827
304
def lower_line_delta(self, delta):
829
306
for start, end, c, lines in delta:
830
out.append(b'%d,%d,%d\n' % (start, end, c))
307
out.append('%d,%d,%d\n' % (start, end, c))
308
out.extend([text for origin, text in lines])
834
def annotate(self, knit, key):
835
annotator = _KnitAnnotator(knit)
836
return annotator.annotate_flat(key)
840
def make_file_factory(annotated, mapper):
841
"""Create a factory for creating a file based KnitVersionedFiles.
843
This is only functional enough to run interface tests, it doesn't try to
844
provide a full pack environment.
846
:param annotated: knit annotations are wanted.
847
:param mapper: The mapper from keys to paths.
849
def factory(transport):
850
index = _KndxIndex(transport, mapper, lambda:None, lambda:True, lambda:True)
851
access = _KnitKeyAccess(transport, mapper)
852
return KnitVersionedFiles(index, access, annotated=annotated)
856
def make_pack_factory(graph, delta, keylength):
857
"""Create a factory for creating a pack based VersionedFiles.
859
This is only functional enough to run interface tests, it doesn't try to
860
provide a full pack environment.
862
:param graph: Store a graph.
863
:param delta: Delta compress contents.
864
:param keylength: How long should keys be.
866
def factory(transport):
867
parents = graph or delta
873
max_delta_chain = 200
876
graph_index = _mod_index.InMemoryGraphIndex(reference_lists=ref_length,
877
key_elements=keylength)
878
stream = transport.open_write_stream('newpack')
879
writer = pack.ContainerWriter(stream.write)
881
index = _KnitGraphIndex(graph_index, lambda:True, parents=parents,
882
deltas=delta, add_callback=graph_index.add_nodes)
883
access = pack_repo._DirectPackAccess({})
884
access.set_writer(writer, graph_index, (transport, 'newpack'))
885
result = KnitVersionedFiles(index, access,
886
max_delta_chain=max_delta_chain)
887
result.stream = stream
888
result.writer = writer
893
def cleanup_pack_knit(versioned_files):
894
versioned_files.stream.close()
895
versioned_files.writer.end()
898
def _get_total_build_size(self, keys, positions):
899
"""Determine the total bytes to build these keys.
901
(helper function because _KnitGraphIndex and _KndxIndex work the same, but
902
don't inherit from a common base.)
904
:param keys: Keys that we want to build
905
:param positions: dict of {key, (info, index_memo, comp_parent)} (such
906
as returned by _get_components_positions)
907
:return: Number of bytes to build those keys
909
all_build_index_memos = {}
913
for key in build_keys:
914
# This is mostly for the 'stacked' case
915
# Where we will be getting the data from a fallback
916
if key not in positions:
918
_, index_memo, compression_parent = positions[key]
919
all_build_index_memos[key] = index_memo
920
if compression_parent not in all_build_index_memos:
921
next_keys.add(compression_parent)
922
build_keys = next_keys
923
return sum(index_memo[2]
924
for index_memo in viewvalues(all_build_index_memos))
927
class KnitVersionedFiles(VersionedFilesWithFallbacks):
928
"""Storage for many versioned files using knit compression.
930
Backend storage is managed by indices and data objects.
932
:ivar _index: A _KnitGraphIndex or similar that can describe the
933
parents, graph, compression and data location of entries in this
934
KnitVersionedFiles. Note that this is only the index for
935
*this* vfs; if there are fallbacks they must be queried separately.
938
def __init__(self, index, data_access, max_delta_chain=200,
939
annotated=False, reload_func=None):
940
"""Create a KnitVersionedFiles with index and data_access.
942
:param index: The index for the knit data.
943
:param data_access: The access object to store and retrieve knit
945
:param max_delta_chain: The maximum number of deltas to permit during
946
insertion. Set to 0 to prohibit the use of deltas.
947
:param annotated: Set to True to cause annotations to be calculated and
948
stored during insertion.
949
:param reload_func: An function that can be called if we think we need
950
to reload the pack listing and try again. See
951
'breezy.bzr.pack_repo.AggregateIndex' for the signature.
312
def make_empty_knit(transport, relpath):
313
"""Construct a empty knit at the specified location."""
314
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
318
class KnitVersionedFile(VersionedFile):
319
"""Weave-like structure with faster random access.
321
A knit stores a number of texts and a summary of the relationships
322
between them. Texts are identified by a string version-id. Texts
323
are normally stored and retrieved as a series of lines, but can
324
also be passed as single strings.
326
Lines are stored with the trailing newline (if any) included, to
327
avoid special cases for files with no final newline. Lines are
328
composed of 8-bit characters, not unicode. The combination of
329
these approaches should mean any 'binary' file can be safely
330
stored and retrieved.
333
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
334
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
335
create=False, create_parent_dir=False, delay_create=False,
337
"""Construct a knit at location specified by relpath.
339
:param create: If not True, only open an existing knit.
340
:param create_parent_dir: If True, create the parent directory if
341
creating the file fails. (This is used for stores with
342
hash-prefixes that may not exist yet)
343
:param delay_create: The calling code is aware that the knit won't
344
actually be created until the first data is stored.
954
self._access = data_access
955
self._max_delta_chain = max_delta_chain
957
self._factory = KnitAnnotateFactory()
959
self._factory = KnitPlainFactory()
960
self._immediate_fallback_vfs = []
961
self._reload_func = reload_func
346
if deprecated_passed(basis_knit):
347
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
348
" deprecated as of bzr 0.9.",
349
DeprecationWarning, stacklevel=2)
350
if access_mode is None:
352
super(KnitVersionedFile, self).__init__(access_mode)
353
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
354
self.transport = transport
355
self.filename = relpath
356
self.factory = factory or KnitAnnotateFactory()
357
self.writable = (access_mode == 'w')
360
self._max_delta_chain = 200
362
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
363
access_mode, create=create, file_mode=file_mode,
364
create_parent_dir=create_parent_dir, delay_create=delay_create,
366
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
367
access_mode, create=create and not len(self), file_mode=file_mode,
368
create_parent_dir=create_parent_dir, delay_create=delay_create,
963
371
def __repr__(self):
964
return "%s(%r, %r)" % (
965
self.__class__.__name__,
969
def without_fallbacks(self):
970
"""Return a clone of this object without any fallbacks configured."""
971
return KnitVersionedFiles(self._index, self._access,
972
self._max_delta_chain, self._factory.annotated,
975
def add_fallback_versioned_files(self, a_versioned_files):
976
"""Add a source of texts for texts not present in this knit.
978
:param a_versioned_files: A VersionedFiles object.
980
self._immediate_fallback_vfs.append(a_versioned_files)
982
def add_lines(self, key, parents, lines, parent_texts=None,
983
left_matching_blocks=None, nostore_sha=None, random_id=False,
985
"""See VersionedFiles.add_lines()."""
986
self._index._check_write_ok()
987
self._check_add(key, lines, random_id, check_content)
989
# The caller might pass None if there is no graph data, but kndx
990
# indexes can't directly store that, so we give them
991
# an empty tuple instead.
993
line_bytes = b''.join(lines)
994
return self._add(key, lines, parents,
995
parent_texts, left_matching_blocks, nostore_sha, random_id,
996
line_bytes=line_bytes)
998
def _add(self, key, lines, parents, parent_texts,
999
left_matching_blocks, nostore_sha, random_id,
1001
"""Add a set of lines on top of version specified by parents.
1003
Any versions not present will be converted into ghosts.
1005
:param lines: A list of strings where each one is a single line (has a
1006
single newline at the end of the string) This is now optional
1007
(callers can pass None). It is left in its location for backwards
1008
compatibility. It should ''.join(lines) must == line_bytes
1009
:param line_bytes: A single string containing the content
1011
We pass both lines and line_bytes because different routes bring the
1012
values to this function. And for memory efficiency, we don't want to
1013
have to split/join on-demand.
1015
# first thing, if the content is something we don't need to store, find
1017
digest = sha_string(line_bytes)
1018
if nostore_sha == digest:
1019
raise errors.ExistingContent
1021
present_parents = []
1022
if parent_texts is None:
1024
# Do a single query to ascertain parent presence; we only compress
1025
# against parents in the same kvf.
1026
present_parent_map = self._index.get_parent_map(parents)
1027
for parent in parents:
1028
if parent in present_parent_map:
1029
present_parents.append(parent)
1031
# Currently we can only compress against the left most present parent.
1032
if (len(present_parents) == 0 or
1033
present_parents[0] != parents[0]):
1036
# To speed the extract of texts the delta chain is limited
1037
# to a fixed number of deltas. This should minimize both
1038
# I/O and the time spend applying deltas.
1039
delta = self._check_should_delta(present_parents[0])
1041
text_length = len(line_bytes)
1044
# Note: line_bytes is not modified to add a newline, that is tracked
1045
# via the no_eol flag. 'lines' *is* modified, because that is the
1046
# general values needed by the Content code.
1047
if line_bytes and not line_bytes.endswith(b'\n'):
1048
options.append(b'no-eol')
1050
# Copy the existing list, or create a new one
1052
lines = osutils.split_lines(line_bytes)
1055
# Replace the last line with one that ends in a final newline
1056
lines[-1] = lines[-1] + b'\n'
1058
lines = osutils.split_lines(line_bytes)
1060
for element in key[:-1]:
1061
if not isinstance(element, bytes):
1062
raise TypeError("key contains non-bytestrings: %r" % (key,))
1064
key = key[:-1] + (b'sha1:' + digest,)
1065
elif not isinstance(key[-1], bytes):
1066
raise TypeError("key contains non-bytestrings: %r" % (key,))
1067
# Knit hunks are still last-element only
1068
version_id = key[-1]
1069
content = self._factory.make(lines, version_id)
1071
# Hint to the content object that its text() call should strip the
1073
content._should_strip_eol = True
1074
if delta or (self._factory.annotated and len(present_parents) > 0):
1075
# Merge annotations from parent texts if needed.
1076
delta_hunks = self._merge_annotations(content, present_parents,
1077
parent_texts, delta, self._factory.annotated,
1078
left_matching_blocks)
1081
options.append(b'line-delta')
1082
store_lines = self._factory.lower_line_delta(delta_hunks)
1083
size, data = self._record_to_data(key, digest,
1086
options.append(b'fulltext')
1087
# isinstance is slower and we have no hierarchy.
1088
if self._factory.__class__ is KnitPlainFactory:
1089
# Use the already joined bytes saving iteration time in
1091
dense_lines = [line_bytes]
1093
dense_lines.append(b'\n')
1094
size, data = self._record_to_data(key, digest,
1097
# get mixed annotation + content and feed it into the
1099
store_lines = self._factory.lower_fulltext(content)
1100
size, data = self._record_to_data(key, digest,
1103
access_memo = self._access.add_raw_records([(key, size)], data)[0]
1104
self._index.add_records(
1105
((key, options, access_memo, parents),),
1106
random_id=random_id)
1107
return digest, text_length, content
1109
def annotate(self, key):
1110
"""See VersionedFiles.annotate."""
1111
return self._factory.annotate(self, key)
1113
def get_annotator(self):
1114
return _KnitAnnotator(self)
1116
def check(self, progress_bar=None, keys=None):
1117
"""See VersionedFiles.check()."""
1119
return self._logical_check()
1121
# At the moment, check does not extra work over get_record_stream
1122
return self.get_record_stream(keys, 'unordered', True)
1124
def _logical_check(self):
1125
# This doesn't actually test extraction of everything, but that will
1126
# impact 'bzr check' substantially, and needs to be integrated with
1127
# care. However, it does check for the obvious problem of a delta with
1129
keys = self._index.keys()
1130
parent_map = self.get_parent_map(keys)
1132
if self._index.get_method(key) != 'fulltext':
1133
compression_parent = parent_map[key][0]
1134
if compression_parent not in parent_map:
1135
raise KnitCorrupt(self,
1136
"Missing basis parent %s for %s" % (
1137
compression_parent, key))
1138
for fallback_vfs in self._immediate_fallback_vfs:
1139
fallback_vfs.check()
1141
def _check_add(self, key, lines, random_id, check_content):
1142
"""check that version_id and lines are safe to add."""
1143
if not all(isinstance(x, bytes) or x is None for x in key):
1144
raise TypeError(key)
1145
version_id = key[-1]
1146
if version_id is not None:
1147
if contains_whitespace(version_id):
1148
raise InvalidRevisionId(version_id, self)
1149
self.check_not_reserved_id(version_id)
1150
# TODO: If random_id==False and the key is already present, we should
1151
# probably check that the existing content is identical to what is
1152
# being inserted, and otherwise raise an exception. This would make
1153
# the bundle code simpler.
1155
self._check_lines_not_unicode(lines)
1156
self._check_lines_are_lines(lines)
1158
def _check_header(self, key, line):
1159
rec = self._split_header(line)
1160
self._check_header_version(rec, key[-1])
1163
def _check_header_version(self, rec, version_id):
1164
"""Checks the header version on original format knit records.
1166
These have the last component of the key embedded in the record.
1168
if rec[1] != version_id:
1169
raise KnitCorrupt(self,
1170
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
1172
def _check_should_delta(self, parent):
372
return '%s(%s)' % (self.__class__.__name__,
373
self.transport.abspath(self.filename))
375
def _check_should_delta(self, first_parents):
1173
376
"""Iterate back through the parent listing, looking for a fulltext.
1175
378
This is used when we want to decide whether to add a delta or a new
1184
387
fulltext_size = None
1185
for count in range(self._max_delta_chain):
1187
# Note that this only looks in the index of this particular
1188
# KnitVersionedFiles, not in the fallbacks. This ensures that
1189
# we won't store a delta spanning physical repository
1191
build_details = self._index.get_build_details([parent])
1192
parent_details = build_details[parent]
1193
except (RevisionNotPresent, KeyError) as e:
1194
# Some basis is not locally present: always fulltext
1196
index_memo, compression_parent, _, _ = parent_details
1197
_, _, size = index_memo
1198
if compression_parent is None:
388
delta_parents = first_parents
389
for count in xrange(self._max_delta_chain):
390
parent = delta_parents[0]
391
method = self._index.get_method(parent)
392
pos, size = self._index.get_position(parent)
393
if method == 'fulltext':
1199
394
fulltext_size = size
1201
396
delta_size += size
1202
# We don't explicitly check for presence because this is in an
1203
# inner loop, and if it's missing it'll fail anyhow.
1204
parent = compression_parent
397
delta_parents = self._index.get_parents(parent)
1206
399
# We couldn't find a fulltext, so we must create a new one
1208
# Simple heuristic - if the total I/O wold be greater as a delta than
1209
# the originally installed fulltext, we create a new fulltext.
1210
402
return fulltext_size > delta_size
1212
def _build_details_to_components(self, build_details):
1213
"""Convert a build_details tuple to a position tuple."""
1214
# record_details, access_memo, compression_parent
1215
return build_details[3], build_details[0], build_details[1]
1217
def _get_components_positions(self, keys, allow_missing=False):
1218
"""Produce a map of position data for the components of keys.
1220
This data is intended to be used for retrieving the knit records.
1222
A dict of key to (record_details, index_memo, next, parents) is
1225
* method is the way referenced data should be applied.
1226
* index_memo is the handle to pass to the data access to actually get
1228
* next is the build-parent of the version, or None for fulltexts.
1229
* parents is the version_ids of the parents of this version
1231
:param allow_missing: If True do not raise an error on a missing
1232
component, just ignore it.
1235
pending_components = keys
1236
while pending_components:
1237
build_details = self._index.get_build_details(pending_components)
1238
current_components = set(pending_components)
1239
pending_components = set()
1240
for key, details in viewitems(build_details):
1241
(index_memo, compression_parent, parents,
1242
record_details) = details
1243
method = record_details[0]
1244
if compression_parent is not None:
1245
pending_components.add(compression_parent)
1246
component_data[key] = self._build_details_to_components(details)
1247
missing = current_components.difference(build_details)
1248
if missing and not allow_missing:
1249
raise errors.RevisionNotPresent(missing.pop(), self)
1250
return component_data
1252
def _get_content(self, key, parent_texts={}):
1253
"""Returns a content object that makes up the specified
1255
cached_version = parent_texts.get(key, None)
1256
if cached_version is not None:
1257
# Ensure the cache dict is valid.
1258
if not self.get_parent_map([key]):
1259
raise RevisionNotPresent(key, self)
1260
return cached_version
1261
generator = _VFContentMapGenerator(self, [key])
1262
return generator._get_content(key)
1264
def get_parent_map(self, keys):
1265
"""Get a map of the graph parents of keys.
