248
753
for start, end, c, lines in delta:
249
754
out.append('%d,%d,%d\n' % (start, end, c))
250
out.extend([text for origin, text in lines])
254
def make_empty_knit(transport, relpath):
255
"""Construct a empty knit at the specified location."""
256
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
260
class KnitVersionedFile(VersionedFile):
261
"""Weave-like structure with faster random access.
263
A knit stores a number of texts and a summary of the relationships
264
between them. Texts are identified by a string version-id. Texts
265
are normally stored and retrieved as a series of lines, but can
266
also be passed as single strings.
268
Lines are stored with the trailing newline (if any) included, to
269
avoid special cases for files with no final newline. Lines are
270
composed of 8-bit characters, not unicode. The combination of
271
these approaches should mean any 'binary' file can be safely
272
stored and retrieved.
275
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
276
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
278
"""Construct a knit at location specified by relpath.
280
:param create: If not True, only open an existing knit.
758
def annotate(self, knit, key):
759
annotator = _KnitAnnotator(knit)
760
return annotator.annotate_flat(key)
764
def make_file_factory(annotated, mapper):
765
"""Create a factory for creating a file based KnitVersionedFiles.
767
This is only functional enough to run interface tests, it doesn't try to
768
provide a full pack environment.
770
:param annotated: knit annotations are wanted.
771
:param mapper: The mapper from keys to paths.
773
def factory(transport):
774
index = _KndxIndex(transport, mapper, lambda:None, lambda:True, lambda:True)
775
access = _KnitKeyAccess(transport, mapper)
776
return KnitVersionedFiles(index, access, annotated=annotated)
780
def make_pack_factory(graph, delta, keylength):
781
"""Create a factory for creating a pack based VersionedFiles.
783
This is only functional enough to run interface tests, it doesn't try to
784
provide a full pack environment.
786
:param graph: Store a graph.
787
:param delta: Delta compress contents.
788
:param keylength: How long should keys be.
790
def factory(transport):
791
parents = graph or delta
797
max_delta_chain = 200
800
graph_index = _mod_index.InMemoryGraphIndex(reference_lists=ref_length,
801
key_elements=keylength)
802
stream = transport.open_write_stream('newpack')
803
writer = pack.ContainerWriter(stream.write)
805
index = _KnitGraphIndex(graph_index, lambda:True, parents=parents,
806
deltas=delta, add_callback=graph_index.add_nodes)
807
access = pack_repo._DirectPackAccess({})
808
access.set_writer(writer, graph_index, (transport, 'newpack'))
809
result = KnitVersionedFiles(index, access,
810
max_delta_chain=max_delta_chain)
811
result.stream = stream
812
result.writer = writer
817
def cleanup_pack_knit(versioned_files):
818
versioned_files.stream.close()
819
versioned_files.writer.end()
822
def _get_total_build_size(self, keys, positions):
823
"""Determine the total bytes to build these keys.
825
(helper function because _KnitGraphIndex and _KndxIndex work the same, but
826
don't inherit from a common base.)
828
:param keys: Keys that we want to build
829
:param positions: dict of {key, (info, index_memo, comp_parent)} (such
830
as returned by _get_components_positions)
831
:return: Number of bytes to build those keys
833
all_build_index_memos = {}
837
for key in build_keys:
838
# This is mostly for the 'stacked' case
839
# Where we will be getting the data from a fallback
840
if key not in positions:
842
_, index_memo, compression_parent = positions[key]
843
all_build_index_memos[key] = index_memo
844
if compression_parent not in all_build_index_memos:
845
next_keys.add(compression_parent)
846
build_keys = next_keys
847
return sum([index_memo[2] for index_memo
848
in all_build_index_memos.itervalues()])
851
class KnitVersionedFiles(VersionedFilesWithFallbacks):
852
"""Storage for many versioned files using knit compression.
854
Backend storage is managed by indices and data objects.
856
:ivar _index: A _KnitGraphIndex or similar that can describe the
857
parents, graph, compression and data location of entries in this
858
KnitVersionedFiles. Note that this is only the index for
859
*this* vfs; if there are fallbacks they must be queried separately.
862
def __init__(self, index, data_access, max_delta_chain=200,
863
annotated=False, reload_func=None):
864
"""Create a KnitVersionedFiles with index and data_access.
866
:param index: The index for the knit data.
867
:param data_access: The access object to store and retrieve knit
869
:param max_delta_chain: The maximum number of deltas to permit during
870
insertion. Set to 0 to prohibit the use of deltas.
871
:param annotated: Set to True to cause annotations to be calculated and
872
stored during insertion.
873
:param reload_func: An function that can be called if we think we need
874
to reload the pack listing and try again. See
875
'bzrlib.repofmt.pack_repo.AggregateIndex' for the signature.
282
if deprecated_passed(basis_knit):
283
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
284
" deprecated as of bzr 0.9.",
285
DeprecationWarning, stacklevel=2)
286
if access_mode is None:
288
super(KnitVersionedFile, self).__init__(access_mode)
289
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
290
self.transport = transport
291
self.filename = relpath
292
self.factory = factory or KnitAnnotateFactory()
293
self.writable = (access_mode == 'w')
296
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
297
access_mode, create=create, file_mode=file_mode)
298
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
299
access_mode, create=create and not len(self), file_mode=file_mode)
878
self._access = data_access
879
self._max_delta_chain = max_delta_chain
881
self._factory = KnitAnnotateFactory()
883
self._factory = KnitPlainFactory()
884
self._immediate_fallback_vfs = []
885
self._reload_func = reload_func
301
887
def __repr__(self):
302
return '%s(%s)' % (self.__class__.__name__,
303
self.transport.abspath(self.filename))
305
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
306
"""See VersionedFile._add_delta()."""
307
self._check_add(version_id, []) # should we check the lines ?
308
self._check_versions_present(parents)
888
return "%s(%r, %r)" % (
889
self.__class__.__name__,
893
def without_fallbacks(self):
894
"""Return a clone of this object without any fallbacks configured."""
895
return KnitVersionedFiles(self._index, self._access,
896
self._max_delta_chain, self._factory.annotated,
899
def add_fallback_versioned_files(self, a_versioned_files):
900
"""Add a source of texts for texts not present in this knit.
902
:param a_versioned_files: A VersionedFiles object.
904
self._immediate_fallback_vfs.append(a_versioned_files)
906
def add_lines(self, key, parents, lines, parent_texts=None,
907
left_matching_blocks=None, nostore_sha=None, random_id=False,
909
"""See VersionedFiles.add_lines()."""
910
self._index._check_write_ok()
911
self._check_add(key, lines, random_id, check_content)
913
# The caller might pass None if there is no graph data, but kndx
914
# indexes can't directly store that, so we give them
915
# an empty tuple instead.
917
line_bytes = ''.join(lines)
918
return self._add(key, lines, parents,
919
parent_texts, left_matching_blocks, nostore_sha, random_id,
920
line_bytes=line_bytes)
922
def _add_text(self, key, parents, text, nostore_sha=None, random_id=False):
923
"""See VersionedFiles._add_text()."""
924
self._index._check_write_ok()
925
self._check_add(key, None, random_id, check_content=False)
926
if text.__class__ is not str:
927
raise errors.BzrBadParameterUnicode("text")
929
# The caller might pass None if there is no graph data, but kndx
930
# indexes can't directly store that, so we give them
931
# an empty tuple instead.
933
return self._add(key, None, parents,
934
None, None, nostore_sha, random_id,
937
def _add(self, key, lines, parents, parent_texts,
938
left_matching_blocks, nostore_sha, random_id,
940
"""Add a set of lines on top of version specified by parents.
942
Any versions not present will be converted into ghosts.
944
:param lines: A list of strings where each one is a single line (has a
945
single newline at the end of the string) This is now optional
946
(callers can pass None). It is left in its location for backwards
947
compatibility. It should ''.join(lines) must == line_bytes
948
:param line_bytes: A single string containing the content
950
We pass both lines and line_bytes because different routes bring the
951
values to this function. And for memory efficiency, we don't want to
952
have to split/join on-demand.
954
# first thing, if the content is something we don't need to store, find
956
digest = sha_string(line_bytes)
957
if nostore_sha == digest:
958
raise errors.ExistingContent
309
960
present_parents = []
961
if parent_texts is None:
963
# Do a single query to ascertain parent presence; we only compress
964
# against parents in the same kvf.
965
present_parent_map = self._index.get_parent_map(parents)
312
966
for parent in parents:
313
if not self.has_version(parent):
314
ghosts.append(parent)
967
if parent in present_parent_map:
316
968
present_parents.append(parent)
318
if delta_parent is None:
319
# reconstitute as full text.
320
assert len(delta) == 1 or len(delta) == 0
322
assert delta[0][0] == 0
323
assert delta[0][1] == 0, delta[0][1]
324
return super(KnitVersionedFile, self)._add_delta(version_id,
335
options.append('no-eol')
337
if delta_parent is not None:
338
# determine the current delta chain length.
970
# Currently we can only compress against the left most present parent.
971
if (len(present_parents) == 0 or
972
present_parents[0] != parents[0]):
339
975
# To speed the extract of texts the delta chain is limited
340
976
# to a fixed number of deltas. This should minimize both
341
977
# I/O and the time spend applying deltas.
343
delta_parents = [delta_parent]
345
parent = delta_parents[0]
346
method = self._index.get_method(parent)
347
if method == 'fulltext':
349
delta_parents = self._index.get_parents(parent)
351
if method == 'line-delta':
352
# did not find a fulltext in the delta limit.
353
# just do a normal insertion.
354
return super(KnitVersionedFile, self)._add_delta(version_id,
361
options.append('line-delta')
362
store_lines = self.factory.lower_line_delta(delta)
364
where, size = self._data.add_record(version_id, digest, store_lines)
365
self._index.add_version(version_id, options, where, size, parents)
367
def _add_raw_records(self, records, data):
368
"""Add all the records 'records' with data pre-joined in 'data'.
370
:param records: A list of tuples(version_id, options, parents, size).
371
:param data: The data for the records. When it is written, the records
372
are adjusted to have pos pointing into data by the sum of
373
the preceding records sizes.
376
pos = self._data.add_raw_record(data)
378
for (version_id, options, parents, size) in records:
379
index_entries.append((version_id, options, pos, size, parents))
381
self._index.add_versions(index_entries)
383
def clear_cache(self):
384
"""Clear the data cache only."""
385
self._data.clear_cache()
387
def copy_to(self, name, transport):
388
"""See VersionedFile.copy_to()."""
389
# copy the current index to a temp index to avoid racing with local
391
transport.put(name + INDEX_SUFFIX + '.tmp', self.transport.get(self._index._filename),)
393
f = self._data._open_file()
395
transport.put(name + DATA_SUFFIX, f)
398
# move the copied index into place
399
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
401
def create_empty(self, name, transport, mode=None):
402
return KnitVersionedFile(name, transport, factory=self.factory, delta=self.delta, create=True)
404
def _fix_parents(self, version, new_parents):
405
"""Fix the parents list for version.
407
This is done by appending a new version to the index
408
with identical data except for the parents list.
409
the parents list must be a superset of the current
412
current_values = self._index._cache[version]
413
assert set(current_values[4]).difference(set(new_parents)) == set()
414
self._index.add_version(version,
420
def get_delta(self, version_id):
421
"""Get a delta for constructing version from some other version."""
422
if not self.has_version(version_id):
423
raise RevisionNotPresent(version_id, self.filename)
425
parents = self.get_parents(version_id)
430
data_pos, data_size = self._index.get_position(version_id)
431
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
432
version_idx = self._index.lookup(version_id)
433
noeol = 'no-eol' in self._index.get_options(version_id)
434
if 'fulltext' == self._index.get_method(version_id):
435
new_content = self.factory.parse_fulltext(data, version_idx)
436
if parent is not None:
437
reference_content = self._get_content(parent)
438
old_texts = reference_content.text()
441
new_texts = new_content.text()
442
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
443
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
445
delta = self.factory.parse_line_delta(data, version_idx)
446
return parent, sha1, noeol, delta
448
def get_graph_with_ghosts(self):
449
"""See VersionedFile.get_graph_with_ghosts()."""
450
graph_items = self._index.get_graph()
451
return dict(graph_items)
453
def get_sha1(self, version_id):
454
"""See VersionedFile.get_sha1()."""
455
record_map = self._get_record_map([version_id])
456
method, content, digest, next = record_map[version_id]
461
"""See VersionedFile.get_suffixes()."""
462
return [DATA_SUFFIX, INDEX_SUFFIX]
464
def has_ghost(self, version_id):
465
"""True if there is a ghost reference in the file to version_id."""
467
if self.has_version(version_id):
978
delta = self._check_should_delta(present_parents[0])
980
text_length = len(line_bytes)
983
# Note: line_bytes is not modified to add a newline, that is tracked
984
# via the no_eol flag. 'lines' *is* modified, because that is the
985
# general values needed by the Content code.
986
if line_bytes and line_bytes[-1] != '\n':
987
options.append('no-eol')
989
# Copy the existing list, or create a new one
991
lines = osutils.split_lines(line_bytes)
994
# Replace the last line with one that ends in a final newline
995
lines[-1] = lines[-1] + '\n'
997
lines = osutils.split_lines(line_bytes)
999
for element in key[:-1]:
1000
if type(element) is not str:
1001
raise TypeError("key contains non-strings: %r" % (key,))
1003
key = key[:-1] + ('sha1:' + digest,)
1004
elif type(key[-1]) is not str:
1005
raise TypeError("key contains non-strings: %r" % (key,))
1006
# Knit hunks are still last-element only
1007
version_id = key[-1]
1008
content = self._factory.make(lines, version_id)
1010
# Hint to the content object that its text() call should strip the
1012
content._should_strip_eol = True
1013
if delta or (self._factory.annotated and len(present_parents) > 0):
1014
# Merge annotations from parent texts if needed.
1015
delta_hunks = self._merge_annotations(content, present_parents,
1016
parent_texts, delta, self._factory.annotated,
1017
left_matching_blocks)
1020
options.append('line-delta')
1021
store_lines = self._factory.lower_line_delta(delta_hunks)
1022
size, bytes = self._record_to_data(key, digest,
1025
options.append('fulltext')
1026
# isinstance is slower and we have no hierarchy.
1027
if self._factory.__class__ is KnitPlainFactory:
1028
# Use the already joined bytes saving iteration time in
1030
dense_lines = [line_bytes]
1032
dense_lines.append('\n')
1033
size, bytes = self._record_to_data(key, digest,
1036
# get mixed annotation + content and feed it into the
1038
store_lines = self._factory.lower_fulltext(content)
1039
size, bytes = self._record_to_data(key, digest,
1042
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
1043
self._index.add_records(
1044
((key, options, access_memo, parents),),
1045
random_id=random_id)
1046
return digest, text_length, content
1048
def annotate(self, key):
1049
"""See VersionedFiles.annotate."""
