828
743
Sets self._text_filter appropriately.
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(
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
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)
851
# eat the iterator to cause it to execute.
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)
745
raise NotImplementedError(self._copy_inventory_texts)
861
747
def _copy_text_texts(self):
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
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
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],
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()
748
raise NotImplementedError(self._copy_text_texts)
888
750
def _create_pack_from_packs(self):
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)
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(
913
signature_nodes = self._index_contents(signature_indices,
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)
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:
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):
939
self.pb.update("Finishing pack", 5)
941
self._pack_collection.allocate(new_pack)
944
def _copy_chks(self, refs=None):
945
# XXX: Todo, recursive follow-pointers facility when fetching some
947
chk_index_map, chk_indices = self._pack_map_and_index_list(
949
chk_nodes = self._index_contents(chk_indices, refs)
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
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)
966
def _copy_nodes(self, nodes, index_map, writer, write_index,
968
"""Copy knit nodes between packs with no graph references.
970
:param output_lines: Output full texts of copied items.
972
pb = ui.ui_factory.nested_progress_bar()
974
return self._do_copy_nodes(nodes, index_map, writer,
975
write_index, pb, output_lines=output_lines)
979
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb,
981
# for record verification
982
knit = KnitVersionedFiles(None, None)
983
# plan a readv on each source 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?
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))
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()
1007
pack_obj = index_map[index]
1008
transport, path = pack_obj.access_tuple()
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:
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])
1023
df, _ = knit._parse_record_header(key, raw_data)
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)
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.
1034
:param output_lines: Return lines present in the copied data as
1035
an iterator of line,version_id.
1037
pb = ui.ui_factory.nested_progress_bar()
1039
for result in self._do_copy_nodes_graph(index_map, writer,
1040
write_index, output_lines, pb, readv_group_iter, total_items):
1043
# Python 2.4 does not permit try:finally: in a generator.
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)
1055
factory = KnitPlainFactory()
1057
pb.update("Copied record", record_index, total_items)
1058
for index, readv_vector, node_vector in readv_group_iter:
1060
pack_obj = index_map[index]
1061
transport, path = pack_obj.access_tuple()
1063
reader = pack.make_readv_reader(transport, path, readv_vector)
1064
except errors.NoSuchFile:
1065
if self._reload_func is not None:
1068
for (names, read_func), (key, eol_flag, references) in \
1069
izip(reader.iter_records(), node_vector):
1070
raw_data = read_func(None)
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)
1077
line_iterator = factory.get_linedelta_content(content)
1078
for line in line_iterator:
1081
# check the header only
1082
df, _ = knit._parse_record_header(key, raw_data)
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)
1089
def _get_text_nodes(self):
1090
text_index_map, text_indices = self._pack_map_and_index_list(
1092
return text_index_map, self._index_contents(text_indices,
1095
def _least_readv_node_readv(self, nodes):
1096
"""Generate request groups for nodes using the least readv's.
1098
:param nodes: An iterable of graph index nodes.
1099
:return: Total node count and an iterator of the data needed to perform
1100
readvs to obtain the data for nodes. Each item yielded by the
1101
iterator is a tuple with:
1102
index, readv_vector, node_vector. readv_vector is a list ready to
1103
hand to the transport readv method, and node_vector is a list of
1104
(key, eol_flag, references) for the node retrieved by the
1105
matching readv_vector.
1107
# group by pack so we do one readv per pack
1108
nodes = sorted(nodes)
1111
for index, key, value, references in nodes:
1112
if index not in request_groups:
1113
request_groups[index] = []
1114
request_groups[index].append((key, value, references))
1116
for index, items in request_groups.iteritems():
1117
pack_readv_requests = []
1118
for key, value, references in items:
1119
# ---- KnitGraphIndex.get_position
1120
bits = value[1:].split(' ')
1121
offset, length = int(bits[0]), int(bits[1])
1122
pack_readv_requests.append(
1123
((offset, length), (key, value[0], references)))
1124
# linear scan up the pack to maximum range combining.
1125
pack_readv_requests.sort()
1126
# split out the readv and the node data.
1127
pack_readv = [readv for readv, node in pack_readv_requests]
1128
node_vector = [node for readv, node in pack_readv_requests]
1129
result.append((index, pack_readv, node_vector))
1130
return total, result
751
raise NotImplementedError(self._create_pack_from_packs)
1132
753
def _log_copied_texts(self):
1133
754
if 'pack' in debug.debug_flags:
1165
767
return new_pack.data_inserted()
1168
class OptimisingPacker(Packer):
1169
"""A packer which spends more time to create better disk layouts."""