1267
:param keys: The keys to look up parents for.
1268
:return: A mapping from keys to parents. Absent keys are absent from
1271
return self._get_parent_map_with_sources(keys)[0]
1273
def _get_parent_map_with_sources(self, keys):
1274
"""Get a map of the parents of keys.
1276
:param keys: The keys to look up parents for.
1277
:return: A tuple. The first element is a mapping from keys to parents.
1278
Absent keys are absent from the mapping. The second element is a
1279
list with the locations each key was found in. The first element
1280
is the in-this-knit parents, the second the first fallback source,
1284
sources = [self._index] + self._immediate_fallback_vfs
1287
for source in sources:
1290
new_result = source.get_parent_map(missing)
1291
source_results.append(new_result)
1292
result.update(new_result)
1293
missing.difference_update(set(new_result))
1294
return result, source_results
1296
def _get_record_map(self, keys, allow_missing=False):
1297
"""Produce a dictionary of knit records.
1299
:return: {key:(record, record_details, digest, next)}
1301
* record: data returned from read_records (a KnitContentobject)
1302
* record_details: opaque information to pass to parse_record
1303
* digest: SHA1 digest of the full text after all steps are done
1304
* next: build-parent of the version, i.e. the leftmost ancestor.
1305
Will be None if the record is not a delta.
1307
:param keys: The keys to build a map for
1308
:param allow_missing: If some records are missing, rather than
1309
error, just return the data that could be generated.
1311
raw_map = self._get_record_map_unparsed(keys,
1312
allow_missing=allow_missing)
1313
return self._raw_map_to_record_map(raw_map)
1315
def _raw_map_to_record_map(self, raw_map):
1316
"""Parse the contents of _get_record_map_unparsed.
1318
:return: see _get_record_map.
1322
data, record_details, next = raw_map[key]
1323
content, digest = self._parse_record(key[-1], data)
1324
result[key] = content, record_details, digest, next
1327
def _get_record_map_unparsed(self, keys, allow_missing=False):
1328
"""Get the raw data for reconstructing keys without parsing it.
1330
:return: A dict suitable for parsing via _raw_map_to_record_map.
1331
key-> raw_bytes, (method, noeol), compression_parent
1333
# This retries the whole request if anything fails. Potentially we
1334
# could be a bit more selective. We could track the keys whose records
1335
# we have successfully found, and then only request the new records
1336
# from there. However, _get_components_positions grabs the whole build
1337
# chain, which means we'll likely try to grab the same records again
1338
# anyway. Also, can the build chains change as part of a pack
1339
# operation? We wouldn't want to end up with a broken chain.
1342
position_map = self._get_components_positions(keys,
1343
allow_missing=allow_missing)
1344
# key = component_id, r = record_details, i_m = index_memo,
1346
records = [(key, i_m) for key, (r, i_m, n)
1347
in viewitems(position_map)]
1348
# Sort by the index memo, so that we request records from the
1349
# same pack file together, and in forward-sorted order
1350
records.sort(key=operator.itemgetter(1))
1352
for key, data in self._read_records_iter_unchecked(records):
1353
(record_details, index_memo, next) = position_map[key]
1354
raw_record_map[key] = data, record_details, next
1355
return raw_record_map
1356
except errors.RetryWithNewPacks as e:
1357
self._access.reload_or_raise(e)
1360
def _split_by_prefix(cls, keys):
1361
"""For the given keys, split them up based on their prefix.
1363
To keep memory pressure somewhat under control, split the
1364
requests back into per-file-id requests, otherwise "bzr co"
1365
extracts the full tree into memory before writing it to disk.
1366
This should be revisited if _get_content_maps() can ever cross
1369
The keys for a given file_id are kept in the same relative order.
1370
Ordering between file_ids is not, though prefix_order will return the
1371
order that the key was first seen.
1373
:param keys: An iterable of key tuples
1374
:return: (split_map, prefix_order)
1375
split_map A dictionary mapping prefix => keys
1376
prefix_order The order that we saw the various prefixes
1378
split_by_prefix = {}
1386
if prefix in split_by_prefix:
1387
split_by_prefix[prefix].append(key)
1389
split_by_prefix[prefix] = [key]
1390
prefix_order.append(prefix)
1391
return split_by_prefix, prefix_order
1393
def _group_keys_for_io(self, keys, non_local_keys, positions,
1394
_min_buffer_size=_STREAM_MIN_BUFFER_SIZE):
1395
"""For the given keys, group them into 'best-sized' requests.
1397
The idea is to avoid making 1 request per file, but to never try to
1398
unpack an entire 1.5GB source tree in a single pass. Also when
1399
possible, we should try to group requests to the same pack file
1402
:return: list of (keys, non_local) tuples that indicate what keys
1403
should be fetched next.
1405
# TODO: Ideally we would group on 2 factors. We want to extract texts
1406
# from the same pack file together, and we want to extract all
1407
# the texts for a given build-chain together. Ultimately it
1408
# probably needs a better global view.
1409
total_keys = len(keys)
1410
prefix_split_keys, prefix_order = self._split_by_prefix(keys)
1411
prefix_split_non_local_keys, _ = self._split_by_prefix(non_local_keys)
1413
cur_non_local = set()
1417
for prefix in prefix_order:
1418
keys = prefix_split_keys[prefix]
1419
non_local = prefix_split_non_local_keys.get(prefix, [])
1421
this_size = self._index._get_total_build_size(keys, positions)
1422
cur_size += this_size
1423
cur_keys.extend(keys)
1424
cur_non_local.update(non_local)
1425
if cur_size > _min_buffer_size:
1426
result.append((cur_keys, cur_non_local))
1427
sizes.append(cur_size)
1429
cur_non_local = set()
1432
result.append((cur_keys, cur_non_local))
1433
sizes.append(cur_size)
1436
def get_record_stream(self, keys, ordering, include_delta_closure):
1437
"""Get a stream of records for keys.
1439
:param keys: The keys to include.
1440
:param ordering: Either 'unordered' or 'topological'. A topologically
1441
sorted stream has compression parents strictly before their
1443
:param include_delta_closure: If True then the closure across any
1444
compression parents will be included (in the opaque data).
1445
:return: An iterator of ContentFactory objects, each of which is only
1446
valid until the iterator is advanced.
1448
# keys might be a generator
1452
if not self._index.has_graph:
1453
# Cannot sort when no graph has been stored.
1454
ordering = 'unordered'
1456
remaining_keys = keys
1459
keys = set(remaining_keys)
1460
for content_factory in self._get_remaining_record_stream(keys,
1461
ordering, include_delta_closure):
1462
remaining_keys.discard(content_factory.key)
1463
yield content_factory
1465
except errors.RetryWithNewPacks as e:
1466
self._access.reload_or_raise(e)
1468
def _get_remaining_record_stream(self, keys, ordering,
1469
include_delta_closure):
1470
"""This function is the 'retry' portion for get_record_stream."""
1471
if include_delta_closure:
1472
positions = self._get_components_positions(keys, allow_missing=True)
1474
build_details = self._index.get_build_details(keys)
1476
# (record_details, access_memo, compression_parent_key)
1477
positions = dict((key, self._build_details_to_components(details))
1478
for key, details in viewitems(build_details))
1479
absent_keys = keys.difference(set(positions))
1480
# There may be more absent keys : if we're missing the basis component
1481
# and are trying to include the delta closure.
1482
# XXX: We should not ever need to examine remote sources because we do
1483
# not permit deltas across versioned files boundaries.
1484
if include_delta_closure:
1485
needed_from_fallback = set()
1486
# Build up reconstructable_keys dict. key:True in this dict means
1487
# the key can be reconstructed.
1488
reconstructable_keys = {}
1492
chain = [key, positions[key][2]]
1494
needed_from_fallback.add(key)
1497
while chain[-1] is not None:
1498
if chain[-1] in reconstructable_keys:
1499
result = reconstructable_keys[chain[-1]]
1503
chain.append(positions[chain[-1]][2])
1505
# missing basis component
1506
needed_from_fallback.add(chain[-1])
1509
for chain_key in chain[:-1]:
1510
reconstructable_keys[chain_key] = result
1512
needed_from_fallback.add(key)
1513
# Double index lookups here : need a unified api ?
1514
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1515
if ordering in ('topological', 'groupcompress'):
1516
if ordering == 'topological':
1517
# Global topological sort
1518
present_keys = tsort.topo_sort(global_map)
1520
present_keys = sort_groupcompress(global_map)
1521
# Now group by source:
1523
current_source = None
1524
for key in present_keys:
1525
for parent_map in parent_maps:
1526
if key in parent_map:
1527
key_source = parent_map
1529
if current_source is not key_source:
1530
source_keys.append((key_source, []))
1531
current_source = key_source
1532
source_keys[-1][1].append(key)
1534
if ordering != 'unordered':
1535
raise AssertionError('valid values for ordering are:'
1536
' "unordered", "groupcompress" or "topological" not: %r'
1538
# Just group by source; remote sources first.
1541
for parent_map in reversed(parent_maps):
1542
source_keys.append((parent_map, []))
1543
for key in parent_map:
1544
present_keys.append(key)
1545
source_keys[-1][1].append(key)
1546
# We have been requested to return these records in an order that
1547
# suits us. So we ask the index to give us an optimally sorted
1549
for source, sub_keys in source_keys:
1550
if source is parent_maps[0]:
1551
# Only sort the keys for this VF
1552
self._index._sort_keys_by_io(sub_keys, positions)
1553
absent_keys = keys - set(global_map)
1554
for key in absent_keys:
1555
yield AbsentContentFactory(key)
1556
# restrict our view to the keys we can answer.
1557
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1558
# XXX: At that point we need to consider the impact of double reads by
1559
# utilising components multiple times.
1560
if include_delta_closure:
1561
# XXX: get_content_maps performs its own index queries; allow state
1563
non_local_keys = needed_from_fallback - absent_keys
1564
for keys, non_local_keys in self._group_keys_for_io(present_keys,
1567
generator = _VFContentMapGenerator(self, keys, non_local_keys,
1570
for record in generator.get_record_stream():
1573
for source, keys in source_keys:
1574
if source is parent_maps[0]:
1575
# this KnitVersionedFiles
1576
records = [(key, positions[key][1]) for key in keys]
1577
for key, raw_data in self._read_records_iter_unchecked(records):
1578
(record_details, index_memo, _) = positions[key]
1579
yield KnitContentFactory(key, global_map[key],
1580
record_details, None, raw_data, self._factory.annotated, None)
1582
vf = self._immediate_fallback_vfs[parent_maps.index(source) - 1]
1583
for record in vf.get_record_stream(keys, ordering,
1584
include_delta_closure):
1587
def get_sha1s(self, keys):
1588
"""See VersionedFiles.get_sha1s()."""
1590
record_map = self._get_record_map(missing, allow_missing=True)
1592
for key, details in viewitems(record_map):
1593
if key not in missing:
1595
# record entry 2 is the 'digest'.
1596
result[key] = details[2]
1597
missing.difference_update(set(result))
1598
for source in self._immediate_fallback_vfs:
1601
new_result = source.get_sha1s(missing)
1602
result.update(new_result)
1603
missing.difference_update(set(new_result))
1606
def insert_record_stream(self, stream):
1607
"""Insert a record stream into this container.
1609
:param stream: A stream of records to insert.
1611
:seealso VersionedFiles.get_record_stream:
1613
def get_adapter(adapter_key):
1615
return adapters[adapter_key]
1617
adapter_factory = adapter_registry.get(adapter_key)
1618
adapter = adapter_factory(self)
1619
adapters[adapter_key] = adapter
1622
if self._factory.annotated:
1623
# self is annotated, we need annotated knits to use directly.
1624
annotated = "annotated-"
1627
# self is not annotated, but we can strip annotations cheaply.
1629
convertibles = {"knit-annotated-ft-gz"}
1630
if self._max_delta_chain:
1631
delta_types.add("knit-annotated-delta-gz")
1632
convertibles.add("knit-annotated-delta-gz")
1633
# The set of types we can cheaply adapt without needing basis texts.
1634
native_types = set()
1635
if self._max_delta_chain:
1636
native_types.add("knit-%sdelta-gz" % annotated)
1637
delta_types.add("knit-%sdelta-gz" % annotated)
1638
native_types.add("knit-%sft-gz" % annotated)
1639
knit_types = native_types.union(convertibles)
1641
# Buffer all index entries that we can't add immediately because their
1642
# basis parent is missing. We don't buffer all because generating
1643
# annotations may require access to some of the new records. However we
1644
# can't generate annotations from new deltas until their basis parent
1645
# is present anyway, so we get away with not needing an index that
1646
# includes the new keys.
1648
# See <http://launchpad.net/bugs/300177> about ordering of compression
1649
# parents in the records - to be conservative, we insist that all
1650
# parents must be present to avoid expanding to a fulltext.
1652
# key = basis_parent, value = index entry to add
1653
buffered_index_entries = {}
1654
for record in stream:
1655
kind = record.storage_kind
1656
if kind.startswith('knit-') and kind.endswith('-gz'):
1657
# Check that the ID in the header of the raw knit bytes matches
1658
# the record metadata.
1659
raw_data = record._raw_record
1660
df, rec = self._parse_record_header(record.key, raw_data)
1663
parents = record.parents
1664
if record.storage_kind in delta_types:
1665
# TODO: eventually the record itself should track
1666
# compression_parent
1667
compression_parent = parents[0]
1669
compression_parent = None
1670
# Raise an error when a record is missing.
1671
if record.storage_kind == 'absent':
1672
raise RevisionNotPresent([record.key], self)
1673
elif ((record.storage_kind in knit_types)
1674
and (compression_parent is None
1675
or not self._immediate_fallback_vfs
1676
or compression_parent in self._index
1677
or compression_parent not in self)):
1678
# we can insert the knit record literally if either it has no
1679
# compression parent OR we already have its basis in this kvf
1680
# OR the basis is not present even in the fallbacks. In the
1681
# last case it will either turn up later in the stream and all
1682
# will be well, or it won't turn up at all and we'll raise an
1685
# TODO: self.__contains__ is somewhat redundant with
1686
# self._index.__contains__; we really want something that directly
1687
# asks if it's only present in the fallbacks. -- mbp 20081119
1688
if record.storage_kind not in native_types:
1690
adapter_key = (record.storage_kind, "knit-delta-gz")
1691
adapter = get_adapter(adapter_key)
1693
adapter_key = (record.storage_kind, "knit-ft-gz")
1694
adapter = get_adapter(adapter_key)
1695
bytes = adapter.get_bytes(record)
1697
# It's a knit record, it has a _raw_record field (even if
1698
# it was reconstituted from a network stream).
1699
bytes = record._raw_record
1700
options = [record._build_details[0].encode('ascii')]
1701
if record._build_details[1]:
1702
options.append(b'no-eol')
1703
# Just blat it across.
1704
# Note: This does end up adding data on duplicate keys. As
1705
# modern repositories use atomic insertions this should not
1706
# lead to excessive growth in the event of interrupted fetches.
1707
# 'knit' repositories may suffer excessive growth, but as a
1708
# deprecated format this is tolerable. It can be fixed if
1709
# needed by in the kndx index support raising on a duplicate
1710
# add with identical parents and options.
1711
access_memo = self._access.add_raw_records(
1712
[(record.key, len(bytes))], bytes)[0]
1713
index_entry = (record.key, options, access_memo, parents)
1714
if b'fulltext' not in options:
1715
# Not a fulltext, so we need to make sure the compression
1716
# parent will also be present.
1717
# Note that pack backed knits don't need to buffer here
1718
# because they buffer all writes to the transaction level,
1719
# but we don't expose that difference at the index level. If
1720
# the query here has sufficient cost to show up in
1721
# profiling we should do that.
1723
# They're required to be physically in this
1724
# KnitVersionedFiles, not in a fallback.
1725
if compression_parent not in self._index:
1726
pending = buffered_index_entries.setdefault(
1727
compression_parent, [])
1728
pending.append(index_entry)
1731
self._index.add_records([index_entry])
1732
elif record.storage_kind == 'chunked':
1733
self.add_lines(record.key, parents,
1734
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1736
# Not suitable for direct insertion as a
1737
# delta, either because it's not the right format, or this
1738
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1739
# 0) or because it depends on a base only present in the
1741
self._access.flush()
1743
# Try getting a fulltext directly from the record.
1744
bytes = record.get_bytes_as('fulltext')
1745
except errors.UnavailableRepresentation:
1746
adapter_key = record.storage_kind, 'fulltext'
1747
adapter = get_adapter(adapter_key)
1748
bytes = adapter.get_bytes(record)
1749
lines = split_lines(bytes)
1751
self.add_lines(record.key, parents, lines)
1752
except errors.RevisionAlreadyPresent:
1754
# Add any records whose basis parent is now available.
1756
added_keys = [record.key]
1758
key = added_keys.pop(0)
1759
if key in buffered_index_entries:
1760
index_entries = buffered_index_entries[key]
1761
self._index.add_records(index_entries)
1763
[index_entry[0] for index_entry in index_entries])
1764
del buffered_index_entries[key]
1765
if buffered_index_entries:
1766
# There were index entries buffered at the end of the stream,
1767
# So these need to be added (if the index supports holding such
1768
# entries for later insertion)
1770
for key in buffered_index_entries:
1771
index_entries = buffered_index_entries[key]
1772
all_entries.extend(index_entries)
1773
self._index.add_records(
1774
all_entries, missing_compression_parents=True)
1776
def get_missing_compression_parent_keys(self):
1777
"""Return an iterable of keys of missing compression parents.
