827
727
Sets self._text_filter appropriately.
829
# select inventory keys
830
inv_keys = self._revision_keys # currently the same keyspace, and note that
831
# querying for keys here could introduce a bug where an inventory item
832
# is missed, so do not change it to query separately without cross
833
# checking like the text key check below.
834
inventory_index_map, inventory_indices = self._pack_map_and_index_list(
836
inv_nodes = self._index_contents(inventory_indices, inv_keys)
837
# copy inventory keys and adjust values
838
# XXX: Should be a helper function to allow different inv representation
840
self.pb.update("Copying inventory texts", 2)
841
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
842
# Only grab the output lines if we will be processing them
843
output_lines = bool(self.revision_ids)
844
inv_lines = self._copy_nodes_graph(inventory_index_map,
845
self.new_pack._writer, self.new_pack.inventory_index,
846
readv_group_iter, total_items, output_lines=output_lines)
847
if self.revision_ids:
848
self._process_inventory_lines(inv_lines)
850
# eat the iterator to cause it to execute.
852
self._text_filter = None
853
if 'pack' in debug.debug_flags:
854
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
855
time.ctime(), self._pack_collection._upload_transport.base,
856
self.new_pack.random_name,
857
self.new_pack.inventory_index.key_count(),
858
time.time() - self.new_pack.start_time)
729
raise NotImplementedError(self._copy_inventory_texts)
860
731
def _copy_text_texts(self):
862
text_index_map, text_nodes = self._get_text_nodes()
863
if self._text_filter is not None:
864
# We could return the keys copied as part of the return value from
865
# _copy_nodes_graph but this doesn't work all that well with the
866
# need to get line output too, so we check separately, and as we're
867
# going to buffer everything anyway, we check beforehand, which
868
# saves reading knit data over the wire when we know there are
870
text_nodes = set(text_nodes)
871
present_text_keys = set(_node[1] for _node in text_nodes)
872
missing_text_keys = set(self._text_filter) - present_text_keys
873
if missing_text_keys:
874
# TODO: raise a specific error that can handle many missing
876
mutter("missing keys during fetch: %r", missing_text_keys)
877
a_missing_key = missing_text_keys.pop()
878
raise errors.RevisionNotPresent(a_missing_key[1],
880
# copy text keys and adjust values
881
self.pb.update("Copying content texts", 3)
882
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
883
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
884
self.new_pack.text_index, readv_group_iter, total_items))
885
self._log_copied_texts()
732
raise NotImplementedError(self._copy_text_texts)
887
734
def _create_pack_from_packs(self):
888
self.pb.update("Opening pack", 0, 5)
889
self.new_pack = self.open_pack()
890
new_pack = self.new_pack
891
# buffer data - we won't be reading-back during the pack creation and
892
# this makes a significant difference on sftp pushes.
893
new_pack.set_write_cache_size(1024*1024)
894
if 'pack' in debug.debug_flags:
895
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
896
for a_pack in self.packs]
897
if self.revision_ids is not None:
898
rev_count = len(self.revision_ids)
901
mutter('%s: create_pack: creating pack from source packs: '
902
'%s%s %s revisions wanted %s t=0',
903
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
904
plain_pack_list, rev_count)
905
self._copy_revision_texts()
906
self._copy_inventory_texts()
907
self._copy_text_texts()
908
# select signature keys
909
signature_filter = self._revision_keys # same keyspace
910
signature_index_map, signature_indices = self._pack_map_and_index_list(
912
signature_nodes = self._index_contents(signature_indices,
914
# copy signature keys and adjust values
915
self.pb.update("Copying signature texts", 4)
916
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
917
new_pack.signature_index)
918
if 'pack' in debug.debug_flags:
919
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
920
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
921
new_pack.signature_index.key_count(),
922
time.time() - new_pack.start_time)