1171
def _revision_node_readv(self, revision_nodes):
1172
"""Return the total revisions and the readv's to issue.
1174
This sort places revisions in topological order with the ancestors
1177
:param revision_nodes: The revision index contents for the packs being
1178
incorporated into the new pack.
1179
:return: As per _least_readv_node_readv.
1181
# build an ancestors dict
1184
for index, key, value, references in revision_nodes:
1185
ancestors[key] = references[0]
1186
by_key[key] = (index, value, references)
1187
order = tsort.topo_sort(ancestors)
1189
# Single IO is pathological, but it will work as a starting point.
1191
for key in reversed(order):
1192
index, value, references = by_key[key]
1193
# ---- KnitGraphIndex.get_position
1194
bits = value[1:].split(' ')
1195
offset, length = int(bits[0]), int(bits[1])
1197
(index, [(offset, length)], [(key, value[0], references)]))
1198
# TODO: combine requests in the same index that are in ascending order.
1199
return total, requests
1201
def open_pack(self):
1202
"""Open a pack for the pack we are creating."""
1203
new_pack = super(OptimisingPacker, self).open_pack()
1204
# Turn on the optimization flags for all the index builders.
1205
new_pack.revision_index.set_optimize(for_size=True)
1206
new_pack.inventory_index.set_optimize(for_size=True)
1207
new_pack.text_index.set_optimize(for_size=True)
1208
new_pack.signature_index.set_optimize(for_size=True)
1212
class ReconcilePacker(Packer):
1213
"""A packer which regenerates indices etc as it copies.
1215
This is used by ``bzr reconcile`` to cause parent text pointers to be
1219
def _extra_init(self):
1220
self._data_changed = False
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._find_text_key_references_from_xml_inventory_lines(
1227
self._text_refs = refs
1228
# during reconcile we:
1229
# - convert unreferenced texts to full texts
1230
# - correct texts which reference a text not copied to be full texts
1231
# - copy all others as-is but with corrected parents.
1232
# - so at this point we don't know enough to decide what becomes a full
1234
self._text_filter = None
1236
def _copy_text_texts(self):
1237
"""generate what texts we should have and then copy."""
1238
self.pb.update("Copying content texts", 3)
1239
# we have three major tasks here:
1240
# 1) generate the ideal index
1241
repo = self._pack_collection.repo
1242
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1243
_1, key, _2, refs in
1244
self.new_pack.revision_index.iter_all_entries()])
1245
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1246
# 2) generate a text_nodes list that contains all the deltas that can
1247
# be used as-is, with corrected parents.
1250
discarded_nodes = []
1251
NULL_REVISION = _mod_revision.NULL_REVISION
1252
text_index_map, text_nodes = self._get_text_nodes()
1253
for node in text_nodes:
1259
ideal_parents = tuple(ideal_index[node[1]])
1261
discarded_nodes.append(node)
1262
self._data_changed = True
1264
if ideal_parents == (NULL_REVISION,):
1266
if ideal_parents == node[3][0]:
1268
ok_nodes.append(node)
1269
elif ideal_parents[0:1] == node[3][0][0:1]:
1270
# the left most parent is the same, or there are no parents
1271
# today. Either way, we can preserve the representation as
1272
# long as we change the refs to be inserted.
1273
self._data_changed = True
1274
ok_nodes.append((node[0], node[1], node[2],
1275
(ideal_parents, node[3][1])))
1276
self._data_changed = True
1278
# Reinsert this text completely
1279
bad_texts.append((node[1], ideal_parents))
1280
self._data_changed = True
1281
# we're finished with some data.
1284
# 3) bulk copy the ok data
1285
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1286
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1287
self.new_pack.text_index, readv_group_iter, total_items))
