829
741
Sets self._text_filter appropriately.
831
# select inventory keys
832
inv_keys = self._revision_keys # currently the same keyspace, and note that
833
# querying for keys here could introduce a bug where an inventory item
834
# is missed, so do not change it to query separately without cross
835
# checking like the text key check below.
836
inventory_index_map, inventory_indices = self._pack_map_and_index_list(
838
inv_nodes = self._index_contents(inventory_indices, inv_keys)
839
# copy inventory keys and adjust values
840
# XXX: Should be a helper function to allow different inv representation
842
self.pb.update("Copying inventory texts", 2)
843
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
844
# Only grab the output lines if we will be processing them
845
output_lines = bool(self.revision_ids)
846
inv_lines = self._copy_nodes_graph(inventory_index_map,
847
self.new_pack._writer, self.new_pack.inventory_index,
848
readv_group_iter, total_items, output_lines=output_lines)
849
if self.revision_ids:
850
self._process_inventory_lines(inv_lines)
852
# eat the iterator to cause it to execute.
854
self._text_filter = None
855
if 'pack' in debug.debug_flags:
856
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
857
time.ctime(), self._pack_collection._upload_transport.base,
858
self.new_pack.random_name,
859
self.new_pack.inventory_index.key_count(),
860
time.time() - self.new_pack.start_time)
743
raise NotImplementedError(self._copy_inventory_texts)
862
745
def _copy_text_texts(self):
864
text_index_map, text_nodes = self._get_text_nodes()
865
if self._text_filter is not None:
866
# We could return the keys copied as part of the return value from
867
# _copy_nodes_graph but this doesn't work all that well with the
868
# need to get line output too, so we check separately, and as we're
869
# going to buffer everything anyway, we check beforehand, which
870
# saves reading knit data over the wire when we know there are
872
text_nodes = set(text_nodes)
873
present_text_keys = set(_node[1] for _node in text_nodes)
874
missing_text_keys = set(self._text_filter) - present_text_keys
875
if missing_text_keys:
876
# TODO: raise a specific error that can handle many missing
878
mutter("missing keys during fetch: %r", missing_text_keys)
879
a_missing_key = missing_text_keys.pop()
880
raise errors.RevisionNotPresent(a_missing_key[1],
882
# copy text keys and adjust values
883
self.pb.update("Copying content texts", 3)
884
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
885
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
886
self.new_pack.text_index, readv_group_iter, total_items))
887
self._log_copied_texts()
746
raise NotImplementedError(self._copy_text_texts)
889
748
def _create_pack_from_packs(self):
890
self.pb.update("Opening pack", 0, 5)
891
self.new_pack = self.open_pack()
892
new_pack = self.new_pack
893
# buffer data - we won't be reading-back during the pack creation and
894
# this makes a significant difference on sftp pushes.
895
new_pack.set_write_cache_size(1024*1024)
896
if 'pack' in debug.debug_flags:
897
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
898
for a_pack in self.packs]
899
if self.revision_ids is not None:
900
rev_count = len(self.revision_ids)
903
mutter('%s: create_pack: creating pack from source packs: '
904
'%s%s %s revisions wanted %s t=0',
905
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
906
plain_pack_list, rev_count)
907
self._copy_revision_texts()
908
self._copy_inventory_texts()
909
self._copy_text_texts()
910
# select signature keys
911
signature_filter = self._revision_keys # same keyspace
912
signature_index_map, signature_indices = self._pack_map_and_index_list(
914
signature_nodes = self._index_contents(signature_indices,
916
# copy signature keys and adjust values
917
self.pb.update("Copying signature texts", 4)
918
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
919
new_pack.signature_index)
920
if 'pack' in debug.debug_flags:
921
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
922
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
923
new_pack.signature_index.key_count(),
924
time.time() - new_pack.start_time)
926
# NB XXX: how to check CHK references are present? perhaps by yielding
927
# the items? How should that interact with stacked repos?
928
if new_pack.chk_index is not None:
930
if 'pack' in debug.debug_flags:
931
mutter('%s: create_pack: chk content copied: %s%s %d items t+%6.3fs',
932
time.ctime(), self._pack_collection._upload_transport.base,
933
new_pack.random_name,
934
new_pack.chk_index.key_count(),
935
time.time() - new_pack.start_time)
936
new_pack._check_references()
937
if not self._use_pack(new_pack):
940
self.pb.update("Finishing pack", 5)
942
self._pack_collection.allocate(new_pack)
945
def _copy_chks(self, refs=None):
946
# XXX: Todo, recursive follow-pointers facility when fetching some
948
chk_index_map, chk_indices = self._pack_map_and_index_list(
950
chk_nodes = self._index_contents(chk_indices, refs)
952
# TODO: This isn't strictly tasteful as we are accessing some private
953
# variables (_serializer). Perhaps a better way would be to have
954
# Repository._deserialise_chk_node()
955
search_key_func = chk_map.search_key_registry.get(
956
self._pack_collection.repo._serializer.search_key_name)
957
def accumlate_refs(lines):
958
# XXX: move to a generic location
960
bytes = ''.join(lines)
961
node = chk_map._deserialise(bytes, ("unknown",), search_key_func)
962
new_refs.update(node.refs())
963
self._copy_nodes(chk_nodes, chk_index_map, self.new_pack._writer,
964
self.new_pack.chk_index, output_lines=accumlate_refs)
967
def _copy_nodes(self, nodes, index_map, writer, write_index,
969
"""Copy knit nodes between packs with no graph references.
