828
732
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)
734
raise NotImplementedError(self._copy_inventory_texts)
861
736
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()
737
raise NotImplementedError(self._copy_text_texts)
888
739
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
740
raise NotImplementedError(self._create_pack_from_packs)
1132
742
def _log_copied_texts(self):
1133
743
if 'pack' in debug.debug_flags:
1165
756
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._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
1233
self._text_filter = None
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.
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:
1258
ideal_parents = tuple(ideal_index[node[1]])
1260
discarded_nodes.append(node)
1261
self._data_changed = True
1263
if ideal_parents == (NULL_REVISION,):
1265
if ideal_parents == node[3][0]:
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
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.
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,
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()
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' %
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()
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
1353
759
class RepositoryPackCollection(object):
1354
760
"""Management of packs within a repository.
1356
762
:ivar _names: map of {pack_name: (index_size,)}
1359
pack_factory = NewPack
1360
resumed_pack_factory = ResumedPack
766
resumed_pack_factory = None
767
normal_packer_class = None
768
optimising_packer_class = None
1362
770
def __init__(self, repo, transport, index_transport, upload_transport,
1363
771
pack_transport, index_builder_class, index_class,
1655
# These attributes are inherited from the Repository base class. Setting
1656
# them to None ensures that if the constructor is changed to not initialize
1657
# them, or a subclass fails to call the constructor, that an error will
1658
# occur rather than the system working but generating incorrect data.
1659
_commit_builder_class = None
2244
1662
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
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,
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,
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,
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,
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,
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
2302
self._reconcile_does_inventory_gc = True
1664
MetaDirRepository.__init__(self, _format, a_bzrdir, control_files)
1665
self._commit_builder_class = _commit_builder_class
1666
self._serializer = _serializer
2303
1667
self._reconcile_fixes_text_parents = True
2304
self._reconcile_backsup_inventory = False
1668
if self._format.supports_external_lookups:
1669
self._unstacked_provider = graph.CachingParentsProvider(
1670
self._make_parents_provider_unstacked())
1672
self._unstacked_provider = graph.CachingParentsProvider(self)
1673
self._unstacked_provider.disable_cache()
1676
def _all_revision_ids(self):
1677
"""See Repository.all_revision_ids()."""
1678
return [key[0] for key in self.revisions.keys()]
2306
1680
def _abort_write_group(self):
2307
1681
self.revisions._index._key_dependencies.clear()
2308
1682
self._pack_collection._abort_write_group()
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)
2315
1684
def _make_parents_provider(self):
2316
return graph.CachingParentsProvider(self)
1685
if not self._format.supports_external_lookups:
1686
return self._unstacked_provider
1687
return graph.StackedParentsProvider(_LazyListJoin(
1688
[self._unstacked_provider], self._fallback_repositories))
2318
1690
def _refresh_data(self):
2319
1691
if not self.is_locked():
2321
1693
self._pack_collection.reload_pack_names()
1694
self._unstacked_provider.disable_cache()
1695
self._unstacked_provider.enable_cache()
2323
1697
def _start_write_group(self):
2324
1698
self._pack_collection._start_write_group()
2445
1825
self.control_files.unlock()
2447
1827
if not self.is_locked():
1828
self._unstacked_provider.disable_cache()
2448
1829
for repo in self._fallback_repositories:
2452
class KnitPackStreamSource(StreamSource):
2453
"""A StreamSource used to transfer data between same-format KnitPack repos.
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
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'
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,
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)
2505
self._text_keys = content_text_keys - parent_text_keys
2506
return ('inventories', _filtered_inv_stream())
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)
2515
def get_stream(self, search):
2516
revision_ids = search.get_keys()
2517
for stream_info in self._fetch_revision_texts(revision_ids):
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()
2525
class RepositoryFormatPack(MetaDirRepositoryFormat):
1833
class RepositoryFormatPack(MetaDirVersionedFileRepositoryFormat):
2526
1834
"""Format logic for pack structured repositories.
2528
1836
This repository format has:
2602
1910
_serializer=self._serializer)
2605
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2606
"""A no-subtrees parameterized Pack repository.
2608
This format was introduced in 0.92.
