832
741
Sets self._text_filter appropriately.
834
# select inventory keys
835
inv_keys = self._revision_keys # currently the same keyspace, and note that
836
# querying for keys here could introduce a bug where an inventory item
837
# is missed, so do not change it to query separately without cross
838
# checking like the text key check below.
839
inventory_index_map, inventory_indices = self._pack_map_and_index_list(
841
inv_nodes = self._index_contents(inventory_indices, inv_keys)
842
# copy inventory keys and adjust values
843
# XXX: Should be a helper function to allow different inv representation
845
self.pb.update("Copying inventory texts", 2)
846
total_items, readv_group_iter = self._least_readv_node_readv(inv_nodes)
847
# Only grab the output lines if we will be processing them
848
output_lines = bool(self.revision_ids)
849
inv_lines = self._copy_nodes_graph(inventory_index_map,
850
self.new_pack._writer, self.new_pack.inventory_index,
851
readv_group_iter, total_items, output_lines=output_lines)
852
if self.revision_ids:
853
self._process_inventory_lines(inv_lines)
855
# eat the iterator to cause it to execute.
857
self._text_filter = None
858
if 'pack' in debug.debug_flags:
859
mutter('%s: create_pack: inventories copied: %s%s %d items t+%6.3fs',
860
time.ctime(), self._pack_collection._upload_transport.base,
861
self.new_pack.random_name,
862
self.new_pack.inventory_index.key_count(),
863
time.time() - self.new_pack.start_time)
743
raise NotImplementedError(self._copy_inventory_texts)
865
745
def _copy_text_texts(self):
867
text_index_map, text_nodes = self._get_text_nodes()
868
if self._text_filter is not None:
869
# We could return the keys copied as part of the return value from
870
# _copy_nodes_graph but this doesn't work all that well with the
871
# need to get line output too, so we check separately, and as we're
872
# going to buffer everything anyway, we check beforehand, which
873
# saves reading knit data over the wire when we know there are
875
text_nodes = set(text_nodes)
876
present_text_keys = set(_node[1] for _node in text_nodes)
877
missing_text_keys = set(self._text_filter) - present_text_keys
878
if missing_text_keys:
879
# TODO: raise a specific error that can handle many missing
881
mutter("missing keys during fetch: %r", missing_text_keys)
882
a_missing_key = missing_text_keys.pop()
883
raise errors.RevisionNotPresent(a_missing_key[1],
885
# copy text keys and adjust values
886
self.pb.update("Copying content texts", 3)
887
total_items, readv_group_iter = self._least_readv_node_readv(text_nodes)
888
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
889
self.new_pack.text_index, readv_group_iter, total_items))
890
self._log_copied_texts()
746
raise NotImplementedError(self._copy_text_texts)
892
748
def _create_pack_from_packs(self):
893
self.pb.update("Opening pack", 0, 5)
894
self.new_pack = self.open_pack()
895
new_pack = self.new_pack
896
# buffer data - we won't be reading-back during the pack creation and
897
# this makes a significant difference on sftp pushes.
898
new_pack.set_write_cache_size(1024*1024)
899
if 'pack' in debug.debug_flags:
900
plain_pack_list = ['%s%s' % (a_pack.pack_transport.base, a_pack.name)
901
for a_pack in self.packs]
902
if self.revision_ids is not None:
903
rev_count = len(self.revision_ids)
906
mutter('%s: create_pack: creating pack from source packs: '
907
'%s%s %s revisions wanted %s t=0',
908
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
909
plain_pack_list, rev_count)
910
self._copy_revision_texts()
911
self._copy_inventory_texts()
912
self._copy_text_texts()
913
# select signature keys
914
signature_filter = self._revision_keys # same keyspace
915
signature_index_map, signature_indices = self._pack_map_and_index_list(
917
signature_nodes = self._index_contents(signature_indices,
919
# copy signature keys and adjust values
920
self.pb.update("Copying signature texts", 4)
921
self._copy_nodes(signature_nodes, signature_index_map, new_pack._writer,
922
new_pack.signature_index)
923
if 'pack' in debug.debug_flags:
924
mutter('%s: create_pack: revision signatures copied: %s%s %d items t+%6.3fs',
925
time.ctime(), self._pack_collection._upload_transport.base, new_pack.random_name,
926
new_pack.signature_index.key_count(),
927
time.time() - new_pack.start_time)
929
# NB XXX: how to check CHK references are present? perhaps by yielding
930
# the items? How should that interact with stacked repos?
931
if new_pack.chk_index is not None:
933
if 'pack' in debug.debug_flags:
934
mutter('%s: create_pack: chk content copied: %s%s %d items t+%6.3fs',
935
time.ctime(), self._pack_collection._upload_transport.base,
936
new_pack.random_name,
937
new_pack.chk_index.key_count(),
938
time.time() - new_pack.start_time)
939
new_pack._check_references()
940
if not self._use_pack(new_pack):
943
self.pb.update("Finishing pack", 5)
945
self._pack_collection.allocate(new_pack)
948
def _copy_chks(self, refs=None):
949
# XXX: Todo, recursive follow-pointers facility when fetching some
951
chk_index_map, chk_indices = self._pack_map_and_index_list(
953
chk_nodes = self._index_contents(chk_indices, refs)
955
# TODO: This isn't strictly tasteful as we are accessing some private
956
# variables (_serializer). Perhaps a better way would be to have
957
# Repository._deserialise_chk_node()
958
search_key_func = chk_map.search_key_registry.get(
959
self._pack_collection.repo._serializer.search_key_name)
960
def accumlate_refs(lines):
961
# XXX: move to a generic location
963
bytes = ''.join(lines)
964
node = chk_map._deserialise(bytes, ("unknown",), search_key_func)
965
new_refs.update(node.refs())
966
self._copy_nodes(chk_nodes, chk_index_map, self.new_pack._writer,
967
self.new_pack.chk_index, output_lines=accumlate_refs)
970
def _copy_nodes(self, nodes, index_map, writer, write_index,
972
"""Copy knit nodes between packs with no graph references.