971
:param output_lines: Output full texts of copied items.
973
pb = ui.ui_factory.nested_progress_bar()
975
return self._do_copy_nodes(nodes, index_map, writer,
976
write_index, pb, output_lines=output_lines)
980
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb,
982
# for record verification
983
knit = KnitVersionedFiles(None, None)
984
# plan a readv on each source pack:
986
nodes = sorted(nodes)
987
# how to map this into knit.py - or knit.py into this?
988
# we don't want the typical knit logic, we want grouping by pack
989
# at this point - perhaps a helper library for the following code
990
# duplication points?
992
for index, key, value in nodes:
993
if index not in request_groups:
994
request_groups[index] = []
995
request_groups[index].append((key, value))
997
pb.update("Copied record", record_index, len(nodes))
998
for index, items in request_groups.iteritems():
999
pack_readv_requests = []
1000
for key, value in items:
1001
# ---- KnitGraphIndex.get_position
1002
bits = value[1:].split(' ')
1003
offset, length = int(bits[0]), int(bits[1])
1004
pack_readv_requests.append((offset, length, (key, value[0])))
1005
# linear scan up the pack
1006
pack_readv_requests.sort()
1008
pack_obj = index_map[index]
1009
transport, path = pack_obj.access_tuple()
1011
reader = pack.make_readv_reader(transport, path,
1012
[offset[0:2] for offset in pack_readv_requests])
1013
except errors.NoSuchFile:
1014
if self._reload_func is not None:
1017
for (names, read_func), (_1, _2, (key, eol_flag)) in \
1018
izip(reader.iter_records(), pack_readv_requests):
1019
raw_data = read_func(None)
1020
# check the header only
1021
if output_lines is not None:
1022
output_lines(knit._parse_record(key[-1], raw_data)[0])
1024
df, _ = knit._parse_record_header(key, raw_data)
1026
pos, size = writer.add_bytes_record(raw_data, names)
1027
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
1028
pb.update("Copied record", record_index)
1031
def _copy_nodes_graph(self, index_map, writer, write_index,
1032
readv_group_iter, total_items, output_lines=False):
1033
"""Copy knit nodes between packs.
1035
:param output_lines: Return lines present in the copied data as
1036
an iterator of line,version_id.
1038
pb = ui.ui_factory.nested_progress_bar()
1040
for result in self._do_copy_nodes_graph(index_map, writer,
1041
write_index, output_lines, pb, readv_group_iter, total_items):
1044
# Python 2.4 does not permit try:finally: in a generator.
1050
def _do_copy_nodes_graph(self, index_map, writer, write_index,
1051
output_lines, pb, readv_group_iter, total_items):
1052
# for record verification
1053
knit = KnitVersionedFiles(None, None)
1054
# for line extraction when requested (inventories only)
1056
factory = KnitPlainFactory()
1058
pb.update("Copied record", record_index, total_items)
1059
for index, readv_vector, node_vector in readv_group_iter:
1061
pack_obj = index_map[index]
1062
transport, path = pack_obj.access_tuple()
1064
reader = pack.make_readv_reader(transport, path, readv_vector)
1065
except errors.NoSuchFile:
1066
if self._reload_func is not None:
1069
for (names, read_func), (key, eol_flag, references) in \
1070
izip(reader.iter_records(), node_vector):
1071
raw_data = read_func(None)
1073
# read the entire thing
1074
content, _ = knit._parse_record(key[-1], raw_data)
1075
if len(references[-1]) == 0:
1076
line_iterator = factory.get_fulltext_content(content)
1078
line_iterator = factory.get_linedelta_content(content)
1079
for line in line_iterator:
1082
# check the header only
1083
df, _ = knit._parse_record_header(key, raw_data)
1085
pos, size = writer.add_bytes_record(raw_data, names)
1086
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1087
pb.update("Copied record", record_index)
1090
def _get_text_nodes(self):
1091
text_index_map, text_indices = self._pack_map_and_index_list(
1093
return text_index_map, self._index_contents(text_indices,
1096
def _least_readv_node_readv(self, nodes):
1097
"""Generate request groups for nodes using the least readv's.
1099
:param nodes: An iterable of graph index nodes.
1100
:return: Total node count and an iterator of the data needed to perform
1101
readvs to obtain the data for nodes. Each item yielded by the
1102
iterator is a tuple with:
1103
index, readv_vector, node_vector. readv_vector is a list ready to
1104
hand to the transport readv method, and node_vector is a list of
1105
(key, eol_flag, references) for the node retrieved by the
1106
matching readv_vector.