1288
# 4) adhoc copy all the other texts.
1289
# We have to topologically insert all texts otherwise we can fail to
1290
# reconcile when parts of a single delta chain are preserved intact,
1291
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1292
# reinserted, and if d3 has incorrect parents it will also be
1293
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1294
# copied), so we will try to delta, but d2 is not currently able to be
1295
# extracted because its basis d1 is not present. Topologically sorting
1296
# addresses this. The following generates a sort for all the texts that
1297
# are being inserted without having to reference the entire text key
1298
# space (we only topo sort the revisions, which is smaller).
1299
topo_order = tsort.topo_sort(ancestors)
1300
rev_order = dict(zip(topo_order, range(len(topo_order))))
1301
bad_texts.sort(key=lambda key:rev_order.get(key[0][1], 0))
1302
transaction = repo.get_transaction()
1303
file_id_index = GraphIndexPrefixAdapter(
1304
self.new_pack.text_index,
1306
add_nodes_callback=self.new_pack.text_index.add_nodes)
1307
data_access = _DirectPackAccess(
1308
{self.new_pack.text_index:self.new_pack.access_tuple()})
1309
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1310
self.new_pack.access_tuple())
1311
output_texts = KnitVersionedFiles(
1312
_KnitGraphIndex(self.new_pack.text_index,
1313
add_callback=self.new_pack.text_index.add_nodes,
1314
deltas=True, parents=True, is_locked=repo.is_locked),
1315
data_access=data_access, max_delta_chain=200)
1316
for key, parent_keys in bad_texts:
1317
# We refer to the new pack to delta data being output.
1318
# A possible improvement would be to catch errors on short reads
1319
# and only flush then.
1320
self.new_pack.flush()
1322
for parent_key in parent_keys:
1323
if parent_key[0] != key[0]:
1324
# Graph parents must match the fileid
1325
raise errors.BzrError('Mismatched key parent %r:%r' %
1327
parents.append(parent_key[1])
1328
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1329
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1330
output_texts.add_lines(key, parent_keys, text_lines,
1331
random_id=True, check_content=False)
1332
# 5) check that nothing inserted has a reference outside the keyspace.
1333
missing_text_keys = self.new_pack.text_index._external_references()
1334
if missing_text_keys:
1335
raise errors.BzrCheckError('Reference to missing compression parents %r'
1336
% (missing_text_keys,))
1337
self._log_copied_texts()
1339
def _use_pack(self, new_pack):
1340
"""Override _use_pack to check for reconcile having changed content."""
1341
# XXX: we might be better checking this at the copy time.
1342
original_inventory_keys = set()
1343
inv_index = self._pack_collection.inventory_index.combined_index
1344
for entry in inv_index.iter_all_entries():
1345
original_inventory_keys.add(entry[1])
1346
new_inventory_keys = set()
1347
for entry in new_pack.inventory_index.iter_all_entries():
1348
new_inventory_keys.add(entry[1])
1349
if new_inventory_keys != original_inventory_keys:
1350
self._data_changed = True
1351
return new_pack.data_inserted() and self._data_changed
1354
770
class RepositoryPackCollection(object):
1355
771
"""Management of packs within a repository.