974
:param output_lines: Output full texts of copied items.
976
pb = ui.ui_factory.nested_progress_bar()
978
return self._do_copy_nodes(nodes, index_map, writer,
979
write_index, pb, output_lines=output_lines)
983
def _do_copy_nodes(self, nodes, index_map, writer, write_index, pb,
985
# for record verification
986
knit = KnitVersionedFiles(None, None)
987
# plan a readv on each source pack:
989
nodes = sorted(nodes)
990
# how to map this into knit.py - or knit.py into this?
991
# we don't want the typical knit logic, we want grouping by pack
992
# at this point - perhaps a helper library for the following code
993
# duplication points?
995
for index, key, value in nodes:
996
if index not in request_groups:
997
request_groups[index] = []
998
request_groups[index].append((key, value))
1000
pb.update("Copied record", record_index, len(nodes))
1001
for index, items in request_groups.iteritems():
1002
pack_readv_requests = []
1003
for key, value in items:
1004
# ---- KnitGraphIndex.get_position
1005
bits = value[1:].split(' ')
1006
offset, length = int(bits[0]), int(bits[1])
1007
pack_readv_requests.append((offset, length, (key, value[0])))
1008
# linear scan up the pack
1009
pack_readv_requests.sort()
1011
pack_obj = index_map[index]
1012
transport, path = pack_obj.access_tuple()
1014
reader = pack.make_readv_reader(transport, path,
1015
[offset[0:2] for offset in pack_readv_requests])
1016
except errors.NoSuchFile:
1017
if self._reload_func is not None:
1020
for (names, read_func), (_1, _2, (key, eol_flag)) in \
1021
izip(reader.iter_records(), pack_readv_requests):
1022
raw_data = read_func(None)
1023
# check the header only
1024
if output_lines is not None:
1025
output_lines(knit._parse_record(key[-1], raw_data)[0])
1027
df, _ = knit._parse_record_header(key, raw_data)
1029
pos, size = writer.add_bytes_record(raw_data, names)
1030
write_index.add_node(key, eol_flag + "%d %d" % (pos, size))
1031
pb.update("Copied record", record_index)
1034
def _copy_nodes_graph(self, index_map, writer, write_index,
1035
readv_group_iter, total_items, output_lines=False):
1036
"""Copy knit nodes between packs.
1038
:param output_lines: Return lines present in the copied data as
1039
an iterator of line,version_id.
1041
pb = ui.ui_factory.nested_progress_bar()
1043
for result in self._do_copy_nodes_graph(index_map, writer,
1044
write_index, output_lines, pb, readv_group_iter, total_items):
1047
# Python 2.4 does not permit try:finally: in a generator.
1053
def _do_copy_nodes_graph(self, index_map, writer, write_index,
1054
output_lines, pb, readv_group_iter, total_items):
1055
# for record verification
1056
knit = KnitVersionedFiles(None, None)
1057
# for line extraction when requested (inventories only)
1059
factory = KnitPlainFactory()
1061
pb.update("Copied record", record_index, total_items)
1062
for index, readv_vector, node_vector in readv_group_iter:
1064
pack_obj = index_map[index]
1065
transport, path = pack_obj.access_tuple()
1067
reader = pack.make_readv_reader(transport, path, readv_vector)
1068
except errors.NoSuchFile:
1069
if self._reload_func is not None:
1072
for (names, read_func), (key, eol_flag, references) in \
1073
izip(reader.iter_records(), node_vector):
1074
raw_data = read_func(None)
1076
# read the entire thing
1077
content, _ = knit._parse_record(key[-1], raw_data)
1078
if len(references[-1]) == 0:
1079
line_iterator = factory.get_fulltext_content(content)
1081
line_iterator = factory.get_linedelta_content(content)
1082
for line in line_iterator:
1085
# check the header only
1086
df, _ = knit._parse_record_header(key, raw_data)
1088
pos, size = writer.add_bytes_record(raw_data, names)
1089
write_index.add_node(key, eol_flag + "%d %d" % (pos, size), references)
1090
pb.update("Copied record", record_index)
1093
def _get_text_nodes(self):
1094
text_index_map, text_indices = self._pack_map_and_index_list(
1096
return text_index_map, self._index_contents(text_indices,
1099
def _least_readv_node_readv(self, nodes):
1100
"""Generate request groups for nodes using the least readv's.
1102
:param nodes: An iterable of graph index nodes.