1108
# group by pack so we do one readv per pack
1109
nodes = sorted(nodes)
1112
for index, key, value, references in nodes:
1113
if index not in request_groups:
1114
request_groups[index] = []
1115
request_groups[index].append((key, value, references))
1117
for index, items in request_groups.iteritems():
1118
pack_readv_requests = []
1119
for key, value, references in items:
1120
# ---- KnitGraphIndex.get_position
1121
bits = value[1:].split(' ')
1122
offset, length = int(bits[0]), int(bits[1])
1123
pack_readv_requests.append(
1124
((offset, length), (key, value[0], references)))
1125
# linear scan up the pack to maximum range combining.
1126
pack_readv_requests.sort()
1127
# split out the readv and the node data.
1128
pack_readv = [readv for readv, node in pack_readv_requests]
1129
node_vector = [node for readv, node in pack_readv_requests]
1130
result.append((index, pack_readv, node_vector))
1131
return total, result
749
raise NotImplementedError(self._create_pack_from_packs)
1133
751
def _log_copied_texts(self):
1134
752
if 'pack' in debug.debug_flags:
1166
765
return new_pack.data_inserted()
1169
class OptimisingPacker(Packer):
1170
"""A packer which spends more time to create better disk layouts."""
1172
def _revision_node_readv(self, revision_nodes):
1173
"""Return the total revisions and the readv's to issue.
1175
This sort places revisions in topological order with the ancestors
1178
:param revision_nodes: The revision index contents for the packs being
1179
incorporated into the new pack.
1180
:return: As per _least_readv_node_readv.
1182
# build an ancestors dict
1185
for index, key, value, references in revision_nodes:
1186
ancestors[key] = references[0]
1187
by_key[key] = (index, value, references)
1188
order = tsort.topo_sort(ancestors)
1190
# Single IO is pathological, but it will work as a starting point.
1192
for key in reversed(order):
1193
index, value, references = by_key[key]
1194
# ---- KnitGraphIndex.get_position
1195
bits = value[1:].split(' ')
1196
offset, length = int(bits[0]), int(bits[1])
1198
(index, [(offset, length)], [(key, value[0], references)]))
1199
# TODO: combine requests in the same index that are in ascending order.
1200
return total, requests
1202
def open_pack(self):
1203
"""Open a pack for the pack we are creating."""
1204
new_pack = super(OptimisingPacker, self).open_pack()
1205
# Turn on the optimization flags for all the index builders.
1206
new_pack.revision_index.set_optimize(for_size=True)
1207
new_pack.inventory_index.set_optimize(for_size=True)
1208
new_pack.text_index.set_optimize(for_size=True)
1209
new_pack.signature_index.set_optimize(for_size=True)
1213
class ReconcilePacker(Packer):
1214
"""A packer which regenerates indices etc as it copies.
1216
This is used by ``bzr reconcile`` to cause parent text pointers to be
1220
def _extra_init(self):
1221
self._data_changed = False
1223
def _process_inventory_lines(self, inv_lines):
1224
"""Generate a text key reference map rather for reconciling with."""
1225
repo = self._pack_collection.repo
1226
refs = repo._find_text_key_references_from_xml_inventory_lines(
1228
self._text_refs = refs
1229
# during reconcile we:
1230
# - convert unreferenced texts to full texts
1231
# - correct texts which reference a text not copied to be full texts
1232
# - copy all others as-is but with corrected parents.
1233
# - so at this point we don't know enough to decide what becomes a full
1235
self._text_filter = None
1237
def _copy_text_texts(self):
1238
"""generate what texts we should have and then copy."""
1239
self.pb.update("Copying content texts", 3)
1240
# we have three major tasks here:
1241
# 1) generate the ideal index
1242
repo = self._pack_collection.repo
1243
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1244
_1, key, _2, refs in
1245
self.new_pack.revision_index.iter_all_entries()])
1246
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1247
# 2) generate a text_nodes list that contains all the deltas that can
1248
# be used as-is, with corrected parents.
1251
discarded_nodes = []
1252
NULL_REVISION = _mod_revision.NULL_REVISION
1253
text_index_map, text_nodes = self._get_text_nodes()
1254
for node in text_nodes:
1260
ideal_parents = tuple(ideal_index[node[1]])
1262
discarded_nodes.append(node)
1263
self._data_changed = True
1265
if ideal_parents == (NULL_REVISION,):
1267
if ideal_parents == node[3][0]:
1269
ok_nodes.append(node)
1270
elif ideal_parents[0:1] == node[3][0][0:1]:
1271
# the left most parent is the same, or there are no parents
1272
# today. Either way, we can preserve the representation as
1273
# long as we change the refs to be inserted.
1274
self._data_changed = True
1275
ok_nodes.append((node[0], node[1], node[2],
1276
(ideal_parents, node[3][1])))
1277
self._data_changed = True
1279
# Reinsert this text completely
1280
bad_texts.append((node[1], ideal_parents))
1281
self._data_changed = True
1282
# we're finished with some data.
1285
# 3) bulk copy the ok data
1286
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1287
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1288
self.new_pack.text_index, readv_group_iter, total_items))