2611
repository_class = KnitPackRepository
2612
_commit_builder_class = PackCommitBuilder
2614
def _serializer(self):
2615
return xml5.serializer_v5
2616
# What index classes to use
2617
index_builder_class = InMemoryGraphIndex
2618
index_class = GraphIndex
2620
def _get_matching_bzrdir(self):
2621
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2623
def _ignore_setting_bzrdir(self, format):
2626
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2628
def get_format_string(self):
2629
"""See RepositoryFormat.get_format_string()."""
2630
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2632
def get_format_description(self):
2633
"""See RepositoryFormat.get_format_description()."""
2634
return "Packs containing knits without subtree support"
2637
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2638
"""A subtrees parameterized Pack repository.
2640
This repository format uses the xml7 serializer to get:
2641
- support for recording full info about the tree root
2642
- support for recording tree-references
2644
This format was introduced in 0.92.
2647
repository_class = KnitPackRepository
2648
_commit_builder_class = PackRootCommitBuilder
2649
rich_root_data = True
2651
supports_tree_reference = True
2653
def _serializer(self):
2654
return xml7.serializer_v7
2655
# What index classes to use
2656
index_builder_class = InMemoryGraphIndex
2657
index_class = GraphIndex
2659
def _get_matching_bzrdir(self):
2660
return bzrdir.format_registry.make_bzrdir(
2661
'pack-0.92-subtree')
2663
def _ignore_setting_bzrdir(self, format):
2666
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2668
def get_format_string(self):
2669
"""See RepositoryFormat.get_format_string()."""
2670
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2672
def get_format_description(self):
2673
"""See RepositoryFormat.get_format_description()."""
2674
return "Packs containing knits with subtree support\n"
2677
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2678
"""A rich-root, no subtrees parameterized Pack repository.
2680
This repository format uses the xml6 serializer to get:
2681
- support for recording full info about the tree root
2683
This format was introduced in 1.0.
2686
repository_class = KnitPackRepository
2687
_commit_builder_class = PackRootCommitBuilder
2688
rich_root_data = True
2689
supports_tree_reference = False
2691
def _serializer(self):
2692
return xml6.serializer_v6
2693
# What index classes to use
2694
index_builder_class = InMemoryGraphIndex
2695
index_class = GraphIndex
2697
def _get_matching_bzrdir(self):
2698
return bzrdir.format_registry.make_bzrdir(
2701
def _ignore_setting_bzrdir(self, format):
2704
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2706
def get_format_string(self):
2707
"""See RepositoryFormat.get_format_string()."""
2708
return ("Bazaar pack repository format 1 with rich root"
2709
" (needs bzr 1.0)\n")
2711
def get_format_description(self):
2712
"""See RepositoryFormat.get_format_description()."""
2713
return "Packs containing knits with rich root support\n"
2716
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2717
"""Repository that supports external references to allow stacking.
2721
Supports external lookups, which results in non-truncated ghosts after
2722
reconcile compared to pack-0.92 formats.
2725
repository_class = KnitPackRepository
2726
_commit_builder_class = PackCommitBuilder
2727
supports_external_lookups = True
2728
# What index classes to use
2729
index_builder_class = InMemoryGraphIndex
2730
index_class = GraphIndex
2733
def _serializer(self):
2734
return xml5.serializer_v5
2736
def _get_matching_bzrdir(self):
2737
return bzrdir.format_registry.make_bzrdir('1.6')
2739
def _ignore_setting_bzrdir(self, format):
2742
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2744
def get_format_string(self):
2745
"""See RepositoryFormat.get_format_string()."""
2746
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2748
def get_format_description(self):
2749
"""See RepositoryFormat.get_format_description()."""
2750
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2753
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2754
"""A repository with rich roots and stacking.
2756
New in release 1.6.1.
2758
Supports stacking on other repositories, allowing data to be accessed
2759
without being stored locally.
2762
repository_class = KnitPackRepository
2763
_commit_builder_class = PackRootCommitBuilder
2764
rich_root_data = True
2765
supports_tree_reference = False # no subtrees
2766
supports_external_lookups = True
2767
# What index classes to use
2768
index_builder_class = InMemoryGraphIndex
2769
index_class = GraphIndex
2772
def _serializer(self):
2773
return xml6.serializer_v6
2775
def _get_matching_bzrdir(self):
2776
return bzrdir.format_registry.make_bzrdir(
2779
def _ignore_setting_bzrdir(self, format):
2782
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2784
def get_format_string(self):
2785
"""See RepositoryFormat.get_format_string()."""
2786
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2788
def get_format_description(self):
2789
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2792
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2793
"""A repository with rich roots and external references.
