727
844
Sets self._text_filter appropriately.
729
raise NotImplementedError(self._copy_inventory_texts)
846
# select inventory keys
847
inv_keys = self._revision_keys # currently the same keyspace, and note that
848
# querying for keys here could introduce a bug where an inventory item
849
# is missed, so do not change it to query separately without cross
850
# checking like the text key check below.
851
inventory_index_map, inventory_indices = self._pack_map_and_index_list(
853
inv_nodes = self._index_contents(inventory_indices, inv_keys)
854
# copy inventory keys and adjust values
855
# XXX: Should be a helper function to allow different inv representation
857
self.pb.update("Copying inventory texts", 2)
858
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
859
# Only grab the output lines if we will be processing them
860
output_lines = bool(self.revision_ids)
861
inv_lines = self._copy_nodes_graph(inventory_index_map,
862
self.new_pack._writer, self.new_pack.inventory_index,
863
readv_group_iter, total_items, output_lines=output_lines)
864
if self.revision_ids:
865
self._process_inventory_lines(inv_lines)
867
# eat the iterator to cause it to execute.
869
self._text_filter = None
870
if 'pack' in debug.debug_flags:
871
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
872
time.ctime(), self._pack_collection._upload_transport.base,
873
self.new_pack.random_name,
874
self.new_pack.inventory_index.key_count(),
875
time.time() - self.new_pack.start_time)
731
877
def _copy_text_texts(self):
732
raise NotImplementedError(self._copy_text_texts)
879
text_index_map, text_nodes = self._get_text_nodes()
880
if self._text_filter is not None:
881
# We could return the keys copied as part of the return value from
882
# _copy_nodes_graph but this doesn't work all that well with the
883
# need to get line output too, so we check separately, and as we're
884
# going to buffer everything anyway, we check beforehand, which
885
# saves reading knit data over the wire when we know there are
887
text_nodes = set(text_nodes)
888
present_text_keys = set(_node[1] for _node in text_nodes)
889
missing_text_keys = set(self._text_filter) - present_text_keys
890
if missing_text_keys:
891
# TODO: raise a specific error that can handle many missing
893
mutter("missing keys during fetch: %r", missing_text_keys)
894
a_missing_key = missing_text_keys.pop()
895
raise errors.RevisionNotPresent(a_missing_key[1],
897
# copy text keys and adjust values
898
self.pb.update("Copying content texts", 3)
899
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
900
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
901
self.new_pack.text_index, readv_group_iter, total_items))
902
self._log_copied_texts()
734
904
def _create_pack_from_packs(self):
735
raise NotImplementedError(self._create_pack_from_packs)
905
self.pb.update("Opening pack", 0, 5)
906
self.new_pack = self.open_pack()
907
new_pack = self.new_pack
908
# buffer data - we won't be reading-back during the pack creation and
909
# this makes a significant difference on sftp pushes.
910
new_pack.set_write_cache_size(1024*1024)
911
if 'pack' in debug.debug_flags:
912
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
913
for a_pack in self.packs]
914
if self.revision_ids is not None:
915
rev_count = len(self.revision_ids)
918
mutter('%s: create_pack: creating pack from source packs: '
919
'%s%s %s revisions wanted %s t=0',
920
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
921
plain_pack_list, rev_count)
922
self._copy_revision_texts()
923
self._copy_inventory_texts()
924
self._copy_text_texts()
925
# select signature keys
926
signature_filter = self._revision_keys # same keyspace
927
signature_index_map, signature_indices = self._pack_map_and_index_list(
929
signature_nodes = self._index_contents(signature_indices,
931
# copy signature keys and adjust values
932
self.pb.update("Copying signature texts", 4)
933
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
934
new_pack.signature_index)
935
if 'pack' in debug.debug_flags:
936
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
937
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
938
new_pack.signature_index.key_count(),
939
time.time() - new_pack.start_time)
941
# NB XXX: how to check CHK references are present? perhaps by yielding
942
# the items? How should that interact with stacked repos?
943
if new_pack.chk_index is not None:
945
if 'pack' in debug.debug_flags:
946
mutter('%s: create_pack: chk content copied: %s%s %d items t+%6.3fs',
947
time.ctime(), self._pack_collection._upload_transport.base,
948
new_pack.random_name,
949
new_pack.chk_index.key_count(),
950
time.time() - new_pack.start_time)
951
new_pack._check_references()
952
if not self._use_pack(new_pack):
955
self.pb.update("Finishing pack", 5)
957
self._pack_collection.allocate(new_pack)
960
def _copy_chks(self, refs=None):
961
# XXX: Todo, recursive follow-pointers facility when fetching some
963
chk_index_map, chk_indices = self._pack_map_and_index_list(
965
chk_nodes = self._index_contents(chk_indices, refs)
967
# TODO: This isn't strictly tasteful as we are accessing some private
968
# variables (_serializer). Perhaps a better way would be to have
969
# Repository._deserialise_chk_node()
970
search_key_func = chk_map.search_key_registry.get(
971
self._pack_collection.repo._serializer.search_key_name)
972
def accumlate_refs(lines):
973
# XXX: move to a generic location
975
bytes = ''.join(lines)
976
node = chk_map._deserialise(bytes, ("unknown",), search_key_func)
977
new_refs.update(node.refs())
978
self._copy_nodes(chk_nodes, chk_index_map, self.new_pack._writer,
979
self.new_pack.chk_index, output_lines=accumlate_refs)
982
def _copy_nodes(self, nodes, index_map, writer, write_index,
984
"""Copy knit nodes between packs with no graph references.