924
# NB XXX: how to check CHK references are present? perhaps by yielding
925
# the items? How should that interact with stacked repos?
926
if new_pack.chk_index is not None:
928
if 'pack' in debug.debug_flags:
929
mutter('%s: create_pack: chk content copied: %s%s %d items t+%6.3fs',
930
time.ctime(), self._pack_collection._upload_transport.base,
931
new_pack.random_name,
932
new_pack.chk_index.key_count(),
933
time.time() - new_pack.start_time)
934
new_pack._check_references()
935
if not self._use_pack(new_pack):
938
self.pb.update("Finishing pack", 5)
940
self._pack_collection.allocate(new_pack)
943
def _copy_chks(self, refs=None):
944
# XXX: Todo, recursive follow-pointers facility when fetching some
946
chk_index_map, chk_indices = self._pack_map_and_index_list(
948
chk_nodes = self._index_contents(chk_indices, refs)
950
# TODO: This isn't strictly tasteful as we are accessing some private
951
# variables (_serializer). Perhaps a better way would be to have
952
# Repository._deserialise_chk_node()
953
search_key_func = chk_map.search_key_registry.get(
954
self._pack_collection.repo._serializer.search_key_name)
955
def accumlate_refs(lines):
956
# XXX: move to a generic location
958
bytes = ''.join(lines)
959
node = chk_map._deserialise(bytes, ("unknown",), search_key_func)
960
new_refs.update(node.refs())
961
self._copy_nodes(chk_nodes, chk_index_map, self.new_pack._writer,
962
self.new_pack.chk_index, output_lines=accumlate_refs)
965
def _copy_nodes(self, nodes, index_map, writer, write_index,
967
"""Copy knit nodes between packs with no graph references.
969
:param output_lines: Output full texts of copied items.
971
pb = ui.ui_factory.nested_progress_bar()
973
return self._do_copy_nodes(nodes, index_map, writer,
974
write_index, pb, output_lines=output_lines)
978
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb,
980
# for record verification
981
knit = KnitVersionedFiles(None, None)
982
# plan a readv on each source pack:
984
nodes = sorted(nodes)
985
# how to map this into knit.py - or knit.py into this?
986
# we don't want the typical knit logic, we want grouping by pack
987
# at this point - perhaps a helper library for the following code
988
# duplication points?
990
for index, key, value in nodes:
991
if index not in request_groups:
992
request_groups[index] = []
993
request_groups[index].append((key, value))
995
pb.update("Copied record", record_index, len(nodes))
996
for index, items in request_groups.iteritems():
997
pack_readv_requests = []
998
for key, value in items:
999
# ---- KnitGraphIndex.get_position
1000
bits = value[1:].split(' ')
1001
offset, length = int(bits[0]), int(bits[1])
1002
pack_readv_requests.append((offset, length, (key, value[0])))
1003
# linear scan up the pack
1004
pack_readv_requests.sort()
1006
pack_obj = index_map[index]
1007
transport, path = pack_obj.access_tuple()
1009
reader = pack.make_readv_reader(transport, path,
1010
[offset[0:2] for offset in pack_readv_requests])
1011
except errors.NoSuchFile:
1012
if self._reload_func is not None:
1015
for (names, read_func), (_1, _2, (key, eol_flag)) in \
1016
izip(reader.iter_records(), pack_readv_requests):
1017
raw_data = read_func(None)
1018
# check the header only
1019
if output_lines is not None:
1020
output_lines(knit._parse_record(key[-1], raw_data)[0])
1022
df, _ = knit._parse_record_header(key, raw_data)
1024
pos, size = writer.add_bytes_record(raw_data, names)
1025
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
1026
pb.update("Copied record", record_index)
1029
def _copy_nodes_graph(self, index_map, writer, write_index,
1030
readv_group_iter, total_items, output_lines=False):
1031
"""Copy knit nodes between packs.
1033
:param output_lines: Return lines present in the copied data as
1034
an iterator of line,version_id.
1036
pb = ui.ui_factory.nested_progress_bar()
1038
for result in self._do_copy_nodes_graph(index_map, writer,
1039
write_index, output_lines, pb, readv_group_iter, total_items):