1357
773
:ivar _names: map of {pack_name: (index_size,)}
1360
pack_factory = NewPack
1361
resumed_pack_factory = ResumedPack
777
resumed_pack_factory = None
778
normal_packer_class = None
779
optimising_packer_class = None
1363
781
def __init__(self, repo, transport, index_transport, upload_transport,
1364
782
pack_transport, index_builder_class, index_class,
1681
# These attributes are inherited from the Repository base class. Setting
1682
# them to None ensures that if the constructor is changed to not initialize
1683
# them, or a subclass fails to call the constructor, that an error will
1684
# occur rather than the system working but generating incorrect data.
1685
_commit_builder_class = None
2245
1688
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
2247
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2248
_commit_builder_class, _serializer)
2249
index_transport = self._transport.clone('indices')
2250
self._pack_collection = RepositoryPackCollection(self, self._transport,
2252
self._transport.clone('upload'),
2253
self._transport.clone('packs'),
2254
_format.index_builder_class,
2255
_format.index_class,
2256
use_chk_index=self._format.supports_chks,
2258
self.inventories = KnitVersionedFiles(
2259
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2260
add_callback=self._pack_collection.inventory_index.add_callback,
2261
deltas=True, parents=True, is_locked=self.is_locked),
2262
data_access=self._pack_collection.inventory_index.data_access,
2263
max_delta_chain=200)
2264
self.revisions = KnitVersionedFiles(
2265
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2266
add_callback=self._pack_collection.revision_index.add_callback,
2267
deltas=False, parents=True, is_locked=self.is_locked,
2268
track_external_parent_refs=True),
2269
data_access=self._pack_collection.revision_index.data_access,
2271
self.signatures = KnitVersionedFiles(
2272
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2273
add_callback=self._pack_collection.signature_index.add_callback,
2274
deltas=False, parents=False, is_locked=self.is_locked),
2275
data_access=self._pack_collection.signature_index.data_access,
2277
self.texts = KnitVersionedFiles(
2278
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2279
add_callback=self._pack_collection.text_index.add_callback,
2280
deltas=True, parents=True, is_locked=self.is_locked),
2281
data_access=self._pack_collection.text_index.data_access,
2282
max_delta_chain=200)
2283
if _format.supports_chks:
2284
# No graph, no compression:- references from chks are between
2285
# different objects not temporal versions of the same; and without
2286
# some sort of temporal structure knit compression will just fail.
2287
self.chk_bytes = KnitVersionedFiles(
2288
_KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2289
add_callback=self._pack_collection.chk_index.add_callback,
2290
deltas=False, parents=False, is_locked=self.is_locked),
2291
data_access=self._pack_collection.chk_index.data_access,
2294
self.chk_bytes = None
2295
# True when the repository object is 'write locked' (as opposed to the
2296
# physical lock only taken out around changes to the pack-names list.)
2297
# Another way to represent this would be a decorator around the control
2298
# files object that presents logical locks as physical ones - if this
2299
# gets ugly consider that alternative design. RBC 20071011
2300
self._write_lock_count = 0
2301
self._transaction = None
2303
self._reconcile_does_inventory_gc = True
1690
MetaDirRepository.__init__(self, _format, a_bzrdir, control_files)
1691
self._commit_builder_class = _commit_builder_class
1692
self._serializer = _serializer
2304
1693
self._reconcile_fixes_text_parents = True
2305
self._reconcile_backsup_inventory = False
1694
if self._format.supports_external_lookups:
1695
self._unstacked_provider = graph.CachingParentsProvider(
1696
self._make_parents_provider_unstacked())
1698
self._unstacked_provider = graph.CachingParentsProvider(self)
1699
self._unstacked_provider.disable_cache()
2307
def _warn_if_deprecated(self, branch=None):
2308
# This class isn't deprecated, but one sub-format is
2309
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2310
super(KnitPackRepository, self)._warn_if_deprecated(branch)
1702
def _all_revision_ids(self):
1703
"""See Repository.all_revision_ids()."""
1704
return [key[0] for key in self.revisions.keys()]
2312
1706
def _abort_write_group(self):
2313
1707
self.revisions._index._key_dependencies.clear()
2314
1708
self._pack_collection._abort_write_group()
2316
def _get_source(self, to_format):
2317
if to_format.network_name() == self._format.network_name():
2318
return KnitPackStreamSource(self, to_format)
2319
return super(KnitPackRepository, self)._get_source(to_format)
2321
1710
def _make_parents_provider(self):
2322
return graph.CachingParentsProvider(self)
1711
if not self._format.supports_external_lookups:
1712
return self._unstacked_provider
1713
return graph.StackedParentsProvider(_LazyListJoin(
1714
[self._unstacked_provider], self._fallback_repositories))
2324
1716
def _refresh_data(self):
2325
1717
if not self.is_locked():
2327
1719
self._pack_collection.reload_pack_names()
1720
self._unstacked_provider.disable_cache()
1721
self._unstacked_provider.enable_cache()
2329
1723
def _start_write_group(self):
2330
1724
self._pack_collection._start_write_group()
2451
1851
self.control_files.unlock()
2453
1853
if not self.is_locked():
1854
self._unstacked_provider.disable_cache()
2454
1855
for repo in self._fallback_repositories:
2458
class KnitPackStreamSource(StreamSource):
2459
"""A StreamSource used to transfer data between same-format KnitPack repos.