1779
Check this after calling insert_record_stream to find out if there are
1780
any missing compression parents. If there are, the records that
1781
depend on them are not able to be inserted safely. For atomic
1782
KnitVersionedFiles built on packs, the transaction should be aborted or
1783
suspended - commit will fail at this point. Nonatomic knits will error
1784
earlier because they have no staging area to put pending entries into.
1786
return self._index.get_missing_compression_parents()
1788
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1789
"""Iterate over the lines in the versioned files from keys.
1791
This may return lines from other keys. Each item the returned
1792
iterator yields is a tuple of a line and a text version that that line
1793
is present in (not introduced in).
1795
Ordering of results is in whatever order is most suitable for the
1796
underlying storage format.
1798
If a progress bar is supplied, it may be used to indicate progress.
1799
The caller is responsible for cleaning up progress bars (because this
1803
* Lines are normalised by the underlying store: they will all have \\n
1805
* Lines are returned in arbitrary order.
1806
* If a requested key did not change any lines (or didn't have any
1807
lines), it may not be mentioned at all in the result.
1809
:param pb: Progress bar supplied by caller.
1810
:return: An iterator over (line, key).
1813
pb = ui.ui_factory.nested_progress_bar()
1819
# we don't care about inclusions, the caller cares.
1820
# but we need to setup a list of records to visit.
1821
# we need key, position, length
1823
build_details = self._index.get_build_details(keys)
1824
for key, details in viewitems(build_details):
1826
key_records.append((key, details[0]))
1827
records_iter = enumerate(self._read_records_iter(key_records))
1828
for (key_idx, (key, data, sha_value)) in records_iter:
1829
pb.update(gettext('Walking content'), key_idx, total)
1830
compression_parent = build_details[key][1]
1831
if compression_parent is None:
1833
line_iterator = self._factory.get_fulltext_content(data)
1836
line_iterator = self._factory.get_linedelta_content(data)
1837
# Now that we are yielding the data for this key, remove it
1840
# XXX: It might be more efficient to yield (key,
1841
# line_iterator) in the future. However for now, this is a
1842
# simpler change to integrate into the rest of the
1843
# codebase. RBC 20071110
1844
for line in line_iterator:
1847
except errors.RetryWithNewPacks as e:
1848
self._access.reload_or_raise(e)
1849
# If there are still keys we've not yet found, we look in the fallback
1850
# vfs, and hope to find them there. Note that if the keys are found
1851
# but had no changes or no content, the fallback may not return
1853
if keys and not self._immediate_fallback_vfs:
1854
# XXX: strictly the second parameter is meant to be the file id
1855
# but it's not easily accessible here.
1856
raise RevisionNotPresent(keys, repr(self))
1857
for source in self._immediate_fallback_vfs:
1861
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1862
source_keys.add(key)
1864
keys.difference_update(source_keys)
1865
pb.update(gettext('Walking content'), total, total)
1867
def _make_line_delta(self, delta_seq, new_content):
1868
"""Generate a line delta from delta_seq and new_content."""
1870
for op in delta_seq.get_opcodes():
1871
if op[0] == 'equal':
1873
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
404
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
405
"""See VersionedFile._add_delta()."""
406
self._check_add(version_id, []) # should we check the lines ?
407
self._check_versions_present(parents)
411
for parent in parents:
412
if not self.has_version(parent):
413
ghosts.append(parent)
415
present_parents.append(parent)
417
if delta_parent is None:
418
# reconstitute as full text.
419
assert len(delta) == 1 or len(delta) == 0
421
assert delta[0][0] == 0
422
assert delta[0][1] == 0, delta[0][1]
423
return super(KnitVersionedFile, self)._add_delta(version_id,
434
options.append('no-eol')
436
if delta_parent is not None:
437
# determine the current delta chain length.
438
# To speed the extract of texts the delta chain is limited
439
# to a fixed number of deltas. This should minimize both
440
# I/O and the time spend applying deltas.
441
# The window was changed to a maximum of 200 deltas, but also added
442
# was a check that the total compressed size of the deltas is
443
# smaller than the compressed size of the fulltext.
444
if not self._check_should_delta([delta_parent]):
445
# We don't want a delta here, just do a normal insertion.
446
return super(KnitVersionedFile, self)._add_delta(version_id,
453
options.append('line-delta')
454
store_lines = self.factory.lower_line_delta(delta)
456
where, size = self._data.add_record(version_id, digest, store_lines)
457
self._index.add_version(version_id, options, where, size, parents)
459
def _add_raw_records(self, records, data):
460
"""Add all the records 'records' with data pre-joined in 'data'.
462
:param records: A list of tuples(version_id, options, parents, size).
463
:param data: The data for the records. When it is written, the records
464
are adjusted to have pos pointing into data by the sum of
465
the preceding records sizes.
468
pos = self._data.add_raw_record(data)
471
for (version_id, options, parents, size) in records:
472
index_entries.append((version_id, options, pos+offset,
474
if self._data._do_cache:
475
self._data._cache[version_id] = data[offset:offset+size]
477
self._index.add_versions(index_entries)
479
def enable_cache(self):
480
"""Start caching data for this knit"""
481
self._data.enable_cache()
483
def clear_cache(self):
484
"""Clear the data cache only."""
485
self._data.clear_cache()
487
def copy_to(self, name, transport):
488
"""See VersionedFile.copy_to()."""
489
# copy the current index to a temp index to avoid racing with local
491
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
492
self.transport.get(self._index._filename))
494
f = self._data._open_file()
496
transport.put_file(name + DATA_SUFFIX, f)
499
# move the copied index into place
500
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
502
def create_empty(self, name, transport, mode=None):
503
return KnitVersionedFile(name, transport, factory=self.factory,
504
delta=self.delta, create=True)
506
def _fix_parents(self, version_id, new_parents):
507
"""Fix the parents list for version.
509
This is done by appending a new version to the index
510
with identical data except for the parents list.
511
the parents list must be a superset of the current
514
current_values = self._index._cache[version_id]
515
assert set(current_values[4]).difference(set(new_parents)) == set()
516
self._index.add_version(version_id,
522
def get_delta(self, version_id):
523
"""Get a delta for constructing version from some other version."""
524
version_id = osutils.safe_revision_id(version_id)
525
self.check_not_reserved_id(version_id)
526
if not self.has_version(version_id):
527
raise RevisionNotPresent(version_id, self.filename)
529
parents = self.get_parents(version_id)
534
data_pos, data_size = self._index.get_position(version_id)
535
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
536
noeol = 'no-eol' in self._index.get_options(version_id)
537
if 'fulltext' == self._index.get_method(version_id):
538
new_content = self.factory.parse_fulltext(data, version_id)
539
if parent is not None:
540
reference_content = self._get_content(parent)
541
old_texts = reference_content.text()
544
new_texts = new_content.text()
545
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
546
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
548
delta = self.factory.parse_line_delta(data, version_id)
549
return parent, sha1, noeol, delta
551
def get_graph_with_ghosts(self):
552
"""See VersionedFile.get_graph_with_ghosts()."""
553
graph_items = self._index.get_graph()
554
return dict(graph_items)
556
def get_sha1(self, version_id):
557
"""See VersionedFile.get_sha1()."""
558
version_id = osutils.safe_revision_id(version_id)
559
record_map = self._get_record_map([version_id])
560
method, content, digest, next = record_map[version_id]
565
"""See VersionedFile.get_suffixes()."""
566
return [DATA_SUFFIX, INDEX_SUFFIX]
568
def has_ghost(self, version_id):
569
"""True if there is a ghost reference in the file to version_id."""
570
version_id = osutils.safe_revision_id(version_id)
572
if self.has_version(version_id):
574
# optimisable if needed by memoising the _ghosts set.
575
items = self._index.get_graph()
576
for node, parents in items:
577
for parent in parents:
578
if parent not in self._index._cache:
579
if parent == version_id:
584
"""See VersionedFile.versions."""
585
return self._index.get_versions()
587
def has_version(self, version_id):
588
"""See VersionedFile.has_version."""
589
version_id = osutils.safe_revision_id(version_id)
590
return self._index.has_version(version_id)
592
__contains__ = has_version
1876
594
def _merge_annotations(self, content, parents, parent_texts={},
1877
delta=None, annotated=None,
1878
left_matching_blocks=None):
1879
"""Merge annotations for content and generate deltas.
1881
This is done by comparing the annotations based on changes to the text
1882
and generating a delta on the resulting full texts. If annotations are
1883
not being created then a simple delta is created.
595
delta=None, annotated=None):
596
"""Merge annotations for content. This is done by comparing
597
the annotations based on changed to the text.
1885
if left_matching_blocks is not None:
1886
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1888
600
delta_seq = None
1890
for parent_key in parents:
1891
merge_content = self._get_content(parent_key, parent_texts)
1892
if (parent_key == parents[0] and delta_seq is not None):
1895
seq = patiencediff.PatienceSequenceMatcher(
1896
None, merge_content.text(), content.text())
601
for parent_id in parents:
602
merge_content = self._get_content(parent_id, parent_texts)
603
seq = patiencediff.PatienceSequenceMatcher(
604
None, merge_content.text(), content.text())
605
if delta_seq is None:
606
# setup a delta seq to reuse.
1897
608
for i, j, n in seq.get_matching_blocks():
1900
# this copies (origin, text) pairs across to the new
1901
# content for any line that matches the last-checked
611
# this appears to copy (origin, text) pairs across to the new
612
# content for any line that matches the last-checked parent.
613
# FIXME: save the sequence control data for delta compression
614
# against the most relevant parent rather than rediffing.
1903
615
content._lines[j:j+n] = merge_content._lines[i:i+n]
1904
# XXX: Robert says the following block is a workaround for a
1905
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1906
if content._lines and not content._lines[-1][1].endswith(b'\n'):
1907
# The copied annotation was from a line without a trailing EOL,
1908
# reinstate one for the content object, to ensure correct
1910
line = content._lines[-1][1] + b'\n'
1911
content._lines[-1] = (content._lines[-1][0], line)
1913
if delta_seq is None:
1914
618
reference_content = self._get_content(parents[0], parent_texts)
1915
619
new_texts = content.text()
1916
620
old_texts = reference_content.text()
1918
622
None, old_texts, new_texts)
1919
623
return self._make_line_delta(delta_seq, content)
1921
def _parse_record(self, version_id, data):
1922
"""Parse an original format knit record.
1924
These have the last element of the key only present in the stored data.
1926
rec, record_contents = self._parse_record_unchecked(data)
1927
self._check_header_version(rec, version_id)
1928
return record_contents, rec[3]
1930
def _parse_record_header(self, key, raw_data):
1931
"""Parse a record header for consistency.
1933
:return: the header and the decompressor stream.
1934
as (stream, header_record)
1936
df = gzip.GzipFile(mode='rb', fileobj=BytesIO(raw_data))
1939
rec = self._check_header(key, df.readline())
1940
except Exception as e:
1941
raise KnitCorrupt(self,
1942
"While reading {%s} got %s(%s)"
1943
% (key, e.__class__.__name__, str(e)))
1946
def _parse_record_unchecked(self, data):
1948
# 4168 calls in 2880 217 internal
1949
# 4168 calls to _parse_record_header in 2121
1950
# 4168 calls to readlines in 330
1951
with gzip.GzipFile(mode='rb', fileobj=BytesIO(data)) as df:
1953
record_contents = df.readlines()
1954
except Exception as e:
1955
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1956
(data, e.__class__.__name__, str(e)))
1957
header = record_contents.pop(0)
1958
rec = self._split_header(header)
1959
last_line = record_contents.pop()
1960
if len(record_contents) != int(rec[2]):
1961
raise KnitCorrupt(self,
1962
'incorrect number of lines %s != %s'
1963
' for version {%s} %s'
1964
% (len(record_contents), int(rec[2]),
1965
rec[1], record_contents))
1966
if last_line != b'end %s\n' % rec[1]:
1967
raise KnitCorrupt(self,
1968
'unexpected version end line %r, wanted %r'
1969
% (last_line, rec[1]))
1970
return rec, record_contents
1972
def _read_records_iter(self, records):
1973
"""Read text records from data file and yield result.
1975
The result will be returned in whatever is the fastest to read.
1976
Not by the order requested. Also, multiple requests for the same
1977
record will only yield 1 response.
1979
:param records: A list of (key, access_memo) entries
1980
:return: Yields (key, contents, digest) in the order
1981
read, not the order requested
1986
# XXX: This smells wrong, IO may not be getting ordered right.
1987
needed_records = sorted(set(records), key=operator.itemgetter(1))
1988
if not needed_records:
1991
# The transport optimizes the fetching as well
1992
# (ie, reads continuous ranges.)
1993
raw_data = self._access.get_raw_records(
1994
[index_memo for key, index_memo in needed_records])
1996
for (key, index_memo), data in zip(needed_records, raw_data):
1997
content, digest = self._parse_record(key[-1], data)
1998
yield key, content, digest
2000
def _read_records_iter_raw(self, records):
2001
"""Read text records from data file and yield raw data.
2003
This unpacks enough of the text record to validate the id is
2004
as expected but thats all.
2006
Each item the iterator yields is (key, bytes,
2007
expected_sha1_of_full_text).
2009
for key, data in self._read_records_iter_unchecked(records):
2010
# validate the header (note that we can only use the suffix in
2011
# current knit records).
2012
df, rec = self._parse_record_header(key, data)
2014
yield key, data, rec[3]
2016
def _read_records_iter_unchecked(self, records):
2017
"""Read text records from data file and yield raw data.
2019
No validation is done.
2021
Yields tuples of (key, data).
2023
# setup an iterator of the external records:
2024
# uses readv so nice and fast we hope.
2026
# grab the disk data needed.
2027
needed_offsets = [index_memo for key, index_memo
2029
raw_records = self._access.get_raw_records(needed_offsets)
2031
for key, index_memo in records:
2032
data = next(raw_records)
2035
def _record_to_data(self, key, digest, lines, dense_lines=None):
2036
"""Convert key, digest, lines into a raw data block.
2038
:param key: The key of the record. Currently keys are always serialised
2039
using just the trailing component.
2040
:param dense_lines: The bytes of lines but in a denser form. For
2041
instance, if lines is a list of 1000 bytestrings each ending in
2042
\\n, dense_lines may be a list with one line in it, containing all
2043
the 1000's lines and their \\n's. Using dense_lines if it is
2044
already known is a win because the string join to create bytes in
2045
this function spends less time resizing the final string.
2046
:return: (len, a BytesIO instance with the raw data ready to read.)
2048
chunks = [b"version %s %d %s\n" % (key[-1], len(lines), digest)]
2049
chunks.extend(dense_lines or lines)
2050
chunks.append(b"end " + key[-1] + b"\n")
2051
for chunk in chunks:
2052
if not isinstance(chunk, bytes):
2053
raise AssertionError(
2054
'data must be plain bytes was %s' % type(chunk))
2055
if lines and not lines[-1].endswith(b'\n'):
2056
raise ValueError('corrupt lines value %r' % lines)
2057
compressed_bytes = b''.join(tuned_gzip.chunks_to_gzip(chunks))
2058
return len(compressed_bytes), compressed_bytes
2060
def _split_header(self, line):
2063
raise KnitCorrupt(self,
2064
'unexpected number of elements in record header')
2068
"""See VersionedFiles.keys."""