1050
return self._factory.annotate(self, key)
1052
def get_annotator(self):
1053
return _KnitAnnotator(self)
1055
def check(self, progress_bar=None, keys=None):
1056
"""See VersionedFiles.check()."""
1058
return self._logical_check()
1060
# At the moment, check does not extra work over get_record_stream
1061
return self.get_record_stream(keys, 'unordered', True)
1063
def _logical_check(self):
1064
# This doesn't actually test extraction of everything, but that will
1065
# impact 'bzr check' substantially, and needs to be integrated with
1066
# care. However, it does check for the obvious problem of a delta with
1068
keys = self._index.keys()
1069
parent_map = self.get_parent_map(keys)
1071
if self._index.get_method(key) != 'fulltext':
1072
compression_parent = parent_map[key][0]
1073
if compression_parent not in parent_map:
1074
raise errors.KnitCorrupt(self,
1075
"Missing basis parent %s for %s" % (
1076
compression_parent, key))
1077
for fallback_vfs in self._immediate_fallback_vfs:
1078
fallback_vfs.check()
1080
def _check_add(self, key, lines, random_id, check_content):
1081
"""check that version_id and lines are safe to add."""
1082
version_id = key[-1]
1083
if version_id is not None:
1084
if contains_whitespace(version_id):
1085
raise InvalidRevisionId(version_id, self)
1086
self.check_not_reserved_id(version_id)
1087
# TODO: If random_id==False and the key is already present, we should
1088
# probably check that the existing content is identical to what is
1089
# being inserted, and otherwise raise an exception. This would make
1090
# the bundle code simpler.
1092
self._check_lines_not_unicode(lines)
1093
self._check_lines_are_lines(lines)
1095
def _check_header(self, key, line):
1096
rec = self._split_header(line)
1097
self._check_header_version(rec, key[-1])
1100
def _check_header_version(self, rec, version_id):
1101
"""Checks the header version on original format knit records.
1103
These have the last component of the key embedded in the record.
1105
if rec[1] != version_id:
1106
raise KnitCorrupt(self,
1107
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
1109
def _check_should_delta(self, parent):
1110
"""Iterate back through the parent listing, looking for a fulltext.
1112
This is used when we want to decide whether to add a delta or a new
1113
fulltext. It searches for _max_delta_chain parents. When it finds a
1114
fulltext parent, it sees if the total size of the deltas leading up to
1115
it is large enough to indicate that we want a new full text anyway.
1117
Return True if we should create a new delta, False if we should use a
1121
fulltext_size = None
1122
for count in xrange(self._max_delta_chain):
1124
# Note that this only looks in the index of this particular
1125
# KnitVersionedFiles, not in the fallbacks. This ensures that
1126
# we won't store a delta spanning physical repository
1128
build_details = self._index.get_build_details([parent])
1129
parent_details = build_details[parent]
1130
except (RevisionNotPresent, KeyError), e:
1131
# Some basis is not locally present: always fulltext
1133
index_memo, compression_parent, _, _ = parent_details
1134
_, _, size = index_memo
1135
if compression_parent is None:
1136
fulltext_size = size
1139
# We don't explicitly check for presence because this is in an
1140
# inner loop, and if it's missing it'll fail anyhow.
1141
parent = compression_parent
1143
# We couldn't find a fulltext, so we must create a new one
469
# optimisable if needed by memoising the _ghosts set.
470
items = self._index.get_graph()
471
for node, parents in items:
472
for parent in parents:
473
if parent not in self._index._cache:
474
if parent == version_id:
479
"""See VersionedFile.versions."""
480
return self._index.get_versions()
482
def has_version(self, version_id):
483
"""See VersionedFile.has_version."""
484
return self._index.has_version(version_id)
486
__contains__ = has_version
1145
# Simple heuristic - if the total I/O wold be greater as a delta than
1146
# the originally installed fulltext, we create a new fulltext.
1147
return fulltext_size > delta_size
1149
def _build_details_to_components(self, build_details):
1150
"""Convert a build_details tuple to a position tuple."""
1151
# record_details, access_memo, compression_parent
1152
return build_details[3], build_details[0], build_details[1]
1154
def _get_components_positions(self, keys, allow_missing=False):
1155
"""Produce a map of position data for the components of keys.
1157
This data is intended to be used for retrieving the knit records.
1159
A dict of key to (record_details, index_memo, next, parents) is
1162
* method is the way referenced data should be applied.
1163
* index_memo is the handle to pass to the data access to actually get
1165
* next is the build-parent of the version, or None for fulltexts.
1166
* parents is the version_ids of the parents of this version
1168
:param allow_missing: If True do not raise an error on a missing
1169
component, just ignore it.
1172
pending_components = keys
1173
while pending_components:
1174
build_details = self._index.get_build_details(pending_components)
1175
current_components = set(pending_components)
1176
pending_components = set()
1177
for key, details in build_details.iteritems():
1178
(index_memo, compression_parent, parents,
1179
record_details) = details
1180
method = record_details[0]
1181
if compression_parent is not None:
1182
pending_components.add(compression_parent)
1183
component_data[key] = self._build_details_to_components(details)
1184
missing = current_components.difference(build_details)
1185
if missing and not allow_missing:
1186
raise errors.RevisionNotPresent(missing.pop(), self)
1187
return component_data
1189
def _get_content(self, key, parent_texts={}):
1190
"""Returns a content object that makes up the specified
1192
cached_version = parent_texts.get(key, None)
1193
if cached_version is not None:
1194
# Ensure the cache dict is valid.
1195
if not self.get_parent_map([key]):
1196
raise RevisionNotPresent(key, self)
1197
return cached_version
1198
generator = _VFContentMapGenerator(self, [key])
1199
return generator._get_content(key)
1201
def get_parent_map(self, keys):
1202
"""Get a map of the graph parents of keys.
1204
:param keys: The keys to look up parents for.
1205
:return: A mapping from keys to parents. Absent keys are absent from
1208
return self._get_parent_map_with_sources(keys)[0]
1210
def _get_parent_map_with_sources(self, keys):
1211
"""Get a map of the parents of keys.
1213
:param keys: The keys to look up parents for.
1214
:return: A tuple. The first element is a mapping from keys to parents.
1215
Absent keys are absent from the mapping. The second element is a
1216
list with the locations each key was found in. The first element
1217
is the in-this-knit parents, the second the first fallback source,
1221
sources = [self._index] + self._immediate_fallback_vfs
1224
for source in sources:
1227
new_result = source.get_parent_map(missing)
1228
source_results.append(new_result)
1229
result.update(new_result)
1230
missing.difference_update(set(new_result))
1231
return result, source_results
1233
def _get_record_map(self, keys, allow_missing=False):
1234
"""Produce a dictionary of knit records.
1236
:return: {key:(record, record_details, digest, next)}
1238
* record: data returned from read_records (a KnitContentobject)
1239
* record_details: opaque information to pass to parse_record
1240
* digest: SHA1 digest of the full text after all steps are done
1241
* next: build-parent of the version, i.e. the leftmost ancestor.
1242
Will be None if the record is not a delta.
1244
:param keys: The keys to build a map for
1245
:param allow_missing: If some records are missing, rather than
1246
error, just return the data that could be generated.
1248
raw_map = self._get_record_map_unparsed(keys,
1249
allow_missing=allow_missing)
1250
return self._raw_map_to_record_map(raw_map)
1252
def _raw_map_to_record_map(self, raw_map):
1253
"""Parse the contents of _get_record_map_unparsed.
1255
:return: see _get_record_map.
1259
data, record_details, next = raw_map[key]
1260
content, digest = self._parse_record(key[-1], data)
1261
result[key] = content, record_details, digest, next
1264
def _get_record_map_unparsed(self, keys, allow_missing=False):
1265
"""Get the raw data for reconstructing keys without parsing it.
1267
:return: A dict suitable for parsing via _raw_map_to_record_map.
1268
key-> raw_bytes, (method, noeol), compression_parent
1270
# This retries the whole request if anything fails. Potentially we
1271
# could be a bit more selective. We could track the keys whose records
1272
# we have successfully found, and then only request the new records
1273
# from there. However, _get_components_positions grabs the whole build
1274
# chain, which means we'll likely try to grab the same records again
1275
# anyway. Also, can the build chains change as part of a pack
1276
# operation? We wouldn't want to end up with a broken chain.
1279
position_map = self._get_components_positions(keys,
1280
allow_missing=allow_missing)
1281
# key = component_id, r = record_details, i_m = index_memo,
1283
records = [(key, i_m) for key, (r, i_m, n)
1284
in position_map.iteritems()]
1285
# Sort by the index memo, so that we request records from the
1286
# same pack file together, and in forward-sorted order
1287
records.sort(key=operator.itemgetter(1))
1289
for key, data in self._read_records_iter_unchecked(records):
1290
(record_details, index_memo, next) = position_map[key]
1291
raw_record_map[key] = data, record_details, next
1292
return raw_record_map
1293
except errors.RetryWithNewPacks, e:
1294
self._access.reload_or_raise(e)
1297
def _split_by_prefix(cls, keys):
1298
"""For the given keys, split them up based on their prefix.
1300
To keep memory pressure somewhat under control, split the
1301
requests back into per-file-id requests, otherwise "bzr co"
1302
extracts the full tree into memory before writing it to disk.
1303
This should be revisited if _get_content_maps() can ever cross
1306
The keys for a given file_id are kept in the same relative order.
1307
Ordering between file_ids is not, though prefix_order will return the
1308
order that the key was first seen.
1310
:param keys: An iterable of key tuples
1311
:return: (split_map, prefix_order)
1312
split_map A dictionary mapping prefix => keys
1313
prefix_order The order that we saw the various prefixes
1315
split_by_prefix = {}
1323
if prefix in split_by_prefix:
1324
split_by_prefix[prefix].append(key)
1326
split_by_prefix[prefix] = [key]
1327
prefix_order.append(prefix)
1328
return split_by_prefix, prefix_order
1330
def _group_keys_for_io(self, keys, non_local_keys, positions,
1331
_min_buffer_size=_STREAM_MIN_BUFFER_SIZE):
1332
"""For the given keys, group them into 'best-sized' requests.
1334
The idea is to avoid making 1 request per file, but to never try to
1335
unpack an entire 1.5GB source tree in a single pass. Also when
1336
possible, we should try to group requests to the same pack file
1339
:return: list of (keys, non_local) tuples that indicate what keys
1340
should be fetched next.
1342
# TODO: Ideally we would group on 2 factors. We want to extract texts
1343
# from the same pack file together, and we want to extract all
1344
# the texts for a given build-chain together. Ultimately it
1345
# probably needs a better global view.
1346
total_keys = len(keys)
1347
prefix_split_keys, prefix_order = self._split_by_prefix(keys)
1348
prefix_split_non_local_keys, _ = self._split_by_prefix(non_local_keys)
1350
cur_non_local = set()
1354
for prefix in prefix_order:
1355
keys = prefix_split_keys[prefix]
1356
non_local = prefix_split_non_local_keys.get(prefix, [])
1358
this_size = self._index._get_total_build_size(keys, positions)
1359
cur_size += this_size
1360
cur_keys.extend(keys)
1361
cur_non_local.update(non_local)
1362
if cur_size > _min_buffer_size:
1363
result.append((cur_keys, cur_non_local))
1364
sizes.append(cur_size)
1366
cur_non_local = set()
1369
result.append((cur_keys, cur_non_local))
1370
sizes.append(cur_size)
1373
def get_record_stream(self, keys, ordering, include_delta_closure):
1374
"""Get a stream of records for keys.
1376
:param keys: The keys to include.
1377
:param ordering: Either 'unordered' or 'topological'. A topologically
1378
sorted stream has compression parents strictly before their
1380
:param include_delta_closure: If True then the closure across any
1381
compression parents will be included (in the opaque data).
1382
:return: An iterator of ContentFactory objects, each of which is only
1383
valid until the iterator is advanced.
1385
# keys might be a generator
1389
if not self._index.has_graph:
1390
# Cannot sort when no graph has been stored.
1391
ordering = 'unordered'
1393
remaining_keys = keys
1396
keys = set(remaining_keys)
1397
for content_factory in self._get_remaining_record_stream(keys,
1398
ordering, include_delta_closure):
1399
remaining_keys.discard(content_factory.key)
1400
yield content_factory
1402
except errors.RetryWithNewPacks, e:
1403
self._access.reload_or_raise(e)
1405
def _get_remaining_record_stream(self, keys, ordering,
1406
include_delta_closure):
1407
"""This function is the 'retry' portion for get_record_stream."""