1103
:return: Total node count and an iterator of the data needed to perform
1104
readvs to obtain the data for nodes. Each item yielded by the
1105
iterator is a tuple with:
1106
index, readv_vector, node_vector. readv_vector is a list ready to
1107
hand to the transport readv method, and node_vector is a list of
1108
(key, eol_flag, references) for the the node retrieved by the
1109
matching readv_vector.
1111
# group by pack so we do one readv per pack
1112
nodes = sorted(nodes)
1115
for index, key, value, references in nodes:
1116
if index not in request_groups:
1117
request_groups[index] = []
1118
request_groups[index].append((key, value, references))
1120
for index, items in request_groups.iteritems():
1121
pack_readv_requests = []
1122
for key, value, references in items:
1123
# ---- KnitGraphIndex.get_position
1124
bits = value[1:].split(' ')
1125
offset, length = int(bits[0]), int(bits[1])
1126
pack_readv_requests.append(
1127
((offset, length), (key, value[0], references)))
1128
# linear scan up the pack to maximum range combining.
1129
pack_readv_requests.sort()
1130
# split out the readv and the node data.
1131
pack_readv = [readv for readv, node in pack_readv_requests]
1132
node_vector = [node for readv, node in pack_readv_requests]
1133
result.append((index, pack_readv, node_vector))
1134
return total, result
749
raise NotImplementedError(self._create_pack_from_packs)
1136
751
def _log_copied_texts(self):
1137
752
if 'pack' in debug.debug_flags:
1169
765
return new_pack.data_inserted()
1172
class OptimisingPacker(Packer):
1173
"""A packer which spends more time to create better disk layouts."""
1175
def _revision_node_readv(self, revision_nodes):
1176
"""Return the total revisions and the readv's to issue.
1178
This sort places revisions in topological order with the ancestors
1181
:param revision_nodes: The revision index contents for the packs being
1182
incorporated into the new pack.
1183
:return: As per _least_readv_node_readv.
1185
# build an ancestors dict
1188
for index, key, value, references in revision_nodes:
1189
ancestors[key] = references[0]
1190
by_key[key] = (index, value, references)
1191
order = tsort.topo_sort(ancestors)
1193
# Single IO is pathological, but it will work as a starting point.
1195
for key in reversed(order):
1196
index, value, references = by_key[key]
1197
# ---- KnitGraphIndex.get_position
1198
bits = value[1:].split(' ')
1199
offset, length = int(bits[0]), int(bits[1])
1201
(index, [(offset, length)], [(key, value[0], references)]))
1202
# TODO: combine requests in the same index that are in ascending order.
1203
return total, requests
1205
def open_pack(self):
1206
"""Open a pack for the pack we are creating."""
1207
new_pack = super(OptimisingPacker, self).open_pack()
1208
# Turn on the optimization flags for all the index builders.
1209
new_pack.revision_index.set_optimize(for_size=True)
1210
new_pack.inventory_index.set_optimize(for_size=True)
1211
new_pack.text_index.set_optimize(for_size=True)
1212
new_pack.signature_index.set_optimize(for_size=True)
1216
class ReconcilePacker(Packer):
1217
"""A packer which regenerates indices etc as it copies.
1219
This is used by ``bzr reconcile`` to cause parent text pointers to be
1223
def _extra_init(self):
1224
self._data_changed = False
1226
def _process_inventory_lines(self, inv_lines):
1227
"""Generate a text key reference map rather for reconciling with."""
1228
repo = self._pack_collection.repo
1229
refs = repo._find_text_key_references_from_xml_inventory_lines(
1231
self._text_refs = refs
1232
# during reconcile we:
1233
# - convert unreferenced texts to full texts
1234
# - correct texts which reference a text not copied to be full texts
1235
# - copy all others as-is but with corrected parents.
1236
# - so at this point we don't know enough to decide what becomes a full
1238
self._text_filter = None
1240
def _copy_text_texts(self):
1241
"""generate what texts we should have and then copy."""
1242
self.pb.update("Copying content texts", 3)
1243
# we have three major tasks here:
1244
# 1) generate the ideal index
1245
repo = self._pack_collection.repo
1246
ancestors = dict([(key[0], tuple(ref[0] for ref in refs[0])) for
1247
_1, key, _2, refs in
1248
self.new_pack.revision_index.iter_all_entries()])
1249
ideal_index = repo._generate_text_key_index(self._text_refs, ancestors)
1250
# 2) generate a text_nodes list that contains all the deltas that can
1251
# be used as-is, with corrected parents.
1254
discarded_nodes = []
1255
NULL_REVISION = _mod_revision.NULL_REVISION
1256
text_index_map, text_nodes = self._get_text_nodes()
1257
for node in text_nodes:
1263
ideal_parents = tuple(ideal_index[node[1]])
1265
discarded_nodes.append(node)
1266
self._data_changed = True
1268
if ideal_parents == (NULL_REVISION,):
1270
if ideal_parents == node[3][0]:
1272
ok_nodes.append(node)
1273
elif ideal_parents[0:1] == node[3][0][0:1]:
1274
# the left most parent is the same, or there are no parents
1275
# today. Either way, we can preserve the representation as
1276
# long as we change the refs to be inserted.
1277
self._data_changed = True
1278
ok_nodes.append((node[0], node[1], node[2],
1279
(ideal_parents, node[3][1])))
1280
self._data_changed = True
1282
# Reinsert this text completely
1283
bad_texts.append((node[1], ideal_parents))