986
:param output_lines: Output full texts of copied items.
988
pb = ui.ui_factory.nested_progress_bar()
990
return self._do_copy_nodes(nodes, index_map, writer,
991
write_index, pb, output_lines=output_lines)
995
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb,
997
# for record verification
998
knit = KnitVersionedFiles(None, None)
999
# plan a readv on each source pack:
1001
nodes = sorted(nodes)
1002
# how to map this into knit.py - or knit.py into this?
1003
# we don't want the typical knit logic, we want grouping by pack
1004
# at this point - perhaps a helper library for the following code
1005
# duplication points?
1007
for index, key, value in nodes:
1008
if index not in request_groups:
1009
request_groups[index] = []
1010
request_groups[index].append((key, value))
1012
pb.update("Copied record", record_index, len(nodes))
1013
for index, items in request_groups.iteritems():
1014
pack_readv_requests = []
1015
for key, value in items:
1016
# ---- KnitGraphIndex.get_position
1017
bits = value[1:].split(' ')
1018
offset, length = int(bits[0]), int(bits[1])
1019
pack_readv_requests.append((offset, length, (key, value[0])))
1020
# linear scan up the pack
1021
pack_readv_requests.sort()
1023
pack_obj = index_map[index]
1024
transport, path = pack_obj.access_tuple()
1026
reader = pack.make_readv_reader(transport, path,
1027
[offset[0:2] for offset in pack_readv_requests])
1028
except errors.NoSuchFile:
1029
if self._reload_func is not None:
1032
for (names, read_func), (_1, _2, (key, eol_flag)) in \
1033
izip(reader.iter_records(), pack_readv_requests):
1034
raw_data = read_func(None)
1035
# check the header only
1036
if output_lines is not None:
1037
output_lines(knit._parse_record(key[-1], raw_data)[0])
1039
df, _ = knit._parse_record_header(key, raw_data)
1041
pos, size = writer.add_bytes_record(raw_data, names)
1042
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
1043
pb.update("Copied record", record_index)
1046
def _copy_nodes_graph(self, index_map, writer, write_index,
1047
readv_group_iter, total_items, output_lines=False):
1048
"""Copy knit nodes between packs.
1050
:param output_lines: Return lines present in the copied data as
1051
an iterator of line,version_id.
1053
pb = ui.ui_factory.nested_progress_bar()
1055
for result in self._do_copy_nodes_graph(index_map, writer,
1056
write_index, output_lines, pb, readv_group_iter, total_items):
1059
# Python 2.4 does not permit try:finally: in a generator.
1065
def _do_copy_nodes_graph(self, index_map, writer, write_index,
1066
output_lines, pb, readv_group_iter, total_items):
1067
# for record verification
1068
knit = KnitVersionedFiles(None, None)
1069
# for line extraction when requested (inventories only)
1071
factory = KnitPlainFactory()
1073
pb.update("Copied record", record_index, total_items)
1074
for index, readv_vector, node_vector in readv_group_iter:
1076
pack_obj = index_map[index]
1077
transport, path = pack_obj.access_tuple()
1079
reader = pack.make_readv_reader(transport, path, readv_vector)
1080
except errors.NoSuchFile:
1081
if self._reload_func is not None:
1084
for (names, read_func), (key, eol_flag, references) in \
1085
izip(reader.iter_records(), node_vector):
1086
raw_data = read_func(None)
1088
# read the entire thing
1089
content, _ = knit._parse_record(key[-1], raw_data)
1090
if len(references[-1]) == 0:
1091
line_iterator = factory.get_fulltext_content(content)
1093
line_iterator = factory.get_linedelta_content(content)
1094
for line in line_iterator:
1097
# check the header only
1098
df, _ = knit._parse_record_header(key, raw_data)
1100
pos, size = writer.add_bytes_record(raw_data, names)
1101
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1102
pb.update("Copied record", record_index)
1105
def _get_text_nodes(self):
1106
text_index_map, text_indices = self._pack_map_and_index_list(
1108
return text_index_map, self._index_contents(text_indices,
1111
def _least_readv_node_readv(self, nodes):
1112
"""Generate request groups for nodes using the least readv's.
1114
:param nodes: An iterable of graph index nodes.