1408
if include_delta_closure:
1409
positions = self._get_components_positions(keys, allow_missing=True)
1411
build_details = self._index.get_build_details(keys)
1413
# (record_details, access_memo, compression_parent_key)
1414
positions = dict((key, self._build_details_to_components(details))
1415
for key, details in build_details.iteritems())
1416
absent_keys = keys.difference(set(positions))
1417
# There may be more absent keys : if we're missing the basis component
1418
# and are trying to include the delta closure.
1419
# XXX: We should not ever need to examine remote sources because we do
1420
# not permit deltas across versioned files boundaries.
1421
if include_delta_closure:
1422
needed_from_fallback = set()
1423
# Build up reconstructable_keys dict. key:True in this dict means
1424
# the key can be reconstructed.
1425
reconstructable_keys = {}
1429
chain = [key, positions[key][2]]
1431
needed_from_fallback.add(key)
1434
while chain[-1] is not None:
1435
if chain[-1] in reconstructable_keys:
1436
result = reconstructable_keys[chain[-1]]
1440
chain.append(positions[chain[-1]][2])
1442
# missing basis component
1443
needed_from_fallback.add(chain[-1])
1446
for chain_key in chain[:-1]:
1447
reconstructable_keys[chain_key] = result
1449
needed_from_fallback.add(key)
1450
# Double index lookups here : need a unified api ?
1451
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1452
if ordering in ('topological', 'groupcompress'):
1453
if ordering == 'topological':
1454
# Global topological sort
1455
present_keys = tsort.topo_sort(global_map)
1457
present_keys = sort_groupcompress(global_map)
1458
# Now group by source:
1460
current_source = None
1461
for key in present_keys:
1462
for parent_map in parent_maps:
1463
if key in parent_map:
1464
key_source = parent_map
1466
if current_source is not key_source:
1467
source_keys.append((key_source, []))
1468
current_source = key_source
1469
source_keys[-1][1].append(key)
1471
if ordering != 'unordered':
1472
raise AssertionError('valid values for ordering are:'
1473
' "unordered", "groupcompress" or "topological" not: %r'
1475
# Just group by source; remote sources first.
1478
for parent_map in reversed(parent_maps):
1479
source_keys.append((parent_map, []))
1480
for key in parent_map:
1481
present_keys.append(key)
1482
source_keys[-1][1].append(key)
1483
# We have been requested to return these records in an order that
1484
# suits us. So we ask the index to give us an optimally sorted
1486
for source, sub_keys in source_keys:
1487
if source is parent_maps[0]:
1488
# Only sort the keys for this VF
1489
self._index._sort_keys_by_io(sub_keys, positions)
1490
absent_keys = keys - set(global_map)
1491
for key in absent_keys:
1492
yield AbsentContentFactory(key)
1493
# restrict our view to the keys we can answer.
1494
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1495
# XXX: At that point we need to consider the impact of double reads by
1496
# utilising components multiple times.
1497
if include_delta_closure:
1498
# XXX: get_content_maps performs its own index queries; allow state
1500
non_local_keys = needed_from_fallback - absent_keys
1501
for keys, non_local_keys in self._group_keys_for_io(present_keys,
1504
generator = _VFContentMapGenerator(self, keys, non_local_keys,
1507
for record in generator.get_record_stream():
1510
for source, keys in source_keys:
1511
if source is parent_maps[0]:
1512
# this KnitVersionedFiles
1513
records = [(key, positions[key][1]) for key in keys]
1514
for key, raw_data in self._read_records_iter_unchecked(records):
1515
(record_details, index_memo, _) = positions[key]
1516
yield KnitContentFactory(key, global_map[key],
1517
record_details, None, raw_data, self._factory.annotated, None)
1519
vf = self._immediate_fallback_vfs[parent_maps.index(source) - 1]
1520
for record in vf.get_record_stream(keys, ordering,
1521
include_delta_closure):
1524
def get_sha1s(self, keys):
1525
"""See VersionedFiles.get_sha1s()."""
1527
record_map = self._get_record_map(missing, allow_missing=True)
1529
for key, details in record_map.iteritems():
1530
if key not in missing:
1532
# record entry 2 is the 'digest'.
1533
result[key] = details[2]
1534
missing.difference_update(set(result))
1535
for source in self._immediate_fallback_vfs:
1538
new_result = source.get_sha1s(missing)
1539
result.update(new_result)
1540
missing.difference_update(set(new_result))
1543
def insert_record_stream(self, stream):
1544
"""Insert a record stream into this container.
1546
:param stream: A stream of records to insert.
1548
:seealso VersionedFiles.get_record_stream:
1550
def get_adapter(adapter_key):
1552
return adapters[adapter_key]
1554
adapter_factory = adapter_registry.get(adapter_key)
1555
adapter = adapter_factory(self)
1556
adapters[adapter_key] = adapter
1559
if self._factory.annotated:
1560
# self is annotated, we need annotated knits to use directly.
1561
annotated = "annotated-"
1564
# self is not annotated, but we can strip annotations cheaply.
1566
convertibles = set(["knit-annotated-ft-gz"])
1567
if self._max_delta_chain:
1568
delta_types.add("knit-annotated-delta-gz")
1569
convertibles.add("knit-annotated-delta-gz")
1570
# The set of types we can cheaply adapt without needing basis texts.
1571
native_types = set()
1572
if self._max_delta_chain:
1573
native_types.add("knit-%sdelta-gz" % annotated)
1574
delta_types.add("knit-%sdelta-gz" % annotated)
1575
native_types.add("knit-%sft-gz" % annotated)
1576
knit_types = native_types.union(convertibles)
1578
# Buffer all index entries that we can't add immediately because their
1579
# basis parent is missing. We don't buffer all because generating
1580
# annotations may require access to some of the new records. However we
1581
# can't generate annotations from new deltas until their basis parent
1582
# is present anyway, so we get away with not needing an index that
1583
# includes the new keys.
1585
# See <http://launchpad.net/bugs/300177> about ordering of compression
1586
# parents in the records - to be conservative, we insist that all
1587
# parents must be present to avoid expanding to a fulltext.
1589
# key = basis_parent, value = index entry to add
1590
buffered_index_entries = {}
1591
for record in stream:
1592
kind = record.storage_kind
1593
if kind.startswith('knit-') and kind.endswith('-gz'):
1594
# Check that the ID in the header of the raw knit bytes matches
1595
# the record metadata.
1596
raw_data = record._raw_record
1597
df, rec = self._parse_record_header(record.key, raw_data)
1600
parents = record.parents
1601
if record.storage_kind in delta_types:
1602
# TODO: eventually the record itself should track
1603
# compression_parent
1604
compression_parent = parents[0]
1606
compression_parent = None
1607
# Raise an error when a record is missing.
1608
if record.storage_kind == 'absent':
1609
raise RevisionNotPresent([record.key], self)
1610
elif ((record.storage_kind in knit_types)
1611
and (compression_parent is None
1612
or not self._immediate_fallback_vfs
1613
or self._index.has_key(compression_parent)
1614
or not self.has_key(compression_parent))):
1615
# we can insert the knit record literally if either it has no
1616
# compression parent OR we already have its basis in this kvf
1617
# OR the basis is not present even in the fallbacks. In the
1618
# last case it will either turn up later in the stream and all
1619
# will be well, or it won't turn up at all and we'll raise an
1622
# TODO: self.has_key is somewhat redundant with
1623
# self._index.has_key; we really want something that directly
1624
# asks if it's only present in the fallbacks. -- mbp 20081119
1625
if record.storage_kind not in native_types:
1627
adapter_key = (record.storage_kind, "knit-delta-gz")
1628
adapter = get_adapter(adapter_key)
1630
adapter_key = (record.storage_kind, "knit-ft-gz")
1631
adapter = get_adapter(adapter_key)
1632
bytes = adapter.get_bytes(record)
1634
# It's a knit record, it has a _raw_record field (even if
1635
# it was reconstituted from a network stream).
1636
bytes = record._raw_record
1637
options = [record._build_details[0]]
1638
if record._build_details[1]:
1639
options.append('no-eol')
1640
# Just blat it across.
1641
# Note: This does end up adding data on duplicate keys. As
1642
# modern repositories use atomic insertions this should not
1643
# lead to excessive growth in the event of interrupted fetches.
1644
# 'knit' repositories may suffer excessive growth, but as a
1645
# deprecated format this is tolerable. It can be fixed if
1646
# needed by in the kndx index support raising on a duplicate
1647
# add with identical parents and options.
1648
access_memo = self._access.add_raw_records(
1649
[(record.key, len(bytes))], bytes)[0]
1650
index_entry = (record.key, options, access_memo, parents)
1651
if 'fulltext' not in options:
1652
# Not a fulltext, so we need to make sure the compression
1653
# parent will also be present.
1654
# Note that pack backed knits don't need to buffer here
1655
# because they buffer all writes to the transaction level,
1656
# but we don't expose that difference at the index level. If
1657
# the query here has sufficient cost to show up in
1658
# profiling we should do that.
1660
# They're required to be physically in this
1661
# KnitVersionedFiles, not in a fallback.
1662
if not self._index.has_key(compression_parent):
1663
pending = buffered_index_entries.setdefault(
1664
compression_parent, [])
1665
pending.append(index_entry)
1668
self._index.add_records([index_entry])
1669
elif record.storage_kind == 'chunked':
1670
self.add_lines(record.key, parents,
1671
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1673
# Not suitable for direct insertion as a
1674
# delta, either because it's not the right format, or this
1675
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1676
# 0) or because it depends on a base only present in the
1678
self._access.flush()
1680
# Try getting a fulltext directly from the record.
1681
bytes = record.get_bytes_as('fulltext')
1682
except errors.UnavailableRepresentation:
1683
adapter_key = record.storage_kind, 'fulltext'
1684
adapter = get_adapter(adapter_key)
1685
bytes = adapter.get_bytes(record)
1686
lines = split_lines(bytes)
1688
self.add_lines(record.key, parents, lines)
1689
except errors.RevisionAlreadyPresent:
1691
# Add any records whose basis parent is now available.
1693
added_keys = [record.key]
1695
key = added_keys.pop(0)
1696
if key in buffered_index_entries:
1697
index_entries = buffered_index_entries[key]
1698
self._index.add_records(index_entries)
1700
[index_entry[0] for index_entry in index_entries])
1701
del buffered_index_entries[key]
1702
if buffered_index_entries:
1703
# There were index entries buffered at the end of the stream,
1704
# So these need to be added (if the index supports holding such
1705
# entries for later insertion)
1707
for key in buffered_index_entries:
1708
index_entries = buffered_index_entries[key]
1709
all_entries.extend(index_entries)
1710
self._index.add_records(
1711
all_entries, missing_compression_parents=True)
1713
def get_missing_compression_parent_keys(self):
1714
"""Return an iterable of keys of missing compression parents.
1716
Check this after calling insert_record_stream to find out if there are
1717
any missing compression parents. If there are, the records that
1718
depend on them are not able to be inserted safely. For atomic
1719
KnitVersionedFiles built on packs, the transaction should be aborted or
1720
suspended - commit will fail at this point. Nonatomic knits will error
1721
earlier because they have no staging area to put pending entries into.
1723
return self._index.get_missing_compression_parents()
1725
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1726
"""Iterate over the lines in the versioned files from keys.
1728
This may return lines from other keys. Each item the returned
1729
iterator yields is a tuple of a line and a text version that that line
1730
is present in (not introduced in).
1732
Ordering of results is in whatever order is most suitable for the
1733
underlying storage format.
1735
If a progress bar is supplied, it may be used to indicate progress.
1736
The caller is responsible for cleaning up progress bars (because this
1740
* Lines are normalised by the underlying store: they will all have \\n
1742
* Lines are returned in arbitrary order.
1743
* If a requested key did not change any lines (or didn't have any
1744
lines), it may not be mentioned at all in the result.
1746
:param pb: Progress bar supplied by caller.
1747
:return: An iterator over (line, key).
1750
pb = ui.ui_factory.nested_progress_bar()
1756
# we don't care about inclusions, the caller cares.
1757
# but we need to setup a list of records to visit.
1758
# we need key, position, length
1760
build_details = self._index.get_build_details(keys)
1761
for key, details in build_details.iteritems():
1763
key_records.append((key, details[0]))
1764
records_iter = enumerate(self._read_records_iter(key_records))
1765
for (key_idx, (key, data, sha_value)) in records_iter:
1766
pb.update(gettext('Walking content'), key_idx, total)
1767
compression_parent = build_details[key][1]
1768
if compression_parent is None:
1770
line_iterator = self._factory.get_fulltext_content(data)
1773
line_iterator = self._factory.get_linedelta_content(data)
1774
# Now that we are yielding the data for this key, remove it
1777
# XXX: It might be more efficient to yield (key,
1778
# line_iterator) in the future. However for now, this is a
1779
# simpler change to integrate into the rest of the
1780
# codebase. RBC 20071110
1781
for line in line_iterator:
1784
except errors.RetryWithNewPacks, e:
1785
self._access.reload_or_raise(e)
1786
# If there are still keys we've not yet found, we look in the fallback
1787
# vfs, and hope to find them there. Note that if the keys are found
1788
# but had no changes or no content, the fallback may not return
1790
if keys and not self._immediate_fallback_vfs:
1791
# XXX: strictly the second parameter is meant to be the file id
1792
# but it's not easily accessible here.
1793
raise RevisionNotPresent(keys, repr(self))
1794
for source in self._immediate_fallback_vfs:
1798
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1799
source_keys.add(key)
1801
keys.difference_update(source_keys)
1802
pb.update(gettext('Walking content'), total, total)
1804
def _make_line_delta(self, delta_seq, new_content):
1805
"""Generate a line delta from delta_seq and new_content."""
1807
for op in delta_seq.get_opcodes():
1808
if op[0] == 'equal':
1810
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
488
1813
def _merge_annotations(self, content, parents, parent_texts={},
489
delta=None, annotated=None):
490
"""Merge annotations for content. This is done by comparing
491
the annotations based on changed to the text.
1814
delta=None, annotated=None,
1815
left_matching_blocks=None):
1816
"""Merge annotations for content and generate deltas.
1818
This is done by comparing the annotations based on changes to the text
1819
and generating a delta on the resulting full texts. If annotations are
1820
not being created then a simple delta is created.
1822
if left_matching_blocks is not None:
1823
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
494
1825
delta_seq = None
495
for parent_id in parents:
496
merge_content = self._get_content(parent_id, parent_texts)
497
seq = KnitSequenceMatcher(None, merge_content.text(), content.text())
498
if delta_seq is None:
499
# setup a delta seq to reuse.
1827
for parent_key in parents:
1828
merge_content = self._get_content(parent_key, parent_texts)
1829
if (parent_key == parents[0] and delta_seq is not None):
1832
seq = patiencediff.PatienceSequenceMatcher(
1833
None, merge_content.text(), content.text())
501
1834
for i, j, n in seq.get_matching_blocks():
504
# this appears to copy (origin, text) pairs across to the new
505
# content for any line that matches the last-checked parent.
506
# FIXME: save the sequence control data for delta compression
507
# against the most relevant parent rather than rediffing.
1837
# this copies (origin, text) pairs across to the new
1838
# content for any line that matches the last-checked
508
1840
content._lines[j:j+n] = merge_content._lines[i:i+n]
1841
# XXX: Robert says the following block is a workaround for a
1842
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1843
if content._lines and content._lines[-1][1][-1] != '\n':
1844
# The copied annotation was from a line without a trailing EOL,
1845
# reinstate one for the content object, to ensure correct
1847
line = content._lines[-1][1] + '\n'
1848
content._lines[-1] = (content._lines[-1][0], line)
1850
if delta_seq is None:
511
1851
reference_content = self._get_content(parents[0], parent_texts)
512
1852
new_texts = content.text()
513
1853
old_texts = reference_content.text()
514
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
1854
delta_seq = patiencediff.PatienceSequenceMatcher(
1855
None, old_texts, new_texts)
515
1856
return self._make_line_delta(delta_seq, content)
517
def _make_line_delta(self, delta_seq, new_content):
518
"""Generate a line delta from delta_seq and new_content."""
520
for op in delta_seq.get_opcodes():
523
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
526
def _get_components_positions(self, version_ids):
527
"""Produce a map of position data for the components of versions.
529
This data is intended to be used for retrieving the knit records.
531
A dict of version_id to (method, data_pos, data_size, next) is
533
method is the way referenced data should be applied.