1284
self._data_changed = True
1285
# we're finished with some data.
1288
# 3) bulk copy the ok data
1289
total_items, readv_group_iter = self._least_readv_node_readv(ok_nodes)
1290
list(self._copy_nodes_graph(text_index_map, self.new_pack._writer,
1291
self.new_pack.text_index, readv_group_iter, total_items))
1292
# 4) adhoc copy all the other texts.
1293
# We have to topologically insert all texts otherwise we can fail to
1294
# reconcile when parts of a single delta chain are preserved intact,
1295
# and other parts are not. E.g. Discarded->d1->d2->d3. d1 will be
1296
# reinserted, and if d3 has incorrect parents it will also be
1297
# reinserted. If we insert d3 first, d2 is present (as it was bulk
1298
# copied), so we will try to delta, but d2 is not currently able to be
1299
# extracted because it's basis d1 is not present. Topologically sorting
1300
# addresses this. The following generates a sort for all the texts that
1301
# are being inserted without having to reference the entire text key
1302
# space (we only topo sort the revisions, which is smaller).
1303
topo_order = tsort.topo_sort(ancestors)
1304
rev_order = dict(zip(topo_order, range(len(topo_order))))
1305
bad_texts.sort(key=lambda key:rev_order.get(key[0][1], 0))
1306
transaction = repo.get_transaction()
1307
file_id_index = GraphIndexPrefixAdapter(
1308
self.new_pack.text_index,
1310
add_nodes_callback=self.new_pack.text_index.add_nodes)
1311
data_access = _DirectPackAccess(
1312
{self.new_pack.text_index:self.new_pack.access_tuple()})
1313
data_access.set_writer(self.new_pack._writer, self.new_pack.text_index,
1314
self.new_pack.access_tuple())
1315
output_texts = KnitVersionedFiles(
1316
_KnitGraphIndex(self.new_pack.text_index,
1317
add_callback=self.new_pack.text_index.add_nodes,
1318
deltas=True, parents=True, is_locked=repo.is_locked),
1319
data_access=data_access, max_delta_chain=200)
1320
for key, parent_keys in bad_texts:
1321
# We refer to the new pack to delta data being output.
1322
# A possible improvement would be to catch errors on short reads
1323
# and only flush then.
1324
self.new_pack.flush()
1326
for parent_key in parent_keys:
1327
if parent_key[0] != key[0]:
1328
# Graph parents must match the fileid
1329
raise errors.BzrError('Mismatched key parent %r:%r' %
1331
parents.append(parent_key[1])
1332
text_lines = osutils.split_lines(repo.texts.get_record_stream(
1333
[key], 'unordered', True).next().get_bytes_as('fulltext'))
1334
output_texts.add_lines(key, parent_keys, text_lines,
1335
random_id=True, check_content=False)
1336
# 5) check that nothing inserted has a reference outside the keyspace.
1337
missing_text_keys = self.new_pack.text_index._external_references()
1338
if missing_text_keys:
1339
raise errors.BzrCheckError('Reference to missing compression parents %r'
1340
% (missing_text_keys,))
1341
self._log_copied_texts()
1343
def _use_pack(self, new_pack):
1344
"""Override _use_pack to check for reconcile having changed content."""
1345
# XXX: we might be better checking this at the copy time.
1346
original_inventory_keys = set()
1347
inv_index = self._pack_collection.inventory_index.combined_index
1348
for entry in inv_index.iter_all_entries():
1349
original_inventory_keys.add(entry[1])
1350
new_inventory_keys = set()
1351
for entry in new_pack.inventory_index.iter_all_entries():
1352
new_inventory_keys.add(entry[1])
1353
if new_inventory_keys != original_inventory_keys:
1354
self._data_changed = True
1355
return new_pack.data_inserted() and self._data_changed
1358
768
class RepositoryPackCollection(object):
1359
769
"""Management of packs within a repository.
1361
771
:ivar _names: map of {pack_name: (index_size,)}
1364
pack_factory = NewPack
1365
resumed_pack_factory = ResumedPack
775
resumed_pack_factory = None
776
normal_packer_class = None
777
optimising_packer_class = None
1367
779
def __init__(self, repo, transport, index_transport, upload_transport,
1368
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
2141
1672
def __init__(self, _format, a_bzrdir, control_files, _commit_builder_class,
2143
KnitRepository.__init__(self, _format, a_bzrdir, control_files,
2144
_commit_builder_class, _serializer)
2145
index_transport = self._transport.clone('indices')
2146
self._pack_collection = RepositoryPackCollection(self, self._transport,
2148
self._transport.clone('upload'),
2149
self._transport.clone('packs'),
2150
_format.index_builder_class,
2151
_format.index_class,
2152
use_chk_index=self._format.supports_chks,
2154
self.inventories = KnitVersionedFiles(
2155
_KnitGraphIndex(self._pack_collection.inventory_index.combined_index,
2156
add_callback=self._pack_collection.inventory_index.add_callback,
2157
deltas=True, parents=True, is_locked=self.is_locked),
2158
data_access=self._pack_collection.inventory_index.data_access,
2159
max_delta_chain=200)
2160
self.revisions = KnitVersionedFiles(
2161
_KnitGraphIndex(self._pack_collection.revision_index.combined_index,
2162
add_callback=self._pack_collection.revision_index.add_callback,
2163
deltas=False, parents=True, is_locked=self.is_locked,
2164
track_external_parent_refs=True),
2165
data_access=self._pack_collection.revision_index.data_access,
2167
self.signatures = KnitVersionedFiles(
2168
_KnitGraphIndex(self._pack_collection.signature_index.combined_index,
2169
add_callback=self._pack_collection.signature_index.add_callback,
2170
deltas=False, parents=False, is_locked=self.is_locked),
2171
data_access=self._pack_collection.signature_index.data_access,
2173
self.texts = KnitVersionedFiles(
2174
_KnitGraphIndex(self._pack_collection.text_index.combined_index,
2175
add_callback=self._pack_collection.text_index.add_callback,
2176
deltas=True, parents=True, is_locked=self.is_locked),
2177
data_access=self._pack_collection.text_index.data_access,
2178
max_delta_chain=200)
2179
if _format.supports_chks:
2180
# No graph, no compression:- references from chks are between
2181
# different objects not temporal versions of the same; and without
2182
# some sort of temporal structure knit compression will just fail.
2183
self.chk_bytes = KnitVersionedFiles(
2184
_KnitGraphIndex(self._pack_collection.chk_index.combined_index,
2185
add_callback=self._pack_collection.chk_index.add_callback,
2186
deltas=False, parents=False, is_locked=self.is_locked),
2187
data_access=self._pack_collection.chk_index.data_access,
2190
self.chk_bytes = None
2191
# True when the repository object is 'write locked' (as opposed to the
2192
# physical lock only taken out around changes to the pack-names list.)
2193
# Another way to represent this would be a decorator around the control
2194
# files object that presents logical locks as physical ones - if this
2195
# gets ugly consider that alternative design. RBC 20071011
2196
self._write_lock_count = 0
2197
self._transaction = None
2199
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
2200
1677
self._reconcile_fixes_text_parents = True
2201
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()
2203
def _warn_if_deprecated(self):
2204
# This class isn't deprecated, but one sub-format is
2205
if isinstance(self._format, RepositoryFormatKnitPack5RichRootBroken):
2206
from bzrlib import repository
2207
if repository._deprecation_warning_done:
2209
repository._deprecation_warning_done = True
2210
warning("Format %s for %s is deprecated - please use"
2211
" 'bzr upgrade --1.6.1-rich-root'"
2212
% (self._format, self.bzrdir.transport.base))
1686
def _all_revision_ids(self):
1687
"""See Repository.all_revision_ids()."""
1688
return [key[0] for key in self.revisions.keys()]
2214
1690
def _abort_write_group(self):
2215
self.revisions._index._key_dependencies.refs.clear()
1691
self.revisions._index._key_dependencies.clear()
2216
1692
self._pack_collection._abort_write_group()
2218
def _find_inconsistent_revision_parents(self):
2219
"""Find revisions with incorrectly cached parents.
2221
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
2222
parents-in-revision).
2224
if not self.is_locked():
2225
raise errors.ObjectNotLocked(self)
2226
pb = ui.ui_factory.nested_progress_bar()
2229
revision_nodes = self._pack_collection.revision_index \
2230
.combined_index.iter_all_entries()
2231
index_positions = []
2232
# Get the cached index values for all revisions, and also the
2233
# location in each index of the revision text so we can perform
2235
for index, key, value, refs in revision_nodes:
2236
node = (index, key, value, refs)
2237
index_memo = self.revisions._index._node_to_position(node)
2238
if index_memo[0] != index:
2239
raise AssertionError('%r != %r' % (index_memo[0], index))
2240
index_positions.append((index_memo, key[0],
2241
tuple(parent[0] for parent in refs[0])))
2242
pb.update("Reading revision index", 0, 0)
2243
index_positions.sort()
2245
pb.update("Checking cached revision graph", 0,
2246
len(index_positions))
2247
for offset in xrange(0, len(index_positions), 1000):
2248
pb.update("Checking cached revision graph", offset)
2249
to_query = index_positions[offset:offset + batch_size]
2252
rev_ids = [item[1] for item in to_query]
2253
revs = self.get_revisions(rev_ids)
2254
for revision, item in zip(revs, to_query):
2255
index_parents = item[2]
2256
rev_parents = tuple(revision.parent_ids)
2257
if index_parents != rev_parents:
2258
result.append((revision.revision_id, index_parents,
2264
def _get_source(self, to_format):
2265
if to_format.network_name() == self._format.network_name():
2266
return KnitPackStreamSource(self, to_format)
2267
return super(KnitPackRepository, self)._get_source(to_format)
2269
1694
def _make_parents_provider(self):
2270
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))
2272
1700
def _refresh_data(self):
2273
1701
if not self.is_locked():
2275
1703
self._pack_collection.reload_pack_names()
1704
self._unstacked_provider.disable_cache()
1705
self._unstacked_provider.enable_cache()
2277
1707
def _start_write_group(self):
2278
1708
self._pack_collection._start_write_group()
2280
1710
def _commit_write_group(self):
2281
self.revisions._index._key_dependencies.refs.clear()
2282
return self._pack_collection._commit_write_group()
1711
hint = self._pack_collection._commit_write_group()
1712
self.revisions._index._key_dependencies.clear()
