
2616
index_class = GraphIndex
2617
2618
def _get_matching_bzrdir(self):
2619
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2620
2621
def _ignore_setting_bzrdir(self, format):
2622
pass
2623
2624
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2625
2626
def get_format_string(self):
2627
"""See RepositoryFormat.get_format_string()."""
2628
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2629
2630
def get_format_description(self):
2631
"""See RepositoryFormat.get_format_description()."""
2632
return "Packs containing knits without subtree support"
2633
2634
2635
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2636
"""A subtrees parameterized Pack repository.
2637
2638
This repository format uses the xml7 serializer to get:
2639
- support for recording full info about the tree root
2640
- support for recording tree-references
2641
2642
This format was introduced in 0.92.
2643
"""
2644
2645
repository_class = KnitPackRepository
2646
_commit_builder_class = PackRootCommitBuilder
2647
rich_root_data = True
2648
experimental = True
2649
supports_tree_reference = True
2650
@property
2651
def _serializer(self):
2652
return xml7.serializer_v7
2653
# What index classes to use
2654
index_builder_class = InMemoryGraphIndex
2655
index_class = GraphIndex
2656
2657
def _get_matching_bzrdir(self):
2658
return bzrdir.format_registry.make_bzrdir(
2659
'pack-0.92-subtree')
2660
2661
def _ignore_setting_bzrdir(self, format):
2662
pass
2663
2664
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2665
2666
def get_format_string(self):
2667
"""See RepositoryFormat.get_format_string()."""
2668
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2669
2670
def get_format_description(self):
2671
"""See RepositoryFormat.get_format_description()."""
2672
return "Packs containing knits with subtree support\n"
2673
2674
2675
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2676
"""A rich-root, no subtrees parameterized Pack repository.
2677
2678
This repository format uses the xml6 serializer to get:
2679
- support for recording full info about the tree root
2680
2681
This format was introduced in 1.0.
2682
"""
2683
2684
repository_class = KnitPackRepository
2685
_commit_builder_class = PackRootCommitBuilder
2686
rich_root_data = True
2687
supports_tree_reference = False
2688
@property
2689
def _serializer(self):
2690
return xml6.serializer_v6
2691
# What index classes to use
2692
index_builder_class = InMemoryGraphIndex
2693
index_class = GraphIndex
2694
2695
def _get_matching_bzrdir(self):
2696
return bzrdir.format_registry.make_bzrdir(
2697
'rich-root-pack')
2698
2699
def _ignore_setting_bzrdir(self, format):
2700
pass
2701
2702
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2703
2704
def get_format_string(self):
2705
"""See RepositoryFormat.get_format_string()."""
2706
return ("Bazaar pack repository format 1 with rich root"
2707
" (needs bzr 1.0)\n")
2708
2709
def get_format_description(self):
2710
"""See RepositoryFormat.get_format_description()."""
2711
return "Packs containing knits with rich root support\n"
2712
2713
2714
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2715
"""Repository that supports external references to allow stacking.
2716
2717
New in release 1.6.
2718
2719
Supports external lookups, which results in non-truncated ghosts after
2720
reconcile compared to pack-0.92 formats.
2721
"""
2722
2723
repository_class = KnitPackRepository
2724
_commit_builder_class = PackCommitBuilder
2725
supports_external_lookups = True
2726
# What index classes to use
2727
index_builder_class = InMemoryGraphIndex
2728
index_class = GraphIndex
2729
2730
@property
2731
def _serializer(self):
2732
return xml5.serializer_v5
2733
2734
def _get_matching_bzrdir(self):
2735
return bzrdir.format_registry.make_bzrdir('1.6')
2736
2737
def _ignore_setting_bzrdir(self, format):
2738
pass
2739
2740
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2741
2742
def get_format_string(self):
2743
"""See RepositoryFormat.get_format_string()."""
2744
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2745
2746
def get_format_description(self):
2747
"""See RepositoryFormat.get_format_description()."""
2748
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2749
2750
2751
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2752
"""A repository with rich roots and stacking.
2753
2754
New in release 1.6.1.
2755
2756
Supports stacking on other repositories, allowing data to be accessed
2757
without being stored locally.
2758
"""
2759
2760
repository_class = KnitPackRepository
2761
_commit_builder_class = PackRootCommitBuilder
2762
rich_root_data = True
2763
supports_tree_reference = False # no subtrees
2764
supports_external_lookups = True
2765
# What index classes to use
2766
index_builder_class = InMemoryGraphIndex
2767
index_class = GraphIndex
2768
2769
@property
2770
def _serializer(self):
2771
return xml6.serializer_v6
2772
2773
def _get_matching_bzrdir(self):
2774
return bzrdir.format_registry.make_bzrdir(
2775
'1.6.1-rich-root')
2776
2777
def _ignore_setting_bzrdir(self, format):
2778
pass
2779
2780
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2781
2782
def get_format_string(self):
2783
"""See RepositoryFormat.get_format_string()."""
2784
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2785
2786
def get_format_description(self):
2787
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2788
2789
2790
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2791
"""A repository with rich roots and external references.