534
data_pos is the position of the data in the knit.
535
data_size is the size of the data in the knit.
536
next is the build-parent of the version, or None for fulltexts.
539
for version_id in version_ids:
542
while cursor is not None and cursor not in component_data:
543
method = self._index.get_method(cursor)
544
if method == 'fulltext':
547
next = self.get_parents(cursor)[0]
548
data_pos, data_size = self._index.get_position(cursor)
549
component_data[cursor] = (method, data_pos, data_size, next)
551
return component_data
553
def _get_content(self, version_id, parent_texts={}):
554
"""Returns a content object that makes up the specified
556
if not self.has_version(version_id):
557
raise RevisionNotPresent(version_id, self.filename)
559
cached_version = parent_texts.get(version_id, None)
560
if cached_version is not None:
561
return cached_version
563
text_map, contents_map = self._get_content_maps([version_id])
564
return contents_map[version_id]
566
def _check_versions_present(self, version_ids):
567
"""Check that all specified versions are present."""
568
version_ids = set(version_ids)
569
for r in list(version_ids):
570
if self._index.has_version(r):
571
version_ids.remove(r)
573
raise RevisionNotPresent(list(version_ids)[0], self.filename)
575
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
576
"""See VersionedFile.add_lines_with_ghosts()."""
577
self._check_add(version_id, lines)
578
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
580
def _add_lines(self, version_id, parents, lines, parent_texts):
581
"""See VersionedFile.add_lines."""
582
self._check_add(version_id, lines)
583
self._check_versions_present(parents)
584
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
586
def _check_add(self, version_id, lines):
587
"""check that version_id and lines are safe to add."""
588
assert self.writable, "knit is not opened for write"
589
### FIXME escape. RBC 20060228
590
if contains_whitespace(version_id):
591
raise InvalidRevisionId(version_id, self.filename)
592
if self.has_version(version_id):
593
raise RevisionAlreadyPresent(version_id, self.filename)
594
self._check_lines_not_unicode(lines)
595
self._check_lines_are_lines(lines)
597
def _add(self, version_id, lines, parents, delta, parent_texts):
598
"""Add a set of lines on top of version specified by parents.
600
If delta is true, compress the text as a line-delta against
603
Any versions not present will be converted into ghosts.
605
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
606
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
607
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
608
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
609
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
610
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
611
# +1383 0 8.0370 8.0370 +<len>
612
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
613
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
614
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
615
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
619
if parent_texts is None:
621
for parent in parents:
622
if not self.has_version(parent):
623
ghosts.append(parent)
625
present_parents.append(parent)
627
if delta and not len(present_parents):
630
digest = sha_strings(lines)
633
if lines[-1][-1] != '\n':
634
options.append('no-eol')
635
lines[-1] = lines[-1] + '\n'
637
if len(present_parents) and delta:
638
# To speed the extract of texts the delta chain is limited
639
# to a fixed number of deltas. This should minimize both
640
# I/O and the time spend applying deltas.
642
delta_parents = present_parents
644
parent = delta_parents[0]
645
method = self._index.get_method(parent)
646
if method == 'fulltext':
648
delta_parents = self._index.get_parents(parent)
650
if method == 'line-delta':
653
lines = self.factory.make(lines, version_id)
654
if delta or (self.factory.annotated and len(present_parents) > 0):
655
# Merge annotations from parent texts if so is needed.
656
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
657
delta, self.factory.annotated)
660
options.append('line-delta')
661
store_lines = self.factory.lower_line_delta(delta_hunks)
1858
def _parse_record(self, version_id, data):
1859
"""Parse an original format knit record.
1861
These have the last element of the key only present in the stored data.
1863
rec, record_contents = self._parse_record_unchecked(data)
1864
self._check_header_version(rec, version_id)
1865
return record_contents, rec[3]
1867
def _parse_record_header(self, key, raw_data):
1868
"""Parse a record header for consistency.
1870
:return: the header and the decompressor stream.
1871
as (stream, header_record)
1873
df = gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1876
rec = self._check_header(key, df.readline())
1877
except Exception, e:
1878
raise KnitCorrupt(self,
1879
"While reading {%s} got %s(%s)"
1880
% (key, e.__class__.__name__, str(e)))
1883
def _parse_record_unchecked(self, data):
1885
# 4168 calls in 2880 217 internal
1886
# 4168 calls to _parse_record_header in 2121
1887
# 4168 calls to readlines in 330
1888
df = gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1890
record_contents = df.readlines()
1891
except Exception, e:
1892
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1893
(data, e.__class__.__name__, str(e)))
1894
header = record_contents.pop(0)
1895
rec = self._split_header(header)
1896
last_line = record_contents.pop()
1897
if len(record_contents) != int(rec[2]):
1898
raise KnitCorrupt(self,
1899
'incorrect number of lines %s != %s'
1900
' for version {%s} %s'
1901
% (len(record_contents), int(rec[2]),
1902
rec[1], record_contents))
1903
if last_line != 'end %s\n' % rec[1]:
1904
raise KnitCorrupt(self,
1905
'unexpected version end line %r, wanted %r'
1906
% (last_line, rec[1]))
1908
return rec, record_contents
1910
def _read_records_iter(self, records):
1911
"""Read text records from data file and yield result.
1913
The result will be returned in whatever is the fastest to read.
1914
Not by the order requested. Also, multiple requests for the same
1915
record will only yield 1 response.
1917
:param records: A list of (key, access_memo) entries
1918
:return: Yields (key, contents, digest) in the order
1919
read, not the order requested
1924
# XXX: This smells wrong, IO may not be getting ordered right.
1925
needed_records = sorted(set(records), key=operator.itemgetter(1))
1926
if not needed_records:
1929
# The transport optimizes the fetching as well
1930
# (ie, reads continuous ranges.)
1931
raw_data = self._access.get_raw_records(
1932
[index_memo for key, index_memo in needed_records])
1934
for (key, index_memo), data in \
1935
izip(iter(needed_records), raw_data):
1936
content, digest = self._parse_record(key[-1], data)
1937
yield key, content, digest
1939
def _read_records_iter_raw(self, records):
1940
"""Read text records from data file and yield raw data.
1942
This unpacks enough of the text record to validate the id is
1943
as expected but thats all.
1945
Each item the iterator yields is (key, bytes,
1946
expected_sha1_of_full_text).
1948
for key, data in self._read_records_iter_unchecked(records):
1949
# validate the header (note that we can only use the suffix in
1950
# current knit records).
1951
df, rec = self._parse_record_header(key, data)
1953
yield key, data, rec[3]
1955
def _read_records_iter_unchecked(self, records):
1956
"""Read text records from data file and yield raw data.
1958
No validation is done.
1960
Yields tuples of (key, data).
1962
# setup an iterator of the external records:
1963
# uses readv so nice and fast we hope.
1965
# grab the disk data needed.
1966
needed_offsets = [index_memo for key, index_memo
1968
raw_records = self._access.get_raw_records(needed_offsets)
1970
for key, index_memo in records:
1971
data = raw_records.next()
1974
def _record_to_data(self, key, digest, lines, dense_lines=None):
1975
"""Convert key, digest, lines into a raw data block.
1977
:param key: The key of the record. Currently keys are always serialised
1978
using just the trailing component.
1979
:param dense_lines: The bytes of lines but in a denser form. For
1980
instance, if lines is a list of 1000 bytestrings each ending in
1981
\\n, dense_lines may be a list with one line in it, containing all
1982
the 1000's lines and their \\n's. Using dense_lines if it is
1983
already known is a win because the string join to create bytes in
1984
this function spends less time resizing the final string.
1985
:return: (len, a StringIO instance with the raw data ready to read.)
1987
chunks = ["version %s %d %s\n" % (key[-1], len(lines), digest)]
1988
chunks.extend(dense_lines or lines)
1989
chunks.append("end %s\n" % key[-1])
1990
for chunk in chunks:
1991
if type(chunk) is not str:
1992
raise AssertionError(
1993
'data must be plain bytes was %s' % type(chunk))
1994
if lines and lines[-1][-1] != '\n':
1995
raise ValueError('corrupt lines value %r' % lines)
1996
compressed_bytes = tuned_gzip.chunks_to_gzip(chunks)
1997
return len(compressed_bytes), compressed_bytes
1999
def _split_header(self, line):
2002
raise KnitCorrupt(self,
2003
'unexpected number of elements in record header')
2007
"""See VersionedFiles.keys."""
2008
if 'evil' in debug.debug_flags:
2009
trace.mutter_callsite(2, "keys scales with size of history")
2010
sources = [self._index] + self._immediate_fallback_vfs
2012
for source in sources:
2013
result.update(source.keys())
2017
class _ContentMapGenerator(object):
2018
"""Generate texts or expose raw deltas for a set of texts."""
2020
def __init__(self, ordering='unordered'):
2021
self._ordering = ordering
2023
def _get_content(self, key):
2024
"""Get the content object for key."""
2025
# Note that _get_content is only called when the _ContentMapGenerator
2026
# has been constructed with just one key requested for reconstruction.
2027
if key in self.nonlocal_keys:
2028
record = self.get_record_stream().next()
2029
# Create a content object on the fly
2030
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
2031
return PlainKnitContent(lines, record.key)
663
options.append('fulltext')
664
store_lines = self.factory.lower_fulltext(lines)
666
where, size = self._data.add_record(version_id, digest, store_lines)
667
self._index.add_version(version_id, options, where, size, parents)
670
def check(self, progress_bar=None):
671
"""See VersionedFile.check()."""
673
def _clone_text(self, new_version_id, old_version_id, parents):
674
"""See VersionedFile.clone_text()."""
675
# FIXME RBC 20060228 make fast by only inserting an index with null
677
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
679
def get_lines(self, version_id):
680
"""See VersionedFile.get_lines()."""
681
return self.get_line_list([version_id])[0]
683
def _get_record_map(self, version_ids):
684
"""Produce a dictionary of knit records.
686
The keys are version_ids, the values are tuples of (method, content,
688
method is the way the content should be applied.
689
content is a KnitContent object.
690
digest is the SHA1 digest of this version id after all steps are done
691
next is the build-parent of the version, i.e. the leftmost ancestor.
692
If the method is fulltext, next will be None.
694
position_map = self._get_components_positions(version_ids)
695
# c = component_id, m = method, p = position, s = size, n = next
696
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
698
for component_id, content, digest in\
699
self._data.read_records_iter(records):
700
method, position, size, next = position_map[component_id]
701
record_map[component_id] = method, content, digest, next
705
def get_text(self, version_id):
706
"""See VersionedFile.get_text"""
707
return self.get_texts([version_id])[0]
709
def get_texts(self, version_ids):
710
return [''.join(l) for l in self.get_line_list(version_ids)]
712
def get_line_list(self, version_ids):
713
"""Return the texts of listed versions as a list of strings."""
714
text_map, content_map = self._get_content_maps(version_ids)
715
return [text_map[v] for v in version_ids]
717
def _get_content_maps(self, version_ids):
718
"""Produce maps of text and KnitContents
2033
# local keys we can ask for directly
2034
return self._get_one_work(key)
2036
def get_record_stream(self):
2037
"""Get a record stream for the keys requested during __init__."""
2038
for record in self._work():
2042
"""Produce maps of text and KnitContents as dicts.
720
2044
:return: (text_map, content_map) where text_map contains the texts for
721
the requested versions and content_map contains the KnitContents.
722
Both dicts take version_ids as their keys.
2045
the requested versions and content_map contains the KnitContents.
724
for version_id in version_ids:
725
if not self.has_version(version_id):
726
raise RevisionNotPresent(version_id, self.filename)
727
record_map = self._get_record_map(version_ids)
732
for version_id in version_ids:
2047
# NB: By definition we never need to read remote sources unless texts
2048
# are requested from them: we don't delta across stores - and we
2049
# explicitly do not want to to prevent data loss situations.
2050
if self.global_map is None:
2051
self.global_map = self.vf.get_parent_map(self.keys)
2052
nonlocal_keys = self.nonlocal_keys
2054
missing_keys = set(nonlocal_keys)
2055
# Read from remote versioned file instances and provide to our caller.
2056
for source in self.vf._immediate_fallback_vfs:
2057
if not missing_keys:
2059
# Loop over fallback repositories asking them for texts - ignore
2060
# any missing from a particular fallback.
2061
for record in source.get_record_stream(missing_keys,
2062
self._ordering, True):
2063
if record.storage_kind == 'absent':
2064
# Not in thie particular stream, may be in one of the
2065
# other fallback vfs objects.
2067
missing_keys.remove(record.key)
2070
if self._raw_record_map is None:
2071
raise AssertionError('_raw_record_map should have been filled')
2073
for key in self.keys:
2074
if key in self.nonlocal_keys:
2076
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2079
def _get_one_work(self, requested_key):
2080
# Now, if we have calculated everything already, just return the
2082
if requested_key in self._contents_map:
2083
return self._contents_map[requested_key]
2084
# To simplify things, parse everything at once - code that wants one text
2085
# probably wants them all.
2086
# FUTURE: This function could be improved for the 'extract many' case
2087
# by tracking each component and only doing the copy when the number of
2088
# children than need to apply delta's to it is > 1 or it is part of the
2090
multiple_versions = len(self.keys) != 1
2091
if self._record_map is None:
2092
self._record_map = self.vf._raw_map_to_record_map(
2093
self._raw_record_map)
2094
record_map = self._record_map
2095
# raw_record_map is key:
2096
# Have read and parsed records at this point.
2097
for key in self.keys:
2098
if key in self.nonlocal_keys:
735
2103
while cursor is not None:
736
method, data, digest, next = record_map[cursor]
737
components.append((cursor, method, data, digest))
738
if cursor in content_map:
2105
record, record_details, digest, next = record_map[cursor]
2107
raise RevisionNotPresent(cursor, self)
2108
components.append((cursor, record, record_details, digest))
2110
if cursor in self._contents_map:
2111
# no need to plan further back
2112
components.append((cursor, None, None, None))
743
for component_id, method, data, digest in reversed(components):
744
if component_id in content_map:
745
content = content_map[component_id]
2116
for (component_id, record, record_details,
2117
digest) in reversed(components):
2118
if component_id in self._contents_map:
2119
content = self._contents_map[component_id]
747
version_idx = self._index.lookup(component_id)
748
if method == 'fulltext':
749
assert content is None
750
content = self.factory.parse_fulltext(data, version_idx)
751
elif method == 'line-delta':
752
delta = self.factory.parse_line_delta(data[:],
754
content = content.copy()
755
content._lines = self._apply_delta(content._lines,
757
content_map[component_id] = content
759
if 'no-eol' in self._index.get_options(version_id):
760
content = content.copy()
761
line = content._lines[-1][1].rstrip('\n')
762
content._lines[-1] = (content._lines[-1][0], line)
763
final_content[version_id] = content
2121
content, delta = self._factory.parse_record(key[-1],
2122
record, record_details, content,
2123
copy_base_content=multiple_versions)
2124
if multiple_versions:
2125
self._contents_map[component_id] = content
765
2127
# digest here is the digest from the last applied component.
766
2128
text = content.text()
767
if sha_strings(text) != digest:
768
raise KnitCorrupt(self.filename,
769
'sha-1 does not match %s' % version_id)
771
text_map[version_id] = text
772
return text_map, final_content
774
def iter_lines_added_or_present_in_versions(self, version_ids=None):
775
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
776
if version_ids is None:
777
version_ids = self.versions()
778
# we don't care about inclusions, the caller cares.
779
# but we need to setup a list of records to visit.
780
# we need version_id, position, length
781
version_id_records = []
782
requested_versions = list(version_ids)
783
# filter for available versions
784
for version_id in requested_versions:
785
if not self.has_version(version_id):
786
raise RevisionNotPresent(version_id, self.filename)
787
# get a in-component-order queue:
789
for version_id in self.versions():
790
if version_id in requested_versions:
791
version_ids.append(version_id)
792
data_pos, length = self._index.get_position(version_id)
793
version_id_records.append((version_id, data_pos, length))
795
pb = bzrlib.ui.ui_factory.nested_progress_bar()
797
total = len(version_id_records)
799
pb.update('Walking content.', count, total)
800
for version_id, data, sha_value in \
801
self._data.read_records_iter(version_id_records):
802
pb.update('Walking content.', count, total)
803
method = self._index.get_method(version_id)
804
version_idx = self._index.lookup(version_id)
805
assert method in ('fulltext', 'line-delta')
806
if method == 'fulltext':
807
content = self.factory.parse_fulltext(data, version_idx)
808
for line in content.text():
811
delta = self.factory.parse_line_delta(data, version_idx)
812
for start, end, count, lines in delta:
813
for origin, line in lines:
816
pb.update('Walking content.', total, total)
819
pb.update('Walking content.', total, total)
823
def num_versions(self):
824
"""See VersionedFile.num_versions()."""
825
return self._index.num_versions()
827
__len__ = num_versions
829
def annotate_iter(self, version_id):
830
"""See VersionedFile.annotate_iter."""
831
content = self._get_content(version_id)
832
for origin, text in content.annotate_iter():
835
def get_parents(self, version_id):
836
"""See VersionedFile.get_parents."""
839
# 52554 calls in 1264 872 internal down from 3674
841
return self._index.get_parents(version_id)
843
raise RevisionNotPresent(version_id, self.filename)
845
def get_parents_with_ghosts(self, version_id):
846
"""See VersionedFile.get_parents."""
848
return self._index.get_parents_with_ghosts(version_id)
850
raise RevisionNotPresent(version_id, self.filename)
852
def get_ancestry(self, versions):
853
"""See VersionedFile.get_ancestry."""
854
if isinstance(versions, basestring):
855
versions = [versions]
858
self._check_versions_present(versions)
859
return self._index.get_ancestry(versions)
861
def get_ancestry_with_ghosts(self, versions):
862
"""See VersionedFile.get_ancestry_with_ghosts."""
863
if isinstance(versions, basestring):
864
versions = [versions]
867
self._check_versions_present(versions)
868
return self._index.get_ancestry_with_ghosts(versions)
870
#@deprecated_method(zero_eight)
871
def walk(self, version_ids):
872
"""See VersionedFile.walk."""
873
# We take the short path here, and extract all relevant texts
874
# and put them in a weave and let that do all the work. Far
875
# from optimal, but is much simpler.
876
# FIXME RB 20060228 this really is inefficient!
877
from bzrlib.weave import Weave
879
w = Weave(self.filename)
880
ancestry = self.get_ancestry(version_ids)
881
sorted_graph = topo_sort(self._index.get_graph())
882
version_list = [vid for vid in sorted_graph if vid in ancestry]
884
for version_id in version_list:
885
lines = self.get_lines(version_id)
886
w.add_lines(version_id, self.get_parents(version_id), lines)
888
for lineno, insert_id, dset, line in w.walk(version_ids):
889
yield lineno, insert_id, dset, line
891
def plan_merge(self, ver_a, ver_b):
892
"""See VersionedFile.plan_merge."""
893
ancestors_b = set(self.get_ancestry(ver_b))
894
def status_a(revision, text):
895
if revision in ancestors_b:
896
return 'killed-b', text
900
ancestors_a = set(self.get_ancestry(ver_a))
901
def status_b(revision, text):
902
if revision in ancestors_a:
903
return 'killed-a', text
907
annotated_a = self.annotate(ver_a)
908
annotated_b = self.annotate(ver_b)
909
plain_a = [t for (a, t) in annotated_a]
910
plain_b = [t for (a, t) in annotated_b]
911
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
914
for ai, bi, l in blocks:
915
# process all mismatched sections
916
# (last mismatched section is handled because blocks always
917
# includes a 0-length last block)
918
for revision, text in annotated_a[a_cur:ai]:
919
yield status_a(revision, text)
920
for revision, text in annotated_b[b_cur:bi]:
921
yield status_b(revision, text)
923
# and now the matched section
926
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
927
assert text_a == text_b
928
yield "unchanged", text_a
931
class _KnitComponentFile(object):
932
"""One of the files used to implement a knit database"""
934
def __init__(self, transport, filename, mode, file_mode=None):
935
self._transport = transport
936
self._filename = filename
938
self._file_mode=file_mode
940
def write_header(self):
941
if self._transport.append(self._filename, StringIO(self.HEADER),
942
mode=self._file_mode):
943
raise KnitCorrupt(self._filename, 'misaligned after writing header')
945
def check_header(self, fp):
947
if line != self.HEADER:
948
raise KnitHeaderError(badline=line)
951
"""Commit is a nop."""
954
return '%s(%s)' % (self.__class__.__name__, self._filename)
957
class _KnitIndex(_KnitComponentFile):
958
"""Manages knit index file.
960
The index is already kept in memory and read on startup, to enable
2129
actual_sha = sha_strings(text)
2130
if actual_sha != digest:
2131
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
2132
if multiple_versions:
2133
return self._contents_map[requested_key]
2137
def _wire_bytes(self):
2138
"""Get the bytes to put on the wire for 'key'.
2140
The first collection of bytes asked for returns the serialised
2141
raw_record_map and the additional details (key, parent) for key.
2142
Subsequent calls return just the additional details (key, parent).
2143
The wire storage_kind given for the first key is 'knit-delta-closure',
2144
For subsequent keys it is 'knit-delta-closure-ref'.
2146
:param key: A key from the content generator.
2147
:return: Bytes to put on the wire.
2150
# kind marker for dispatch on the far side,
2151
lines.append('knit-delta-closure')
2153
if self.vf._factory.annotated:
2154
lines.append('annotated')
2157
# then the list of keys
2158
lines.append('\t'.join(['\x00'.join(key) for key in self.keys
2159
if key not in self.nonlocal_keys]))
2160
# then the _raw_record_map in serialised form:
2162
# for each item in the map:
2164
# 1 line with parents if the key is to be yielded (None: for None, '' for ())
2165
# one line with method
2166
# one line with noeol
2167
# one line with next ('' for None)
2168
# one line with byte count of the record bytes
2170
for key, (record_bytes, (method, noeol), next) in \
2171
self._raw_record_map.iteritems():
2172
key_bytes = '\x00'.join(key)
2173
parents = self.global_map.get(key, None)
2175
parent_bytes = 'None:'
2177
parent_bytes = '\t'.join('\x00'.join(key) for key in parents)
2178
method_bytes = method
2184
next_bytes = '\x00'.join(next)
2187
map_byte_list.append('%s\n%s\n%s\n%s\n%s\n%d\n%s' % (
2188
key_bytes, parent_bytes, method_bytes, noeol_bytes, next_bytes,
2189
len(record_bytes), record_bytes))
2190
map_bytes = ''.join(map_byte_list)
2191
lines.append(map_bytes)
2192
bytes = '\n'.join(lines)
2196
class _VFContentMapGenerator(_ContentMapGenerator):
2197
"""Content map generator reading from a VersionedFiles object."""
2199
def __init__(self, versioned_files, keys, nonlocal_keys=None,
2200
global_map=None, raw_record_map=None, ordering='unordered'):
2201
"""Create a _ContentMapGenerator.
2203
:param versioned_files: The versioned files that the texts are being
2205
:param keys: The keys to produce content maps for.
2206
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
2207
which are known to not be in this knit, but rather in one of the
2209
:param global_map: The result of get_parent_map(keys) (or a supermap).
2210
This is required if get_record_stream() is to be used.
2211
:param raw_record_map: A unparsed raw record map to use for answering
2214
_ContentMapGenerator.__init__(self, ordering=ordering)
2215
# The vf to source data from
2216
self.vf = versioned_files
2218
self.keys = list(keys)
2219
# Keys known to be in fallback vfs objects
2220
if nonlocal_keys is None:
2221
self.nonlocal_keys = set()
2223
self.nonlocal_keys = frozenset(nonlocal_keys)
2224
# Parents data for keys to be returned in get_record_stream
2225
self.global_map = global_map
2226
# The chunked lists for self.keys in text form
2228
# A cache of KnitContent objects used in extracting texts.
2229
self._contents_map = {}
2230
# All the knit records needed to assemble the requested keys as full
2232
self._record_map = None
2233
if raw_record_map is None:
2234
self._raw_record_map = self.vf._get_record_map_unparsed(keys,
2237
self._raw_record_map = raw_record_map
2238
# the factory for parsing records
2239
self._factory = self.vf._factory
2242
class _NetworkContentMapGenerator(_ContentMapGenerator):
2243
"""Content map generator sourced from a network stream."""
2245
def __init__(self, bytes, line_end):
2246
"""Construct a _NetworkContentMapGenerator from a bytes block."""
2248
self.global_map = {}
2249
self._raw_record_map = {}
2250
self._contents_map = {}
2251
self._record_map = None
2252
self.nonlocal_keys = []
2253
# Get access to record parsing facilities
2254
self.vf = KnitVersionedFiles(None, None)
2257
line_end = bytes.find('\n', start)
2258
line = bytes[start:line_end]
2259
start = line_end + 1
2260
if line == 'annotated':
2261
self._factory = KnitAnnotateFactory()
2263
self._factory = KnitPlainFactory()
2264
# list of keys to emit in get_record_stream
2265
line_end = bytes.find('\n', start)
2266
line = bytes[start:line_end]
2267
start = line_end + 1
2269
tuple(segment.split('\x00')) for segment in line.split('\t')
2271
# now a loop until the end. XXX: It would be nice if this was just a
2272
# bunch of the same records as get_record_stream(..., False) gives, but
2273
# there is a decent sized gap stopping that at the moment.
2277
line_end = bytes.find('\n', start)
2278
key = tuple(bytes[start:line_end].split('\x00'))
2279
start = line_end + 1
2280
# 1 line with parents (None: for None, '' for ())
2281
line_end = bytes.find('\n', start)
2282
line = bytes[start:line_end]
2287
[tuple(segment.split('\x00')) for segment in line.split('\t')
2289
self.global_map[key] = parents
2290
start = line_end + 1
2291
# one line with method
2292
line_end = bytes.find('\n', start)
2293
line = bytes[start:line_end]
2295
start = line_end + 1
2296
# one line with noeol
2297
line_end = bytes.find('\n', start)
2298
line = bytes[start:line_end]
2300
start = line_end + 1
2301
# one line with next ('' for None)
2302
line_end = bytes.find('\n', start)
2303
line = bytes[start:line_end]
2307
next = tuple(bytes[start:line_end].split('\x00'))
2308
start = line_end + 1
2309
# one line with byte count of the record bytes
2310
line_end = bytes.find('\n', start)
2311
line = bytes[start:line_end]
2313
start = line_end + 1
2315
record_bytes = bytes[start:start+count]
2316
start = start + count
2318
self._raw_record_map[key] = (record_bytes, (method, noeol), next)
2320
def get_record_stream(self):
2321
"""Get a record stream for for keys requested by the bytestream."""
2323
for key in self.keys:
2324
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2327
def _wire_bytes(self):
2331
class _KndxIndex(object):
2332
"""Manages knit index files
2334
The index is kept in memory and read on startup, to enable
961
2335
fast lookups of revision information. The cursor of the index
962
2336
file is always pointing to the end, making it easy to append
974
2348
Duplicate entries may be written to the index for a single version id
975
2349
if this is done then the latter one completely replaces the former:
976
this allows updates to correct version and parent information.
2350
this allows updates to correct version and parent information.
977
2351
Note that the two entries may share the delta, and that successive
978
2352
annotations and references MUST point to the first entry.
980
2354
The index file on disc contains a header, followed by one line per knit
981
2355
record. The same revision can be present in an index file more than once.
982
The first occurrence gets assigned a sequence number starting from 0.
2356
The first occurrence gets assigned a sequence number starting from 0.
984
2358
The format of a single line is
985
2359
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
986
2360
REVISION_ID is a utf8-encoded revision id
987
FLAGS is a comma separated list of flags about the record. Values include
2361
FLAGS is a comma separated list of flags about the record. Values include
988
2362
no-eol, line-delta, fulltext.
989
2363
BYTE_OFFSET is the ascii representation of the byte offset in the data file
990
that the the compressed data starts at.
2364
that the compressed data starts at.
991
2365
LENGTH is the ascii representation of the length of the data file.
992
2366
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
994
2368
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
995
2369
revision id already in the knit that is a parent of REVISION_ID.
996
2370
The ' :' marker is the end of record marker.
999
when a write is interrupted to the index file, it will result in a line that
1000
does not end in ' :'. If the ' :' is not present at the end of a line, or at
1001
the end of the file, then the record that is missing it will be ignored by
2373
when a write is interrupted to the index file, it will result in a line
2374
that does not end in ' :'. If the ' :' is not present at the end of a line,
2375
or at the end of the file, then the record that is missing it will be
2376
ignored by the parser.
1004
2378
When writing new records to the index file, the data is preceded by '\n'
1005
2379
to ensure that records always start on new lines even if the last write was
1006
2380
interrupted. As a result its normal for the last line in the index to be
1007
2381
missing a trailing newline. One can be added with no harmful effects.
2383
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
2384
where prefix is e.g. the (fileid,) for .texts instances or () for
2385
constant-mapped things like .revisions, and the old state is
2386
tuple(history_vector, cache_dict). This is used to prevent having an
2387
ABI change with the C extension that reads .kndx files.