1042
# Python 2.4 does not permit try:finally: in a generator.
1048
def _do_copy_nodes_graph(self, index_map, writer, write_index,
1049
output_lines, pb, readv_group_iter, total_items):
1050
# for record verification
1051
knit = KnitVersionedFiles(None, None)
1052
# for line extraction when requested (inventories only)
1054
factory = KnitPlainFactory()
1056
pb.update("Copied record", record_index, total_items)
1057
for index, readv_vector, node_vector in readv_group_iter:
1059
pack_obj = index_map[index]
1060
transport, path = pack_obj.access_tuple()
1062
reader = pack.make_readv_reader(transport, path, readv_vector)
1063
except errors.NoSuchFile:
1064
if self._reload_func is not None:
1067
for (names, read_func), (key, eol_flag, references) in \
1068
izip(reader.iter_records(), node_vector):
1069
raw_data = read_func(None)
1071
# read the entire thing
1072
content, _ = knit._parse_record(key[-1], raw_data)
1073
if len(references[-1]) == 0:
1074
line_iterator = factory.get_fulltext_content(content)
1076
line_iterator = factory.get_linedelta_content(content)
1077
for line in line_iterator:
1080
# check the header only
1081
df, _ = knit._parse_record_header(key, raw_data)
1083
pos, size = writer.add_bytes_record(raw_data, names)
1084
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1085
pb.update("Copied record", record_index)
1088
def _get_text_nodes(self):
1089
text_index_map, text_indices = self._pack_map_and_index_list(
1091
return text_index_map, self._index_contents(text_indices,
1094
def _least_readv_node_readv(self, nodes):
1095
"""Generate request groups for nodes using the least readv's.
1097
:param nodes: An iterable of graph index nodes.
1098
:return: Total node count and an iterator of the data needed to perform
1099
readvs to obtain the data for nodes. Each item yielded by the
1100
iterator is a tuple with:
1101
index, readv_vector, node_vector. readv_vector is a list ready to
1102
hand to the transport readv method, and node_vector is a list of
1103
(key, eol_flag, references) for the node retrieved by the
1104
matching readv_vector.
1106
# group by pack so we do one readv per pack
1107
nodes = sorted(nodes)
1110
for index, key, value, references in nodes:
1111
if index not in request_groups:
1112
request_groups[index] = []
1113
request_groups[index].append((key, value, references))
1115
for index, items in request_groups.iteritems():
1116
pack_readv_requests = []
1117
for key, value, references in items:
1118
# ---- KnitGraphIndex.get_position
1119
bits = value[1:].split(' ')
1120
offset, length = int(bits[0]), int(bits[1])
1121
pack_readv_requests.append(
1122
((offset, length), (key, value[0], references)))
1123
# linear scan up the pack to maximum range combining.
1124
pack_readv_requests.sort()
1125
# split out the readv and the node data.
1126
pack_readv = [readv for readv, node in pack_readv_requests]
1127
node_vector = [node for readv, node in pack_readv_requests]
1128
result.append((index, pack_readv, node_vector))
1129
return total, result
735
raise NotImplementedError(self._create_pack_from_packs)
1131
737
def _log_copied_texts(self):
1132
738
if 'pack' in debug.debug_flags:
1164
751
return new_pack.data_inserted()
1167
class OptimisingPacker(Packer):
1168
"""A packer which spends more time to create better disk layouts."""
1170
def _revision_node_readv(self, revision_nodes):
1171
"""Return the total revisions and the readv's to issue.
1173
This sort places revisions in topological order with the ancestors
1176
:param revision_nodes: The revision index contents for the packs being
1177
incorporated into the new pack.
1178
:return: As per _least_readv_node_readv.
1180
# build an ancestors dict
1183
for index, key, value, references in revision_nodes:
1184
ancestors[key] = references[0]
1185
by_key[key] = (index, value, references)
1186
order = tsort.topo_sort(ancestors)
1188
# Single IO is pathological, but it will work as a starting point.
1190
for key in reversed(order):
1191
index, value, references = by_key[key]
1192
# ---- KnitGraphIndex.get_position
1193
bits = value[1:].split(' ')
1194
offset, length = int(bits[0]), int(bits[1])
1196
(index, [(offset, length)], [(key, value[0], references)]))
1197
# TODO: combine requests in the same index that are in ascending order.
1198
return total, requests
1200
def open_pack(self):
1201
"""Open a pack for the pack we are creating."""