2797
Supports external lookups, which results in non-truncated ghosts after
2798
reconcile compared to pack-0.92 formats.
2800
This format was deprecated because the serializer it uses accidentally
2801
supported subtrees, when the format was not intended to. This meant that
2802
someone could accidentally fetch from an incorrect repository.
2805
repository_class = KnitPackRepository
2806
_commit_builder_class = PackRootCommitBuilder
2807
rich_root_data = True
2808
supports_tree_reference = False # no subtrees
2810
supports_external_lookups = True
2811
# What index classes to use
2812
index_builder_class = InMemoryGraphIndex
2813
index_class = GraphIndex
2816
def _serializer(self):
2817
return xml7.serializer_v7
2819
def _get_matching_bzrdir(self):
2820
matching = bzrdir.format_registry.make_bzrdir(
2822
matching.repository_format = self
2825
def _ignore_setting_bzrdir(self, format):
2828
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2830
def get_format_string(self):
2831
"""See RepositoryFormat.get_format_string()."""
2832
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2834
def get_format_description(self):
2835
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2838
def is_deprecated(self):
2842
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2843
"""A repository with stacking and btree indexes,
2844
without rich roots or subtrees.
2846
This is equivalent to pack-1.6 with B+Tree indices.
2849
repository_class = KnitPackRepository
2850
_commit_builder_class = PackCommitBuilder
2851
supports_external_lookups = True
2852
# What index classes to use
2853
index_builder_class = btree_index.BTreeBuilder
2854
index_class = btree_index.BTreeGraphIndex
2857
def _serializer(self):
2858
return xml5.serializer_v5
2860
def _get_matching_bzrdir(self):
2861
return bzrdir.format_registry.make_bzrdir('1.9')
2863
def _ignore_setting_bzrdir(self, format):
2866
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2868
def get_format_string(self):
2869
"""See RepositoryFormat.get_format_string()."""
2870
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2872
def get_format_description(self):
2873
"""See RepositoryFormat.get_format_description()."""
2874
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2877
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2878
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2880
1.6-rich-root with B+Tree indices.
2883
repository_class = KnitPackRepository
2884
_commit_builder_class = PackRootCommitBuilder
2885
rich_root_data = True
2886
supports_tree_reference = False # no subtrees
2887
supports_external_lookups = True
2888
# What index classes to use
2889
index_builder_class = btree_index.BTreeBuilder
2890
index_class = btree_index.BTreeGraphIndex
2893
def _serializer(self):
2894
return xml6.serializer_v6
2896
def _get_matching_bzrdir(self):
2897
return bzrdir.format_registry.make_bzrdir(
2900
def _ignore_setting_bzrdir(self, format):
2903
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2905
def get_format_string(self):
2906
"""See RepositoryFormat.get_format_string()."""
2907
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2909
def get_format_description(self):
2910
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2913
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2914
"""A subtrees development repository.
2916
This format should be retained in 2.3, to provide an upgrade path from this
2917
to RepositoryFormat2aSubtree. It can be removed in later releases.
2919
1.6.1-subtree[as it might have been] with B+Tree indices.
2922
repository_class = KnitPackRepository
2923
_commit_builder_class = PackRootCommitBuilder
2924
rich_root_data = True
2926
supports_tree_reference = True
2927
supports_external_lookups = True
2928
# What index classes to use
2929
index_builder_class = btree_index.BTreeBuilder
2930
index_class = btree_index.BTreeGraphIndex
2933
def _serializer(self):
2934
return xml7.serializer_v7
2936
def _get_matching_bzrdir(self):
2937
return bzrdir.format_registry.make_bzrdir(
2938
'development5-subtree')
2940
def _ignore_setting_bzrdir(self, format):
2943
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2945
def get_format_string(self):
2946
"""See RepositoryFormat.get_format_string()."""
2947
return ("Bazaar development format 2 with subtree support "
2948
"(needs bzr.dev from before 1.8)\n")
2950
def get_format_description(self):
2951
"""See RepositoryFormat.get_format_description()."""
2952
return ("Development repository format, currently the same as "
2953
"1.6.1-subtree with B+Tree indices.\n")
2956
1913
class RetryPackOperations(errors.RetryWithNewPacks):
2957
1914
"""Raised when we are packing and we find a missing file.
2966
1923
" context: %(context)s %(orig_error)s")
1926
class _DirectPackAccess(object):
1927
"""Access to data in one or more packs with less translation."""