1010
2390
HEADER = "# bzr knit index 8\n"
1012
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1013
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1015
def _cache_version(self, version_id, options, pos, size, parents):
2392
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
2393
"""Create a _KndxIndex on transport using mapper."""
2394
self._transport = transport
2395
self._mapper = mapper
2396
self._get_scope = get_scope
2397
self._allow_writes = allow_writes
2398
self._is_locked = is_locked
2400
self.has_graph = True
2402
def add_records(self, records, random_id=False, missing_compression_parents=False):
2403
"""Add multiple records to the index.
2405
:param records: a list of tuples:
2406
(key, options, access_memo, parents).
2407
:param random_id: If True the ids being added were randomly generated
2408
and no check for existence will be performed.
2409
:param missing_compression_parents: If True the records being added are
2410
only compressed against texts already in the index (or inside
2411
records). If False the records all refer to unavailable texts (or
2412
texts inside records) as compression parents.
2414
if missing_compression_parents:
2415
# It might be nice to get the edge of the records. But keys isn't
2417
keys = sorted(record[0] for record in records)
2418
raise errors.RevisionNotPresent(keys, self)
2420
for record in records:
2423
path = self._mapper.map(key) + '.kndx'
2424
path_keys = paths.setdefault(path, (prefix, []))
2425
path_keys[1].append(record)
2426
for path in sorted(paths):
2427
prefix, path_keys = paths[path]
2428
self._load_prefixes([prefix])
2430
orig_history = self._kndx_cache[prefix][1][:]
2431
orig_cache = self._kndx_cache[prefix][0].copy()
2434
for key, options, (_, pos, size), parents in path_keys:
2436
# kndx indices cannot be parentless.
2438
line = "\n%s %s %s %s %s :" % (
2439
key[-1], ','.join(options), pos, size,
2440
self._dictionary_compress(parents))
2441
if type(line) is not str:
2442
raise AssertionError(
2443
'data must be utf8 was %s' % type(line))
2445
self._cache_key(key, options, pos, size, parents)
2446
if len(orig_history):
2447
self._transport.append_bytes(path, ''.join(lines))
2449
self._init_index(path, lines)
2451
# If any problems happen, restore the original values and re-raise
2452
self._kndx_cache[prefix] = (orig_cache, orig_history)
2455
def scan_unvalidated_index(self, graph_index):
2456
"""See _KnitGraphIndex.scan_unvalidated_index."""
2457
# Because kndx files do not support atomic insertion via separate index
2458
# files, they do not support this method.
2459
raise NotImplementedError(self.scan_unvalidated_index)
2461
def get_missing_compression_parents(self):
2462
"""See _KnitGraphIndex.get_missing_compression_parents."""
2463
# Because kndx files do not support atomic insertion via separate index
2464
# files, they do not support this method.
2465
raise NotImplementedError(self.get_missing_compression_parents)
2467
def _cache_key(self, key, options, pos, size, parent_keys):
1016
2468
"""Cache a version record in the history array and index cache.
1018
This is inlined into __init__ for performance. KEEP IN SYNC.
2470
This is inlined into _load_data for performance. KEEP IN SYNC.
1019
2471
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
2475
version_id = key[-1]
2476
# last-element only for compatibilty with the C load_data.
2477
parents = tuple(parent[-1] for parent in parent_keys)
2478
for parent in parent_keys:
2479
if parent[:-1] != prefix:
2480
raise ValueError("mismatched prefixes for %r, %r" % (
2482
cache, history = self._kndx_cache[prefix]
1022
2483
# only want the _history index to reference the 1st index entry
1023
2484
# for version_id
1024
if version_id not in self._cache:
1025
index = len(self._history)
1026
self._history.append(version_id)
2485
if version_id not in cache:
2486
index = len(history)
2487
history.append(version_id)
1028
index = self._cache[version_id][5]
1029
self._cache[version_id] = (version_id,
2489
index = cache[version_id][5]
2490
cache[version_id] = (version_id,
1036
def __init__(self, transport, filename, mode, create=False, file_mode=None):
1037
_KnitComponentFile.__init__(self, transport, filename, mode, file_mode)
1039
# position in _history is the 'official' index for a revision
1040
# but the values may have come from a newer entry.
1041
# so - wc -l of a knit index is != the number of unique names
1044
pb = bzrlib.ui.ui_factory.nested_progress_bar()
2497
def check_header(self, fp):
2498
line = fp.readline()
2500
# An empty file can actually be treated as though the file doesn't
2502
raise errors.NoSuchFile(self)
2503
if line != self.HEADER:
2504
raise KnitHeaderError(badline=line, filename=self)
2506
def _check_read(self):
2507
if not self._is_locked():
2508
raise errors.ObjectNotLocked(self)
2509
if self._get_scope() != self._scope:
2512
def _check_write_ok(self):
2513
"""Assert if not writes are permitted."""
2514
if not self._is_locked():
2515
raise errors.ObjectNotLocked(self)
2516
if self._get_scope() != self._scope:
2518
if self._mode != 'w':
2519
raise errors.ReadOnlyObjectDirtiedError(self)
2521
def get_build_details(self, keys):
2522
"""Get the method, index_memo and compression parent for keys.
2524
Ghosts are omitted from the result.
2526
:param keys: An iterable of keys.
2527
:return: A dict of key:(index_memo, compression_parent, parents,
2530
opaque structure to pass to read_records to extract the raw
2533
Content that this record is built upon, may be None
2535
Logical parents of this node
2537
extra information about the content which needs to be passed to
2538
Factory.parse_record
2540
parent_map = self.get_parent_map(keys)
2543
if key not in parent_map:
2545
method = self.get_method(key)
2546
parents = parent_map[key]
2547
if method == 'fulltext':
2548
compression_parent = None
2550
compression_parent = parents[0]
2551
noeol = 'no-eol' in self.get_options(key)
2552
index_memo = self.get_position(key)
2553
result[key] = (index_memo, compression_parent,
2554
parents, (method, noeol))
2557
def get_method(self, key):
2558
"""Return compression method of specified key."""
2559
options = self.get_options(key)
2560
if 'fulltext' in options:
2562
elif 'line-delta' in options:
2565
raise errors.KnitIndexUnknownMethod(self, options)
2567
def get_options(self, key):
2568
"""Return a list representing options.
2572
prefix, suffix = self._split_key(key)
2573
self._load_prefixes([prefix])
1049
pb.update('read knit index', count, total)
1050
fp = self._transport.get(self._filename)
2575
return self._kndx_cache[prefix][0][suffix][1]
2577
raise RevisionNotPresent(key, self)
2579
def find_ancestry(self, keys):
2580
"""See CombinedGraphIndex.find_ancestry()"""
2581
prefixes = set(key[:-1] for key in keys)
2582
self._load_prefixes(prefixes)
2585
missing_keys = set()
2586
pending_keys = list(keys)
2587
# This assumes that keys will not reference parents in a different
2588
# prefix, which is accurate so far.
2590
key = pending_keys.pop()
2591
if key in parent_map:
2595
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2597
missing_keys.add(key)
2599
parent_keys = tuple([prefix + (suffix,)
2600
for suffix in suffix_parents])
2601
parent_map[key] = parent_keys
2602
pending_keys.extend([p for p in parent_keys
2603
if p not in parent_map])
2604
return parent_map, missing_keys
2606
def get_parent_map(self, keys):
2607
"""Get a map of the parents of keys.
2609
:param keys: The keys to look up parents for.
2610
:return: A mapping from keys to parents. Absent keys are absent from
2613
# Parse what we need to up front, this potentially trades off I/O
2614
# locality (.kndx and .knit in the same block group for the same file
2615
# id) for less checking in inner loops.
2616
prefixes = set(key[:-1] for key in keys)
2617
self._load_prefixes(prefixes)
2622
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2626
result[key] = tuple(prefix + (suffix,) for
2627
suffix in suffix_parents)
2630
def get_position(self, key):
2631
"""Return details needed to access the version.
2633
:return: a tuple (key, data position, size) to hand to the access
2634
logic to get the record.
2636
prefix, suffix = self._split_key(key)
2637
self._load_prefixes([prefix])
2638
entry = self._kndx_cache[prefix][0][suffix]
2639
return key, entry[2], entry[3]
2641
has_key = _mod_index._has_key_from_parent_map
2643
def _init_index(self, path, extra_lines=[]):
2644
"""Initialize an index."""
2646
sio.write(self.HEADER)
2647
sio.writelines(extra_lines)
2649
self._transport.put_file_non_atomic(path, sio,
2650
create_parent_dir=True)
2651
# self._create_parent_dir)
2652
# mode=self._file_mode,
2653
# dir_mode=self._dir_mode)
2656
"""Get all the keys in the collection.
2658
The keys are not ordered.
2661
# Identify all key prefixes.
2662
# XXX: A bit hacky, needs polish.
2663
if type(self._mapper) is ConstantMapper:
2667
for quoted_relpath in self._transport.iter_files_recursive():
2668
path, ext = os.path.splitext(quoted_relpath)
2670
prefixes = [self._mapper.unmap(path) for path in relpaths]
2671
self._load_prefixes(prefixes)
2672
for prefix in prefixes:
2673
for suffix in self._kndx_cache[prefix][1]:
2674
result.add(prefix + (suffix,))
2677
def _load_prefixes(self, prefixes):
2678
"""Load the indices for prefixes."""
2680
for prefix in prefixes:
2681
if prefix not in self._kndx_cache:
2682
# the load_data interface writes to these variables.
2685
self._filename = prefix
1052
self.check_header(fp)
1053
# readlines reads the whole file at once:
1054
# bad for transports like http, good for local disk
1055
# we save 60 ms doing this one change (
1056
# from calling readline each time to calling
1058
# probably what we want for nice behaviour on
1059
# http is a incremental readlines that yields, or
1060
# a check for local vs non local indexes,
1061
for l in fp.readlines():
1063
if len(rec) < 5 or rec[-1] != ':':
1065
# FIXME: in the future we should determine if its a
1066
# short write - and ignore it
1067
# or a different failure, and raise. RBC 20060407
1071
#pb.update('read knit index', count, total)
1072
# See self._parse_parents
1074
for value in rec[4:-1]:
1076
# uncompressed reference
1077
parents.append(value[1:])
1079
# this is 15/4000ms faster than isinstance,
1081
# this function is called thousands of times a
1082
# second so small variations add up.
1083
assert value.__class__ is str
1084
parents.append(self._history[int(value)])
1085
# end self._parse_parents
1086
# self._cache_version(rec[0],
1087
# rec[1].split(','),
1091
# --- self._cache_version
1092
# only want the _history index to reference the 1st
1093
# index entry for version_id
1095
if version_id not in self._cache:
1096
index = len(self._history)
1097
self._history.append(version_id)
1099
index = self._cache[version_id][5]
1100
self._cache[version_id] = (version_id,
1106
# --- self._cache_version
1109
except NoSuchFile, e:
1110
if mode != 'w' or not create:
1114
pb.update('read knit index', total, total)
1117
def _parse_parents(self, compressed_parents):
1118
"""convert a list of string parent values into version ids.
1120
ints are looked up in the index.
1121
.FOO values are ghosts and converted in to FOO.
1123
NOTE: the function is retained here for clarity, and for possible
1124
use in partial index reads. However bulk processing now has
1125
it inlined in __init__ for inner-loop optimisation.
2687
path = self._mapper.map(prefix) + '.kndx'
2688
fp = self._transport.get(path)
2690
# _load_data may raise NoSuchFile if the target knit is
2692
_load_data(self, fp)
2695
self._kndx_cache[prefix] = (self._cache, self._history)
2700
self._kndx_cache[prefix] = ({}, [])
2701
if type(self._mapper) is ConstantMapper:
2702
# preserve behaviour for revisions.kndx etc.
2703
self._init_index(path)
2708
missing_keys = _mod_index._missing_keys_from_parent_map
2710
def _partition_keys(self, keys):
2711
"""Turn keys into a dict of prefix:suffix_list."""
2714
prefix_keys = result.setdefault(key[:-1], [])
2715
prefix_keys.append(key[-1])
2718
def _dictionary_compress(self, keys):
2719
"""Dictionary compress keys.
2721
:param keys: The keys to generate references to.
2722
:return: A string representation of keys. keys which are present are
2723
dictionary compressed, and others are emitted as fulltext with a
1128
for value in compressed_parents:
1129
if value[-1] == '.':
1130
# uncompressed reference
1131
result.append(value[1:])
1133
# this is 15/4000ms faster than isinstance,
1134
# this function is called thousands of times a
1135
# second so small variations add up.
1136
assert value.__class__ is str
1137
result.append(self._history[int(value)])
1140
def get_graph(self):
1142
for version_id, index in self._cache.iteritems():
1143
graph.append((version_id, index[4]))
1146
def get_ancestry(self, versions):
1147
"""See VersionedFile.get_ancestry."""
1148
# get a graph of all the mentioned versions:
1150
pending = set(versions)
1152
version = pending.pop()
1153
parents = self._cache[version][4]
1154
# got the parents ok
1156
parents = [parent for parent in parents if parent in self._cache]
1157
for parent in parents:
1158
# if not completed and not a ghost
1159
if parent not in graph:
1161
graph[version] = parents
1162
return topo_sort(graph.items())
1164
def get_ancestry_with_ghosts(self, versions):
1165
"""See VersionedFile.get_ancestry_with_ghosts."""
1166
# get a graph of all the mentioned versions:
1168
pending = set(versions)
1170
version = pending.pop()
1172
parents = self._cache[version][4]
1178
# got the parents ok
1179
for parent in parents:
1180
if parent not in graph:
1182
graph[version] = parents
1183
return topo_sort(graph.items())
1185
def num_versions(self):
1186
return len(self._history)
1188
__len__ = num_versions
1190
def get_versions(self):
1191
return self._history
1193
def idx_to_name(self, idx):
1194
return self._history[idx]
1196
def lookup(self, version_id):
1197
assert version_id in self._cache
1198
return self._cache[version_id][5]
1200
def _version_list_to_index(self, versions):
1201
2728
result_list = []
1202
for version in versions:
1203
if version in self._cache:
2729
prefix = keys[0][:-1]
2730
cache = self._kndx_cache[prefix][0]
2732
if key[:-1] != prefix:
2733
# kndx indices cannot refer across partitioned storage.
2734
raise ValueError("mismatched prefixes for %r" % keys)
2735
if key[-1] in cache:
1204
2736
# -- inlined lookup() --
1205
result_list.append(str(self._cache[version][5]))
2737
result_list.append(str(cache[key[-1]][5]))
1206
2738
# -- end lookup () --
1208
result_list.append('.' + version.encode('utf-8'))
2740
result_list.append('.' + key[-1])
1209
2741
return ' '.join(result_list)
1211
def add_version(self, version_id, options, pos, size, parents):
1212
"""Add a version record to the index."""