1289
# 4) adhoc copy all the other texts.
1290
# We have to topologically insert all texts otherwise we can fail to
1291
# reconcile when parts of a single delta chain are preserved intact,
1292
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1293
# reinserted, and if d3 has incorrect parents it will also be
1294
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1295
# copied), so we will try to delta, but d2 is not currently able to be
1296
# extracted because it's basis d1 is not present. Topologically sorting
1297
# addresses this. The following generates a sort for all the texts that
1298
# are being inserted without having to reference the entire text key
1299
# space (we only topo sort the revisions, which is smaller).
1300
topo_order = tsort.topo_sort(ancestors)
1301
rev_order = dict(zip(topo_order, range(len(topo_order))))
1302
bad_texts.sort(key=lambda key:rev_order.get(key[0][1], 0))
1303
transaction = repo.get_transaction()
1304
file_id_index = GraphIndexPrefixAdapter(
1305
self.new_pack.text_index,
1307
add_nodes_callback=self.new_pack.text_index.add_nodes)
1308
data_access = _DirectPackAccess(
1309
{self.new_pack.text_index:self.new_pack.access_tuple()})
1310
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1311
self.new_pack.access_tuple())
1312
output_texts = KnitVersionedFiles(
1313
_KnitGraphIndex(self.new_pack.text_index,
1314
add_callback=self.new_pack.text_index.add_nodes,
1315
deltas=True, parents=True, is_locked=repo.is_locked),
1316
data_access=data_access, max_delta_chain=200)
1317
for key, parent_keys in bad_texts:
1318
# We refer to the new pack to delta data being output.
1319
# A possible improvement would be to catch errors on short reads
1320
# and only flush then.
1321
self.new_pack.flush()
1323
for parent_key in parent_keys:
1324
if parent_key[0] != key[0]:
1325
# Graph parents must match the fileid
1326
raise errors.BzrError('Mismatched key parent %r:%r' %
1328
parents.append(parent_key[1])
1329
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1330
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1331
output_texts.add_lines(key, parent_keys, text_lines,
1332
random_id=True, check_content=False)
1333
# 5) check that nothing inserted has a reference outside the keyspace.
1334
missing_text_keys = self.new_pack.text_index._external_references()
1335
if missing_text_keys:
1336
raise errors.BzrCheckError('Reference to missing compression parents %r'
1337
% (missing_text_keys,))
1338
self._log_copied_texts()
1340
def _use_pack(self, new_pack):
1341
"""Override _use_pack to check for reconcile having changed content."""
1342
# XXX: we might be better checking this at the copy time.
1343
original_inventory_keys = set()
1344
inv_index = self._pack_collection.inventory_index.combined_index
1345
for entry in inv_index.iter_all_entries():
1346
original_inventory_keys.add(entry[1])
1347
new_inventory_keys = set()
1348
for entry in new_pack.inventory_index.iter_all_entries():
1349
new_inventory_keys.add(entry[1])
1350
if new_inventory_keys != original_inventory_keys:
1351
self._data_changed = True
1352
return new_pack.data_inserted() and self._data_changed
1355
768
class RepositoryPackCollection(object):
1356
769
"""Management of packs within a repository.
1358
771
:ivar _names: map of {pack_name: (index_size,)}
1361
pack_factory = NewPack
1362
resumed_pack_factory = ResumedPack
775
resumed_pack_factory = None
776
normal_packer_class = None
777
optimising_packer_class = None
1364
779
def __init__(self, repo, transport, index_transport, upload_transport,
1365
780
pack_transport, index_builder_class, index_class,
1665
# These attributes are inherited from the Repository base class. Setting
1666
# them to None ensures that if the constructor is changed to not initialize
1667
# them, or a subclass fails to call the constructor, that an error will
1668
# occur rather than the system working but generating incorrect data.
1669
_commit_builder_class = None
2223
1672
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
2225
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2226
_commit_builder_class, _serializer)
2227
index_transport = self._transport.clone('indices')
2228
self._pack_collection = RepositoryPackCollection(self, self._transport,
2230
self._transport.clone('upload'),
2231
self._transport.clone('packs'),
2232
_format.index_builder_class,
2233
_format.index_class,
2234
use_chk_index=self._format.supports_chks,
2236
self.inventories = KnitVersionedFiles(
2237
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2238
add_callback=self._pack_collection.inventory_index.add_callback,
2239
deltas=True, parents=True, is_locked=self.is_locked),
2240
data_access=self._pack_collection.inventory_index.data_access,
2241
max_delta_chain=200)
2242
self.revisions = KnitVersionedFiles(
2243
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2244
add_callback=self._pack_collection.revision_index.add_callback,
2245
deltas=False, parents=True, is_locked=self.is_locked,
2246
track_external_parent_refs=True),
2247
data_access=self._pack_collection.revision_index.data_access,
2249
self.signatures = KnitVersionedFiles(
2250
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2251
add_callback=self._pack_collection.signature_index.add_callback,
2252
deltas=False, parents=False, is_locked=self.is_locked),
2253
data_access=self._pack_collection.signature_index.data_access,
2255
self.texts = KnitVersionedFiles(
2256
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2257
add_callback=self._pack_collection.text_index.add_callback,
2258
deltas=True, parents=True, is_locked=self.is_locked),
2259
data_access=self._pack_collection.text_index.data_access,
2260
max_delta_chain=200)
2261
if _format.supports_chks:
2262
# No graph, no compression:- references from chks are between
2263
# different objects not temporal versions of the same; and without
2264
# some sort of temporal structure knit compression will just fail.
2265
self.chk_bytes = KnitVersionedFiles(
2266
_KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2267
add_callback=self._pack_collection.chk_index.add_callback,
2268
deltas=False, parents=False, is_locked=self.is_locked),
2269
data_access=self._pack_collection.chk_index.data_access,
2272
self.chk_bytes = None
2273
# True when the repository object is 'write locked' (as opposed to the
2274
# physical lock only taken out around changes to the pack-names list.)
2275
# Another way to represent this would be a decorator around the control
2276
# files object that presents logical locks as physical ones - if this
2277
# gets ugly consider that alternative design. RBC 20071011
2278
self._write_lock_count = 0
2279
self._transaction = None
2281
self._reconcile_does_inventory_gc = True
1674
MetaDirRepository.__init__(self, _format, a_bzrdir, control_files)
1675
self._commit_builder_class = _commit_builder_class
1676
self._serializer = _serializer
2282
1677
self._reconcile_fixes_text_parents = True
2283
self._reconcile_backsup_inventory = False
1678
if self._format.supports_external_lookups:
1679
self._unstacked_provider = graph.CachingParentsProvider(
1680
self._make_parents_provider_unstacked())
1682
self._unstacked_provider = graph.CachingParentsProvider(self)
1683
self._unstacked_provider.disable_cache()
2285
def _warn_if_deprecated(self, branch=None):
2286
# This class isn't deprecated, but one sub-format is
2287
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2288
super(KnitPackRepository, self)._warn_if_deprecated(branch)
1686
def _all_revision_ids(self):
1687
"""See Repository.all_revision_ids()."""
1688
return [key[0] for key in self.revisions.keys()]
2290
1690
def _abort_write_group(self):
2291
1691
self.revisions._index._key_dependencies.clear()
2292
1692
self._pack_collection._abort_write_group()
2294
def _get_source(self, to_format):
2295
if to_format.network_name() == self._format.network_name():
2296
return KnitPackStreamSource(self, to_format)
2297
return super(KnitPackRepository, self)._get_source(to_format)
2299
1694
def _make_parents_provider(self):
2300
return graph.CachingParentsProvider(self)
1695
if not self._format.supports_external_lookups:
1696
return self._unstacked_provider
1697
return graph.StackedParentsProvider(_LazyListJoin(
1698
[self._unstacked_provider], self._fallback_repositories))
2302
1700
def _refresh_data(self):
2303
1701
if not self.is_locked():
2305
1703
self._pack_collection.reload_pack_names()
1704
self._unstacked_provider.disable_cache()
1705
self._unstacked_provider.enable_cache()
2307
1707
def _start_write_group(self):
2308
1708
self._pack_collection._start_write_group()
2429
1835
self.control_files.unlock()
2431
1837
if not self.is_locked():
1838
self._unstacked_provider.disable_cache()
2432
1839
for repo in self._fallback_repositories:
2436
class KnitPackStreamSource(StreamSource):
2437
"""A StreamSource used to transfer data between same-format KnitPack repos.
2439
This source assumes:
2440
1) Same serialization format for all objects
2441
2) Same root information
2442
3) XML format inventories
2443
4) Atomic inserts (so we can stream inventory texts before text
2448
def __init__(self, from_repository, to_format):
2449
super(KnitPackStreamSource, self).__init__(from_repository, to_format)
2450
self._text_keys = None
2451
self._text_fetch_order = 'unordered'
2453
def _get_filtered_inv_stream(self, revision_ids):
2454
from_repo = self.from_repository
2455
parent_ids = from_repo._find_parent_ids_of_revisions(revision_ids)
2456
parent_keys = [(p,) for p in parent_ids]
2457
find_text_keys = from_repo._find_text_key_references_from_xml_inventory_lines
2458
parent_text_keys = set(find_text_keys(
2459
from_repo._inventory_xml_lines_for_keys(parent_keys)))
2460
content_text_keys = set()
2461
knit = KnitVersionedFiles(None, None)
2462
factory = KnitPlainFactory()
2463
def find_text_keys_from_content(record):
2464
if record.storage_kind not in ('knit-delta-gz', 'knit-ft-gz'):
2465
raise ValueError("Unknown content storage kind for"
2466
" inventory text: %s" % (record.storage_kind,))