2461
This source assumes:
2462
1) Same serialization format for all objects
2463
2) Same root information
2464
3) XML format inventories
2465
4) Atomic inserts (so we can stream inventory texts before text
2470
def __init__(self, from_repository, to_format):
2471
super(KnitPackStreamSource, self).__init__(from_repository, to_format)
2472
self._text_keys = None
2473
self._text_fetch_order = 'unordered'
2475
def _get_filtered_inv_stream(self, revision_ids):
2476
from_repo = self.from_repository
2477
parent_ids = from_repo._find_parent_ids_of_revisions(revision_ids)
2478
parent_keys = [(p,) for p in parent_ids]
2479
find_text_keys = from_repo._find_text_key_references_from_xml_inventory_lines
2480
parent_text_keys = set(find_text_keys(
2481
from_repo._inventory_xml_lines_for_keys(parent_keys)))
2482
content_text_keys = set()
2483
knit = KnitVersionedFiles(None, None)
2484
factory = KnitPlainFactory()
2485
def find_text_keys_from_content(record):
2486
if record.storage_kind not in ('knit-delta-gz', 'knit-ft-gz'):
2487
raise ValueError("Unknown content storage kind for"
2488
" inventory text: %s" % (record.storage_kind,))
2489
# It's a knit record, it has a _raw_record field (even if it was
2490
# reconstituted from a network stream).
2491
raw_data = record._raw_record
2492
# read the entire thing
2493
revision_id = record.key[-1]
2494
content, _ = knit._parse_record(revision_id, raw_data)
2495
if record.storage_kind == 'knit-delta-gz':
2496
line_iterator = factory.get_linedelta_content(content)
2497
elif record.storage_kind == 'knit-ft-gz':
2498
line_iterator = factory.get_fulltext_content(content)
2499
content_text_keys.update(find_text_keys(
2500
[(line, revision_id) for line in line_iterator]))
2501
revision_keys = [(r,) for r in revision_ids]
2502
def _filtered_inv_stream():
2503
source_vf = from_repo.inventories
2504
stream = source_vf.get_record_stream(revision_keys,
2506
for record in stream:
2507
if record.storage_kind == 'absent':
2508
raise errors.NoSuchRevision(from_repo, record.key)
2509
find_text_keys_from_content(record)
2511
self._text_keys = content_text_keys - parent_text_keys
2512
return ('inventories', _filtered_inv_stream())
2514
def _get_text_stream(self):
2515
# Note: We know we don't have to handle adding root keys, because both
2516
# the source and target are the identical network name.
2517
text_stream = self.from_repository.texts.get_record_stream(
2518
self._text_keys, self._text_fetch_order, False)
2519
return ('texts', text_stream)
2521
def get_stream(self, search):
2522
revision_ids = search.get_keys()
2523
for stream_info in self._fetch_revision_texts(revision_ids):
2525
self._revision_keys = [(rev_id,) for rev_id in revision_ids]
2526
yield self._get_filtered_inv_stream(revision_ids)
2527
yield self._get_text_stream()
2531
class RepositoryFormatPack(MetaDirRepositoryFormat):
1859
class RepositoryFormatPack(MetaDirVersionedFileRepositoryFormat):
2532
1860
"""Format logic for pack structured repositories.
2534
1862
This repository format has:
2606
1936
_serializer=self._serializer)
2609
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2610
"""A no-subtrees parameterized Pack repository.
2612
This format was introduced in 0.92.
2615
repository_class = KnitPackRepository
2616
_commit_builder_class = PackCommitBuilder
2618
def _serializer(self):
2619
return xml5.serializer_v5
2620
# What index classes to use
2621
index_builder_class = InMemoryGraphIndex
2622
index_class = GraphIndex
2624
def _get_matching_bzrdir(self):
2625
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2627
def _ignore_setting_bzrdir(self, format):
2630
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2632
def get_format_string(self):
2633
"""See RepositoryFormat.get_format_string()."""
2634
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2636
def get_format_description(self):
2637
"""See RepositoryFormat.get_format_description()."""
2638
return "Packs containing knits without subtree support"
2641
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2642
"""A subtrees parameterized Pack repository.