1213
self.add_versions(((version_id, options, pos, size, parents),))
1215
def add_versions(self, versions):
1216
"""Add multiple versions to the index.
1218
:param versions: a list of tuples:
1219
(version_id, options, pos, size, parents).
1222
for version_id, options, pos, size, parents in versions:
1223
line = "\n%s %s %s %s %s :" % (version_id.encode('utf-8'),
1227
self._version_list_to_index(parents))
1228
assert isinstance(line, str), \
1229
'content must be utf-8 encoded: %r' % (line,)
1231
self._transport.append(self._filename, StringIO(''.join(lines)))
1232
# cache after writing, so that a failed write leads to missing cache
1233
# entries not extra ones. XXX TODO: RBC 20060502 in the event of a
1234
# failure, reload the index or flush it or some such, to prevent
1235
# writing records that did complete twice.
1236
for version_id, options, pos, size, parents in versions:
1237
self._cache_version(version_id, options, pos, size, parents)
1239
def has_version(self, version_id):
1240
"""True if the version is in the index."""
1241
return self._cache.has_key(version_id)
1243
def get_position(self, version_id):
1244
"""Return data position and size of specified version."""
1245
return (self._cache[version_id][2], \
1246
self._cache[version_id][3])
1248
def get_method(self, version_id):
1249
"""Return compression method of specified version."""
1250
options = self._cache[version_id][1]
1251
if 'fulltext' in options:
2743
def _reset_cache(self):
2744
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2745
# (cache_dict, history_vector) for parsed kndx files.
2746
self._kndx_cache = {}
2747
self._scope = self._get_scope()
2748
allow_writes = self._allow_writes()
2754
def _sort_keys_by_io(self, keys, positions):
2755
"""Figure out an optimal order to read the records for the given keys.
2757
Sort keys, grouped by index and sorted by position.
2759
:param keys: A list of keys whose records we want to read. This will be
2761
:param positions: A dict, such as the one returned by
2762
_get_components_positions()
2765
def get_sort_key(key):
2766
index_memo = positions[key][1]
2767
# Group by prefix and position. index_memo[0] is the key, so it is
2768
# (file_id, revision_id) and we don't want to sort on revision_id,
2769
# index_memo[1] is the position, and index_memo[2] is the size,
2770
# which doesn't matter for the sort
2771
return index_memo[0][:-1], index_memo[1]
2772
return keys.sort(key=get_sort_key)
2774
_get_total_build_size = _get_total_build_size
2776
def _split_key(self, key):
2777
"""Split key into a prefix and suffix."""
2778
return key[:-1], key[-1]
2781
class _KnitGraphIndex(object):
2782
"""A KnitVersionedFiles index layered on GraphIndex."""
2784
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2785
add_callback=None, track_external_parent_refs=False):
2786
"""Construct a KnitGraphIndex on a graph_index.
2788
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2789
:param is_locked: A callback to check whether the object should answer
2791
:param deltas: Allow delta-compressed records.
2792
:param parents: If True, record knits parents, if not do not record
2794
:param add_callback: If not None, allow additions to the index and call
2795
this callback with a list of added GraphIndex nodes:
2796
[(node, value, node_refs), ...]
2797
:param is_locked: A callback, returns True if the index is locked and
2799
:param track_external_parent_refs: If True, record all external parent
2800
references parents from added records. These can be retrieved
2801
later by calling get_missing_parents().
2803
self._add_callback = add_callback
2804
self._graph_index = graph_index
2805
self._deltas = deltas
2806
self._parents = parents
2807
if deltas and not parents:
2808
# XXX: TODO: Delta tree and parent graph should be conceptually
2810
raise KnitCorrupt(self, "Cannot do delta compression without "
2812
self.has_graph = parents
2813
self._is_locked = is_locked
2814
self._missing_compression_parents = set()
2815
if track_external_parent_refs:
2816
self._key_dependencies = _KeyRefs()
2818
self._key_dependencies = None
2821
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2823
def add_records(self, records, random_id=False,
2824
missing_compression_parents=False):
2825
"""Add multiple records to the index.
2827
This function does not insert data into the Immutable GraphIndex
2828
backing the KnitGraphIndex, instead it prepares data for insertion by
2829
the caller and checks that it is safe to insert then calls
2830
self._add_callback with the prepared GraphIndex nodes.
2832
:param records: a list of tuples:
2833
(key, options, access_memo, parents).
2834
:param random_id: If True the ids being added were randomly generated
2835
and no check for existence will be performed.
2836
:param missing_compression_parents: If True the records being added are
2837
only compressed against texts already in the index (or inside
2838
records). If False the records all refer to unavailable texts (or
2839
texts inside records) as compression parents.
2841
if not self._add_callback:
2842
raise errors.ReadOnlyError(self)
2843
# we hope there are no repositories with inconsistent parentage
2847
compression_parents = set()
2848
key_dependencies = self._key_dependencies
2849
for (key, options, access_memo, parents) in records:
2851
parents = tuple(parents)
2852
if key_dependencies is not None:
2853
key_dependencies.add_references(key, parents)
2854
index, pos, size = access_memo
2855
if 'no-eol' in options:
2859
value += "%d %d" % (pos, size)
2860
if not self._deltas:
2861
if 'line-delta' in options:
2862
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2865
if 'line-delta' in options:
2866
node_refs = (parents, (parents[0],))
2867
if missing_compression_parents:
2868
compression_parents.add(parents[0])
2870
node_refs = (parents, ())
2872
node_refs = (parents, )
2875
raise KnitCorrupt(self, "attempt to add node with parents "
2876
"in parentless index.")
2878
keys[key] = (value, node_refs)
2881
present_nodes = self._get_entries(keys)
2882
for (index, key, value, node_refs) in present_nodes:
2883
parents = node_refs[:1]
2884
# Sometimes these are passed as a list rather than a tuple
2885
passed = static_tuple.as_tuples(keys[key])
2886
passed_parents = passed[1][:1]
2887
if (value[0] != keys[key][0][0] or
2888
parents != passed_parents):
2889
node_refs = static_tuple.as_tuples(node_refs)
2890
raise KnitCorrupt(self, "inconsistent details in add_records"
2891
": %s %s" % ((value, node_refs), passed))
2895
for key, (value, node_refs) in keys.iteritems():
2896
result.append((key, value, node_refs))
2898
for key, (value, node_refs) in keys.iteritems():
2899
result.append((key, value))
2900
self._add_callback(result)
2901
if missing_compression_parents:
2902
# This may appear to be incorrect (it does not check for
2903
# compression parents that are in the existing graph index),
2904
# but such records won't have been buffered, so this is
2905
# actually correct: every entry when
2906
# missing_compression_parents==True either has a missing parent, or
2907
# a parent that is one of the keys in records.
2908
compression_parents.difference_update(keys)
2909
self._missing_compression_parents.update(compression_parents)
2910
# Adding records may have satisfied missing compression parents.
2911
self._missing_compression_parents.difference_update(keys)
2913
def scan_unvalidated_index(self, graph_index):
2914
"""Inform this _KnitGraphIndex that there is an unvalidated index.
2916
This allows this _KnitGraphIndex to keep track of any missing
2917
compression parents we may want to have filled in to make those
2920
:param graph_index: A GraphIndex
2923
new_missing = graph_index.external_references(ref_list_num=1)
2924
new_missing.difference_update(self.get_parent_map(new_missing))
2925
self._missing_compression_parents.update(new_missing)
2926
if self._key_dependencies is not None:
2927
# Add parent refs from graph_index (and discard parent refs that
2928
# the graph_index has).
2929
for node in graph_index.iter_all_entries():
2930
self._key_dependencies.add_references(node[1], node[3][0])
2932
def get_missing_compression_parents(self):
2933
"""Return the keys of missing compression parents.
2935
Missing compression parents occur when a record stream was missing
2936
basis texts, or a index was scanned that had missing basis texts.
2938
return frozenset(self._missing_compression_parents)
2940
def get_missing_parents(self):
2941
"""Return the keys of missing parents."""
2942
# If updating this, you should also update
2943
# groupcompress._GCGraphIndex.get_missing_parents
2944
# We may have false positives, so filter those out.
2945
self._key_dependencies.satisfy_refs_for_keys(
2946
self.get_parent_map(self._key_dependencies.get_unsatisfied_refs()))
2947
return frozenset(self._key_dependencies.get_unsatisfied_refs())
2949
def _check_read(self):
2950
"""raise if reads are not permitted."""
2951
if not self._is_locked():
2952
raise errors.ObjectNotLocked(self)
2954
def _check_write_ok(self):
2955
"""Assert if writes are not permitted."""
2956
if not self._is_locked():
2957
raise errors.ObjectNotLocked(self)
2959
def _compression_parent(self, an_entry):
2960
# return the key that an_entry is compressed against, or None
2961
# Grab the second parent list (as deltas implies parents currently)
2962
compression_parents = an_entry[3][1]
2963
if not compression_parents:
2965
if len(compression_parents) != 1:
2966
raise AssertionError(
2967
"Too many compression parents: %r" % compression_parents)
2968
return compression_parents[0]
2970
def get_build_details(self, keys):
2971
"""Get the method, index_memo and compression parent for version_ids.
2973
Ghosts are omitted from the result.
2975
:param keys: An iterable of keys.
2976
:return: A dict of key:
2977
(index_memo, compression_parent, parents, record_details).
2979
opaque structure to pass to read_records to extract the raw
2982
Content that this record is built upon, may be None
2984
Logical parents of this node
2986
extra information about the content which needs to be passed to
2987
Factory.parse_record
2991
entries = self._get_entries(keys, False)
2992
for entry in entries:
2994
if not self._parents:
2997
parents = entry[3][0]
2998
if not self._deltas:
2999
compression_parent_key = None
3001
compression_parent_key = self._compression_parent(entry)
3002
noeol = (entry[2][0] == 'N')
3003
if compression_parent_key:
3004
method = 'line-delta'
3007
result[key] = (self._node_to_position(entry),
3008
compression_parent_key, parents,
3012
def _get_entries(self, keys, check_present=False):
3013
"""Get the entries for keys.
3015
:param keys: An iterable of index key tuples.
3020
for node in self._graph_index.iter_entries(keys):
3022
found_keys.add(node[1])
3024
# adapt parentless index to the rest of the code.
3025
for node in self._graph_index.iter_entries(keys):
3026
yield node[0], node[1], node[2], ()
3027
found_keys.add(node[1])
3029
missing_keys = keys.difference(found_keys)
3031
raise RevisionNotPresent(missing_keys.pop(), self)
3033
def get_method(self, key):
3034
"""Return compression method of specified key."""
3035
return self._get_method(self._get_node(key))
3037
def _get_method(self, node):
3038
if not self._deltas:
1252
3039
return 'fulltext'
1254
assert 'line-delta' in options
3040
if self._compression_parent(node):
1255
3041
return 'line-delta'
1257
def get_options(self, version_id):
1258
return self._cache[version_id][1]
1260
def get_parents(self, version_id):
1261
"""Return parents of specified version ignoring ghosts."""
1262
return [parent for parent in self._cache[version_id][4]
1263
if parent in self._cache]
1265
def get_parents_with_ghosts(self, version_id):
1266
"""Return parents of specified version with ghosts."""
1267
return self._cache[version_id][4]
1269
def check_versions_present(self, version_ids):
1270
"""Check that all specified versions are present."""
1271
version_ids = set(version_ids)
1272
for version_id in list(version_ids):
1273
if version_id in self._cache:
1274
version_ids.remove(version_id)
1276
raise RevisionNotPresent(list(version_ids)[0], self.filename)
1279
class _KnitData(_KnitComponentFile):
1280
"""Contents of the knit data file"""
1282
HEADER = "# bzr knit data 8\n"
1284
def __init__(self, transport, filename, mode, create=False, file_mode=None):
1285
_KnitComponentFile.__init__(self, transport, filename, mode)
1286
self._checked = False
1288
self._transport.put(self._filename, StringIO(''), mode=file_mode)
1290
def clear_cache(self):
1291
"""Clear the record cache."""
3045
def _get_node(self, key):
3047
return list(self._get_entries([key]))[0]
3049
raise RevisionNotPresent(key, self)
3051
def get_options(self, key):
3052
"""Return a list representing options.
3056
node = self._get_node(key)
3057
options = [self._get_method(node)]
3058
if node[2][0] == 'N':
3059
options.append('no-eol')
3062
def find_ancestry(self, keys):
3063
"""See CombinedGraphIndex.find_ancestry()"""
3064
return self._graph_index.find_ancestry(keys, 0)
3066
def get_parent_map(self, keys):
3067
"""Get a map of the parents of keys.
3069
:param keys: The keys to look up parents for.
3070
:return: A mapping from keys to parents. Absent keys are absent from
3074
nodes = self._get_entries(keys)
3078
result[node[1]] = node[3][0]
3081
result[node[1]] = None
3084
def get_position(self, key):
3085
"""Return details needed to access the version.
3087
:return: a tuple (index, data position, size) to hand to the access
3088
logic to get the record.
3090
node = self._get_node(key)
3091
return self._node_to_position(node)
3093
has_key = _mod_index._has_key_from_parent_map
3096
"""Get all the keys in the collection.
3098
The keys are not ordered.
3101
return [node[1] for node in self._graph_index.iter_all_entries()]
3103
missing_keys = _mod_index._missing_keys_from_parent_map
3105
def _node_to_position(self, node):
3106
"""Convert an index value to position details."""
3107
bits = node[2][1:].split(' ')
3108
return node[0], int(bits[0]), int(bits[1])
3110
def _sort_keys_by_io(self, keys, positions):
3111
"""Figure out an optimal order to read the records for the given keys.
3113
Sort keys, grouped by index and sorted by position.
3115
:param keys: A list of keys whose records we want to read. This will be
3117
:param positions: A dict, such as the one returned by
3118
_get_components_positions()
3121
def get_index_memo(key):
3122
# index_memo is at offset [1]. It is made up of (GraphIndex,
3123
# position, size). GI is an object, which will be unique for each
3124
# pack file. This causes us to group by pack file, then sort by
3125
# position. Size doesn't matter, but it isn't worth breaking up the
3127
return positions[key][1]
3128
return keys.sort(key=get_index_memo)
3130
_get_total_build_size = _get_total_build_size
3133
class _KnitKeyAccess(object):
3134
"""Access to records in .knit files."""