2792
2793
New in release 1.6.
2794
2795
Supports external lookups, which results in non-truncated ghosts after
2796
reconcile compared to pack-0.92 formats.
2797
2798
This format was deprecated because the serializer it uses accidentally
2799
supported subtrees, when the format was not intended to. This meant that
2800
someone could accidentally fetch from an incorrect repository.
2801
"""
2802
2803
repository_class = KnitPackRepository
2804
_commit_builder_class = PackRootCommitBuilder
2805
rich_root_data = True
2806
supports_tree_reference = False # no subtrees
2807
2808
supports_external_lookups = True
2809
# What index classes to use
2810
index_builder_class = InMemoryGraphIndex
2811
index_class = GraphIndex
2812
2813
@property
2814
def _serializer(self):
2815
return xml7.serializer_v7
2816
2817
def _get_matching_bzrdir(self):
2818
matching = bzrdir.format_registry.make_bzrdir(
2819
'1.6.1-rich-root')
2820
matching.repository_format = self
2821
return matching
2822
2823
def _ignore_setting_bzrdir(self, format):
2824
pass
2825
2826
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2827
2828
def get_format_string(self):
2829
"""See RepositoryFormat.get_format_string()."""
2830
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2831
2832
def get_format_description(self):
2833
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2834
" (deprecated)")
2835
2836
def is_deprecated(self):
2837
return True
2838
2839
2840
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2841
"""A repository with stacking and btree indexes,
2842
without rich roots or subtrees.
2843
2844
This is equivalent to pack-1.6 with B+Tree indices.
2845
"""
2846
2847
repository_class = KnitPackRepository
2848
_commit_builder_class = PackCommitBuilder
2849
supports_external_lookups = True
2850
# What index classes to use
2851
index_builder_class = btree_index.BTreeBuilder
2852
index_class = btree_index.BTreeGraphIndex
2853
2854
@property
2855
def _serializer(self):
2856
return xml5.serializer_v5
2857
2858
def _get_matching_bzrdir(self):
2859
return bzrdir.format_registry.make_bzrdir('1.9')
2860
2861
def _ignore_setting_bzrdir(self, format):
2862
pass
2863
2864
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2865
2866
def get_format_string(self):
2867
"""See RepositoryFormat.get_format_string()."""
2868
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2869
2870
def get_format_description(self):
2871
"""See RepositoryFormat.get_format_description()."""
2872
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2873
2874
2875
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2876
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2877
2878
1.6-rich-root with B+Tree indices.
2879
"""
2880
2881
repository_class = KnitPackRepository
2882
_commit_builder_class = PackRootCommitBuilder
2883
rich_root_data = True
2884
supports_tree_reference = False # no subtrees
2885
supports_external_lookups = True
2886
# What index classes to use
2887
index_builder_class = btree_index.BTreeBuilder
2888
index_class = btree_index.BTreeGraphIndex
2889
2890
@property
2891
def _serializer(self):
2892
return xml6.serializer_v6
2893
2894
def _get_matching_bzrdir(self):
2895
return bzrdir.format_registry.make_bzrdir(
2896
'1.9-rich-root')
2897
2898
def _ignore_setting_bzrdir(self, format):
2899
pass
2900
2901
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2902
2903
def get_format_string(self):
2904
"""See RepositoryFormat.get_format_string()."""
2905
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2906
2907
def get_format_description(self):
2908
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2909
2910
2911
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2912
"""A subtrees development repository.
2913
2914
This format should be retained in 2.3, to provide an upgrade path from this
2915
to RepositoryFormat2aSubtree. It can be removed in later releases.
2916
2917
1.6.1-subtree[as it might have been] with B+Tree indices.
2918
"""
2919
2920
repository_class = KnitPackRepository
2921
_commit_builder_class = PackRootCommitBuilder
2922
rich_root_data = True
2923
experimental = True
2924
supports_tree_reference = True
2925
supports_external_lookups = True
2926
# What index classes to use
2927
index_builder_class = btree_index.BTreeBuilder
2928
index_class = btree_index.BTreeGraphIndex
2929
2930
@property
2931
def _serializer(self):
2932
return xml7.serializer_v7
2933
2934
def _get_matching_bzrdir(self):
2935
return bzrdir.format_registry.make_bzrdir(
2936
'development5-subtree')
2937
2938
def _ignore_setting_bzrdir(self, format):
2939
pass
2940
2941
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2942
2943
def get_format_string(self):
2944
"""See RepositoryFormat.get_format_string()."""
2945
return ("Bazaar development format 2 with subtree support "
2946
"(needs bzr.dev from before 1.8)\n")
2947
2948
def get_format_description(self):
2949
"""See RepositoryFormat.get_format_description()."""
2950
return ("Development repository format, currently the same as "
2951
"1.6.1-subtree with B+Tree indices.\n")
2952
2953
2033
2954
class RetryPackOperations(errors.RetryWithNewPacks):
2034
2955
"""Raised when we are packing and we find a missing file.
2035
2956
Loggerhead is a web-based interface for Breezy
Version: 2.0.1