2644
This repository format uses the xml7 serializer to get:
2645
- support for recording full info about the tree root
2646
- support for recording tree-references
2648
This format was introduced in 0.92.
2651
repository_class = KnitPackRepository
2652
_commit_builder_class = PackRootCommitBuilder
2653
rich_root_data = True
2655
supports_tree_reference = True
2657
def _serializer(self):
2658
return xml7.serializer_v7
2659
# What index classes to use
2660
index_builder_class = InMemoryGraphIndex
2661
index_class = GraphIndex
2663
def _get_matching_bzrdir(self):
2664
return bzrdir.format_registry.make_bzrdir(
2665
'pack-0.92-subtree')
2667
def _ignore_setting_bzrdir(self, format):
2670
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2672
def get_format_string(self):
2673
"""See RepositoryFormat.get_format_string()."""
2674
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2676
def get_format_description(self):
2677
"""See RepositoryFormat.get_format_description()."""
2678
return "Packs containing knits with subtree support\n"
2681
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2682
"""A rich-root, no subtrees parameterized Pack repository.
2684
This repository format uses the xml6 serializer to get:
2685
- support for recording full info about the tree root
2687
This format was introduced in 1.0.
2690
repository_class = KnitPackRepository
2691
_commit_builder_class = PackRootCommitBuilder
2692
rich_root_data = True
2693
supports_tree_reference = False
2695
def _serializer(self):
2696
return xml6.serializer_v6
2697
# What index classes to use
2698
index_builder_class = InMemoryGraphIndex
2699
index_class = GraphIndex
2701
def _get_matching_bzrdir(self):
2702
return bzrdir.format_registry.make_bzrdir(
2705
def _ignore_setting_bzrdir(self, format):
2708
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2710
def get_format_string(self):
2711
"""See RepositoryFormat.get_format_string()."""
2712
return ("Bazaar pack repository format 1 with rich root"
2713
" (needs bzr 1.0)\n")
2715
def get_format_description(self):
2716
"""See RepositoryFormat.get_format_description()."""
2717
return "Packs containing knits with rich root support\n"
2720
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2721
"""Repository that supports external references to allow stacking.
2725
Supports external lookups, which results in non-truncated ghosts after
2726
reconcile compared to pack-0.92 formats.
2729
repository_class = KnitPackRepository
2730
_commit_builder_class = PackCommitBuilder
2731
supports_external_lookups = True
2732
# What index classes to use
2733
index_builder_class = InMemoryGraphIndex
2734
index_class = GraphIndex
2737
def _serializer(self):
2738
return xml5.serializer_v5
2740
def _get_matching_bzrdir(self):
2741
return bzrdir.format_registry.make_bzrdir('1.6')
2743
def _ignore_setting_bzrdir(self, format):
2746
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2748
def get_format_string(self):
2749
"""See RepositoryFormat.get_format_string()."""
2750
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2752
def get_format_description(self):
2753
"""See RepositoryFormat.get_format_description()."""
2754
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2757
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2758
"""A repository with rich roots and stacking.
2760
New in release 1.6.1.
2762
Supports stacking on other repositories, allowing data to be accessed
2763
without being stored locally.
2766
repository_class = KnitPackRepository
2767
_commit_builder_class = PackRootCommitBuilder
2768
rich_root_data = True
2769
supports_tree_reference = False # no subtrees
2770
supports_external_lookups = True
2771
# What index classes to use
2772
index_builder_class = InMemoryGraphIndex
2773
index_class = GraphIndex
2776
def _serializer(self):
2777
return xml6.serializer_v6
2779
def _get_matching_bzrdir(self):
2780
return bzrdir.format_registry.make_bzrdir(
2783
def _ignore_setting_bzrdir(self, format):
2786
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2788
def get_format_string(self):
2789
"""See RepositoryFormat.get_format_string()."""
2790
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2792
def get_format_description(self):
2793
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2796
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2797
"""A repository with rich roots and external references.
2801
Supports external lookups, which results in non-truncated ghosts after
2802
reconcile compared to pack-0.92 formats.
2804
This format was deprecated because the serializer it uses accidentally
2805
supported subtrees, when the format was not intended to. This meant that
2806
someone could accidentally fetch from an incorrect repository.