1115
:return: Total node count and an iterator of the data needed to perform
1116
readvs to obtain the data for nodes. Each item yielded by the
1117
iterator is a tuple with:
1118
index, readv_vector, node_vector. readv_vector is a list ready to
1119
hand to the transport readv method, and node_vector is a list of
1120
(key, eol_flag, references) for the node retrieved by the
1121
matching readv_vector.
1123
# group by pack so we do one readv per pack
1124
nodes = sorted(nodes)
1127
for index, key, value, references in nodes:
1128
if index not in request_groups:
1129
request_groups[index] = []
1130
request_groups[index].append((key, value, references))
1132
for index, items in request_groups.iteritems():
1133
pack_readv_requests = []
1134
for key, value, references in items:
1135
# ---- KnitGraphIndex.get_position
1136
bits = value[1:].split(' ')
1137
offset, length = int(bits[0]), int(bits[1])
1138
pack_readv_requests.append(
1139
((offset, length), (key, value[0], references)))
1140
# linear scan up the pack to maximum range combining.
1141
pack_readv_requests.sort()
1142
# split out the readv and the node data.
1143
pack_readv = [readv for readv, node in pack_readv_requests]
1144
node_vector = [node for readv, node in pack_readv_requests]
1145
result.append((index, pack_readv, node_vector))
1146
return total, result
737
1148
def _log_copied_texts(self):
738
1149
if 'pack' in debug.debug_flags:
751
1181
return new_pack.data_inserted()
1184
class OptimisingPacker(Packer):
1185
"""A packer which spends more time to create better disk layouts."""
1187
def _revision_node_readv(self, revision_nodes):
1188
"""Return the total revisions and the readv's to issue.
1190
This sort places revisions in topological order with the ancestors
1193
:param revision_nodes: The revision index contents for the packs being
1194
incorporated into the new pack.
1195
:return: As per _least_readv_node_readv.
1197
# build an ancestors dict
1200
for index, key, value, references in revision_nodes:
1201
ancestors[key] = references[0]
1202
by_key[key] = (index, value, references)
1203
order = tsort.topo_sort(ancestors)
1205
# Single IO is pathological, but it will work as a starting point.
1207
for key in reversed(order):
1208
index, value, references = by_key[key]
1209
# ---- KnitGraphIndex.get_position
1210
bits = value[1:].split(' ')
1211
offset, length = int(bits[0]), int(bits[1])
1213
(index, [(offset, length)], [(key, value[0], references)]))
1214
# TODO: combine requests in the same index that are in ascending order.
1215
return total, requests
1217
def open_pack(self):
1218
"""Open a pack for the pack we are creating."""
1219
new_pack = super(OptimisingPacker, self).open_pack()
1220
# Turn on the optimization flags for all the index builders.
1221
new_pack.revision_index.set_optimize(for_size=True)
1222
new_pack.inventory_index.set_optimize(for_size=True)
1223
new_pack.text_index.set_optimize(for_size=True)
1224
new_pack.signature_index.set_optimize(for_size=True)
1228
class ReconcilePacker(Packer):
1229
"""A packer which regenerates indices etc as it copies.
1231
This is used by ``bzr reconcile`` to cause parent text pointers to be
1235
def _extra_init(self):
1236
self._data_changed = False
1238
def _process_inventory_lines(self, inv_lines):
1239
"""Generate a text key reference map rather for reconciling with."""
1240
repo = self._pack_collection.repo
1241
refs = repo._find_text_key_references_from_xml_inventory_lines(
1243
self._text_refs = refs
1244
# during reconcile we:
1245
# - convert unreferenced texts to full texts
1246
# - correct texts which reference a text not copied to be full texts
1247
# - copy all others as-is but with corrected parents.
1248
# - so at this point we don't know enough to decide what becomes a full
1250
self._text_filter = None
1252
def _copy_text_texts(self):
1253
"""generate what texts we should have and then copy."""
1254
self.pb.update("Copying content texts", 3)
1255
# we have three major tasks here:
1256
# 1) generate the ideal index
1257
repo = self._pack_collection.repo
1258
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1259
_1, key, _2, refs in
1260
self.new_pack.revision_index.iter_all_entries()])
1261
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1262
# 2) generate a text_nodes list that contains all the deltas that can
1263
# be used as-is, with corrected parents.
1266
discarded_nodes = []
1267
NULL_REVISION = _mod_revision.NULL_REVISION
1268
text_index_map, text_nodes = self._get_text_nodes()
1269
for node in text_nodes:
1275
ideal_parents = tuple(ideal_index[node[1]])
1277
discarded_nodes.append(node)
1278
self._data_changed = True
1280
if ideal_parents == (NULL_REVISION,):
1282
if ideal_parents == node[3][0]:
1284
ok_nodes.append(node)
1285
elif ideal_parents[0:1] == node[3][0][0:1]:
1286
# the left most parent is the same, or there are no parents
1287
# today. Either way, we can preserve the representation as
1288
# long as we change the refs to be inserted.
1289
self._data_changed = True
1290
ok_nodes.append((node[0], node[1], node[2],
1291
(ideal_parents, node[3][1])))
1292
self._data_changed = True
1294
# Reinsert this text completely
1295
bad_texts.append((node[1], ideal_parents))
1296
self._data_changed = True
1297
# we're finished with some data.
1300
# 3) bulk copy the ok data
1301
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1302
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1303
self.new_pack.text_index, readv_group_iter, total_items))
1304
# 4) adhoc copy all the other texts.
1305
# We have to topologically insert all texts otherwise we can fail to
1306
# reconcile when parts of a single delta chain are preserved intact,
1307
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1308
# reinserted, and if d3 has incorrect parents it will also be
1309
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1310
# copied), so we will try to delta, but d2 is not currently able to be
1311
# extracted because it's basis d1 is not present. Topologically sorting
1312
# addresses this. The following generates a sort for all the texts that
1313
# are being inserted without having to reference the entire text key
1314
# space (we only topo sort the revisions, which is smaller).
1315
topo_order = tsort.topo_sort(ancestors)
1316
rev_order = dict(zip(topo_order, range(len(topo_order))))
1317
bad_texts.sort(key=lambda key:rev_order.get(key[0][1], 0))
1318
transaction = repo.get_transaction()
1319
file_id_index = GraphIndexPrefixAdapter(
1320
self.new_pack.text_index,
1322
add_nodes_callback=self.new_pack.text_index.add_nodes)
1323
data_access = _DirectPackAccess(
1324
{self.new_pack.text_index:self.new_pack.access_tuple()})
1325
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1326
self.new_pack.access_tuple())
1327
output_texts = KnitVersionedFiles(
1328
_KnitGraphIndex(self.new_pack.text_index,
1329
add_callback=self.new_pack.text_index.add_nodes,
1330
deltas=True, parents=True, is_locked=repo.is_locked),
1331
data_access=data_access, max_delta_chain=200)
1332
for key, parent_keys in bad_texts:
1333
# We refer to the new pack to delta data being output.
1334
# A possible improvement would be to catch errors on short reads
1335
# and only flush then.
1336
self.new_pack.flush()
1338
for parent_key in parent_keys:
1339
if parent_key[0] != key[0]:
1340
# Graph parents must match the fileid
1341
raise errors.BzrError('Mismatched key parent %r:%r' %
1343
parents.append(parent_key[1])
1344
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1345
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1346
output_texts.add_lines(key, parent_keys, text_lines,
1347
random_id=True, check_content=False)
1348
# 5) check that nothing inserted has a reference outside the keyspace.
1349
missing_text_keys = self.new_pack.text_index._external_references()
1350
if missing_text_keys:
1351
raise errors.BzrCheckError('Reference to missing compression parents %r'
1352
% (missing_text_keys,))
1353
self._log_copied_texts()
1355
def _use_pack(self, new_pack):
1356
"""Override _use_pack to check for reconcile having changed content."""
1357
# XXX: we might be better checking this at the copy time.
1358
original_inventory_keys = set()
1359
inv_index = self._pack_collection.inventory_index.combined_index
1360
for entry in inv_index.iter_all_entries():
1361
original_inventory_keys.add(entry[1])
1362
new_inventory_keys = set()
1363
for entry in new_pack.inventory_index.iter_all_entries():
1364
new_inventory_keys.add(entry[1])
1365
if new_inventory_keys != original_inventory_keys:
1366
self._data_changed = True
1367
return new_pack.data_inserted() and self._data_changed
754
1370
class RepositoryPackCollection(object):
755
1371
"""Management of packs within a repository.