1202
new_pack = super(OptimisingPacker, self).open_pack()
1203
# Turn on the optimization flags for all the index builders.
1204
new_pack.revision_index.set_optimize(for_size=True)
1205
new_pack.inventory_index.set_optimize(for_size=True)
1206
new_pack.text_index.set_optimize(for_size=True)
1207
new_pack.signature_index.set_optimize(for_size=True)
1211
class ReconcilePacker(Packer):
1212
"""A packer which regenerates indices etc as it copies.
1214
This is used by ``bzr reconcile`` to cause parent text pointers to be
1218
def _extra_init(self):
1219
self._data_changed = False
1221
def _process_inventory_lines(self, inv_lines):
1222
"""Generate a text key reference map rather for reconciling with."""
1223
repo = self._pack_collection.repo
1224
refs = repo._find_text_key_references_from_xml_inventory_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 it's 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
754
class RepositoryPackCollection(object):
1354
755
"""Management of packs within a repository.
1356
757
:ivar _names: map of {pack_name: (index_size,)}
1359
pack_factory = NewPack
1360
resumed_pack_factory = ResumedPack
761
resumed_pack_factory = None
762
normal_packer_class = None
763
optimising_packer_class = None
1362
765
def __init__(self, repo, transport, index_transport, upload_transport,
1363
766
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
2221
1657
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
2223
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2224
_commit_builder_class, _serializer)
2225
index_transport = self._transport.clone('indices')
2226
self._pack_collection = RepositoryPackCollection(self, self._transport,
2228
self._transport.clone('upload'),
2229
self._transport.clone('packs'),
2230
_format.index_builder_class,
2231
_format.index_class,
2232
use_chk_index=self._format.supports_chks,
2234
self.inventories = KnitVersionedFiles(
2235
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2236
add_callback=self._pack_collection.inventory_index.add_callback,
2237
deltas=True, parents=True, is_locked=self.is_locked),
2238
data_access=self._pack_collection.inventory_index.data_access,
2239
max_delta_chain=200)
2240
self.revisions = KnitVersionedFiles(
2241
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2242
add_callback=self._pack_collection.revision_index.add_callback,
2243
deltas=False, parents=True, is_locked=self.is_locked,
2244
track_external_parent_refs=True),
2245
data_access=self._pack_collection.revision_index.data_access,
2247
self.signatures = KnitVersionedFiles(
2248
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2249
add_callback=self._pack_collection.signature_index.add_callback,
2250
deltas=False, parents=False, is_locked=self.is_locked),
2251
data_access=self._pack_collection.signature_index.data_access,
2253
self.texts = KnitVersionedFiles(
2254
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2255
add_callback=self._pack_collection.text_index.add_callback,
2256
deltas=True, parents=True, is_locked=self.is_locked),
2257
data_access=self._pack_collection.text_index.data_access,
2258
max_delta_chain=200)
2259
if _format.supports_chks:
2260
# No graph, no compression:- references from chks are between
2261
# different objects not temporal versions of the same; and without
2262
# some sort of temporal structure knit compression will just fail.
2263
self.chk_bytes = KnitVersionedFiles(
2264
_KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2265
add_callback=self._pack_collection.chk_index.add_callback,
2266
deltas=False, parents=False, is_locked=self.is_locked),
2267
data_access=self._pack_collection.chk_index.data_access,
2270
self.chk_bytes = None
2271
# True when the repository object is 'write locked' (as opposed to the
2272
# physical lock only taken out around changes to the pack-names list.)
2273
# Another way to represent this would be a decorator around the control
2274
# files object that presents logical locks as physical ones - if this
2275
# gets ugly consider that alternative design. RBC 20071011
2276
self._write_lock_count = 0
2277
self._transaction = None
2279
self._reconcile_does_inventory_gc = True
1659
MetaDirRepository.__init__(self, _format, a_bzrdir, control_files)
1660
self._commit_builder_class = _commit_builder_class
1661
self._serializer = _serializer
2280
1662
self._reconcile_fixes_text_parents = True
2281
self._reconcile_backsup_inventory = False
2283
def _warn_if_deprecated(self, branch=None):
2284
# This class isn't deprecated, but one sub-format is
2285
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2286
super(KnitPackRepository, self)._warn_if_deprecated(branch)
1665
def _all_revision_ids(self):
1666
"""See Repository.all_revision_ids()."""