2809
repository_class = KnitPackRepository
2810
_commit_builder_class = PackRootCommitBuilder
2811
rich_root_data = True
2812
supports_tree_reference = False # no subtrees
2814
supports_external_lookups = True
2815
# What index classes to use
2816
index_builder_class = InMemoryGraphIndex
2817
index_class = GraphIndex
2820
def _serializer(self):
2821
return xml7.serializer_v7
2823
def _get_matching_bzrdir(self):
2824
matching = bzrdir.format_registry.make_bzrdir(
2826
matching.repository_format = self
2829
def _ignore_setting_bzrdir(self, format):
2832
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2834
def get_format_string(self):
2835
"""See RepositoryFormat.get_format_string()."""
2836
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2838
def get_format_description(self):
2839
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2843
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2844
"""A repository with stacking and btree indexes,
2845
without rich roots or subtrees.
2847
This is equivalent to pack-1.6 with B+Tree indices.
2850
repository_class = KnitPackRepository
2851
_commit_builder_class = PackCommitBuilder
2852
supports_external_lookups = True
2853
# What index classes to use
2854
index_builder_class = btree_index.BTreeBuilder
2855
index_class = btree_index.BTreeGraphIndex
2858
def _serializer(self):
2859
return xml5.serializer_v5
2861
def _get_matching_bzrdir(self):
2862
return bzrdir.format_registry.make_bzrdir('1.9')
2864
def _ignore_setting_bzrdir(self, format):
2867
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2869
def get_format_string(self):
2870
"""See RepositoryFormat.get_format_string()."""
2871
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2873
def get_format_description(self):
2874
"""See RepositoryFormat.get_format_description()."""
2875
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2878
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2879
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2881
1.6-rich-root with B+Tree indices.
2884
repository_class = KnitPackRepository
2885
_commit_builder_class = PackRootCommitBuilder
2886
rich_root_data = True
2887
supports_tree_reference = False # no subtrees
2888
supports_external_lookups = True
2889
# What index classes to use
2890
index_builder_class = btree_index.BTreeBuilder
2891
index_class = btree_index.BTreeGraphIndex
2894
def _serializer(self):
2895
return xml6.serializer_v6
2897
def _get_matching_bzrdir(self):
2898
return bzrdir.format_registry.make_bzrdir(
2901
def _ignore_setting_bzrdir(self, format):
2904
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2906
def get_format_string(self):
2907
"""See RepositoryFormat.get_format_string()."""
2908
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2910
def get_format_description(self):
2911
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2914
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2915
"""A subtrees development repository.
2917
This format should be retained in 2.3, to provide an upgrade path from this
2918
to RepositoryFormat2aSubtree. It can be removed in later releases.
2920
1.6.1-subtree[as it might have been] with B+Tree indices.
2923
repository_class = KnitPackRepository
2924
_commit_builder_class = PackRootCommitBuilder
2925
rich_root_data = True
2927
supports_tree_reference = True
2928
supports_external_lookups = True
2929
# What index classes to use
2930
index_builder_class = btree_index.BTreeBuilder
2931
index_class = btree_index.BTreeGraphIndex
2934
def _serializer(self):
2935
return xml7.serializer_v7
2937
def _get_matching_bzrdir(self):
2938
return bzrdir.format_registry.make_bzrdir(
2939
'development5-subtree')
2941
def _ignore_setting_bzrdir(self, format):
2944
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2946
def get_format_string(self):
2947
"""See RepositoryFormat.get_format_string()."""
2948
return ("Bazaar development format 2 with subtree support "
2949
"(needs bzr.dev from before 1.8)\n")
2951
def get_format_description(self):
2952
"""See RepositoryFormat.get_format_description()."""
2953
return ("Development repository format, currently the same as "
2954
"1.6.1-subtree with B+Tree indices.\n")
2957
1939
class RetryPackOperations(errors.RetryWithNewPacks):
2958
1940
"""Raised when we are packing and we find a missing file.
2967
1949
" context: %(context)s %(orig_error)s")
1952
class _DirectPackAccess(object):
1953
"""Access to data in one or more packs with less translation."""
1955
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
1956
"""Create a _DirectPackAccess object.
1958
:param index_to_packs: A dict mapping index objects to the transport
1959
and file names for obtaining data.