757
1373
:ivar _names: map of {pack_name: (index_size,)}
761
resumed_pack_factory = None
762
normal_packer_class = None
763
optimising_packer_class = None
1376
pack_factory = NewPack
1377
resumed_pack_factory = ResumedPack
765
1379
def __init__(self, repo, transport, index_transport, upload_transport,
766
1380
pack_transport, index_builder_class, index_class,
1650
# These attributes are inherited from the Repository base class. Setting
1651
# them to None ensures that if the constructor is changed to not initialize
1652
# them, or a subclass fails to call the constructor, that an error will
1653
# occur rather than the system working but generating incorrect data.
1654
_commit_builder_class = None
1657
2226
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
1659
MetaDirRepository.__init__(self, _format, a_bzrdir, control_files)
1660
self._commit_builder_class = _commit_builder_class
1661
self._serializer = _serializer
2228
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2229
_commit_builder_class, _serializer)
2230
index_transport = self._transport.clone('indices')
2231
self._pack_collection = RepositoryPackCollection(self, self._transport,
2233
self._transport.clone('upload'),
2234
self._transport.clone('packs'),
2235
_format.index_builder_class,
2236
_format.index_class,
2237
use_chk_index=self._format.supports_chks,
2239
self.inventories = KnitVersionedFiles(
2240
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2241
add_callback=self._pack_collection.inventory_index.add_callback,
2242
deltas=True, parents=True, is_locked=self.is_locked),
2243
data_access=self._pack_collection.inventory_index.data_access,
2244
max_delta_chain=200)
2245
self.revisions = KnitVersionedFiles(
2246
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2247
add_callback=self._pack_collection.revision_index.add_callback,
2248
deltas=False, parents=True, is_locked=self.is_locked,
2249
track_external_parent_refs=True),
2250
data_access=self._pack_collection.revision_index.data_access,
2252
self.signatures = KnitVersionedFiles(
2253
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2254
add_callback=self._pack_collection.signature_index.add_callback,
2255
deltas=False, parents=False, is_locked=self.is_locked),
2256
data_access=self._pack_collection.signature_index.data_access,
2258
self.texts = KnitVersionedFiles(
2259
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2260
add_callback=self._pack_collection.text_index.add_callback,
2261
deltas=True, parents=True, is_locked=self.is_locked),
2262
data_access=self._pack_collection.text_index.data_access,
2263
max_delta_chain=200)
2264
if _format.supports_chks:
2265
# No graph, no compression:- references from chks are between
2266
# different objects not temporal versions of the same; and without
2267
# some sort of temporal structure knit compression will just fail.
2268
self.chk_bytes = KnitVersionedFiles(
2269
_KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2270
add_callback=self._pack_collection.chk_index.add_callback,
2271
deltas=False, parents=False, is_locked=self.is_locked),
2272
data_access=self._pack_collection.chk_index.data_access,
2275
self.chk_bytes = None
2276
# True when the repository object is 'write locked' (as opposed to the
2277
# physical lock only taken out around changes to the pack-names list.)
2278
# Another way to represent this would be a decorator around the control
2279
# files object that presents logical locks as physical ones - if this
2280
# gets ugly consider that alternative design. RBC 20071011
2281
self._write_lock_count = 0
2282
self._transaction = None
2284
self._reconcile_does_inventory_gc = True
1662
2285
self._reconcile_fixes_text_parents = True
2286
self._reconcile_backsup_inventory = False
1665
def _all_revision_ids(self):
1666
"""See Repository.all_revision_ids()."""
1667
return [key[0] for key in self.revisions.keys()]
2288
def _warn_if_deprecated(self, branch=None):
2289
# This class isn't deprecated, but one sub-format is
2290
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2291
super(KnitPackRepository, self)._warn_if_deprecated(branch)
1669
2293
def _abort_write_group(self):
1670
2294
self.revisions._index._key_dependencies.clear()
1671
2295
self._pack_collection._abort_write_group()
2297
def _get_source(self, to_format):
2298
if to_format.network_name() == self._format.network_name():
2299
return KnitPackStreamSource(self, to_format)
2300
return super(KnitPackRepository, self)._get_source(to_format)
1673
2302
def _make_parents_provider(self):
1674
2303
return graph.CachingParentsProvider(self)
2429
class KnitPackStreamSource(StreamSource):
2430
"""A StreamSource used to transfer data between same-format KnitPack repos.
2432
This source assumes:
2433
1) Same serialization format for all objects
2434
2) Same root information
2435
3) XML format inventories
2436
4) Atomic inserts (so we can stream inventory texts before text
2441
def __init__(self, from_repository, to_format):
2442
super(KnitPackStreamSource, self).__init__(from_repository, to_format)
2443
self._text_keys = None
2444
self._text_fetch_order = 'unordered'
2446
def _get_filtered_inv_stream(self, revision_ids):
2447
from_repo = self.from_repository
2448
parent_ids = from_repo._find_parent_ids_of_revisions(revision_ids)
2449
parent_keys = [(p,) for p in parent_ids]
2450
find_text_keys = from_repo._find_text_key_references_from_xml_inventory_lines
2451
parent_text_keys = set(find_text_keys(
2452
from_repo._inventory_xml_lines_for_keys(parent_keys)))
2453
content_text_keys = set()
2454
knit = KnitVersionedFiles(None, None)
2455
factory = KnitPlainFactory()
2456
def find_text_keys_from_content(record):
2457
if record.storage_kind not in ('knit-delta-gz', 'knit-ft-gz'):
2458
raise ValueError("Unknown content storage kind for"
2459
" inventory text: %s" % (record.storage_kind,))
2460
# It's a knit record, it has a _raw_record field (even if it was
2461
# reconstituted from a network stream).
2462
raw_data = record._raw_record
2463
# read the entire thing
2464
revision_id = record.key[-1]
2465
content, _ = knit._parse_record(revision_id, raw_data)
2466
if record.storage_kind == 'knit-delta-gz':
2467
line_iterator = factory.get_linedelta_content(content)
2468
elif record.storage_kind == 'knit-ft-gz':
2469
line_iterator = factory.get_fulltext_content(content)
2470
content_text_keys.update(find_text_keys(
2471
[(line, revision_id) for line in line_iterator]))
2472
revision_keys = [(r,) for r in revision_ids]
2473
def _filtered_inv_stream():
2474
source_vf = from_repo.inventories
2475
stream = source_vf.get_record_stream(revision_keys,
2477
for record in stream:
2478
if record.storage_kind == 'absent':
2479
raise errors.NoSuchRevision(from_repo, record.key)
2480
find_text_keys_from_content(record)
2482
self._text_keys = content_text_keys - parent_text_keys
2483
return ('inventories', _filtered_inv_stream())
2485
def _get_text_stream(self):