2467
# It's a knit record, it has a _raw_record field (even if it was
2468
# reconstituted from a network stream).
2469
raw_data = record._raw_record
2470
# read the entire thing
2471
revision_id = record.key[-1]
2472
content, _ = knit._parse_record(revision_id, raw_data)
2473
if record.storage_kind == 'knit-delta-gz':
2474
line_iterator = factory.get_linedelta_content(content)
2475
elif record.storage_kind == 'knit-ft-gz':
2476
line_iterator = factory.get_fulltext_content(content)
2477
content_text_keys.update(find_text_keys(
2478
[(line, revision_id) for line in line_iterator]))
2479
revision_keys = [(r,) for r in revision_ids]
2480
def _filtered_inv_stream():
2481
source_vf = from_repo.inventories
2482
stream = source_vf.get_record_stream(revision_keys,
2484
for record in stream:
2485
if record.storage_kind == 'absent':
2486
raise errors.NoSuchRevision(from_repo, record.key)
2487
find_text_keys_from_content(record)
2489
self._text_keys = content_text_keys - parent_text_keys
2490
return ('inventories', _filtered_inv_stream())
2492
def _get_text_stream(self):
2493
# Note: We know we don't have to handle adding root keys, because both
2494
# the source and target are the identical network name.
2495
text_stream = self.from_repository.texts.get_record_stream(
2496
self._text_keys, self._text_fetch_order, False)
2497
return ('texts', text_stream)
2499
def get_stream(self, search):
2500
revision_ids = search.get_keys()
2501
for stream_info in self._fetch_revision_texts(revision_ids):
2503
self._revision_keys = [(rev_id,) for rev_id in revision_ids]
2504
yield self._get_filtered_inv_stream(revision_ids)
2505
yield self._get_text_stream()
2509
class RepositoryFormatPack(MetaDirRepositoryFormat):
1843
class RepositoryFormatPack(MetaDirVersionedFileRepositoryFormat):
2510
1844
"""Format logic for pack structured repositories.
2512
1846
This repository format has:
2584
1920
_serializer=self._serializer)
2587
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2588
"""A no-subtrees parameterized Pack repository.
2590
This format was introduced in 0.92.
2593
repository_class = KnitPackRepository
2594
_commit_builder_class = PackCommitBuilder
2596
def _serializer(self):
2597
return xml5.serializer_v5
2598
# What index classes to use
2599
index_builder_class = InMemoryGraphIndex
2600
index_class = GraphIndex
2602
def _get_matching_bzrdir(self):
2603
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2605
def _ignore_setting_bzrdir(self, format):
2608
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2610
def get_format_string(self):
2611
"""See RepositoryFormat.get_format_string()."""
2612
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2614
def get_format_description(self):
2615
"""See RepositoryFormat.get_format_description()."""
2616
return "Packs containing knits without subtree support"
2619
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2620
"""A subtrees parameterized Pack repository.
2622
This repository format uses the xml7 serializer to get:
2623
- support for recording full info about the tree root
2624
- support for recording tree-references
2626
This format was introduced in 0.92.
2629
repository_class = KnitPackRepository
2630
_commit_builder_class = PackRootCommitBuilder
2631
rich_root_data = True
2633
supports_tree_reference = True
2635
def _serializer(self):
2636
return xml7.serializer_v7
2637
# What index classes to use
2638
index_builder_class = InMemoryGraphIndex
2639
index_class = GraphIndex
2641
def _get_matching_bzrdir(self):
2642
return bzrdir.format_registry.make_bzrdir(
2643
'pack-0.92-subtree')
2645
def _ignore_setting_bzrdir(self, format):
2648
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2650
def get_format_string(self):
2651
"""See RepositoryFormat.get_format_string()."""
2652
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2654
def get_format_description(self):
2655
"""See RepositoryFormat.get_format_description()."""
2656
return "Packs containing knits with subtree support\n"
2659
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2660
"""A rich-root, no subtrees parameterized Pack repository.
2662
This repository format uses the xml6 serializer to get:
2663
- support for recording full info about the tree root
2665
This format was introduced in 1.0.
2668
repository_class = KnitPackRepository
2669
_commit_builder_class = PackRootCommitBuilder
2670
rich_root_data = True
2671
supports_tree_reference = False
2673
def _serializer(self):
2674
return xml6.serializer_v6
2675
# What index classes to use
2676
index_builder_class = InMemoryGraphIndex
2677
index_class = GraphIndex
2679
def _get_matching_bzrdir(self):
2680
return bzrdir.format_registry.make_bzrdir(
2683
def _ignore_setting_bzrdir(self, format):
2686
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2688
def get_format_string(self):
2689
"""See RepositoryFormat.get_format_string()."""
2690
return ("Bazaar pack repository format 1 with rich root"
2691
" (needs bzr 1.0)\n")
2693
def get_format_description(self):
2694
"""See RepositoryFormat.get_format_description()."""
2695
return "Packs containing knits with rich root support\n"
2698
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2699
"""Repository that supports external references to allow stacking.
2703
Supports external lookups, which results in non-truncated ghosts after
2704
reconcile compared to pack-0.92 formats.
2707
repository_class = KnitPackRepository
2708
_commit_builder_class = PackCommitBuilder
2709
supports_external_lookups = True
2710
# What index classes to use
2711
index_builder_class = InMemoryGraphIndex
2712
index_class = GraphIndex
2715
def _serializer(self):
2716
return xml5.serializer_v5
2718
def _get_matching_bzrdir(self):
2719
return bzrdir.format_registry.make_bzrdir('1.6')
2721
def _ignore_setting_bzrdir(self, format):
2724
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2726
def get_format_string(self):
2727
"""See RepositoryFormat.get_format_string()."""