1960
:param reload_func: A function to call if we determine that the pack
1961
files have moved and we need to reload our caches. See
1962
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
1964
self._container_writer = None
1965
self._write_index = None
1966
self._indices = index_to_packs
1967
self._reload_func = reload_func
1968
self._flush_func = flush_func
1970
def add_raw_records(self, key_sizes, raw_data):
1971
"""Add raw knit bytes to a storage area.
1973
The data is spooled to the container writer in one bytes-record per
1976
:param sizes: An iterable of tuples containing the key and size of each
1978
:param raw_data: A bytestring containing the data.
1979
:return: A list of memos to retrieve the record later. Each memo is an
1980
opaque index memo. For _DirectPackAccess the memo is (index, pos,
1981
length), where the index field is the write_index object supplied
1982
to the PackAccess object.
1984
if type(raw_data) is not str:
1985
raise AssertionError(
1986
'data must be plain bytes was %s' % type(raw_data))
1989
for key, size in key_sizes:
1990
p_offset, p_length = self._container_writer.add_bytes_record(
1991
raw_data[offset:offset+size], [])
1993
result.append((self._write_index, p_offset, p_length))
1997
"""Flush pending writes on this access object.
1999
This will flush any buffered writes to a NewPack.
2001
if self._flush_func is not None:
2004
def get_raw_records(self, memos_for_retrieval):
2005
"""Get the raw bytes for a records.
2007
:param memos_for_retrieval: An iterable containing the (index, pos,
2008
length) memo for retrieving the bytes. The Pack access method
2009
looks up the pack to use for a given record in its index_to_pack
2011
:return: An iterator over the bytes of the records.
2013
# first pass, group into same-index requests
2015
current_index = None
2016
for (index, offset, length) in memos_for_retrieval:
2017
if current_index == index:
2018
current_list.append((offset, length))
2020
if current_index is not None:
2021
request_lists.append((current_index, current_list))
2022
current_index = index
2023
current_list = [(offset, length)]
2024
# handle the last entry
2025
if current_index is not None:
2026
request_lists.append((current_index, current_list))
2027
for index, offsets in request_lists:
2029
transport, path = self._indices[index]
2031
# A KeyError here indicates that someone has triggered an index
2032
# reload, and this index has gone missing, we need to start
2034
if self._reload_func is None:
2035
# If we don't have a _reload_func there is nothing that can
2038
raise errors.RetryWithNewPacks(index,
2039
reload_occurred=True,
2040
exc_info=sys.exc_info())
2042
reader = pack.make_readv_reader(transport, path, offsets)
2043
for names, read_func in reader.iter_records():
2044
yield read_func(None)
2045
except errors.NoSuchFile:
2046
# A NoSuchFile error indicates that a pack file has gone
2047
# missing on disk, we need to trigger a reload, and start over.
2048
if self._reload_func is None:
2050
raise errors.RetryWithNewPacks(transport.abspath(path),
2051
reload_occurred=False,
2052
exc_info=sys.exc_info())
2054
def set_writer(self, writer, index, transport_packname):
2055
"""Set a writer to use for adding data."""
2056
if index is not None:
2057
self._indices[index] = transport_packname
2058
self._container_writer = writer
2059
self._write_index = index
2061
def reload_or_raise(self, retry_exc):
2062
"""Try calling the reload function, or re-raise the original exception.
2064
This should be called after _DirectPackAccess raises a
2065
RetryWithNewPacks exception. This function will handle the common logic
2066
of determining when the error is fatal versus being temporary.
2067
It will also make sure that the original exception is raised, rather
2068
than the RetryWithNewPacks exception.
2070
If this function returns, then the calling function should retry
2071
whatever operation was being performed. Otherwise an exception will
2074
:param retry_exc: A RetryWithNewPacks exception.
2077
if self._reload_func is None:
2079
elif not self._reload_func():
2080
# The reload claimed that nothing changed
2081
if not retry_exc.reload_occurred:
2082
# If there wasn't an earlier reload, then we really were
2083
# expecting to find changes. We didn't find them, so this is a
2087
exc_class, exc_value, exc_traceback = retry_exc.exc_info
2088
raise exc_class, exc_value, exc_traceback
added
removed
bzrlib/repofmt/pack_repo.py