2069
if 'evil' in debug.debug_flags:
2070
trace.mutter_callsite(2, "keys scales with size of history")
2071
sources = [self._index] + self._immediate_fallback_vfs
2073
for source in sources:
2074
result.update(source.keys())
2078
class _ContentMapGenerator(object):
2079
"""Generate texts or expose raw deltas for a set of texts."""
2081
def __init__(self, ordering='unordered'):
2082
self._ordering = ordering
2084
def _get_content(self, key):
2085
"""Get the content object for key."""
2086
# Note that _get_content is only called when the _ContentMapGenerator
2087
# has been constructed with just one key requested for reconstruction.
2088
if key in self.nonlocal_keys:
2089
record = next(self.get_record_stream())
2090
# Create a content object on the fly
2091
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
2092
return PlainKnitContent(lines, record.key)
625
def _make_line_delta(self, delta_seq, new_content):
626
"""Generate a line delta from delta_seq and new_content."""
628
for op in delta_seq.get_opcodes():
631
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
634
def _get_components_positions(self, version_ids):
635
"""Produce a map of position data for the components of versions.
637
This data is intended to be used for retrieving the knit records.
639
A dict of version_id to (method, data_pos, data_size, next) is
641
method is the way referenced data should be applied.
642
data_pos is the position of the data in the knit.
643
data_size is the size of the data in the knit.
644
next is the build-parent of the version, or None for fulltexts.
647
for version_id in version_ids:
650
while cursor is not None and cursor not in component_data:
651
method = self._index.get_method(cursor)
652
if method == 'fulltext':
655
next = self.get_parents(cursor)[0]
656
data_pos, data_size = self._index.get_position(cursor)
657
component_data[cursor] = (method, data_pos, data_size, next)
659
return component_data
661
def _get_content(self, version_id, parent_texts={}):
662
"""Returns a content object that makes up the specified
664
if not self.has_version(version_id):
665
raise RevisionNotPresent(version_id, self.filename)
667
cached_version = parent_texts.get(version_id, None)
668
if cached_version is not None:
669
return cached_version
671
text_map, contents_map = self._get_content_maps([version_id])
672
return contents_map[version_id]
674
def _check_versions_present(self, version_ids):
675
"""Check that all specified versions are present."""
676
self._index.check_versions_present(version_ids)
678
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
679
"""See VersionedFile.add_lines_with_ghosts()."""
680
self._check_add(version_id, lines)
681
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
683
def _add_lines(self, version_id, parents, lines, parent_texts):
684
"""See VersionedFile.add_lines."""
685
self._check_add(version_id, lines)
686
self._check_versions_present(parents)
687
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
689
def _check_add(self, version_id, lines):
690
"""check that version_id and lines are safe to add."""
691
assert self.writable, "knit is not opened for write"
692
### FIXME escape. RBC 20060228
693
if contains_whitespace(version_id):
694
raise InvalidRevisionId(version_id, self.filename)
695
self.check_not_reserved_id(version_id)
696
if self.has_version(version_id):
697
raise RevisionAlreadyPresent(version_id, self.filename)
698
self._check_lines_not_unicode(lines)
699
self._check_lines_are_lines(lines)
701
def _add(self, version_id, lines, parents, delta, parent_texts):
702
"""Add a set of lines on top of version specified by parents.
704
If delta is true, compress the text as a line-delta against
707
Any versions not present will be converted into ghosts.
709
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
710
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
711
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
712
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
713
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
714
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
715
# +1383 0 8.0370 8.0370 +<len>
716
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
717
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
718
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
719
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
723
if parent_texts is None:
725
for parent in parents:
726
if not self.has_version(parent):
727
ghosts.append(parent)
729
present_parents.append(parent)
731
if delta and not len(present_parents):
734
digest = sha_strings(lines)
737
if lines[-1][-1] != '\n':
738
options.append('no-eol')
739
lines[-1] = lines[-1] + '\n'
741
if len(present_parents) and delta:
742
# To speed the extract of texts the delta chain is limited
743
# to a fixed number of deltas. This should minimize both
744
# I/O and the time spend applying deltas.
745
delta = self._check_should_delta(present_parents)
747
assert isinstance(version_id, str)
748
lines = self.factory.make(lines, version_id)
749
if delta or (self.factory.annotated and len(present_parents) > 0):
750
# Merge annotations from parent texts if so is needed.
751
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
752
delta, self.factory.annotated)
755
options.append('line-delta')
756
store_lines = self.factory.lower_line_delta(delta_hunks)
2094
# local keys we can ask for directly
2095
return self._get_one_work(key)
2097
def get_record_stream(self):
2098
"""Get a record stream for the keys requested during __init__."""
2099
for record in self._work():
2103
"""Produce maps of text and KnitContents as dicts.
758
options.append('fulltext')
759
store_lines = self.factory.lower_fulltext(lines)
761
where, size = self._data.add_record(version_id, digest, store_lines)
762
self._index.add_version(version_id, options, where, size, parents)
765
def check(self, progress_bar=None):
766
"""See VersionedFile.check()."""
768
def _clone_text(self, new_version_id, old_version_id, parents):
769
"""See VersionedFile.clone_text()."""
770
# FIXME RBC 20060228 make fast by only inserting an index with null
772
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
774
def get_lines(self, version_id):
775
"""See VersionedFile.get_lines()."""
776
return self.get_line_list([version_id])[0]
778
def _get_record_map(self, version_ids):
779
"""Produce a dictionary of knit records.
781
The keys are version_ids, the values are tuples of (method, content,
783
method is the way the content should be applied.
784
content is a KnitContent object.
785
digest is the SHA1 digest of this version id after all steps are done
786
next is the build-parent of the version, i.e. the leftmost ancestor.
787
If the method is fulltext, next will be None.
789
position_map = self._get_components_positions(version_ids)
790
# c = component_id, m = method, p = position, s = size, n = next
791
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
793
for component_id, content, digest in \
794
self._data.read_records_iter(records):
795
method, position, size, next = position_map[component_id]
796
record_map[component_id] = method, content, digest, next
800
def get_text(self, version_id):
801
"""See VersionedFile.get_text"""
802
return self.get_texts([version_id])[0]
804
def get_texts(self, version_ids):
805
return [''.join(l) for l in self.get_line_list(version_ids)]
807
def get_line_list(self, version_ids):
808
"""Return the texts of listed versions as a list of strings."""
809
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
810
for version_id in version_ids:
811
self.check_not_reserved_id(version_id)
812
text_map, content_map = self._get_content_maps(version_ids)
813
return [text_map[v] for v in version_ids]
815
def _get_content_maps(self, version_ids):
816
"""Produce maps of text and KnitContents
2105
818
:return: (text_map, content_map) where text_map contains the texts for
2106
the requested versions and content_map contains the KnitContents.
819
the requested versions and content_map contains the KnitContents.
820
Both dicts take version_ids as their keys.
2108
# NB: By definition we never need to read remote sources unless texts
2109
# are requested from them: we don't delta across stores - and we
2110
# explicitly do not want to to prevent data loss situations.
2111
if self.global_map is None:
2112
self.global_map = self.vf.get_parent_map(self.keys)
2113
nonlocal_keys = self.nonlocal_keys
2115
missing_keys = set(nonlocal_keys)
2116
# Read from remote versioned file instances and provide to our caller.
2117
for source in self.vf._immediate_fallback_vfs:
2118
if not missing_keys:
2120
# Loop over fallback repositories asking them for texts - ignore
2121
# any missing from a particular fallback.
2122
for record in source.get_record_stream(missing_keys,
2123
self._ordering, True):
2124
if record.storage_kind == 'absent':
2125
# Not in thie particular stream, may be in one of the
2126
# other fallback vfs objects.
2128
missing_keys.remove(record.key)
2131
if self._raw_record_map is None:
2132
raise AssertionError('_raw_record_map should have been filled')
2134
for key in self.keys:
2135
if key in self.nonlocal_keys:
2137
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2140
def _get_one_work(self, requested_key):
2141
# Now, if we have calculated everything already, just return the
2143
if requested_key in self._contents_map:
2144
return self._contents_map[requested_key]
2145
# To simplify things, parse everything at once - code that wants one text
2146
# probably wants them all.
2147
# FUTURE: This function could be improved for the 'extract many' case
2148
# by tracking each component and only doing the copy when the number of
2149
# children than need to apply delta's to it is > 1 or it is part of the
2151
multiple_versions = len(self.keys) != 1
2152
if self._record_map is None:
2153
self._record_map = self.vf._raw_map_to_record_map(
2154
self._raw_record_map)
2155
record_map = self._record_map
2156
# raw_record_map is key:
2157
# Have read and parsed records at this point.
2158
for key in self.keys:
2159
if key in self.nonlocal_keys:
822
for version_id in version_ids:
823
if not self.has_version(version_id):
824
raise RevisionNotPresent(version_id, self.filename)
825
record_map = self._get_record_map(version_ids)
830
for version_id in version_ids:
2164
833
while cursor is not None:
2166
record, record_details, digest, next = record_map[cursor]
2168
raise RevisionNotPresent(cursor, self)
2169
components.append((cursor, record, record_details, digest))
2171
if cursor in self._contents_map:
2172
# no need to plan further back
2173
components.append((cursor, None, None, None))
834
method, data, digest, next = record_map[cursor]
835
components.append((cursor, method, data, digest))
836
if cursor in content_map:
2177
for (component_id, record, record_details,
2178
digest) in reversed(components):
2179
if component_id in self._contents_map:
2180
content = self._contents_map[component_id]
841
for component_id, method, data, digest in reversed(components):
842
if component_id in content_map:
843
content = content_map[component_id]
2182
content, delta = self._factory.parse_record(key[-1],
2183
record, record_details, content,
2184
copy_base_content=multiple_versions)
2185
if multiple_versions:
2186
self._contents_map[component_id] = content
845
if method == 'fulltext':
846
assert content is None
847
content = self.factory.parse_fulltext(data, version_id)
848
elif method == 'line-delta':
849
delta = self.factory.parse_line_delta(data, version_id)
850
content = content.copy()
851
content._lines = self._apply_delta(content._lines,
853
content_map[component_id] = content
855
if 'no-eol' in self._index.get_options(version_id):
856
content = content.copy()
857
line = content._lines[-1][1].rstrip('\n')
858
content._lines[-1] = (content._lines[-1][0], line)
859
final_content[version_id] = content
2188
861
# digest here is the digest from the last applied component.
2189
862
text = content.text()
2190
actual_sha = sha_strings(text)
2191
if actual_sha != digest:
2192
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
2193
if multiple_versions:
2194
return self._contents_map[requested_key]
2198
def _wire_bytes(self):
2199
"""Get the bytes to put on the wire for 'key'.
2201
The first collection of bytes asked for returns the serialised
2202
raw_record_map and the additional details (key, parent) for key.
2203
Subsequent calls return just the additional details (key, parent).
2204
The wire storage_kind given for the first key is 'knit-delta-closure',
2205
For subsequent keys it is 'knit-delta-closure-ref'.
2207
:param key: A key from the content generator.
2208
:return: Bytes to put on the wire.
2211
# kind marker for dispatch on the far side,
2212
lines.append(b'knit-delta-closure')
2214
if self.vf._factory.annotated:
2215
lines.append(b'annotated')
2218
# then the list of keys
2219
lines.append(b'\t'.join(b'\x00'.join(key) for key in self.keys
2220
if key not in self.nonlocal_keys))
2221
# then the _raw_record_map in serialised form:
2223
# for each item in the map:
2225
# 1 line with parents if the key is to be yielded (None: for None, '' for ())
2226
# one line with method
2227
# one line with noeol
2228
# one line with next ('' for None)
2229
# one line with byte count of the record bytes
2231
for key, (record_bytes, (method, noeol), next) in viewitems(
2232
self._raw_record_map):
2233
key_bytes = b'\x00'.join(key)
2234
parents = self.global_map.get(key, None)
2236
parent_bytes = b'None:'
2238
parent_bytes = b'\t'.join(b'\x00'.join(key) for key in parents)
2239
method_bytes = method.encode('ascii')
2245
next_bytes = b'\x00'.join(next)
2248
map_byte_list.append(b'\n'.join(
2249
[key_bytes, parent_bytes, method_bytes, noeol_bytes, next_bytes,
2250
b'%d' % len(record_bytes), record_bytes]))
2251
map_bytes = b''.join(map_byte_list)
2252
lines.append(map_bytes)
2253
bytes = b'\n'.join(lines)
2257
class _VFContentMapGenerator(_ContentMapGenerator):
2258
"""Content map generator reading from a VersionedFiles object."""
2260
def __init__(self, versioned_files, keys, nonlocal_keys=None,
2261
global_map=None, raw_record_map=None, ordering='unordered'):
2262
"""Create a _ContentMapGenerator.
2264
:param versioned_files: The versioned files that the texts are being
2266
:param keys: The keys to produce content maps for.
2267
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
2268
which are known to not be in this knit, but rather in one of the
2270
:param global_map: The result of get_parent_map(keys) (or a supermap).
2271
This is required if get_record_stream() is to be used.
2272
:param raw_record_map: A unparsed raw record map to use for answering
2275
_ContentMapGenerator.__init__(self, ordering=ordering)
2276
# The vf to source data from
2277
self.vf = versioned_files
2279
self.keys = list(keys)
2280
# Keys known to be in fallback vfs objects
2281
if nonlocal_keys is None:
2282
self.nonlocal_keys = set()
2284
self.nonlocal_keys = frozenset(nonlocal_keys)
2285
# Parents data for keys to be returned in get_record_stream
2286
self.global_map = global_map
2287
# The chunked lists for self.keys in text form
2289
# A cache of KnitContent objects used in extracting texts.
2290
self._contents_map = {}
2291
# All the knit records needed to assemble the requested keys as full
2293
self._record_map = None
2294
if raw_record_map is None:
2295
self._raw_record_map = self.vf._get_record_map_unparsed(keys,
2298
self._raw_record_map = raw_record_map
2299
# the factory for parsing records
2300
self._factory = self.vf._factory
2303
class _NetworkContentMapGenerator(_ContentMapGenerator):
2304
"""Content map generator sourced from a network stream."""
2306
def __init__(self, bytes, line_end):
2307
"""Construct a _NetworkContentMapGenerator from a bytes block."""
2309
self.global_map = {}
2310
self._raw_record_map = {}
2311
self._contents_map = {}
2312
self._record_map = None
2313
self.nonlocal_keys = []
2314
# Get access to record parsing facilities
2315
self.vf = KnitVersionedFiles(None, None)
2318
line_end = bytes.find(b'\n', start)
2319
line = bytes[start:line_end]
2320
start = line_end + 1
2321
if line == b'annotated':
2322
self._factory = KnitAnnotateFactory()
2324
self._factory = KnitPlainFactory()
2325
# list of keys to emit in get_record_stream
2326
line_end = bytes.find(b'\n', start)
2327
line = bytes[start:line_end]
2328
start = line_end + 1
2330
tuple(segment.split(b'\x00')) for segment in line.split(b'\t')
2332
# now a loop until the end. XXX: It would be nice if this was just a
2333
# bunch of the same records as get_record_stream(..., False) gives, but
2334
# there is a decent sized gap stopping that at the moment.
2338
line_end = bytes.find(b'\n', start)
2339
key = tuple(bytes[start:line_end].split(b'\x00'))
2340
start = line_end + 1
2341
# 1 line with parents (None: for None, '' for ())
2342
line_end = bytes.find(b'\n', start)
2343
line = bytes[start:line_end]
2344
if line == b'None:':
2348
tuple(segment.split(b'\x00')) for segment in line.split(b'\t')
2350
self.global_map[key] = parents
2351
start = line_end + 1
2352
# one line with method
2353
line_end = bytes.find(b'\n', start)
2354
line = bytes[start:line_end]
2356
start = line_end + 1
2357
# one line with noeol
2358
line_end = bytes.find(b'\n', start)
2359
line = bytes[start:line_end]
2360
noeol = line == b"T"
2361
start = line_end + 1
2362
# one line with next (b'' for None)
2363
line_end = bytes.find(b'\n', start)
2364
line = bytes[start:line_end]
2368
next = tuple(bytes[start:line_end].split(b'\x00'))
2369
start = line_end + 1
2370
# one line with byte count of the record bytes
2371
line_end = bytes.find(b'\n', start)
2372
line = bytes[start:line_end]
2374
start = line_end + 1
2376
record_bytes = bytes[start:start+count]
2377
start = start + count
2379
self._raw_record_map[key] = (record_bytes, (method, noeol), next)
2381
def get_record_stream(self):
2382
"""Get a record stream for for keys requested by the bytestream."""