2728
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2730
def get_format_description(self):
2731
"""See RepositoryFormat.get_format_description()."""
2732
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2735
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2736
"""A repository with rich roots and stacking.
2738
New in release 1.6.1.
2740
Supports stacking on other repositories, allowing data to be accessed
2741
without being stored locally.
2744
repository_class = KnitPackRepository
2745
_commit_builder_class = PackRootCommitBuilder
2746
rich_root_data = True
2747
supports_tree_reference = False # no subtrees
2748
supports_external_lookups = True
2749
# What index classes to use
2750
index_builder_class = InMemoryGraphIndex
2751
index_class = GraphIndex
2754
def _serializer(self):
2755
return xml6.serializer_v6
2757
def _get_matching_bzrdir(self):
2758
return bzrdir.format_registry.make_bzrdir(
2761
def _ignore_setting_bzrdir(self, format):
2764
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2766
def get_format_string(self):
2767
"""See RepositoryFormat.get_format_string()."""
2768
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2770
def get_format_description(self):
2771
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2774
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2775
"""A repository with rich roots and external references.
2779
Supports external lookups, which results in non-truncated ghosts after
2780
reconcile compared to pack-0.92 formats.
2782
This format was deprecated because the serializer it uses accidentally
2783
supported subtrees, when the format was not intended to. This meant that
2784
someone could accidentally fetch from an incorrect repository.
2787
repository_class = KnitPackRepository
2788
_commit_builder_class = PackRootCommitBuilder
2789
rich_root_data = True
2790
supports_tree_reference = False # no subtrees
2792
supports_external_lookups = True
2793
# What index classes to use
2794
index_builder_class = InMemoryGraphIndex
2795
index_class = GraphIndex
2798
def _serializer(self):
2799
return xml7.serializer_v7
2801
def _get_matching_bzrdir(self):
2802
matching = bzrdir.format_registry.make_bzrdir(
2804
matching.repository_format = self
2807
def _ignore_setting_bzrdir(self, format):
2810
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2812
def get_format_string(self):
2813
"""See RepositoryFormat.get_format_string()."""
2814
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2816
def get_format_description(self):
2817
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2821
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2822
"""A repository with stacking and btree indexes,
2823
without rich roots or subtrees.
2825
This is equivalent to pack-1.6 with B+Tree indices.
2828
repository_class = KnitPackRepository
2829
_commit_builder_class = PackCommitBuilder
2830
supports_external_lookups = True
2831
# What index classes to use
2832
index_builder_class = BTreeBuilder
2833
index_class = BTreeGraphIndex
2836
def _serializer(self):
2837
return xml5.serializer_v5
2839
def _get_matching_bzrdir(self):
2840
return bzrdir.format_registry.make_bzrdir('1.9')
2842
def _ignore_setting_bzrdir(self, format):
2845
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2847
def get_format_string(self):
2848
"""See RepositoryFormat.get_format_string()."""
2849
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2851
def get_format_description(self):
2852
"""See RepositoryFormat.get_format_description()."""
2853
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2856
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2857
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2859
1.6-rich-root with B+Tree indices.
2862
repository_class = KnitPackRepository
2863
_commit_builder_class = PackRootCommitBuilder
2864
rich_root_data = True
2865
supports_tree_reference = False # no subtrees
2866
supports_external_lookups = True
2867
# What index classes to use
2868
index_builder_class = BTreeBuilder
2869
index_class = BTreeGraphIndex
2872
def _serializer(self):
2873
return xml6.serializer_v6
2875
def _get_matching_bzrdir(self):
2876
return bzrdir.format_registry.make_bzrdir(
2879
def _ignore_setting_bzrdir(self, format):
2882
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2884
def get_format_string(self):
2885
"""See RepositoryFormat.get_format_string()."""
2886
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2888
def get_format_description(self):
2889
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2892
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2893
"""A subtrees development repository.
2895
This format should be retained until the second release after bzr 1.7.
2897
1.6.1-subtree[as it might have been] with B+Tree indices.
2899
This is [now] retained until we have a CHK based subtree format in
2903
repository_class = KnitPackRepository
2904
_commit_builder_class = PackRootCommitBuilder
2905
rich_root_data = True
2907
supports_tree_reference = True
2908
supports_external_lookups = True
2909
# What index classes to use
2910
index_builder_class = BTreeBuilder
2911
index_class = BTreeGraphIndex
2914
def _serializer(self):
2915
return xml7.serializer_v7
2917
def _get_matching_bzrdir(self):
2918
return bzrdir.format_registry.make_bzrdir(
2919
'development-subtree')
2921
def _ignore_setting_bzrdir(self, format):
2924
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2926
def get_format_string(self):
2927
"""See RepositoryFormat.get_format_string()."""
2928
return ("Bazaar development format 2 with subtree support "
2929
"(needs bzr.dev from before 1.8)\n")
2931
def get_format_description(self):
2932
"""See RepositoryFormat.get_format_description()."""