1667
return [key[0] for key in self.revisions.keys()]
2288
1669
def _abort_write_group(self):
2289
1670
self.revisions._index._key_dependencies.clear()
2290
1671
self._pack_collection._abort_write_group()
2292
def _get_source(self, to_format):
2293
if to_format.network_name() == self._format.network_name():
2294
return KnitPackStreamSource(self, to_format)
2295
return super(KnitPackRepository, self)._get_source(to_format)
2297
1673
def _make_parents_provider(self):
2298
1674
return graph.CachingParentsProvider(self)
2424
class KnitPackStreamSource(StreamSource):
2425
"""A StreamSource used to transfer data between same-format KnitPack repos.
2427
This source assumes:
2428
1) Same serialization format for all objects
2429
2) Same root information
2430
3) XML format inventories
2431
4) Atomic inserts (so we can stream inventory texts before text
2436
def __init__(self, from_repository, to_format):
2437
super(KnitPackStreamSource, self).__init__(from_repository, to_format)
2438
self._text_keys = None
2439
self._text_fetch_order = 'unordered'
2441
def _get_filtered_inv_stream(self, revision_ids):
2442
from_repo = self.from_repository
2443
parent_ids = from_repo._find_parent_ids_of_revisions(revision_ids)
2444
parent_keys = [(p,) for p in parent_ids]
2445
find_text_keys = from_repo._find_text_key_references_from_xml_inventory_lines
2446
parent_text_keys = set(find_text_keys(
2447
from_repo._inventory_xml_lines_for_keys(parent_keys)))
2448
content_text_keys = set()
2449
knit = KnitVersionedFiles(None, None)
2450
factory = KnitPlainFactory()
2451
def find_text_keys_from_content(record):
2452
if record.storage_kind not in ('knit-delta-gz', 'knit-ft-gz'):
2453
raise ValueError("Unknown content storage kind for"
2454
" inventory text: %s" % (record.storage_kind,))
2455
# It's a knit record, it has a _raw_record field (even if it was
2456
# reconstituted from a network stream).
2457
raw_data = record._raw_record
2458
# read the entire thing
2459
revision_id = record.key[-1]
2460
content, _ = knit._parse_record(revision_id, raw_data)
2461
if record.storage_kind == 'knit-delta-gz':
2462
line_iterator = factory.get_linedelta_content(content)
2463
elif record.storage_kind == 'knit-ft-gz':
2464
line_iterator = factory.get_fulltext_content(content)
2465
content_text_keys.update(find_text_keys(
2466
[(line, revision_id) for line in line_iterator]))
2467
revision_keys = [(r,) for r in revision_ids]
2468
def _filtered_inv_stream():
2469
source_vf = from_repo.inventories
2470
stream = source_vf.get_record_stream(revision_keys,
2472
for record in stream:
2473
if record.storage_kind == 'absent':
2474
raise errors.NoSuchRevision(from_repo, record.key)
2475
find_text_keys_from_content(record)
2477
self._text_keys = content_text_keys - parent_text_keys
2478
return ('inventories', _filtered_inv_stream())
2480
def _get_text_stream(self):
2481
# Note: We know we don't have to handle adding root keys, because both
2482
# the source and target are the identical network name.
2483
text_stream = self.from_repository.texts.get_record_stream(
2484
self._text_keys, self._text_fetch_order, False)
2485
return ('texts', text_stream)
2487
def get_stream(self, search):
2488
revision_ids = search.get_keys()
2489
for stream_info in self._fetch_revision_texts(revision_ids):
2491
self._revision_keys = [(rev_id,) for rev_id in revision_ids]
2492
yield self._get_filtered_inv_stream(revision_ids)
2493
yield self._get_text_stream()
2497
1809
class RepositoryFormatPack(MetaDirRepositoryFormat):
2498
1810
"""Format logic for pack structured repositories.
2572
1887
_serializer=self._serializer)
2575
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2576
"""A no-subtrees parameterized Pack repository.
2578
This format was introduced in 0.92.
2581
repository_class = KnitPackRepository
2582
_commit_builder_class = PackCommitBuilder
2584
def _serializer(self):
2585
return xml5.serializer_v5
2586
# What index classes to use
2587
index_builder_class = InMemoryGraphIndex
2588
index_class = GraphIndex
2590
def _get_matching_bzrdir(self):
2591
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2593
def _ignore_setting_bzrdir(self, format):
2596
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2598
def get_format_string(self):
2599
"""See RepositoryFormat.get_format_string()."""