2384
for key in self.keys:
2385
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2388
def _wire_bytes(self):
2392
class _KndxIndex(object):
2393
"""Manages knit index files
2395
The index is kept in memory and read on startup, to enable
863
if sha_strings(text) != digest:
864
raise KnitCorrupt(self.filename,
865
'sha-1 does not match %s' % version_id)
867
text_map[version_id] = text
868
return text_map, final_content
870
def iter_lines_added_or_present_in_versions(self, version_ids=None,
872
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
873
if version_ids is None:
874
version_ids = self.versions()
876
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
878
pb = progress.DummyProgress()
879
# we don't care about inclusions, the caller cares.
880
# but we need to setup a list of records to visit.
881
# we need version_id, position, length
882
version_id_records = []
883
requested_versions = set(version_ids)
884
# filter for available versions
885
for version_id in requested_versions:
886
if not self.has_version(version_id):
887
raise RevisionNotPresent(version_id, self.filename)
888
# get a in-component-order queue:
889
for version_id in self.versions():
890
if version_id in requested_versions:
891
data_pos, length = self._index.get_position(version_id)
892
version_id_records.append((version_id, data_pos, length))
894
total = len(version_id_records)
895
for version_idx, (version_id, data, sha_value) in \
896
enumerate(self._data.read_records_iter(version_id_records)):
897
pb.update('Walking content.', version_idx, total)
898
method = self._index.get_method(version_id)
900
assert method in ('fulltext', 'line-delta')
901
if method == 'fulltext':
902
line_iterator = self.factory.get_fulltext_content(data)
904
line_iterator = self.factory.get_linedelta_content(data)
905
for line in line_iterator:
908
pb.update('Walking content.', total, total)
910
def num_versions(self):
911
"""See VersionedFile.num_versions()."""
912
return self._index.num_versions()
914
__len__ = num_versions
916
def annotate_iter(self, version_id):
917
"""See VersionedFile.annotate_iter."""
918
version_id = osutils.safe_revision_id(version_id)
919
content = self._get_content(version_id)
920
for origin, text in content.annotate_iter():
923
def get_parents(self, version_id):
924
"""See VersionedFile.get_parents."""
927
# 52554 calls in 1264 872 internal down from 3674
928
version_id = osutils.safe_revision_id(version_id)
930
return self._index.get_parents(version_id)
932
raise RevisionNotPresent(version_id, self.filename)
934
def get_parents_with_ghosts(self, version_id):
935
"""See VersionedFile.get_parents."""
936
version_id = osutils.safe_revision_id(version_id)
938
return self._index.get_parents_with_ghosts(version_id)
940
raise RevisionNotPresent(version_id, self.filename)
942
def get_ancestry(self, versions):
943
"""See VersionedFile.get_ancestry."""
944
if isinstance(versions, basestring):
945
versions = [versions]
948
versions = [osutils.safe_revision_id(v) for v in versions]
949
return self._index.get_ancestry(versions)
951
def get_ancestry_with_ghosts(self, versions):
952
"""See VersionedFile.get_ancestry_with_ghosts."""
953
if isinstance(versions, basestring):
954
versions = [versions]
957
versions = [osutils.safe_revision_id(v) for v in versions]
958
return self._index.get_ancestry_with_ghosts(versions)
960
#@deprecated_method(zero_eight)
961
def walk(self, version_ids):
962
"""See VersionedFile.walk."""
963
# We take the short path here, and extract all relevant texts
964
# and put them in a weave and let that do all the work. Far
965
# from optimal, but is much simpler.
966
# FIXME RB 20060228 this really is inefficient!
967
from bzrlib.weave import Weave
969
w = Weave(self.filename)
970
ancestry = self.get_ancestry(version_ids)
971
sorted_graph = topo_sort(self._index.get_graph())
972
version_list = [vid for vid in sorted_graph if vid in ancestry]
974
for version_id in version_list:
975
lines = self.get_lines(version_id)
976
w.add_lines(version_id, self.get_parents(version_id), lines)
978
for lineno, insert_id, dset, line in w.walk(version_ids):
979
yield lineno, insert_id, dset, line
981
def plan_merge(self, ver_a, ver_b):
982
"""See VersionedFile.plan_merge."""
983
ver_a = osutils.safe_revision_id(ver_a)
984
ver_b = osutils.safe_revision_id(ver_b)
985
ancestors_b = set(self.get_ancestry(ver_b))
986
def status_a(revision, text):
987
if revision in ancestors_b:
988
return 'killed-b', text
992
ancestors_a = set(self.get_ancestry(ver_a))
993
def status_b(revision, text):
994
if revision in ancestors_a:
995
return 'killed-a', text
999
annotated_a = self.annotate(ver_a)
1000
annotated_b = self.annotate(ver_b)
1001
plain_a = [t for (a, t) in annotated_a]
1002
plain_b = [t for (a, t) in annotated_b]
1003
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
1006
for ai, bi, l in blocks:
1007
# process all mismatched sections
1008
# (last mismatched section is handled because blocks always
1009
# includes a 0-length last block)
1010
for revision, text in annotated_a[a_cur:ai]:
1011
yield status_a(revision, text)
1012
for revision, text in annotated_b[b_cur:bi]:
1013
yield status_b(revision, text)
1015
# and now the matched section
1018
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
1019
assert text_a == text_b
1020
yield "unchanged", text_a
1023
class _KnitComponentFile(object):
1024
"""One of the files used to implement a knit database"""
1026
def __init__(self, transport, filename, mode, file_mode=None,
1027
create_parent_dir=False, dir_mode=None):
1028
self._transport = transport
1029
self._filename = filename
1031
self._file_mode = file_mode
1032
self._dir_mode = dir_mode
1033
self._create_parent_dir = create_parent_dir
1034
self._need_to_create = False
1036
def _full_path(self):
1037
"""Return the full path to this file."""
1038
return self._transport.base + self._filename
1040
def check_header(self, fp):
1041
line = fp.readline()
1043
# An empty file can actually be treated as though the file doesn't
1045
raise errors.NoSuchFile(self._full_path())
1046
if line != self.HEADER:
1047
raise KnitHeaderError(badline=line,
1048
filename=self._transport.abspath(self._filename))
1051
"""Commit is a nop."""
1054
return '%s(%s)' % (self.__class__.__name__, self._filename)
1057
class _KnitIndex(_KnitComponentFile):
1058
"""Manages knit index file.
1060
The index is already kept in memory and read on startup, to enable
2396
1061
fast lookups of revision information. The cursor of the index
2397
1062
file is always pointing to the end, making it easy to append
2440
1105
to ensure that records always start on new lines even if the last write was
2441
1106
interrupted. As a result its normal for the last line in the index to be
2442
1107
missing a trailing newline. One can be added with no harmful effects.
2444
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
2445
where prefix is e.g. the (fileid,) for .texts instances or () for
2446
constant-mapped things like .revisions, and the old state is
2447
tuple(history_vector, cache_dict). This is used to prevent having an
2448
ABI change with the C extension that reads .kndx files.
2451
HEADER = b"# bzr knit index 8\n"
2453
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
2454
"""Create a _KndxIndex on transport using mapper."""
2455
self._transport = transport
2456
self._mapper = mapper
2457
self._get_scope = get_scope
2458
self._allow_writes = allow_writes
2459
self._is_locked = is_locked
2461
self.has_graph = True
2463
def add_records(self, records, random_id=False, missing_compression_parents=False):
2464
"""Add multiple records to the index.
2466
:param records: a list of tuples:
2467
(key, options, access_memo, parents).
2468
:param random_id: If True the ids being added were randomly generated
2469
and no check for existence will be performed.
2470
:param missing_compression_parents: If True the records being added are
2471
only compressed against texts already in the index (or inside
2472
records). If False the records all refer to unavailable texts (or
2473
texts inside records) as compression parents.
2475
if missing_compression_parents:
2476
# It might be nice to get the edge of the records. But keys isn't
2478
keys = sorted(record[0] for record in records)
2479
raise errors.RevisionNotPresent(keys, self)
2481
for record in records:
2484
path = self._mapper.map(key) + '.kndx'
2485
path_keys = paths.setdefault(path, (prefix, []))
2486
path_keys[1].append(record)
2487
for path in sorted(paths):
2488
prefix, path_keys = paths[path]
2489
self._load_prefixes([prefix])
2491
orig_history = self._kndx_cache[prefix][1][:]
2492
orig_cache = self._kndx_cache[prefix][0].copy()
2495
for key, options, (_, pos, size), parents in path_keys:
2496
if not all(isinstance(option, bytes) for option in options):
2497
raise TypeError(options)
2499
# kndx indices cannot be parentless.
2502
b'\n' + key[-1], b','.join(options), b'%d' % pos, b'%d' % size,
2503
self._dictionary_compress(parents), b':'])
2504
if not isinstance(line, bytes):
2505
raise AssertionError(
2506
'data must be utf8 was %s' % type(line))
2508
self._cache_key(key, options, pos, size, parents)
2509
if len(orig_history):
2510
self._transport.append_bytes(path, b''.join(lines))
2512
self._init_index(path, lines)
2514
# If any problems happen, restore the original values and re-raise
2515
self._kndx_cache[prefix] = (orig_cache, orig_history)
2518
def scan_unvalidated_index(self, graph_index):
2519
"""See _KnitGraphIndex.scan_unvalidated_index."""
2520
# Because kndx files do not support atomic insertion via separate index
2521
# files, they do not support this method.
2522
raise NotImplementedError(self.scan_unvalidated_index)
2524
def get_missing_compression_parents(self):
2525
"""See _KnitGraphIndex.get_missing_compression_parents."""
2526
# Because kndx files do not support atomic insertion via separate index
2527
# files, they do not support this method.
2528
raise NotImplementedError(self.get_missing_compression_parents)
2530
def _cache_key(self, key, options, pos, size, parent_keys):
1110
HEADER = "# bzr knit index 8\n"
1112
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1113
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1115
def _cache_version(self, version_id, options, pos, size, parents):
2531
1116
"""Cache a version record in the history array and index cache.
2533
1118
This is inlined into _load_data for performance. KEEP IN SYNC.
2534
1119
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
2538
version_id = key[-1]
2539
# last-element only for compatibilty with the C load_data.
2540
parents = tuple(parent[-1] for parent in parent_keys)
2541
for parent in parent_keys:
2542
if parent[:-1] != prefix:
2543
raise ValueError("mismatched prefixes for %r, %r" % (
2545
cache, history = self._kndx_cache[prefix]
2546
1122
# only want the _history index to reference the 1st index entry
2547
1123
# for version_id
2548
if version_id not in cache:
2549
index = len(history)
2550
history.append(version_id)
1124
if version_id not in self._cache:
1125
index = len(self._history)
1126
self._history.append(version_id)
2552
index = cache[version_id][5]
2553
cache[version_id] = (version_id,
1128
index = self._cache[version_id][5]
1129
self._cache[version_id] = (version_id,
2560
def check_header(self, fp):
2561
line = fp.readline()
2563
# An empty file can actually be treated as though the file doesn't
2565
raise errors.NoSuchFile(self)
2566
if line != self.HEADER:
2567
raise KnitHeaderError(badline=line, filename=self)
2569
def _check_read(self):
2570
if not self._is_locked():
2571
raise errors.ObjectNotLocked(self)
2572
if self._get_scope() != self._scope:
2575
def _check_write_ok(self):
2576
"""Assert if not writes are permitted."""
2577
if not self._is_locked():
2578
raise errors.ObjectNotLocked(self)
2579
if self._get_scope() != self._scope:
2581
if self._mode != 'w':
2582
raise errors.ReadOnlyObjectDirtiedError(self)
2584
def get_build_details(self, keys):
2585
"""Get the method, index_memo and compression parent for keys.
2587
Ghosts are omitted from the result.
2589
:param keys: An iterable of keys.
2590
:return: A dict of key:(index_memo, compression_parent, parents,
2593
opaque structure to pass to read_records to extract the raw
2596
Content that this record is built upon, may be None
2598
Logical parents of this node
2600
extra information about the content which needs to be passed to
2601
Factory.parse_record
2603
parent_map = self.get_parent_map(keys)
2606
if key not in parent_map:
2608
method = self.get_method(key)
2609
if not isinstance(method, str):
2610
raise TypeError(method)
2611
parents = parent_map[key]
2612
if method == 'fulltext':
2613
compression_parent = None
2615
compression_parent = parents[0]
2616
noeol = b'no-eol' in self.get_options(key)
2617
index_memo = self.get_position(key)
2618
result[key] = (index_memo, compression_parent,
2619
parents, (method, noeol))
2622
def get_method(self, key):
2623
"""Return compression method of specified key."""
2624
options = self.get_options(key)
2625
if b'fulltext' in options:
2627
elif b'line-delta' in options:
2630
raise KnitIndexUnknownMethod(self, options)
2632
def get_options(self, key):
2633
"""Return a list representing options.
2637
prefix, suffix = self._split_key(key)
2638
self._load_prefixes([prefix])
1136
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1137
create_parent_dir=False, delay_create=False, dir_mode=None):
1138
_KnitComponentFile.__init__(self, transport, filename, mode,
1139
file_mode=file_mode,
1140
create_parent_dir=create_parent_dir,
1143
# position in _history is the 'official' index for a revision
1144
# but the values may have come from a newer entry.
1145
# so - wc -l of a knit index is != the number of unique names
2640
return self._kndx_cache[prefix][0][suffix][1]
2642
raise RevisionNotPresent(key, self)
2644
def find_ancestry(self, keys):
2645
"""See CombinedGraphIndex.find_ancestry()"""
2646
prefixes = set(key[:-1] for key in keys)
2647
self._load_prefixes(prefixes)
2650
missing_keys = set()
2651
pending_keys = list(keys)
2652
# This assumes that keys will not reference parents in a different
2653
# prefix, which is accurate so far.
2655
key = pending_keys.pop()
2656
if key in parent_map:
1149
fp = self._transport.get(self._filename)
1151
# _load_data may raise NoSuchFile if the target knit is
1157
if mode != 'w' or not create:
1160
self._need_to_create = True
1162
self._transport.put_bytes_non_atomic(
1163
self._filename, self.HEADER, mode=self._file_mode)
1165
def _load_data(self, fp):
1167
history = self._history
1169
self.check_header(fp)
1170
# readlines reads the whole file at once:
1171
# bad for transports like http, good for local disk
1172
# we save 60 ms doing this one change (
1173
# from calling readline each time to calling
1175
# probably what we want for nice behaviour on
1176
# http is a incremental readlines that yields, or
1177
# a check for local vs non local indexes,
1178
history_top = len(history) - 1
1179
for line in fp.readlines():
1181
if len(rec) < 5 or rec[-1] != ':':
1183
# FIXME: in the future we should determine if its a
1184
# short write - and ignore it
1185
# or a different failure, and raise. RBC 20060407
2660
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2662
missing_keys.add(key)
2664
parent_keys = tuple([prefix + (suffix,)
2665
for suffix in suffix_parents])
2666
parent_map[key] = parent_keys
2667
pending_keys.extend([p for p in parent_keys
2668
if p not in parent_map])
2669
return parent_map, missing_keys
2671
def get_parent_map(self, keys):
2672
"""Get a map of the parents of keys.
2674
:param keys: The keys to look up parents for.
2675
:return: A mapping from keys to parents. Absent keys are absent from
2678
# Parse what we need to up front, this potentially trades off I/O
2679
# locality (.kndx and .knit in the same block group for the same file
2680
# id) for less checking in inner loops.
2681
prefixes = set(key[:-1] for key in keys)
2682
self._load_prefixes(prefixes)
2687
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2691
result[key] = tuple(prefix + (suffix,) for
2692
suffix in suffix_parents)
2695
def get_position(self, key):
2696
"""Return details needed to access the version.