2933
return ("Development repository format, currently the same as "
2934
"1.6.1-subtree with B+Tree indices.\n")
1923
class RetryPackOperations(errors.RetryWithNewPacks):
1924
"""Raised when we are packing and we find a missing file.
1926
Meant as a signaling exception, to tell the RepositoryPackCollection.pack
1927
code it should try again.
1930
internal_error = True
1932
_fmt = ("Pack files have changed, reload and try pack again."
1933
" context: %(context)s %(orig_error)s")
1936
class _DirectPackAccess(object):
1937
"""Access to data in one or more packs with less translation."""
1939
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
1940
"""Create a _DirectPackAccess object.
1942
:param index_to_packs: A dict mapping index objects to the transport
1943
and file names for obtaining data.
1944
:param reload_func: A function to call if we determine that the pack
1945
files have moved and we need to reload our caches. See
1946
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
1948
self._container_writer = None
1949
self._write_index = None
1950
self._indices = index_to_packs
1951
self._reload_func = reload_func
1952
self._flush_func = flush_func
1954
def add_raw_records(self, key_sizes, raw_data):
1955
"""Add raw knit bytes to a storage area.
1957
The data is spooled to the container writer in one bytes-record per
1960
:param sizes: An iterable of tuples containing the key and size of each
1962
:param raw_data: A bytestring containing the data.
1963
:return: A list of memos to retrieve the record later. Each memo is an
1964
opaque index memo. For _DirectPackAccess the memo is (index, pos,
1965
length), where the index field is the write_index object supplied
1966
to the PackAccess object.
1968
if type(raw_data) is not str:
1969
raise AssertionError(
1970
'data must be plain bytes was %s' % type(raw_data))
1973
for key, size in key_sizes:
1974
p_offset, p_length = self._container_writer.add_bytes_record(
1975
raw_data[offset:offset+size], [])
1977
result.append((self._write_index, p_offset, p_length))
1981
"""Flush pending writes on this access object.
1983
This will flush any buffered writes to a NewPack.
1985
if self._flush_func is not None:
1988
def get_raw_records(self, memos_for_retrieval):
1989
"""Get the raw bytes for a records.
1991
:param memos_for_retrieval: An iterable containing the (index, pos,
1992
length) memo for retrieving the bytes. The Pack access method
1993
looks up the pack to use for a given record in its index_to_pack
1995
:return: An iterator over the bytes of the records.
1997
# first pass, group into same-index requests
1999
current_index = None
2000
for (index, offset, length) in memos_for_retrieval:
2001
if current_index == index:
2002
current_list.append((offset, length))
2004
if current_index is not None:
2005
request_lists.append((current_index, current_list))
2006
current_index = index
2007
current_list = [(offset, length)]
2008
# handle the last entry
2009
if current_index is not None:
2010
request_lists.append((current_index, current_list))
2011
for index, offsets in request_lists:
2013
transport, path = self._indices[index]
2015
# A KeyError here indicates that someone has triggered an index
2016
# reload, and this index has gone missing, we need to start
2018
if self._reload_func is None:
2019
# If we don't have a _reload_func there is nothing that can
2022
raise errors.RetryWithNewPacks(index,
2023
reload_occurred=True,
2024
exc_info=sys.exc_info())
2026
reader = pack.make_readv_reader(transport, path, offsets)
2027
for names, read_func in reader.iter_records():
2028
yield read_func(None)
2029
except errors.NoSuchFile:
2030
# A NoSuchFile error indicates that a pack file has gone
2031
# missing on disk, we need to trigger a reload, and start over.
2032
if self._reload_func is None:
2034
raise errors.RetryWithNewPacks(transport.abspath(path),
2035
reload_occurred=False,
2036
exc_info=sys.exc_info())
2038
def set_writer(self, writer, index, transport_packname):
2039
"""Set a writer to use for adding data."""
2040
if index is not None:
2041
self._indices[index] = transport_packname
2042
self._container_writer = writer
2043
self._write_index = index
2045
def reload_or_raise(self, retry_exc):
2046
"""Try calling the reload function, or re-raise the original exception.
2048
This should be called after _DirectPackAccess raises a
2049
RetryWithNewPacks exception. This function will handle the common logic
2050
of determining when the error is fatal versus being temporary.
2051
It will also make sure that the original exception is raised, rather
2052
than the RetryWithNewPacks exception.
2054
If this function returns, then the calling function should retry
2055
whatever operation was being performed. Otherwise an exception will
2058
:param retry_exc: A RetryWithNewPacks exception.
2061
if self._reload_func is None:
2063
elif not self._reload_func():
2064
# The reload claimed that nothing changed
2065
if not retry_exc.reload_occurred:
2066
# If there wasn't an earlier reload, then we really were
2067
# expecting to find changes. We didn't find them, so this is a
2071
exc_class, exc_value, exc_traceback = retry_exc.exc_info
2072
raise exc_class, exc_value, exc_traceback
added
removed
bzrlib/repofmt/pack_repo.py