2600
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2602
def get_format_description(self):
2603
"""See RepositoryFormat.get_format_description()."""
2604
return "Packs containing knits without subtree support"
2607
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2608
"""A subtrees parameterized Pack repository.
2610
This repository format uses the xml7 serializer to get:
2611
- support for recording full info about the tree root
2612
- support for recording tree-references
2614
This format was introduced in 0.92.
2617
repository_class = KnitPackRepository
2618
_commit_builder_class = PackRootCommitBuilder
2619
rich_root_data = True
2621
supports_tree_reference = True
2623
def _serializer(self):
2624
return xml7.serializer_v7
2625
# What index classes to use
2626
index_builder_class = InMemoryGraphIndex
2627
index_class = GraphIndex
2629
def _get_matching_bzrdir(self):
2630
return bzrdir.format_registry.make_bzrdir(
2631
'pack-0.92-subtree')
2633
def _ignore_setting_bzrdir(self, format):
2636
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2638
def get_format_string(self):
2639
"""See RepositoryFormat.get_format_string()."""
2640
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2642
def get_format_description(self):
2643
"""See RepositoryFormat.get_format_description()."""
2644
return "Packs containing knits with subtree support\n"
2647
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2648
"""A rich-root, no subtrees parameterized Pack repository.
2650
This repository format uses the xml6 serializer to get:
2651
- support for recording full info about the tree root
2653
This format was introduced in 1.0.
2656
repository_class = KnitPackRepository
2657
_commit_builder_class = PackRootCommitBuilder
2658
rich_root_data = True
2659
supports_tree_reference = False
2661
def _serializer(self):
2662
return xml6.serializer_v6
2663
# What index classes to use
2664
index_builder_class = InMemoryGraphIndex
2665
index_class = GraphIndex
2667
def _get_matching_bzrdir(self):
2668
return bzrdir.format_registry.make_bzrdir(
2671
def _ignore_setting_bzrdir(self, format):
2674
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2676
def get_format_string(self):
2677
"""See RepositoryFormat.get_format_string()."""
2678
return ("Bazaar pack repository format 1 with rich root"
2679
" (needs bzr 1.0)\n")
2681
def get_format_description(self):
2682
"""See RepositoryFormat.get_format_description()."""
2683
return "Packs containing knits with rich root support\n"
2686
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2687
"""Repository that supports external references to allow stacking.
2691
Supports external lookups, which results in non-truncated ghosts after
2692
reconcile compared to pack-0.92 formats.
2695
repository_class = KnitPackRepository
2696
_commit_builder_class = PackCommitBuilder
2697
supports_external_lookups = True
2698
# What index classes to use
2699
index_builder_class = InMemoryGraphIndex
2700
index_class = GraphIndex
2703
def _serializer(self):
2704
return xml5.serializer_v5
2706
def _get_matching_bzrdir(self):
2707
return bzrdir.format_registry.make_bzrdir('1.6')
2709
def _ignore_setting_bzrdir(self, format):
2712
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2714
def get_format_string(self):
2715
"""See RepositoryFormat.get_format_string()."""
2716
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2718
def get_format_description(self):
2719
"""See RepositoryFormat.get_format_description()."""
2720
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2723
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2724
"""A repository with rich roots and stacking.
2726
New in release 1.6.1.
2728
Supports stacking on other repositories, allowing data to be accessed
2729
without being stored locally.
2732
repository_class = KnitPackRepository
2733
_commit_builder_class = PackRootCommitBuilder
2734
rich_root_data = True
2735
supports_tree_reference = False # no subtrees
2736
supports_external_lookups = True
2737
# What index classes to use
2738
index_builder_class = InMemoryGraphIndex
2739
index_class = GraphIndex
2742
def _serializer(self):
2743
return xml6.serializer_v6
2745
def _get_matching_bzrdir(self):
2746
return bzrdir.format_registry.make_bzrdir(
2749
def _ignore_setting_bzrdir(self, format):
2752
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2754
def get_format_string(self):
2755
"""See RepositoryFormat.get_format_string()."""