2486
# Note: We know we don't have to handle adding root keys, because both
2487
# the source and target are the identical network name.
2488
text_stream = self.from_repository.texts.get_record_stream(
2489
self._text_keys, self._text_fetch_order, False)
2490
return ('texts', text_stream)
2492
def get_stream(self, search):
2493
revision_ids = search.get_keys()
2494
for stream_info in self._fetch_revision_texts(revision_ids):
2496
self._revision_keys = [(rev_id,) for rev_id in revision_ids]
2497
yield self._get_filtered_inv_stream(revision_ids)
2498
yield self._get_text_stream()
1809
2502
class RepositoryFormatPack(MetaDirRepositoryFormat):
1810
2503
"""Format logic for pack structured repositories.
1887
2575
_serializer=self._serializer)
1890
class RetryPackOperations(errors.RetryWithNewPacks):
1891
"""Raised when we are packing and we find a missing file.
1893
Meant as a signaling exception, to tell the RepositoryPackCollection.pack
1894
code it should try again.
1897
internal_error = True
1899
_fmt = ("Pack files have changed, reload and try pack again."
1900
" context: %(context)s %(orig_error)s")
1903
class _DirectPackAccess(object):
1904
"""Access to data in one or more packs with less translation."""
1906
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
1907
"""Create a _DirectPackAccess object.
1909
:param index_to_packs: A dict mapping index objects to the transport
1910
and file names for obtaining data.
1911
:param reload_func: A function to call if we determine that the pack
1912
files have moved and we need to reload our caches. See
1913
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
1915
self._container_writer = None
1916
self._write_index = None
1917
self._indices = index_to_packs
1918
self._reload_func = reload_func
1919
self._flush_func = flush_func
1921
def add_raw_records(self, key_sizes, raw_data):
1922
"""Add raw knit bytes to a storage area.
1924
The data is spooled to the container writer in one bytes-record per
1927
:param sizes: An iterable of tuples containing the key and size of each
1929
:param raw_data: A bytestring containing the data.
1930
:return: A list of memos to retrieve the record later. Each memo is an
1931
opaque index memo. For _DirectPackAccess the memo is (index, pos,
1932
length), where the index field is the write_index object supplied
1933
to the PackAccess object.
1935
if type(raw_data) is not str:
1936
raise AssertionError(
1937
'data must be plain bytes was %s' % type(raw_data))
1940
for key, size in key_sizes:
1941
p_offset, p_length = self._container_writer.add_bytes_record(
1942
raw_data[offset:offset+size], [])
1944
result.append((self._write_index, p_offset, p_length))
1948
"""Flush pending writes on this access object.
1950
This will flush any buffered writes to a NewPack.
1952
if self._flush_func is not None:
1955
def get_raw_records(self, memos_for_retrieval):
1956
"""Get the raw bytes for a records.
1958
:param memos_for_retrieval: An iterable containing the (index, pos,
1959
length) memo for retrieving the bytes. The Pack access method
1960
looks up the pack to use for a given record in its index_to_pack
1962
:return: An iterator over the bytes of the records.
1964
# first pass, group into same-index requests
1966
current_index = None
1967
for (index, offset, length) in memos_for_retrieval:
1968
if current_index == index:
1969
current_list.append((offset, length))
1971
if current_index is not None:
1972
request_lists.append((current_index, current_list))
1973
current_index = index
1974
current_list = [(offset, length)]
1975
# handle the last entry
1976
if current_index is not None:
1977
request_lists.append((current_index, current_list))
1978
for index, offsets in request_lists:
1980
transport, path = self._indices[index]
1982
# A KeyError here indicates that someone has triggered an index
1983
# reload, and this index has gone missing, we need to start
1985
if self._reload_func is None:
1986
# If we don't have a _reload_func there is nothing that can
1989
raise errors.RetryWithNewPacks(index,
1990
reload_occurred=True,
1991
exc_info=sys.exc_info())
1993
reader = pack.make_readv_reader(transport, path, offsets)
1994
for names, read_func in reader.iter_records():
1995
yield read_func(None)
1996
except errors.NoSuchFile:
1997
# A NoSuchFile error indicates that a pack file has gone
1998
# missing on disk, we need to trigger a reload, and start over.
1999
if self._reload_func is None:
2001
raise errors.RetryWithNewPacks(transport.abspath(path),
2002
reload_occurred=False,
2003
exc_info=sys.exc_info())
2005
def set_writer(self, writer, index, transport_packname):
2006
"""Set a writer to use for adding data."""