2698
:return: a tuple (key, data position, size) to hand to the access
2699
logic to get the record.
2701
prefix, suffix = self._split_key(key)
2702
self._load_prefixes([prefix])
2703
entry = self._kndx_cache[prefix][0][suffix]
2704
return key, entry[2], entry[3]
2706
__contains__ = _mod_index._has_key_from_parent_map
2708
def _init_index(self, path, extra_lines=[]):
2709
"""Initialize an index."""
2711
sio.write(self.HEADER)
2712
sio.writelines(extra_lines)
2714
self._transport.put_file_non_atomic(path, sio,
2715
create_parent_dir=True)
2716
# self._create_parent_dir)
2717
# mode=self._file_mode,
2718
# dir_mode=self._dir_mode)
2721
"""Get all the keys in the collection.
2723
The keys are not ordered.
2726
# Identify all key prefixes.
2727
# XXX: A bit hacky, needs polish.
2728
if isinstance(self._mapper, ConstantMapper):
2732
for quoted_relpath in self._transport.iter_files_recursive():
2733
path, ext = os.path.splitext(quoted_relpath)
2735
prefixes = [self._mapper.unmap(path) for path in relpaths]
2736
self._load_prefixes(prefixes)
2737
for prefix in prefixes:
2738
for suffix in self._kndx_cache[prefix][1]:
2739
result.add(prefix + (suffix,))
2742
def _load_prefixes(self, prefixes):
2743
"""Load the indices for prefixes."""
2745
for prefix in prefixes:
2746
if prefix not in self._kndx_cache:
2747
# the load_data interface writes to these variables.
2750
self._filename = prefix
2752
path = self._mapper.map(prefix) + '.kndx'
2753
with self._transport.get(path) as fp:
2754
# _load_data may raise NoSuchFile if the target knit is
2756
_load_data(self, fp)
2757
self._kndx_cache[prefix] = (self._cache, self._history)
2762
self._kndx_cache[prefix] = ({}, [])
2763
if isinstance(self._mapper, ConstantMapper):
2764
# preserve behaviour for revisions.kndx etc.
2765
self._init_index(path)
2770
missing_keys = _mod_index._missing_keys_from_parent_map
2772
def _partition_keys(self, keys):
2773
"""Turn keys into a dict of prefix:suffix_list."""
2776
prefix_keys = result.setdefault(key[:-1], [])
2777
prefix_keys.append(key[-1])
2780
def _dictionary_compress(self, keys):
2781
"""Dictionary compress keys.
2783
:param keys: The keys to generate references to.
2784
:return: A string representation of keys. keys which are present are
2785
dictionary compressed, and others are emitted as fulltext with a
1189
for value in rec[4:-1]:
1191
# uncompressed reference
1192
parent_id = value[1:]
1194
parent_id = history[int(value)]
1195
parents.append(parent_id)
1197
version_id, options, pos, size = rec[:4]
1198
version_id = version_id
1200
# See self._cache_version
1201
# only want the _history index to reference the 1st
1202
# index entry for version_id
1203
if version_id not in cache:
1206
history.append(version_id)
1208
index = cache[version_id][5]
1209
cache[version_id] = (version_id,
1215
# end self._cache_version
1217
def get_graph(self):
1218
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1220
def get_ancestry(self, versions):
1221
"""See VersionedFile.get_ancestry."""
1222
# get a graph of all the mentioned versions:
1224
pending = set(versions)
1227
version = pending.pop()
1230
parents = [p for p in cache[version][4] if p in cache]
1232
raise RevisionNotPresent(version, self._filename)
1233
# if not completed and not a ghost
1234
pending.update([p for p in parents if p not in graph])
1235
graph[version] = parents
1236
return topo_sort(graph.items())
1238
def get_ancestry_with_ghosts(self, versions):
1239
"""See VersionedFile.get_ancestry_with_ghosts."""
1240
# get a graph of all the mentioned versions:
1241
self.check_versions_present(versions)
1244
pending = set(versions)
1246
version = pending.pop()
1248
parents = cache[version][4]
1254
pending.update([p for p in parents if p not in graph])
1255
graph[version] = parents
1256
return topo_sort(graph.items())
1258
def num_versions(self):
1259
return len(self._history)
1261
__len__ = num_versions
1263
def get_versions(self):
1264
return self._history
1266
def idx_to_name(self, idx):
1267
return self._history[idx]
1269
def lookup(self, version_id):
1270
assert version_id in self._cache
1271
return self._cache[version_id][5]
1273
def _version_list_to_index(self, versions):
2790
1274
result_list = []
2791
prefix = keys[0][:-1]
2792
cache = self._kndx_cache[prefix][0]
2794
if key[:-1] != prefix:
2795
# kndx indices cannot refer across partitioned storage.
2796
raise ValueError("mismatched prefixes for %r" % keys)
2797
if key[-1] in cache:
1276
for version in versions:
1277
if version in cache:
2798
1278
# -- inlined lookup() --
2799
result_list.append(b'%d' % cache[key[-1]][5])
1279
result_list.append(str(cache[version][5]))
2800
1280
# -- end lookup () --
2802
result_list.append(b'.' + key[-1])
2803
return b' '.join(result_list)
2805
def _reset_cache(self):
2806
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2807
# (cache_dict, history_vector) for parsed kndx files.
2808
self._kndx_cache = {}
2809
self._scope = self._get_scope()
2810
allow_writes = self._allow_writes()
2816
def _sort_keys_by_io(self, keys, positions):
2817
"""Figure out an optimal order to read the records for the given keys.
2819
Sort keys, grouped by index and sorted by position.
2821
:param keys: A list of keys whose records we want to read. This will be
2823
:param positions: A dict, such as the one returned by
2824
_get_components_positions()
2827
def get_sort_key(key):
2828
index_memo = positions[key][1]
2829
# Group by prefix and position. index_memo[0] is the key, so it is
2830
# (file_id, revision_id) and we don't want to sort on revision_id,
2831
# index_memo[1] is the position, and index_memo[2] is the size,
2832
# which doesn't matter for the sort
2833
return index_memo[0][:-1], index_memo[1]
2834
return keys.sort(key=get_sort_key)
2836
_get_total_build_size = _get_total_build_size
2838
def _split_key(self, key):
2839
"""Split key into a prefix and suffix."""
2840
# GZ 2018-07-03: This is intentionally either a sequence or bytes?
2841
if isinstance(key, bytes):
2842
return key[:-1], key[-1:]
2843
return key[:-1], key[-1]
2846
class _KnitGraphIndex(object):
2847
"""A KnitVersionedFiles index layered on GraphIndex."""
2849
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2850
add_callback=None, track_external_parent_refs=False):
2851
"""Construct a KnitGraphIndex on a graph_index.
2853
:param graph_index: An implementation of breezy.index.GraphIndex.
2854
:param is_locked: A callback to check whether the object should answer
2856
:param deltas: Allow delta-compressed records.
2857
:param parents: If True, record knits parents, if not do not record
2859
:param add_callback: If not None, allow additions to the index and call
2860
this callback with a list of added GraphIndex nodes:
2861
[(node, value, node_refs), ...]
2862
:param is_locked: A callback, returns True if the index is locked and
2864
:param track_external_parent_refs: If True, record all external parent
2865
references parents from added records. These can be retrieved
2866
later by calling get_missing_parents().
2868
self._add_callback = add_callback
2869
self._graph_index = graph_index
2870
self._deltas = deltas
2871
self._parents = parents
2872
if deltas and not parents:
2873
# XXX: TODO: Delta tree and parent graph should be conceptually
2875
raise KnitCorrupt(self, "Cannot do delta compression without "
2877
self.has_graph = parents
2878
self._is_locked = is_locked
2879
self._missing_compression_parents = set()
2880
if track_external_parent_refs:
2881
self._key_dependencies = _KeyRefs()
2883
self._key_dependencies = None
2886
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2888
def add_records(self, records, random_id=False,
2889
missing_compression_parents=False):
2890
"""Add multiple records to the index.
2892
This function does not insert data into the Immutable GraphIndex
2893
backing the KnitGraphIndex, instead it prepares data for insertion by
2894
the caller and checks that it is safe to insert then calls
2895
self._add_callback with the prepared GraphIndex nodes.
2897
:param records: a list of tuples:
2898
(key, options, access_memo, parents).
2899
:param random_id: If True the ids being added were randomly generated
2900
and no check for existence will be performed.
2901
:param missing_compression_parents: If True the records being added are
2902
only compressed against texts already in the index (or inside
2903
records). If False the records all refer to unavailable texts (or
2904
texts inside records) as compression parents.
2906
if not self._add_callback:
2907
raise errors.ReadOnlyError(self)
2908
# we hope there are no repositories with inconsistent parentage
2912
compression_parents = set()
2913
key_dependencies = self._key_dependencies
2914
for (key, options, access_memo, parents) in records:
2916
parents = tuple(parents)
2917
if key_dependencies is not None:
2918
key_dependencies.add_references(key, parents)
2919
index, pos, size = access_memo
2920
if b'no-eol' in options:
2924
value += b"%d %d" % (pos, size)
2925
if not self._deltas:
2926
if b'line-delta' in options:
2927
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2930
if b'line-delta' in options:
2931
node_refs = (parents, (parents[0],))
2932
if missing_compression_parents:
2933
compression_parents.add(parents[0])
2935
node_refs = (parents, ())
1282
result_list.append('.' + version)
1283
return ' '.join(result_list)
1285
def add_version(self, version_id, options, pos, size, parents):
1286
"""Add a version record to the index."""
1287
self.add_versions(((version_id, options, pos, size, parents),))
1289
def add_versions(self, versions):
1290
"""Add multiple versions to the index.
1292
:param versions: a list of tuples:
1293
(version_id, options, pos, size, parents).
1296
orig_history = self._history[:]
1297
orig_cache = self._cache.copy()
1300
for version_id, options, pos, size, parents in versions:
1301
line = "\n%s %s %s %s %s :" % (version_id,
1305
self._version_list_to_index(parents))
1306
assert isinstance(line, str), \
1307
'content must be utf-8 encoded: %r' % (line,)
1309
self._cache_version(version_id, options, pos, size, parents)
1310
if not self._need_to_create:
1311
self._transport.append_bytes(self._filename, ''.join(lines))
1314
sio.write(self.HEADER)
1315
sio.writelines(lines)
1317
self._transport.put_file_non_atomic(self._filename, sio,
1318
create_parent_dir=self._create_parent_dir,
1319
mode=self._file_mode,
1320
dir_mode=self._dir_mode)
1321
self._need_to_create = False
1323
# If any problems happen, restore the original values and re-raise
1324
self._history = orig_history
1325
self._cache = orig_cache
1328
def has_version(self, version_id):
1329
"""True if the version is in the index."""
1330
return version_id in self._cache
1332
def get_position(self, version_id):
1333
"""Return data position and size of specified version."""
1334
entry = self._cache[version_id]
1335
return entry[2], entry[3]
1337
def get_method(self, version_id):
1338
"""Return compression method of specified version."""
1339
options = self._cache[version_id][1]
1340
if 'fulltext' in options:
1343
if 'line-delta' not in options:
1344
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1347
def get_options(self, version_id):
1348
return self._cache[version_id][1]
1350
def get_parents(self, version_id):
1351
"""Return parents of specified version ignoring ghosts."""
1352
return [parent for parent in self._cache[version_id][4]
1353
if parent in self._cache]
1355
def get_parents_with_ghosts(self, version_id):
1356
"""Return parents of specified version with ghosts."""
1357
return self._cache[version_id][4]
1359
def check_versions_present(self, version_ids):
1360
"""Check that all specified versions are present."""
1362
for version_id in version_ids:
1363
if version_id not in cache:
1364
raise RevisionNotPresent(version_id, self._filename)
1367
class _KnitData(_KnitComponentFile):
1368
"""Contents of the knit data file"""
1370
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1371
create_parent_dir=False, delay_create=False,
1373
_KnitComponentFile.__init__(self, transport, filename, mode,
1374
file_mode=file_mode,
1375
create_parent_dir=create_parent_dir,
1377
self._checked = False
1378
# TODO: jam 20060713 conceptually, this could spill to disk
1379
# if the cached size gets larger than a certain amount
1380
# but it complicates the model a bit, so for now just use
1381
# a simple dictionary
1383
self._do_cache = False
1386
self._need_to_create = create
1388
self._transport.put_bytes_non_atomic(self._filename, '',
1389
mode=self._file_mode)
1391
def enable_cache(self):
1392
"""Enable caching of reads."""
1393
self._do_cache = True
1395
def clear_cache(self):
1396
"""Clear the record cache."""
1397
self._do_cache = False
1400
def _open_file(self):
1402
return self._transport.get(self._filename)
1407
def _record_to_data(self, version_id, digest, lines):
1408
"""Convert version_id, digest, lines into a raw data block.
1410
:return: (len, a StringIO instance with the raw data ready to read.)
1413
data_file = GzipFile(None, mode='wb', fileobj=sio)
1415
assert isinstance(version_id, str)
1416
data_file.writelines(chain(
1417
["version %s %d %s\n" % (version_id,
1421
["end %s\n" % version_id]))
1428
def add_raw_record(self, raw_data):
1429
"""Append a prepared record to the data file.
1431
:return: the offset in the data file raw_data was written.
1433
assert isinstance(raw_data, str), 'data must be plain bytes'
1434
if not self._need_to_create:
1435
return self._transport.append_bytes(self._filename, raw_data)
1437
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1438
create_parent_dir=self._create_parent_dir,
1439
mode=self._file_mode,
1440
dir_mode=self._dir_mode)
1441
self._need_to_create = False
1444
def add_record(self, version_id, digest, lines):
1445
"""Write new text record to disk. Returns the position in the
1446
file where it was written."""
1447
size, sio = self._record_to_data(version_id, digest, lines)
1449
if not self._need_to_create:
1450
start_pos = self._transport.append_file(self._filename, sio)
1452
self._transport.put_file_non_atomic(self._filename, sio,
1453
create_parent_dir=self._create_parent_dir,
1454
mode=self._file_mode,
1455
dir_mode=self._dir_mode)
1456
self._need_to_create = False
1459
self._cache[version_id] = sio.getvalue()
1460
return start_pos, size
1462
def _parse_record_header(self, version_id, raw_data):
1463
"""Parse a record header for consistency.
1465
:return: the header and the decompressor stream.
1466
as (stream, header_record)
1468
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1470
rec = self._check_header(version_id, df.readline())
1471
except Exception, e:
1472
raise KnitCorrupt(self._filename,
1473
"While reading {%s} got %s(%s)"
1474
% (version_id, e.__class__.__name__, str(e)))
1477
def _check_header(self, version_id, line):
1480
raise KnitCorrupt(self._filename,
1481
'unexpected number of elements in record header')
1482
if rec[1] != version_id:
1483
raise KnitCorrupt(self._filename,
1484
'unexpected version, wanted %r, got %r'
1485
% (version_id, rec[1]))
1488
def _parse_record(self, version_id, data):
1490
# 4168 calls in 2880 217 internal
1491
# 4168 calls to _parse_record_header in 2121
1492
# 4168 calls to readlines in 330
1493
df = GzipFile(mode='rb', fileobj=StringIO(data))
1496
record_contents = df.readlines()
1497
except Exception, e:
1498
raise KnitCorrupt(self._filename,
1499
"While reading {%s} got %s(%s)"
1500
% (version_id, e.__class__.__name__, str(e)))
1501
header = record_contents.pop(0)
1502
rec = self._check_header(version_id, header)
1504
last_line = record_contents.pop()
1505
if len(record_contents) != int(rec[2]):
1506
raise KnitCorrupt(self._filename,
1507
'incorrect number of lines %s != %s'
1509
% (len(record_contents), int(rec[2]),
1511
if last_line != 'end %s\n' % rec[1]:
1512
raise KnitCorrupt(self._filename,
1513
'unexpected version end line %r, wanted %r'
1514
% (last_line, version_id))
1516
return record_contents, rec[3]
1518
def read_records_iter_raw(self, records):
1519
"""Read text records from data file and yield raw data.
1521
This unpacks enough of the text record to validate the id is
1522
as expected but thats all.