1713
# The commit may have added keys that were previously cached as
1714
# missing, so reset the cache.
1715
self._unstacked_provider.disable_cache()
1716
self._unstacked_provider.enable_cache()
2284
1719
def suspend_write_group(self):
2285
1720
# XXX check self._write_group is self.get_transaction()?
2286
1721
tokens = self._pack_collection._suspend_write_group()
2287
self.revisions._index._key_dependencies.refs.clear()
1722
self.revisions._index._key_dependencies.clear()
2288
1723
self._write_group = None
2381
1835
self.control_files.unlock()
2383
1837
if not self.is_locked():
1838
self._unstacked_provider.disable_cache()
2384
1839
for repo in self._fallback_repositories:
2388
class KnitPackStreamSource(StreamSource):
2389
"""A StreamSource used to transfer data between same-format KnitPack repos.
2391
This source assumes:
2392
1) Same serialization format for all objects
2393
2) Same root information
2394
3) XML format inventories
2395
4) Atomic inserts (so we can stream inventory texts before text
2400
def __init__(self, from_repository, to_format):
2401
super(KnitPackStreamSource, self).__init__(from_repository, to_format)
2402
self._text_keys = None
2403
self._text_fetch_order = 'unordered'
2405
def _get_filtered_inv_stream(self, revision_ids):
2406
from_repo = self.from_repository
2407
parent_ids = from_repo._find_parent_ids_of_revisions(revision_ids)
2408
parent_keys = [(p,) for p in parent_ids]
2409
find_text_keys = from_repo._find_text_key_references_from_xml_inventory_lines
2410
parent_text_keys = set(find_text_keys(
2411
from_repo._inventory_xml_lines_for_keys(parent_keys)))
2412
content_text_keys = set()
2413
knit = KnitVersionedFiles(None, None)
2414
factory = KnitPlainFactory()
2415
def find_text_keys_from_content(record):
2416
if record.storage_kind not in ('knit-delta-gz', 'knit-ft-gz'):
2417
raise ValueError("Unknown content storage kind for"
2418
" inventory text: %s" % (record.storage_kind,))
2419
# It's a knit record, it has a _raw_record field (even if it was
2420
# reconstituted from a network stream).
2421
raw_data = record._raw_record
2422
# read the entire thing
2423
revision_id = record.key[-1]
2424
content, _ = knit._parse_record(revision_id, raw_data)
2425
if record.storage_kind == 'knit-delta-gz':
2426
line_iterator = factory.get_linedelta_content(content)
2427
elif record.storage_kind == 'knit-ft-gz':
2428
line_iterator = factory.get_fulltext_content(content)
2429
content_text_keys.update(find_text_keys(
2430
[(line, revision_id) for line in line_iterator]))
2431
revision_keys = [(r,) for r in revision_ids]
2432
def _filtered_inv_stream():
2433
source_vf = from_repo.inventories
2434
stream = source_vf.get_record_stream(revision_keys,
2436
for record in stream:
2437
if record.storage_kind == 'absent':
2438
raise errors.NoSuchRevision(from_repo, record.key)
2439
find_text_keys_from_content(record)
2441
self._text_keys = content_text_keys - parent_text_keys
2442
return ('inventories', _filtered_inv_stream())
2444
def _get_text_stream(self):
2445
# Note: We know we don't have to handle adding root keys, because both
2446
# the source and target are the identical network name.
2447
text_stream = self.from_repository.texts.get_record_stream(
2448
self._text_keys, self._text_fetch_order, False)
2449
return ('texts', text_stream)
2451
def get_stream(self, search):
2452
revision_ids = search.get_keys()
2453
for stream_info in self._fetch_revision_texts(revision_ids):
2455
self._revision_keys = [(rev_id,) for rev_id in revision_ids]
2456
yield self._get_filtered_inv_stream(revision_ids)
2457
yield self._get_text_stream()
2461
class RepositoryFormatPack(MetaDirRepositoryFormat):
1843
class RepositoryFormatPack(MetaDirVersionedFileRepositoryFormat):
2462
1844
"""Format logic for pack structured repositories.
2464
1846
This repository format has:
2534
1920
_serializer=self._serializer)
2537
class RepositoryFormatKnitPack1(RepositoryFormatPack):
2538
"""A no-subtrees parameterized Pack repository.
2540
This format was introduced in 0.92.
2543
repository_class = KnitPackRepository
2544
_commit_builder_class = PackCommitBuilder
2546
def _serializer(self):
2547
return xml5.serializer_v5
2548
# What index classes to use
2549
index_builder_class = InMemoryGraphIndex
2550
index_class = GraphIndex
2552
def _get_matching_bzrdir(self):
2553
return bzrdir.format_registry.make_bzrdir('pack-0.92')
2555
def _ignore_setting_bzrdir(self, format):
2558
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2560
def get_format_string(self):
2561
"""See RepositoryFormat.get_format_string()."""
2562
return "Bazaar pack repository format 1 (needs bzr 0.92)\n"
2564
def get_format_description(self):
2565
"""See RepositoryFormat.get_format_description()."""
2566
return "Packs containing knits without subtree support"
2568
def check_conversion_target(self, target_format):
2572
class RepositoryFormatKnitPack3(RepositoryFormatPack):
2573
"""A subtrees parameterized Pack repository.