2007
if index is not None:
2008
self._indices[index] = transport_packname
2009
self._container_writer = writer
2010
self._write_index = index
2012
def reload_or_raise(self, retry_exc):
2013
"""Try calling the reload function, or re-raise the original exception.
2015
This should be called after _DirectPackAccess raises a
2016
RetryWithNewPacks exception. This function will handle the common logic
2017
of determining when the error is fatal versus being temporary.
2018
It will also make sure that the original exception is raised, rather
2019
than the RetryWithNewPacks exception.
2021
If this function returns, then the calling function should retry
2022
whatever operation was being performed. Otherwise an exception will
2025
:param retry_exc: A RetryWithNewPacks exception.
2028
if self._reload_func is None:
2030
elif not self._reload_func():
2031
# The reload claimed that nothing changed
2032
if not retry_exc.reload_occurred:
2033
# If there wasn't an earlier reload, then we really were
2034
# expecting to find changes. We didn't find them, so this is a
2038
exc_class, exc_value, exc_traceback = retry_exc.exc_info
2039
raise exc_class, exc_value, exc_traceback
2578
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2579
"""A no-subtrees parameterized Pack repository.
2581
This format was introduced in 0.92.
2584
repository_class = KnitPackRepository
2585
_commit_builder_class = PackCommitBuilder
2587
def _serializer(self):
2588
return xml5.serializer_v5
2589
# What index classes to use
2590
index_builder_class = InMemoryGraphIndex
2591
index_class = GraphIndex
2593
def _get_matching_bzrdir(self):
2594
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2596
def _ignore_setting_bzrdir(self, format):
2599
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2601
def get_format_string(self):
2602
"""See RepositoryFormat.get_format_string()."""
2603
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2605
def get_format_description(self):
2606
"""See RepositoryFormat.get_format_description()."""
2607
return "Packs containing knits without subtree support"
2610
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2611
"""A subtrees parameterized Pack repository.
2613
This repository format uses the xml7 serializer to get:
2614
- support for recording full info about the tree root
2615
- support for recording tree-references
2617
This format was introduced in 0.92.
2620
repository_class = KnitPackRepository
2621
_commit_builder_class = PackRootCommitBuilder
2622
rich_root_data = True
2624
supports_tree_reference = True
2626
def _serializer(self):
2627
return xml7.serializer_v7
2628
# What index classes to use
2629
index_builder_class = InMemoryGraphIndex
2630
index_class = GraphIndex
2632
def _get_matching_bzrdir(self):
2633
return bzrdir.format_registry.make_bzrdir(
2634
'pack-0.92-subtree')
2636
def _ignore_setting_bzrdir(self, format):
2639
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2641
def get_format_string(self):
2642
"""See RepositoryFormat.get_format_string()."""
2643
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2645
def get_format_description(self):
2646
"""See RepositoryFormat.get_format_description()."""
2647
return "Packs containing knits with subtree support\n"
2650
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2651
"""A rich-root, no subtrees parameterized Pack repository.
2653
This repository format uses the xml6 serializer to get:
2654
- support for recording full info about the tree root
2656
This format was introduced in 1.0.
2659
repository_class = KnitPackRepository
2660
_commit_builder_class = PackRootCommitBuilder
2661
rich_root_data = True
2662
supports_tree_reference = False
2664
def _serializer(self):
2665
return xml6.serializer_v6
2666
# What index classes to use
2667
index_builder_class = InMemoryGraphIndex
2668
index_class = GraphIndex
2670
def _get_matching_bzrdir(self):
2671
return bzrdir.format_registry.make_bzrdir(
2674
def _ignore_setting_bzrdir(self, format):
2677
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2679
def get_format_string(self):
2680
"""See RepositoryFormat.get_format_string()."""
2681
return ("Bazaar pack repository format 1 with rich root"
2682
" (needs bzr 1.0)\n")
2684
def get_format_description(self):
2685
"""See RepositoryFormat.get_format_description()."""
2686
return "Packs containing knits with rich root support\n"
2689
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2690
"""Repository that supports external references to allow stacking.
2694
Supports external lookups, which results in non-truncated ghosts after
2695
reconcile compared to pack-0.92 formats.
2698
repository_class = KnitPackRepository
2699
_commit_builder_class = PackCommitBuilder
2700
supports_external_lookups = True
2701
# What index classes to use
2702
index_builder_class = InMemoryGraphIndex
2703
index_class = GraphIndex
2706
def _serializer(self):
2707
return xml5.serializer_v5
2709
def _get_matching_bzrdir(self):
2710
return bzrdir.format_registry.make_bzrdir('1.6')
2712
def _ignore_setting_bzrdir(self, format):
2715
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2717
def get_format_string(self):
2718
"""See RepositoryFormat.get_format_string()."""
2719
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2721
def get_format_description(self):
2722
"""See RepositoryFormat.get_format_description()."""
2723
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2726
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2727
"""A repository with rich roots and stacking.