1524
# setup an iterator of the external records:
1525
# uses readv so nice and fast we hope.
1527
# grab the disk data needed.
1529
# Don't check _cache if it is empty
1530
needed_offsets = [(pos, size) for version_id, pos, size
1532
if version_id not in self._cache]
1534
needed_offsets = [(pos, size) for version_id, pos, size
1537
raw_records = self._transport.readv(self._filename, needed_offsets)
1539
for version_id, pos, size in records:
1540
if version_id in self._cache:
1541
# This data has already been validated
1542
data = self._cache[version_id]
1544
pos, data = raw_records.next()
1546
self._cache[version_id] = data
1548
# validate the header
1549
df, rec = self._parse_record_header(version_id, data)
1551
yield version_id, data
1553
def read_records_iter(self, records):
1554
"""Read text records from data file and yield result.
1556
The result will be returned in whatever is the fastest to read.
1557
Not by the order requested. Also, multiple requests for the same
1558
record will only yield 1 response.
1559
:param records: A list of (version_id, pos, len) entries
1560
:return: Yields (version_id, contents, digest) in the order
1561
read, not the order requested
1567
# Skip records we have alread seen
1568
yielded_records = set()
1569
needed_records = set()
1570
for record in records:
1571
if record[0] in self._cache:
1572
if record[0] in yielded_records:
1574
yielded_records.add(record[0])
1575
data = self._cache[record[0]]
1576
content, digest = self._parse_record(record[0], data)
1577
yield (record[0], content, digest)
2937
node_refs = (parents, )
2940
raise KnitCorrupt(self, "attempt to add node with parents "
2941
"in parentless index.")
2943
keys[key] = (value, node_refs)
2946
present_nodes = self._get_entries(keys)
2947
for (index, key, value, node_refs) in present_nodes:
2948
parents = node_refs[:1]
2949
# Sometimes these are passed as a list rather than a tuple
2950
passed = static_tuple.as_tuples(keys[key])
2951
passed_parents = passed[1][:1]
2952
if (value[0:1] != keys[key][0][0:1] or
2953
parents != passed_parents):
2954
node_refs = static_tuple.as_tuples(node_refs)
2955
raise KnitCorrupt(self, "inconsistent details in add_records"
2956
": %s %s" % ((value, node_refs), passed))
2960
for key, (value, node_refs) in viewitems(keys):
2961
result.append((key, value, node_refs))
2963
for key, (value, node_refs) in viewitems(keys):
2964
result.append((key, value))
2965
self._add_callback(result)
2966
if missing_compression_parents:
2967
# This may appear to be incorrect (it does not check for
2968
# compression parents that are in the existing graph index),
2969
# but such records won't have been buffered, so this is
2970
# actually correct: every entry when
2971
# missing_compression_parents==True either has a missing parent, or
2972
# a parent that is one of the keys in records.
2973
compression_parents.difference_update(keys)
2974
self._missing_compression_parents.update(compression_parents)
2975
# Adding records may have satisfied missing compression parents.
2976
self._missing_compression_parents.difference_update(keys)
2978
def scan_unvalidated_index(self, graph_index):
2979
"""Inform this _KnitGraphIndex that there is an unvalidated index.
2981
This allows this _KnitGraphIndex to keep track of any missing
2982
compression parents we may want to have filled in to make those
2985
:param graph_index: A GraphIndex
2988
new_missing = graph_index.external_references(ref_list_num=1)
2989
new_missing.difference_update(self.get_parent_map(new_missing))
2990
self._missing_compression_parents.update(new_missing)
2991
if self._key_dependencies is not None:
2992
# Add parent refs from graph_index (and discard parent refs that
2993
# the graph_index has).
2994
for node in graph_index.iter_all_entries():
2995
self._key_dependencies.add_references(node[1], node[3][0])
2997
def get_missing_compression_parents(self):
2998
"""Return the keys of missing compression parents.
3000
Missing compression parents occur when a record stream was missing
3001
basis texts, or a index was scanned that had missing basis texts.
3003
return frozenset(self._missing_compression_parents)
3005
def get_missing_parents(self):
3006
"""Return the keys of missing parents."""
3007
# If updating this, you should also update
3008
# groupcompress._GCGraphIndex.get_missing_parents
3009
# We may have false positives, so filter those out.
3010
self._key_dependencies.satisfy_refs_for_keys(
3011
self.get_parent_map(self._key_dependencies.get_unsatisfied_refs()))
3012
return frozenset(self._key_dependencies.get_unsatisfied_refs())
3014
def _check_read(self):
3015
"""raise if reads are not permitted."""
3016
if not self._is_locked():
3017
raise errors.ObjectNotLocked(self)
3019
def _check_write_ok(self):
3020
"""Assert if writes are not permitted."""
3021
if not self._is_locked():
3022
raise errors.ObjectNotLocked(self)
3024
def _compression_parent(self, an_entry):
3025
# return the key that an_entry is compressed against, or None
3026
# Grab the second parent list (as deltas implies parents currently)
3027
compression_parents = an_entry[3][1]
3028
if not compression_parents:
3030
if len(compression_parents) != 1:
3031
raise AssertionError(
3032
"Too many compression parents: %r" % compression_parents)
3033
return compression_parents[0]
3035
def get_build_details(self, keys):
3036
"""Get the method, index_memo and compression parent for version_ids.
3038
Ghosts are omitted from the result.
3040
:param keys: An iterable of keys.
3041
:return: A dict of key:
3042
(index_memo, compression_parent, parents, record_details).
3044
opaque structure to pass to read_records to extract the raw
3047
Content that this record is built upon, may be None
3049
Logical parents of this node
3051
extra information about the content which needs to be passed to
3052
Factory.parse_record
3056
entries = self._get_entries(keys, False)
3057
for entry in entries:
3059
if not self._parents:
3062
parents = entry[3][0]
3063
if not self._deltas:
3064
compression_parent_key = None
3066
compression_parent_key = self._compression_parent(entry)
3067
noeol = (entry[2][0:1] == b'N')
3068
if compression_parent_key:
3069
method = 'line-delta'
3072
result[key] = (self._node_to_position(entry),
3073
compression_parent_key, parents,
3077
def _get_entries(self, keys, check_present=False):
3078
"""Get the entries for keys.
3080
:param keys: An iterable of index key tuples.
3085
for node in self._graph_index.iter_entries(keys):
3087
found_keys.add(node[1])
3089
# adapt parentless index to the rest of the code.
3090
for node in self._graph_index.iter_entries(keys):
3091
yield node[0], node[1], node[2], ()
3092
found_keys.add(node[1])
3094
missing_keys = keys.difference(found_keys)
3096
raise RevisionNotPresent(missing_keys.pop(), self)
3098
def get_method(self, key):
3099
"""Return compression method of specified key."""
3100
return self._get_method(self._get_node(key))
3102
def _get_method(self, node):
3103
if not self._deltas:
3105
if self._compression_parent(node):
3110
def _get_node(self, key):
3112
return list(self._get_entries([key]))[0]
3114
raise RevisionNotPresent(key, self)
3116
def get_options(self, key):
3117
"""Return a list representing options.
3121
node = self._get_node(key)
3122
options = [self._get_method(node).encode('ascii')]
3123
if node[2][0:1] == b'N':
3124
options.append(b'no-eol')
3127
def find_ancestry(self, keys):
3128
"""See CombinedGraphIndex.find_ancestry()"""
3129
return self._graph_index.find_ancestry(keys, 0)
3131
def get_parent_map(self, keys):
3132
"""Get a map of the parents of keys.
3134
:param keys: The keys to look up parents for.
3135
:return: A mapping from keys to parents. Absent keys are absent from
3139
nodes = self._get_entries(keys)
3143
result[node[1]] = node[3][0]
3146
result[node[1]] = None
3149
def get_position(self, key):
3150
"""Return details needed to access the version.
3152
:return: a tuple (index, data position, size) to hand to the access
3153
logic to get the record.
3155
node = self._get_node(key)
3156
return self._node_to_position(node)
3158
__contains__ = _mod_index._has_key_from_parent_map
3161
"""Get all the keys in the collection.
3163
The keys are not ordered.
3166
return [node[1] for node in self._graph_index.iter_all_entries()]
3168
missing_keys = _mod_index._missing_keys_from_parent_map
3170
def _node_to_position(self, node):
3171
"""Convert an index value to position details."""
3172
bits = node[2][1:].split(b' ')
3173
return node[0], int(bits[0]), int(bits[1])
3175
def _sort_keys_by_io(self, keys, positions):
3176
"""Figure out an optimal order to read the records for the given keys.
3178
Sort keys, grouped by index and sorted by position.
3180
:param keys: A list of keys whose records we want to read. This will be
3182
:param positions: A dict, such as the one returned by
3183
_get_components_positions()
3186
def get_index_memo(key):
3187
# index_memo is at offset [1]. It is made up of (GraphIndex,
3188
# position, size). GI is an object, which will be unique for each
3189
# pack file. This causes us to group by pack file, then sort by
3190
# position. Size doesn't matter, but it isn't worth breaking up the
3192
return positions[key][1]
3193
return keys.sort(key=get_index_memo)
3195
_get_total_build_size = _get_total_build_size
3198
class _KnitKeyAccess(object):
3199
"""Access to records in .knit files."""
3201
def __init__(self, transport, mapper):
3202
"""Create a _KnitKeyAccess with transport and mapper.
3204
:param transport: The transport the access object is rooted at.
3205
:param mapper: The mapper used to map keys to .knit files.
3207
self._transport = transport
3208
self._mapper = mapper
3210
def add_raw_records(self, key_sizes, raw_data):
3211
"""Add raw knit bytes to a storage area.
3213
The data is spooled to the container writer in one bytes-record per
3216
:param sizes: An iterable of tuples containing the key and size of each
3218
:param raw_data: A bytestring containing the data.
3219
:return: A list of memos to retrieve the record later. Each memo is an
3220
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
3221
length), where the key is the record key.
3223
if not isinstance(raw_data, bytes):
3224
raise AssertionError(
3225
'data must be plain bytes was %s' % type(raw_data))
3228
# TODO: This can be tuned for writing to sftp and other servers where
3229
# append() is relatively expensive by grouping the writes to each key
3231
for key, size in key_sizes:
3232
path = self._mapper.map(key)
3234
base = self._transport.append_bytes(path + '.knit',
3235
raw_data[offset:offset+size])
3236
except errors.NoSuchFile:
3237
self._transport.mkdir(osutils.dirname(path))
3238
base = self._transport.append_bytes(path + '.knit',
3239
raw_data[offset:offset+size])
3243
result.append((key, base, size))
3247
"""Flush pending writes on this access object.
3249
For .knit files this is a no-op.
3253
def get_raw_records(self, memos_for_retrieval):
3254
"""Get the raw bytes for a records.
3256
:param memos_for_retrieval: An iterable containing the access memo for
3257
retrieving the bytes.
3258
:return: An iterator over the bytes of the records.
3260
# first pass, group into same-index request to minimise readv's issued.
3262
current_prefix = None
3263
for (key, offset, length) in memos_for_retrieval:
3264
if current_prefix == key[:-1]:
3265
current_list.append((offset, length))
3267
if current_prefix is not None:
3268
request_lists.append((current_prefix, current_list))
3269
current_prefix = key[:-1]
3270
current_list = [(offset, length)]
3271
# handle the last entry
3272
if current_prefix is not None:
3273
request_lists.append((current_prefix, current_list))
3274
for prefix, read_vector in request_lists:
3275
path = self._mapper.map(prefix) + '.knit'
3276
for pos, data in self._transport.readv(path, read_vector):
3280
def annotate_knit(knit, revision_id):
3281
"""Annotate a knit with no cached annotations.
3283
This implementation is for knits with no cached annotations.
3284
It will work for knits with cached annotations, but this is not
1579
needed_records.add(record)
1580
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1582
needed_records = sorted(set(records), key=operator.itemgetter(1))
1584
if not needed_records:
1587
# The transport optimizes the fetching as well
1588
# (ie, reads continuous ranges.)
1589
readv_response = self._transport.readv(self._filename,
1590
[(pos, size) for version_id, pos, size in needed_records])
1592
for (version_id, pos, size), (pos, data) in \
1593
izip(iter(needed_records), readv_response):
1594
content, digest = self._parse_record(version_id, data)
1596
self._cache[version_id] = data
1597
yield version_id, content, digest
1599
def read_records(self, records):
1600
"""Read records into a dictionary."""
1602
for record_id, content, digest in \
1603
self.read_records_iter(records):
1604
components[record_id] = (content, digest)
1608
class InterKnit(InterVersionedFile):
1609
"""Optimised code paths for knit to knit operations."""
1611
_matching_file_from_factory = KnitVersionedFile
1612
_matching_file_to_factory = KnitVersionedFile
1615
def is_compatible(source, target):
1616
"""Be compatible with knits. """
1618
return (isinstance(source, KnitVersionedFile) and
1619
isinstance(target, KnitVersionedFile))
1620
except AttributeError:
1623
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1624
"""See InterVersionedFile.join."""
1625
assert isinstance(self.source, KnitVersionedFile)
1626
assert isinstance(self.target, KnitVersionedFile)
1628
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1633
pb = ui.ui_factory.nested_progress_bar()
1635
version_ids = list(version_ids)
1636
if None in version_ids:
1637
version_ids.remove(None)
1639
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1640
this_versions = set(self.target._index.get_versions())
1641
needed_versions = self.source_ancestry - this_versions
1642
cross_check_versions = self.source_ancestry.intersection(this_versions)
1643
mismatched_versions = set()
1644
for version in cross_check_versions:
1645
# scan to include needed parents.
1646
n1 = set(self.target.get_parents_with_ghosts(version))
1647
n2 = set(self.source.get_parents_with_ghosts(version))
1649
# FIXME TEST this check for cycles being introduced works
1650
# the logic is we have a cycle if in our graph we are an
1651
# ancestor of any of the n2 revisions.
1657
parent_ancestors = self.source.get_ancestry(parent)
1658
if version in parent_ancestors:
1659
raise errors.GraphCycleError([parent, version])
1660
# ensure this parent will be available later.
1661
new_parents = n2.difference(n1)
1662
needed_versions.update(new_parents.difference(this_versions))
1663
mismatched_versions.add(version)
1665
if not needed_versions and not mismatched_versions:
1667
full_list = topo_sort(self.source.get_graph())
1669
version_list = [i for i in full_list if (not self.target.has_version(i)
1670
and i in needed_versions)]
1674
copy_queue_records = []
1676
for version_id in version_list:
1677
options = self.source._index.get_options(version_id)
1678
parents = self.source._index.get_parents_with_ghosts(version_id)
1679
# check that its will be a consistent copy:
1680
for parent in parents:
1681
# if source has the parent, we must :
1682
# * already have it or
1683
# * have it scheduled already
1684
# otherwise we don't care
1685
assert (self.target.has_version(parent) or
1686
parent in copy_set or
1687
not self.source.has_version(parent))
1688
data_pos, data_size = self.source._index.get_position(version_id)
1689
copy_queue_records.append((version_id, data_pos, data_size))
1690
copy_queue.append((version_id, options, parents))
1691
copy_set.add(version_id)
1693
# data suck the join:
1695
total = len(version_list)
1698
for (version_id, raw_data), \
1699
(version_id2, options, parents) in \
1700
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1702
assert version_id == version_id2, 'logic error, inconsistent results'
1704
pb.update("Joining knit", count, total)
1705
raw_records.append((version_id, options, parents, len(raw_data)))
1706
raw_datum.append(raw_data)
1707
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1709
for version in mismatched_versions:
1710
# FIXME RBC 20060309 is this needed?
1711
n1 = set(self.target.get_parents_with_ghosts(version))
1712
n2 = set(self.source.get_parents_with_ghosts(version))
1713
# write a combined record to our history preserving the current
1714
# parents as first in the list
1715
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1716
self.target.fix_parents(version, new_parents)
1722
InterVersionedFile.register_optimiser(InterKnit)
1725
class WeaveToKnit(InterVersionedFile):
1726
"""Optimised code paths for weave to knit operations."""
1728
_matching_file_from_factory = bzrlib.weave.WeaveFile
1729
_matching_file_to_factory = KnitVersionedFile
1732
def is_compatible(source, target):
1733
"""Be compatible with weaves to knits."""
1735
return (isinstance(source, bzrlib.weave.Weave) and
1736
isinstance(target, KnitVersionedFile))
1737
except AttributeError:
1740
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1741
"""See InterVersionedFile.join."""