2575
This repository format uses the xml7 serializer to get:
2576
- support for recording full info about the tree root
2577
- support for recording tree-references
2579
This format was introduced in 0.92.
2582
repository_class = KnitPackRepository
2583
_commit_builder_class = PackRootCommitBuilder
2584
rich_root_data = True
2585
supports_tree_reference = True
2587
def _serializer(self):
2588
return xml7.serializer_v7
2589
# What index classes to use
2590
index_builder_class = InMemoryGraphIndex
2591
index_class = GraphIndex
2593
def _get_matching_bzrdir(self):
2594
return bzrdir.format_registry.make_bzrdir(
2595
'pack-0.92-subtree')
2597
def _ignore_setting_bzrdir(self, format):
2600
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2602
def check_conversion_target(self, target_format):
2603
if not target_format.rich_root_data:
2604
raise errors.BadConversionTarget(
2605
'Does not support rich root data.', target_format)
2606
if not getattr(target_format, 'supports_tree_reference', False):
2607
raise errors.BadConversionTarget(
2608
'Does not support nested trees', target_format)
2610
def get_format_string(self):
2611
"""See RepositoryFormat.get_format_string()."""
2612
return "Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n"
2614
def get_format_description(self):
2615
"""See RepositoryFormat.get_format_description()."""
2616
return "Packs containing knits with subtree support\n"
2619
class RepositoryFormatKnitPack4(RepositoryFormatPack):
2620
"""A rich-root, no subtrees parameterized Pack repository.
2622
This repository format uses the xml6 serializer to get:
2623
- support for recording full info about the tree root
2625
This format was introduced in 1.0.
2628
repository_class = KnitPackRepository
2629
_commit_builder_class = PackRootCommitBuilder
2630
rich_root_data = True
2631
supports_tree_reference = False
2633
def _serializer(self):
2634
return xml6.serializer_v6
2635
# What index classes to use
2636
index_builder_class = InMemoryGraphIndex
2637
index_class = GraphIndex
2639
def _get_matching_bzrdir(self):
2640
return bzrdir.format_registry.make_bzrdir(
2643
def _ignore_setting_bzrdir(self, format):
2646
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2648
def check_conversion_target(self, target_format):
2649
if not target_format.rich_root_data:
2650
raise errors.BadConversionTarget(
2651
'Does not support rich root data.', target_format)
2653
def get_format_string(self):
2654
"""See RepositoryFormat.get_format_string()."""
2655
return ("Bazaar pack repository format 1 with rich root"
2656
" (needs bzr 1.0)\n")
2658
def get_format_description(self):
2659
"""See RepositoryFormat.get_format_description()."""
2660
return "Packs containing knits with rich root support\n"
2663
class RepositoryFormatKnitPack5(RepositoryFormatPack):
2664
"""Repository that supports external references to allow stacking.
2668
Supports external lookups, which results in non-truncated ghosts after
2669
reconcile compared to pack-0.92 formats.
2672
repository_class = KnitPackRepository
2673
_commit_builder_class = PackCommitBuilder
2674
supports_external_lookups = True
2675
# What index classes to use
2676
index_builder_class = InMemoryGraphIndex
2677
index_class = GraphIndex
2680
def _serializer(self):
2681
return xml5.serializer_v5
2683
def _get_matching_bzrdir(self):
2684
return bzrdir.format_registry.make_bzrdir('1.6')
2686
def _ignore_setting_bzrdir(self, format):
2689
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2691
def get_format_string(self):
2692
"""See RepositoryFormat.get_format_string()."""
2693
return "Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n"
2695
def get_format_description(self):
2696
"""See RepositoryFormat.get_format_description()."""
2697
return "Packs 5 (adds stacking support, requires bzr 1.6)"
2699
def check_conversion_target(self, target_format):
2703
class RepositoryFormatKnitPack5RichRoot(RepositoryFormatPack):
2704
"""A repository with rich roots and stacking.
2706
New in release 1.6.1.
2708
Supports stacking on other repositories, allowing data to be accessed
2709
without being stored locally.
2712
repository_class = KnitPackRepository
2713
_commit_builder_class = PackRootCommitBuilder
2714
rich_root_data = True
2715
supports_tree_reference = False # no subtrees
2716
supports_external_lookups = True
2717
# What index classes to use
2718
index_builder_class = InMemoryGraphIndex
2719
index_class = GraphIndex
2722
def _serializer(self):
2723
return xml6.serializer_v6
2725
def _get_matching_bzrdir(self):
2726
return bzrdir.format_registry.make_bzrdir(
2729
def _ignore_setting_bzrdir(self, format):
2732
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2734
def check_conversion_target(self, target_format):
2735
if not target_format.rich_root_data:
2736
raise errors.BadConversionTarget(
2737
'Does not support rich root data.', target_format)
2739
def get_format_string(self):
2740
"""See RepositoryFormat.get_format_string()."""
2741
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n"
2743
def get_format_description(self):
2744
return "Packs 5 rich-root (adds stacking support, requires bzr 1.6.1)"
2747
class RepositoryFormatKnitPack5RichRootBroken(RepositoryFormatPack):
2748
"""A repository with rich roots and external references.
2752
Supports external lookups, which results in non-truncated ghosts after
2753
reconcile compared to pack-0.92 formats.