2729
New in release 1.6.1.
2731
Supports stacking on other repositories, allowing data to be accessed
2732
without being stored locally.
2735
repository_class = KnitPackRepository
2736
_commit_builder_class = PackRootCommitBuilder
2737
rich_root_data = True
2738
supports_tree_reference = False # no subtrees
2739
supports_external_lookups = True
2740
# What index classes to use
2741
index_builder_class = InMemoryGraphIndex
2742
index_class = GraphIndex
2745
def _serializer(self):
2746
return xml6.serializer_v6
2748
def _get_matching_bzrdir(self):
2749
return bzrdir.format_registry.make_bzrdir(
2752
def _ignore_setting_bzrdir(self, format):
2755
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2757
def get_format_string(self):
2758
"""See RepositoryFormat.get_format_string()."""
2759
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2761
def get_format_description(self):
2762
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2765
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2766
"""A repository with rich roots and external references.
2770
Supports external lookups, which results in non-truncated ghosts after
2771
reconcile compared to pack-0.92 formats.
2773
This format was deprecated because the serializer it uses accidentally
2774
supported subtrees, when the format was not intended to. This meant that
2775
someone could accidentally fetch from an incorrect repository.
2778
repository_class = KnitPackRepository
2779
_commit_builder_class = PackRootCommitBuilder
2780
rich_root_data = True
2781
supports_tree_reference = False # no subtrees
2783
supports_external_lookups = True
2784
# What index classes to use
2785
index_builder_class = InMemoryGraphIndex
2786
index_class = GraphIndex
2789
def _serializer(self):
2790
return xml7.serializer_v7
2792
def _get_matching_bzrdir(self):
2793
matching = bzrdir.format_registry.make_bzrdir(
2795
matching.repository_format = self
2798
def _ignore_setting_bzrdir(self, format):
2801
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2803
def get_format_string(self):
2804
"""See RepositoryFormat.get_format_string()."""
2805
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2807
def get_format_description(self):
2808
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2812
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2813
"""A repository with stacking and btree indexes,
2814
without rich roots or subtrees.
2816
This is equivalent to pack-1.6 with B+Tree indices.
2819
repository_class = KnitPackRepository
2820
_commit_builder_class = PackCommitBuilder
2821
supports_external_lookups = True
2822
# What index classes to use
2823
index_builder_class = BTreeBuilder
2824
index_class = BTreeGraphIndex
2827
def _serializer(self):
2828
return xml5.serializer_v5
2830
def _get_matching_bzrdir(self):
2831
return bzrdir.format_registry.make_bzrdir('1.9')
2833
def _ignore_setting_bzrdir(self, format):
2836
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2838
def get_format_string(self):
2839
"""See RepositoryFormat.get_format_string()."""
2840
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2842
def get_format_description(self):
2843
"""See RepositoryFormat.get_format_description()."""
2844
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2847
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2848
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2850
1.6-rich-root with B+Tree indices.
2853
repository_class = KnitPackRepository
2854
_commit_builder_class = PackRootCommitBuilder
2855
rich_root_data = True
2856
supports_tree_reference = False # no subtrees
2857
supports_external_lookups = True
2858
# What index classes to use
2859
index_builder_class = BTreeBuilder
2860
index_class = BTreeGraphIndex
2863
def _serializer(self):
2864
return xml6.serializer_v6
2866
def _get_matching_bzrdir(self):
2867
return bzrdir.format_registry.make_bzrdir(
2870
def _ignore_setting_bzrdir(self, format):
2873
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2875
def get_format_string(self):
2876
"""See RepositoryFormat.get_format_string()."""
2877
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2879
def get_format_description(self):
2880
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2883
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2884
"""A subtrees development repository.
2886
This format should be retained until the second release after bzr 1.7.
2888
1.6.1-subtree[as it might have been] with B+Tree indices.
2890
This is [now] retained until we have a CHK based subtree format in
2894
repository_class = KnitPackRepository
2895
_commit_builder_class = PackRootCommitBuilder
2896
rich_root_data = True
2898
supports_tree_reference = True
2899
supports_external_lookups = True
2900
# What index classes to use
2901
index_builder_class = BTreeBuilder
2902
index_class = BTreeGraphIndex
2905
def _serializer(self):
2906
return xml7.serializer_v7
2908
def _get_matching_bzrdir(self):
2909
return bzrdir.format_registry.make_bzrdir(
2910
'development-subtree')
2912
def _ignore_setting_bzrdir(self, format):
2915
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2917
def get_format_string(self):
2918
"""See RepositoryFormat.get_format_string()."""
2919
return ("Bazaar development format 2 with subtree support "
2920
"(needs bzr.dev from before 1.8)\n")
2922
def get_format_description(self):
2923
"""See RepositoryFormat.get_format_description()."""
2924
return ("Development repository format, currently the same as "
2925
"1.6.1-subtree with B+Tree indices.\n")
added
removed
bzrlib/repofmt/pack_repo.py