2756
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2758
def get_format_description(self):
2759
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2762
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2763
"""A repository with rich roots and external references.
2767
Supports external lookups, which results in non-truncated ghosts after
2768
reconcile compared to pack-0.92 formats.
2770
This format was deprecated because the serializer it uses accidentally
2771
supported subtrees, when the format was not intended to. This meant that
2772
someone could accidentally fetch from an incorrect repository.
2775
repository_class = KnitPackRepository
2776
_commit_builder_class = PackRootCommitBuilder
2777
rich_root_data = True
2778
supports_tree_reference = False # no subtrees
2780
supports_external_lookups = True
2781
# What index classes to use
2782
index_builder_class = InMemoryGraphIndex
2783
index_class = GraphIndex
2786
def _serializer(self):
2787
return xml7.serializer_v7
2789
def _get_matching_bzrdir(self):
2790
matching = bzrdir.format_registry.make_bzrdir(
2792
matching.repository_format = self
2795
def _ignore_setting_bzrdir(self, format):
2798
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2800
def get_format_string(self):
2801
"""See RepositoryFormat.get_format_string()."""
2802
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2804
def get_format_description(self):
2805
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2809
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2810
"""A repository with stacking and btree indexes,
2811
without rich roots or subtrees.
2813
This is equivalent to pack-1.6 with B+Tree indices.
2816
repository_class = KnitPackRepository
2817
_commit_builder_class = PackCommitBuilder
2818
supports_external_lookups = True
2819
# What index classes to use
2820
index_builder_class = BTreeBuilder
2821
index_class = BTreeGraphIndex
2824
def _serializer(self):
2825
return xml5.serializer_v5
2827
def _get_matching_bzrdir(self):
2828
return bzrdir.format_registry.make_bzrdir('1.9')
2830
def _ignore_setting_bzrdir(self, format):
2833
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2835
def get_format_string(self):
2836
"""See RepositoryFormat.get_format_string()."""
2837
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2839
def get_format_description(self):
2840
"""See RepositoryFormat.get_format_description()."""
2841
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2844
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2845
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2847
1.6-rich-root with B+Tree indices.
2850
repository_class = KnitPackRepository
2851
_commit_builder_class = PackRootCommitBuilder
2852
rich_root_data = True
2853
supports_tree_reference = False # no subtrees
2854
supports_external_lookups = True
2855
# What index classes to use
2856
index_builder_class = BTreeBuilder
2857
index_class = BTreeGraphIndex
2860
def _serializer(self):
2861
return xml6.serializer_v6
2863
def _get_matching_bzrdir(self):
2864
return bzrdir.format_registry.make_bzrdir(
2867
def _ignore_setting_bzrdir(self, format):
2870
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2872
def get_format_string(self):
2873
"""See RepositoryFormat.get_format_string()."""
2874
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2876
def get_format_description(self):
2877
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2880
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2881
"""A subtrees development repository.
2883
This format should be retained until the second release after bzr 1.7.
2885
1.6.1-subtree[as it might have been] with B+Tree indices.
2887
This is [now] retained until we have a CHK based subtree format in
2891
repository_class = KnitPackRepository
2892
_commit_builder_class = PackRootCommitBuilder
2893
rich_root_data = True
2895
supports_tree_reference = True
2896
supports_external_lookups = True
2897
# What index classes to use
2898
index_builder_class = BTreeBuilder
2899
index_class = BTreeGraphIndex
2902
def _serializer(self):
2903
return xml7.serializer_v7
2905
def _get_matching_bzrdir(self):
2906
return bzrdir.format_registry.make_bzrdir(
2907
'development-subtree')
2909
def _ignore_setting_bzrdir(self, format):
2912
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2914
def get_format_string(self):
2915
"""See RepositoryFormat.get_format_string()."""
2916
return ("Bazaar development format 2 with subtree support "
2917
"(needs bzr.dev from before 1.8)\n")
2919
def get_format_description(self):
2920
"""See RepositoryFormat.get_format_description()."""
2921
return ("Development repository format, currently the same as "
2922
"1.6.1-subtree with B+Tree indices.\n")
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
added
removed
bzrlib/repofmt/pack_repo.py