2755
This format was deprecated because the serializer it uses accidentally
2756
supported subtrees, when the format was not intended to. This meant that
2757
someone could accidentally fetch from an incorrect repository.
2760
repository_class = KnitPackRepository
2761
_commit_builder_class = PackRootCommitBuilder
2762
rich_root_data = True
2763
supports_tree_reference = False # no subtrees
2765
supports_external_lookups = True
2766
# What index classes to use
2767
index_builder_class = InMemoryGraphIndex
2768
index_class = GraphIndex
2771
def _serializer(self):
2772
return xml7.serializer_v7
2774
def _get_matching_bzrdir(self):
2775
matching = bzrdir.format_registry.make_bzrdir(
2777
matching.repository_format = self
2780
def _ignore_setting_bzrdir(self, format):
2783
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2785
def check_conversion_target(self, target_format):
2786
if not target_format.rich_root_data:
2787
raise errors.BadConversionTarget(
2788
'Does not support rich root data.', target_format)
2790
def get_format_string(self):
2791
"""See RepositoryFormat.get_format_string()."""
2792
return "Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n"
2794
def get_format_description(self):
2795
return ("Packs 5 rich-root (adds stacking support, requires bzr 1.6)"
2799
class RepositoryFormatKnitPack6(RepositoryFormatPack):
2800
"""A repository with stacking and btree indexes,
2801
without rich roots or subtrees.
2803
This is equivalent to pack-1.6 with B+Tree indices.
2806
repository_class = KnitPackRepository
2807
_commit_builder_class = PackCommitBuilder
2808
supports_external_lookups = True
2809
# What index classes to use
2810
index_builder_class = BTreeBuilder
2811
index_class = BTreeGraphIndex
2814
def _serializer(self):
2815
return xml5.serializer_v5
2817
def _get_matching_bzrdir(self):
2818
return bzrdir.format_registry.make_bzrdir('1.9')
2820
def _ignore_setting_bzrdir(self, format):
2823
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2825
def get_format_string(self):
2826
"""See RepositoryFormat.get_format_string()."""
2827
return "Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n"
2829
def get_format_description(self):
2830
"""See RepositoryFormat.get_format_description()."""
2831
return "Packs 6 (uses btree indexes, requires bzr 1.9)"
2833
def check_conversion_target(self, target_format):
2837
class RepositoryFormatKnitPack6RichRoot(RepositoryFormatPack):
2838
"""A repository with rich roots, no subtrees, stacking and btree indexes.
2840
1.6-rich-root with B+Tree indices.
2843
repository_class = KnitPackRepository
2844
_commit_builder_class = PackRootCommitBuilder
2845
rich_root_data = True
2846
supports_tree_reference = False # no subtrees
2847
supports_external_lookups = True
2848
# What index classes to use
2849
index_builder_class = BTreeBuilder
2850
index_class = BTreeGraphIndex
2853
def _serializer(self):
2854
return xml6.serializer_v6
2856
def _get_matching_bzrdir(self):
2857
return bzrdir.format_registry.make_bzrdir(
2860
def _ignore_setting_bzrdir(self, format):
2863
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2865
def check_conversion_target(self, target_format):
2866
if not target_format.rich_root_data:
2867
raise errors.BadConversionTarget(
2868
'Does not support rich root data.', target_format)
2870
def get_format_string(self):
2871
"""See RepositoryFormat.get_format_string()."""
2872
return "Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n"
2874
def get_format_description(self):
2875
return "Packs 6 rich-root (uses btree indexes, requires bzr 1.9)"
2878
class RepositoryFormatPackDevelopment2Subtree(RepositoryFormatPack):
2879
"""A subtrees development repository.
2881
This format should be retained until the second release after bzr 1.7.
2883
1.6.1-subtree[as it might have been] with B+Tree indices.
2885
This is [now] retained until we have a CHK based subtree format in
2889
repository_class = KnitPackRepository
2890
_commit_builder_class = PackRootCommitBuilder
2891
rich_root_data = True
2892
supports_tree_reference = True
2893
supports_external_lookups = True
2894
# What index classes to use
2895
index_builder_class = BTreeBuilder
2896
index_class = BTreeGraphIndex
2899
def _serializer(self):
2900
return xml7.serializer_v7
2902
def _get_matching_bzrdir(self):
2903
return bzrdir.format_registry.make_bzrdir(
2904
'development-subtree')
2906
def _ignore_setting_bzrdir(self, format):
2909
_matchingbzrdir = property(_get_matching_bzrdir, _ignore_setting_bzrdir)
2911
def check_conversion_target(self, target_format):
2912
if not target_format.rich_root_data:
2913
raise errors.BadConversionTarget(
2914
'Does not support rich root data.', target_format)
2915
if not getattr(target_format, 'supports_tree_reference', False):
2916
raise errors.BadConversionTarget(
2917
'Does not support nested trees', target_format)
2919
def get_format_string(self):
2920
"""See RepositoryFormat.get_format_string()."""
2921
return ("Bazaar development format 2 with subtree support "
2922
"(needs bzr.dev from before 1.8)\n")
2924
def get_format_description(self):
2925
"""See RepositoryFormat.get_format_description()."""
2926
return ("Development repository format, currently the same as "
2927
"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