1929
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
1930
"""Create a _DirectPackAccess object.
1932
:param index_to_packs: A dict mapping index objects to the transport
1933
and file names for obtaining data.
1934
:param reload_func: A function to call if we determine that the pack
1935
files have moved and we need to reload our caches. See
1936
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
1938
self._container_writer = None
1939
self._write_index = None
1940
self._indices = index_to_packs
1941
self._reload_func = reload_func
1942
self._flush_func = flush_func
1944
def add_raw_records(self, key_sizes, raw_data):
1945
"""Add raw knit bytes to a storage area.
1947
The data is spooled to the container writer in one bytes-record per
1950
:param sizes: An iterable of tuples containing the key and size of each
1952
:param raw_data: A bytestring containing the data.
1953
:return: A list of memos to retrieve the record later. Each memo is an
1954
opaque index memo. For _DirectPackAccess the memo is (index, pos,
1955
length), where the index field is the write_index object supplied
1956
to the PackAccess object.
1958
if type(raw_data) is not str:
1959
raise AssertionError(
1960
'data must be plain bytes was %s' % type(raw_data))
1963
for key, size in key_sizes:
1964
p_offset, p_length = self._container_writer.add_bytes_record(
1965
raw_data[offset:offset+size], [])
1967
result.append((self._write_index, p_offset, p_length))
1971
"""Flush pending writes on this access object.
1973
This will flush any buffered writes to a NewPack.
1975
if self._flush_func is not None:
1978
def get_raw_records(self, memos_for_retrieval):
1979
"""Get the raw bytes for a records.
1981
:param memos_for_retrieval: An iterable containing the (index, pos,
1982
length) memo for retrieving the bytes. The Pack access method
1983
looks up the pack to use for a given record in its index_to_pack
1985
:return: An iterator over the bytes of the records.
1987
# first pass, group into same-index requests
1989
current_index = None
1990
for (index, offset, length) in memos_for_retrieval:
1991
if current_index == index:
1992
current_list.append((offset, length))
1994
if current_index is not None:
1995
request_lists.append((current_index, current_list))
1996
current_index = index
1997
current_list = [(offset, length)]
1998
# handle the last entry
1999
if current_index is not None:
2000
request_lists.append((current_index, current_list))
2001
for index, offsets in request_lists:
2003
transport, path = self._indices[index]
2005
# A KeyError here indicates that someone has triggered an index
2006
# reload, and this index has gone missing, we need to start
2008
if self._reload_func is None:
2009
# If we don't have a _reload_func there is nothing that can
2012
raise errors.RetryWithNewPacks(index,
2013
reload_occurred=True,
2014
exc_info=sys.exc_info())
2016
reader = pack.make_readv_reader(transport, path, offsets)
2017
for names, read_func in reader.iter_records():
2018
yield read_func(None)
2019
except errors.NoSuchFile:
2020
# A NoSuchFile error indicates that a pack file has gone
2021
# missing on disk, we need to trigger a reload, and start over.
2022
if self._reload_func is None:
2024
raise errors.RetryWithNewPacks(transport.abspath(path),
2025
reload_occurred=False,
2026
exc_info=sys.exc_info())
2028
def set_writer(self, writer, index, transport_packname):
2029
"""Set a writer to use for adding data."""
2030
if index is not None:
2031
self._indices[index] = transport_packname
2032
self._container_writer = writer
2033
self._write_index = index
2035
def reload_or_raise(self, retry_exc):
2036
"""Try calling the reload function, or re-raise the original exception.
2038
This should be called after _DirectPackAccess raises a
2039
RetryWithNewPacks exception. This function will handle the common logic
2040
of determining when the error is fatal versus being temporary.
2041
It will also make sure that the original exception is raised, rather
2042
than the RetryWithNewPacks exception.
2044
If this function returns, then the calling function should retry
2045
whatever operation was being performed. Otherwise an exception will
2048
:param retry_exc: A RetryWithNewPacks exception.
2051
if self._reload_func is None:
2053
elif not self._reload_func():
2054
# The reload claimed that nothing changed
2055
if not retry_exc.reload_occurred:
2056
# If there wasn't an earlier reload, then we really were
2057
# expecting to find changes. We didn't find them, so this is a
2061
exc_class, exc_value, exc_traceback = retry_exc.exc_info
2062
raise exc_class, exc_value, exc_traceback