1742
assert isinstance(self.source, bzrlib.weave.Weave)
1743
assert isinstance(self.target, KnitVersionedFile)
1745
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1750
pb = ui.ui_factory.nested_progress_bar()
1752
version_ids = list(version_ids)
1754
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1755
this_versions = set(self.target._index.get_versions())
1756
needed_versions = self.source_ancestry - this_versions
1757
cross_check_versions = self.source_ancestry.intersection(this_versions)
1758
mismatched_versions = set()
1759
for version in cross_check_versions:
1760
# scan to include needed parents.
1761
n1 = set(self.target.get_parents_with_ghosts(version))
1762
n2 = set(self.source.get_parents(version))
1763
# if all of n2's parents are in n1, then its fine.
1764
if n2.difference(n1):
1765
# FIXME TEST this check for cycles being introduced works
1766
# the logic is we have a cycle if in our graph we are an
1767
# ancestor of any of the n2 revisions.
1773
parent_ancestors = self.source.get_ancestry(parent)
1774
if version in parent_ancestors:
1775
raise errors.GraphCycleError([parent, version])
1776
# ensure this parent will be available later.
1777
new_parents = n2.difference(n1)
1778
needed_versions.update(new_parents.difference(this_versions))
1779
mismatched_versions.add(version)
1781
if not needed_versions and not mismatched_versions:
1783
full_list = topo_sort(self.source.get_graph())
1785
version_list = [i for i in full_list if (not self.target.has_version(i)
1786
and i in needed_versions)]
1790
total = len(version_list)
1791
for version_id in version_list:
1792
pb.update("Converting to knit", count, total)
1793
parents = self.source.get_parents(version_id)
1794
# check that its will be a consistent copy:
1795
for parent in parents:
1796
# if source has the parent, we must already have it
1797
assert (self.target.has_version(parent))
1798
self.target.add_lines(
1799
version_id, parents, self.source.get_lines(version_id))
1802
for version in mismatched_versions:
1803
# FIXME RBC 20060309 is this needed?
1804
n1 = set(self.target.get_parents_with_ghosts(version))
1805
n2 = set(self.source.get_parents(version))
1806
# write a combined record to our history preserving the current
1807
# parents as first in the list
1808
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1809
self.target.fix_parents(version, new_parents)
1815
InterVersionedFile.register_optimiser(WeaveToKnit)
1818
class KnitSequenceMatcher(difflib.SequenceMatcher):
1819
"""Knit tuned sequence matcher.
1821
This is based on profiling of difflib which indicated some improvements
1822
for our usage pattern.
3287
annotator = _KnitAnnotator(knit)
3288
return iter(annotator.annotate_flat(revision_id))
3291
class _KnitAnnotator(annotate.Annotator):
3292
"""Build up the annotations for a text."""
3294
def __init__(self, vf):
3295
annotate.Annotator.__init__(self, vf)
3297
# TODO: handle Nodes which cannot be extracted
3298
# self._ghosts = set()
3300
# Map from (key, parent_key) => matching_blocks, should be 'use once'
3301
self._matching_blocks = {}
3303
# KnitContent objects
3304
self._content_objects = {}
3305
# The number of children that depend on this fulltext content object
3306
self._num_compression_children = {}
3307
# Delta records that need their compression parent before they can be
3309
self._pending_deltas = {}
3310
# Fulltext records that are waiting for their parents fulltexts before
3311
# they can be yielded for annotation
3312
self._pending_annotation = {}
3314
self._all_build_details = {}
3316
def _get_build_graph(self, key):
3317
"""Get the graphs for building texts and annotations.
3319
The data you need for creating a full text may be different than the
3320
data you need to annotate that text. (At a minimum, you need both
3321
parents to create an annotation, but only need 1 parent to generate the
3324
:return: A list of (key, index_memo) records, suitable for
3325
passing to read_records_iter to start reading in the raw data from
1825
def find_longest_match(self, alo, ahi, blo, bhi):
1826
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1828
If isjunk is not defined:
1830
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1831
alo <= i <= i+k <= ahi
1832
blo <= j <= j+k <= bhi
1833
and for all (i',j',k') meeting those conditions,
1836
and if i == i', j <= j'
1838
In other words, of all maximal matching blocks, return one that
1839
starts earliest in a, and of all those maximal matching blocks that
1840
start earliest in a, return the one that starts earliest in b.
1842
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1843
>>> s.find_longest_match(0, 5, 0, 9)
1846
If isjunk is defined, first the longest matching block is
1847
determined as above, but with the additional restriction that no
1848
junk element appears in the block. Then that block is extended as
1849
far as possible by matching (only) junk elements on both sides. So
1850
the resulting block never matches on junk except as identical junk
1851
happens to be adjacent to an "interesting" match.
1853
Here's the same example as before, but considering blanks to be
1854
junk. That prevents " abcd" from matching the " abcd" at the tail
1855
end of the second sequence directly. Instead only the "abcd" can
1856
match, and matches the leftmost "abcd" in the second sequence:
1858
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1859
>>> s.find_longest_match(0, 5, 0, 9)
1862
If no blocks match, return (alo, blo, 0).
1864
>>> s = SequenceMatcher(None, "ab", "c")
1865
>>> s.find_longest_match(0, 2, 0, 1)
3331
self._num_needed_children[key] = 1
3333
# get all pending nodes
3334
this_iteration = pending
3335
build_details = self._vf._index.get_build_details(this_iteration)
3336
self._all_build_details.update(build_details)
3337
# new_nodes = self._vf._index._get_entries(this_iteration)
3339
for key, details in viewitems(build_details):
3340
(index_memo, compression_parent, parent_keys,
3341
record_details) = details
3342
self._parent_map[key] = parent_keys
3343
self._heads_provider = None
3344
records.append((key, index_memo))
3345
# Do we actually need to check _annotated_lines?
3346
pending.update([p for p in parent_keys
3347
if p not in self._all_build_details])
3349
for parent_key in parent_keys:
3350
if parent_key in self._num_needed_children:
3351
self._num_needed_children[parent_key] += 1
3353
self._num_needed_children[parent_key] = 1
3354
if compression_parent:
3355
if compression_parent in self._num_compression_children:
3356
self._num_compression_children[compression_parent] += 1
3358
self._num_compression_children[compression_parent] = 1
3360
missing_versions = this_iteration.difference(build_details)
3361
if missing_versions:
3362
for key in missing_versions:
3363
if key in self._parent_map and key in self._text_cache:
3364
# We already have this text ready, we just need to
3365
# yield it later so we get it annotated
3367
parent_keys = self._parent_map[key]
3368
for parent_key in parent_keys:
3369
if parent_key in self._num_needed_children:
3370
self._num_needed_children[parent_key] += 1
3372
self._num_needed_children[parent_key] = 1
3373
pending.update([p for p in parent_keys
3374
if p not in self._all_build_details])
3376
raise errors.RevisionNotPresent(key, self._vf)
3377
# Generally we will want to read the records in reverse order, because
3378
# we find the parent nodes after the children
3380
return records, ann_keys
3382
def _get_needed_texts(self, key, pb=None):
3383
# if True or len(self._vf._immediate_fallback_vfs) > 0:
3384
if len(self._vf._immediate_fallback_vfs) > 0:
3385
# If we have fallbacks, go to the generic path
3386
for v in annotate.Annotator._get_needed_texts(self, key, pb=pb):
1869
# CAUTION: stripping common prefix or suffix would be incorrect.
1873
# Longest matching block is "ab", but if common prefix is
1874
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1875
# strip, so ends up claiming that ab is changed to acab by
1876
# inserting "ca" in the middle. That's minimal but unintuitive:
1877
# "it's obvious" that someone inserted "ac" at the front.
1878
# Windiff ends up at the same place as diff, but by pairing up
1879
# the unique 'b's and then matching the first two 'a's.
1881
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1882
besti, bestj, bestsize = alo, blo, 0
1883
# find longest junk-free match
1884
# during an iteration of the loop, j2len[j] = length of longest
1885
# junk-free match ending with a[i-1] and b[j]
1889
for i in xrange(alo, ahi):
1890
# look at all instances of a[i] in b; note that because
1891
# b2j has no junk keys, the loop is skipped if a[i] is junk
1892
j2lenget = j2len.get
1895
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1896
# following improvement
1897
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1898
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1899
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1901
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1902
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1903
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
3391
records, ann_keys = self._get_build_graph(key)
3392
for idx, (sub_key, text, num_lines) in enumerate(
3393
self._extract_texts(records)):
3395
pb.update(gettext('annotating'), idx, len(records))
3396
yield sub_key, text, num_lines
3397
for sub_key in ann_keys:
3398
text = self._text_cache[sub_key]
3399
num_lines = len(text) # bad assumption
3400
yield sub_key, text, num_lines
3402
except errors.RetryWithNewPacks as e:
3403
self._vf._access.reload_or_raise(e)
3404
# The cached build_details are no longer valid
3405
self._all_build_details.clear()
3407
def _cache_delta_blocks(self, key, compression_parent, delta, lines):
3408
parent_lines = self._text_cache[compression_parent]
3409
blocks = list(KnitContent.get_line_delta_blocks(delta, parent_lines, lines))
3410
self._matching_blocks[(key, compression_parent)] = blocks
3412
def _expand_record(self, key, parent_keys, compression_parent, record,
3415
if compression_parent:
3416
if compression_parent not in self._content_objects:
3417
# Waiting for the parent
3418
self._pending_deltas.setdefault(compression_parent, []).append(
3419
(key, parent_keys, record, record_details))
3421
# We have the basis parent, so expand the delta
3422
num = self._num_compression_children[compression_parent]
3425
base_content = self._content_objects.pop(compression_parent)
3426
self._num_compression_children.pop(compression_parent)
3428
self._num_compression_children[compression_parent] = num
3429
base_content = self._content_objects[compression_parent]
3430
# It is tempting to want to copy_base_content=False for the last
3431
# child object. However, whenever noeol=False,
3432
# self._text_cache[parent_key] is content._lines. So mutating it
3433
# gives very bad results.
3434
# The alternative is to copy the lines into text cache, but then we
3435
# are copying anyway, so just do it here.
3436
content, delta = self._vf._factory.parse_record(
3437
key, record, record_details, base_content,
3438
copy_base_content=True)
3441
content, _ = self._vf._factory.parse_record(
3442
key, record, record_details, None)
3443
if self._num_compression_children.get(key, 0) > 0:
3444
self._content_objects[key] = content
3445
lines = content.text()
3446
self._text_cache[key] = lines
3447
if delta is not None:
3448
self._cache_delta_blocks(key, compression_parent, delta, lines)
3451
def _get_parent_annotations_and_matches(self, key, text, parent_key):
3452
"""Get the list of annotations for the parent, and the matching lines.
3454
:param text: The opaque value given by _get_needed_texts
3455
:param parent_key: The key for the parent text
3456
:return: (parent_annotations, matching_blocks)
3457
parent_annotations is a list as long as the number of lines in
3459
matching_blocks is a list of (parent_idx, text_idx, len) tuples
3460
indicating which lines match between the two texts
3462
block_key = (key, parent_key)
3463
if block_key in self._matching_blocks:
3464
blocks = self._matching_blocks.pop(block_key)
3465
parent_annotations = self._annotations_cache[parent_key]
3466
return parent_annotations, blocks
3467
return annotate.Annotator._get_parent_annotations_and_matches(self,
3468
key, text, parent_key)
3470
def _process_pending(self, key):
3471
"""The content for 'key' was just processed.
3473
Determine if there is any more pending work to be processed.
3476
if key in self._pending_deltas:
3477
compression_parent = key
3478
children = self._pending_deltas.pop(key)
3479
for child_key, parent_keys, record, record_details in children:
3480
lines = self._expand_record(child_key, parent_keys,
3482
record, record_details)
3483
if self._check_ready_for_annotations(child_key, parent_keys):
3484
to_return.append(child_key)
3485
# Also check any children that are waiting for this parent to be
3487
if key in self._pending_annotation:
3488
children = self._pending_annotation.pop(key)
3489
to_return.extend([c for c, p_keys in children
3490
if self._check_ready_for_annotations(c, p_keys)])
3493
def _check_ready_for_annotations(self, key, parent_keys):
3494
"""return true if this text is ready to be yielded.
3496
Otherwise, this will return False, and queue the text into
3497
self._pending_annotation
3499
for parent_key in parent_keys:
3500
if parent_key not in self._annotations_cache:
3501
# still waiting on at least one parent text, so queue it up
3502
# Note that if there are multiple parents, we need to wait
3504
self._pending_annotation.setdefault(parent_key,
3505
[]).append((key, parent_keys))
3509
def _extract_texts(self, records):
3510
"""Extract the various texts needed based on records"""
3511
# We iterate in the order read, rather than a strict order requested
3512
# However, process what we can, and put off to the side things that
3513
# still need parents, cleaning them up when those parents are
3516
# 1) As 'records' are read, see if we can expand these records into
3517
# Content objects (and thus lines)
3518
# 2) If a given line-delta is waiting on its compression parent, it
3519
# gets queued up into self._pending_deltas, otherwise we expand
3520
# it, and put it into self._text_cache and self._content_objects
3521
# 3) If we expanded the text, we will then check to see if all
3522
# parents have also been processed. If so, this text gets yielded,
3523
# else this record gets set aside into pending_annotation
3524
# 4) Further, if we expanded the text in (2), we will then check to
3525
# see if there are any children in self._pending_deltas waiting to
3526
# also be processed. If so, we go back to (2) for those
3527
# 5) Further again, if we yielded the text, we can then check if that
3528
# 'unlocks' any of the texts in pending_annotations, which should
3529
# then get yielded as well
3530
# Note that both steps 4 and 5 are 'recursive' in that unlocking one
3531
# compression child could unlock yet another, and yielding a fulltext
3532
# will also 'unlock' the children that are waiting on that annotation.
3533
# (Though also, unlocking 1 parent's fulltext, does not unlock a child
3534
# if other parents are also waiting.)
3535
# We want to yield content before expanding child content objects, so
3536
# that we know when we can re-use the content lines, and the annotation
3537
# code can know when it can stop caching fulltexts, as well.
3539
# Children that are missing their compression parent
3541
for (key, record, digest) in self._vf._read_records_iter(records):
3543
details = self._all_build_details[key]
3544
(_, compression_parent, parent_keys, record_details) = details
3545
lines = self._expand_record(key, parent_keys, compression_parent,
3546
record, record_details)
3548
# Pending delta should be queued up
3550
# At this point, we may be able to yield this content, if all
3551
# parents are also finished
3552
yield_this_text = self._check_ready_for_annotations(key,
3555
# All parents present
3556
yield key, lines, len(lines)
3557
to_process = self._process_pending(key)
3559
this_process = to_process
3561
for key in this_process:
3562
lines = self._text_cache[key]
3563
yield key, lines, len(lines)
3564
to_process.extend(self._process_pending(key))
3567
from ._knit_load_data_pyx import _load_data_c as _load_data
3568
except ImportError as e:
3569
osutils.failed_to_load_extension(e)
3570
from ._knit_load_data_py import _load_data_py as _load_data
1915
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1917
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1920
# Extend the best by non-junk elements on each end. In particular,
1921
# "popular" non-junk elements aren't in b2j, which greatly speeds
1922
# the inner loop above, but also means "the best" match so far
1923
# doesn't contain any junk *or* popular non-junk elements.
1924
while besti > alo and bestj > blo and \
1925
not isbjunk(b[bestj-1]) and \
1926
a[besti-1] == b[bestj-1]:
1927
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1928
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1929
not isbjunk(b[bestj+bestsize]) and \
1930
a[besti+bestsize] == b[bestj+bestsize]:
1933
# Now that we have a wholly interesting match (albeit possibly
1934
# empty!), we may as well suck up the matching junk on each
1935
# side of it too. Can't think of a good reason not to, and it
1936
# saves post-processing the (possibly considerable) expense of
1937
# figuring out what to do with it. In the case of an empty
1938
# interesting match, this is clearly the right thing to do,
1939
# because no other kind of match is possible in the regions.
1940
while besti > alo and bestj > blo and \
1941
isbjunk(b[bestj-1]) and \
1942
a[besti-1] == b[bestj-1]:
1943
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1944
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1945
isbjunk(b[bestj+bestsize]) and \
1946
a[besti+bestsize] == b[bestj+bestsize]:
1947
bestsize = bestsize + 1
1949
return besti, bestj, bestsize
added
removed
bzrlib/knit.py