3136
def __init__(self, transport, mapper):
3137
"""Create a _KnitKeyAccess with transport and mapper.
3139
:param transport: The transport the access object is rooted at.
3140
:param mapper: The mapper used to map keys to .knit files.
3142
self._transport = transport
3143
self._mapper = mapper
3145
def add_raw_records(self, key_sizes, raw_data):
3146
"""Add raw knit bytes to a storage area.
3148
The data is spooled to the container writer in one bytes-record per
3151
:param sizes: An iterable of tuples containing the key and size of each
3153
:param raw_data: A bytestring containing the data.
3154
:return: A list of memos to retrieve the record later. Each memo is an
3155
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
3156
length), where the key is the record key.
3158
if type(raw_data) is not str:
3159
raise AssertionError(
3160
'data must be plain bytes was %s' % type(raw_data))
3163
# TODO: This can be tuned for writing to sftp and other servers where
3164
# append() is relatively expensive by grouping the writes to each key
3166
for key, size in key_sizes:
3167
path = self._mapper.map(key)
3169
base = self._transport.append_bytes(path + '.knit',
3170
raw_data[offset:offset+size])
3171
except errors.NoSuchFile:
3172
self._transport.mkdir(osutils.dirname(path))
3173
base = self._transport.append_bytes(path + '.knit',
3174
raw_data[offset:offset+size])
3178
result.append((key, base, size))
3182
"""Flush pending writes on this access object.
3184
For .knit files this is a no-op.
1294
def _open_file(self):
1296
return self._transport.get(self._filename)
1301
def _record_to_data(self, version_id, digest, lines):
1302
"""Convert version_id, digest, lines into a raw data block.
1304
:return: (len, a StringIO instance with the raw data ready to read.)
1307
data_file = GzipFile(None, mode='wb', fileobj=sio)
1308
data_file.writelines(chain(
1309
["version %s %d %s\n" % (version_id.encode('utf-8'),
1313
["end %s\n" % version_id.encode('utf-8')]))
1320
def add_raw_record(self, raw_data):
1321
"""Append a prepared record to the data file.
1323
:return: the offset in the data file raw_data was written.
1325
assert isinstance(raw_data, str), 'data must be plain bytes'
1326
return self._transport.append(self._filename, StringIO(raw_data))
1328
def add_record(self, version_id, digest, lines):
1329
"""Write new text record to disk. Returns the position in the
1330
file where it was written."""
1331
size, sio = self._record_to_data(version_id, digest, lines)
1333
start_pos = self._transport.append(self._filename, sio)
1334
return start_pos, size
1336
def _parse_record_header(self, version_id, raw_data):
1337
"""Parse a record header for consistency.
1339
:return: the header and the decompressor stream.
1340
as (stream, header_record)
1342
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1343
rec = df.readline().split()
1345
raise KnitCorrupt(self._filename, 'unexpected number of elements in record header')
1346
if rec[1].decode('utf-8')!= version_id:
1347
raise KnitCorrupt(self._filename,
1348
'unexpected version, wanted %r, got %r' % (
1349
version_id, rec[1]))
1352
def _parse_record(self, version_id, data):
1354
# 4168 calls in 2880 217 internal
1355
# 4168 calls to _parse_record_header in 2121
1356
# 4168 calls to readlines in 330
1357
df, rec = self._parse_record_header(version_id, data)
1358
record_contents = df.readlines()
1359
l = record_contents.pop()
1360
assert len(record_contents) == int(rec[2])
1361
if l.decode('utf-8') != 'end %s\n' % version_id:
1362
raise KnitCorrupt(self._filename, 'unexpected version end line %r, wanted %r'
1365
return record_contents, rec[3]
1367
def read_records_iter_raw(self, records):
1368
"""Read text records from data file and yield raw data.
1370
This unpacks enough of the text record to validate the id is
1371
as expected but thats all.
1373
It will actively recompress currently cached records on the
1374
basis that that is cheaper than I/O activity.
1376
# setup an iterator of the external records:
1377
# uses readv so nice and fast we hope.
1379
# grab the disk data needed.
1380
raw_records = self._transport.readv(self._filename,
1381
[(pos, size) for version_id, pos, size in records])
1383
for version_id, pos, size in records:
1384
pos, data = raw_records.next()
1385
# validate the header
1386
df, rec = self._parse_record_header(version_id, data)
1388
yield version_id, data
1390
def read_records_iter(self, records):
1391
"""Read text records from data file and yield result.
1393
Each passed record is a tuple of (version_id, pos, len) and
1394
will be read in the given order. Yields (version_id,
1397
if len(records) == 0:
3188
def get_raw_records(self, memos_for_retrieval):
3189
"""Get the raw bytes for a records.
3191
:param memos_for_retrieval: An iterable containing the access memo for
3192
retrieving the bytes.
3193
:return: An iterator over the bytes of the records.
3195
# first pass, group into same-index request to minimise readv's issued.
3197
current_prefix = None
3198
for (key, offset, length) in memos_for_retrieval:
3199
if current_prefix == key[:-1]:
3200
current_list.append((offset, length))
3202
if current_prefix is not None:
3203
request_lists.append((current_prefix, current_list))
3204
current_prefix = key[:-1]
3205
current_list = [(offset, length)]
3206
# handle the last entry
3207
if current_prefix is not None:
3208
request_lists.append((current_prefix, current_list))
3209
for prefix, read_vector in request_lists:
3210
path = self._mapper.map(prefix) + '.knit'
3211
for pos, data in self._transport.readv(path, read_vector):
3215
def annotate_knit(knit, revision_id):
3216
"""Annotate a knit with no cached annotations.
3218
This implementation is for knits with no cached annotations.
3219
It will work for knits with cached annotations, but this is not
3222
annotator = _KnitAnnotator(knit)
3223
return iter(annotator.annotate_flat(revision_id))
3226
class _KnitAnnotator(annotate.Annotator):
3227
"""Build up the annotations for a text."""
3229
def __init__(self, vf):
3230
annotate.Annotator.__init__(self, vf)
3232
# TODO: handle Nodes which cannot be extracted
3233
# self._ghosts = set()
3235
# Map from (key, parent_key) => matching_blocks, should be 'use once'
3236
self._matching_blocks = {}
3238
# KnitContent objects
3239
self._content_objects = {}
3240
# The number of children that depend on this fulltext content object
3241
self._num_compression_children = {}
3242
# Delta records that need their compression parent before they can be
3244
self._pending_deltas = {}
3245
# Fulltext records that are waiting for their parents fulltexts before
3246
# they can be yielded for annotation
3247
self._pending_annotation = {}
3249
self._all_build_details = {}
3251
def _get_build_graph(self, key):
3252
"""Get the graphs for building texts and annotations.
3254
The data you need for creating a full text may be different than the
3255
data you need to annotate that text. (At a minimum, you need both
3256
parents to create an annotation, but only need 1 parent to generate the
3259
:return: A list of (key, index_memo) records, suitable for
3260
passing to read_records_iter to start reading in the raw data from
3263
pending = set([key])
3266
self._num_needed_children[key] = 1
3268
# get all pending nodes
3269
this_iteration = pending
3270
build_details = self._vf._index.get_build_details(this_iteration)
3271
self._all_build_details.update(build_details)
3272
# new_nodes = self._vf._index._get_entries(this_iteration)
3274
for key, details in build_details.iteritems():
3275
(index_memo, compression_parent, parent_keys,
3276
record_details) = details
3277
self._parent_map[key] = parent_keys
3278
self._heads_provider = None
3279
records.append((key, index_memo))
3280
# Do we actually need to check _annotated_lines?
3281
pending.update([p for p in parent_keys
3282
if p not in self._all_build_details])
3284
for parent_key in parent_keys:
3285
if parent_key in self._num_needed_children:
3286
self._num_needed_children[parent_key] += 1
3288
self._num_needed_children[parent_key] = 1
3289
if compression_parent:
3290
if compression_parent in self._num_compression_children:
3291
self._num_compression_children[compression_parent] += 1
3293
self._num_compression_children[compression_parent] = 1
3295
missing_versions = this_iteration.difference(build_details.keys())
3296
if missing_versions:
3297
for key in missing_versions:
3298
if key in self._parent_map and key in self._text_cache:
3299
# We already have this text ready, we just need to
3300
# yield it later so we get it annotated
3302
parent_keys = self._parent_map[key]
3303
for parent_key in parent_keys:
3304
if parent_key in self._num_needed_children:
3305
self._num_needed_children[parent_key] += 1
3307
self._num_needed_children[parent_key] = 1
3308
pending.update([p for p in parent_keys
3309
if p not in self._all_build_details])
3311
raise errors.RevisionNotPresent(key, self._vf)
3312
# Generally we will want to read the records in reverse order, because
3313
# we find the parent nodes after the children
3315
return records, ann_keys
3317
def _get_needed_texts(self, key, pb=None):
3318
# if True or len(self._vf._immediate_fallback_vfs) > 0:
3319
if len(self._vf._immediate_fallback_vfs) > 0:
3320
# If we have fallbacks, go to the generic path
3321
for v in annotate.Annotator._get_needed_texts(self, key, pb=pb):
1400
# 60890 calls for 4168 extractions in 5045, 683 internal.
1401
# 4168 calls to readv in 1411
1402
# 4168 calls to parse_record in 2880
1404
# Get unique records, sorted by position
1405
needed_records = sorted(set(records), key=operator.itemgetter(1))
1407
# We take it that the transport optimizes the fetching as good
1408
# as possible (ie, reads continuous ranges.)
1409
response = self._transport.readv(self._filename,
1410
[(pos, size) for version_id, pos, size in needed_records])
1413
for (record_id, pos, size), (pos, data) in \
1414
izip(iter(needed_records), response):
1415
content, digest = self._parse_record(record_id, data)
1416
record_map[record_id] = (digest, content)
1418
for version_id, pos, size in records:
1419
digest, content = record_map[version_id]
1420
yield version_id, content, digest
1422
def read_records(self, records):
1423
"""Read records into a dictionary."""
1425
for record_id, content, digest in self.read_records_iter(records):
1426
components[record_id] = (content, digest)
1430
class InterKnit(InterVersionedFile):
1431
"""Optimised code paths for knit to knit operations."""
1433
_matching_file_from_factory = KnitVersionedFile
1434
_matching_file_to_factory = KnitVersionedFile
1437
def is_compatible(source, target):
1438
"""Be compatible with knits. """
1440
return (isinstance(source, KnitVersionedFile) and
1441
isinstance(target, KnitVersionedFile))
1442
except AttributeError:
1445
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1446
"""See InterVersionedFile.join."""
1447
assert isinstance(self.source, KnitVersionedFile)
1448
assert isinstance(self.target, KnitVersionedFile)
1450
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1455
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1457
version_ids = list(version_ids)
1458
if None in version_ids:
1459
version_ids.remove(None)
1461
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1462
this_versions = set(self.target._index.get_versions())
1463
needed_versions = self.source_ancestry - this_versions
1464
cross_check_versions = self.source_ancestry.intersection(this_versions)
1465
mismatched_versions = set()
1466
for version in cross_check_versions:
1467
# scan to include needed parents.
1468
n1 = set(self.target.get_parents_with_ghosts(version))
1469
n2 = set(self.source.get_parents_with_ghosts(version))
1471
# FIXME TEST this check for cycles being introduced works
1472
# the logic is we have a cycle if in our graph we are an
1473
# ancestor of any of the n2 revisions.
1479
parent_ancestors = self.source.get_ancestry(parent)
1480
if version in parent_ancestors:
1481
raise errors.GraphCycleError([parent, version])
1482
# ensure this parent will be available later.
1483
new_parents = n2.difference(n1)
1484
needed_versions.update(new_parents.difference(this_versions))
1485
mismatched_versions.add(version)
1487
if not needed_versions and not mismatched_versions:
1489
full_list = topo_sort(self.source.get_graph())
1491
version_list = [i for i in full_list if (not self.target.has_version(i)
1492
and i in needed_versions)]
1496
copy_queue_records = []
1498
for version_id in version_list:
1499
options = self.source._index.get_options(version_id)
1500
parents = self.source._index.get_parents_with_ghosts(version_id)
1501
# check that its will be a consistent copy:
1502
for parent in parents:
1503
# if source has the parent, we must :
1504
# * already have it or
1505
# * have it scheduled already
1506
# otherwise we don't care
1507
assert (self.target.has_version(parent) or
1508
parent in copy_set or
1509
not self.source.has_version(parent))
1510
data_pos, data_size = self.source._index.get_position(version_id)
1511
copy_queue_records.append((version_id, data_pos, data_size))
1512
copy_queue.append((version_id, options, parents))
1513
copy_set.add(version_id)
1515
# data suck the join:
1517
total = len(version_list)
1520
for (version_id, raw_data), \
1521
(version_id2, options, parents) in \
1522
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1524
assert version_id == version_id2, 'logic error, inconsistent results'
1526
pb.update("Joining knit", count, total)
1527
raw_records.append((version_id, options, parents, len(raw_data)))
1528
raw_datum.append(raw_data)
1529
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1531
for version in mismatched_versions:
1532
# FIXME RBC 20060309 is this needed?
1533
n1 = set(self.target.get_parents_with_ghosts(version))
1534
n2 = set(self.source.get_parents_with_ghosts(version))
1535
# write a combined record to our history preserving the current
1536
# parents as first in the list
1537
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1538
self.target.fix_parents(version, new_parents)
1544
InterVersionedFile.register_optimiser(InterKnit)
1547
class WeaveToKnit(InterVersionedFile):
1548
"""Optimised code paths for weave to knit operations."""
1550
_matching_file_from_factory = bzrlib.weave.WeaveFile
1551
_matching_file_to_factory = KnitVersionedFile
1554
def is_compatible(source, target):
1555
"""Be compatible with weaves to knits."""
1557
return (isinstance(source, bzrlib.weave.Weave) and
1558
isinstance(target, KnitVersionedFile))
1559
except AttributeError:
1562
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1563
"""See InterVersionedFile.join."""
1564
assert isinstance(self.source, bzrlib.weave.Weave)
1565
assert isinstance(self.target, KnitVersionedFile)
1567
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1572
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1574
version_ids = list(version_ids)
1576
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1577
this_versions = set(self.target._index.get_versions())
1578
needed_versions = self.source_ancestry - this_versions
1579
cross_check_versions = self.source_ancestry.intersection(this_versions)
1580
mismatched_versions = set()
1581
for version in cross_check_versions:
1582
# scan to include needed parents.
1583
n1 = set(self.target.get_parents_with_ghosts(version))
1584
n2 = set(self.source.get_parents(version))
1585
# if all of n2's parents are in n1, then its fine.
1586
if n2.difference(n1):
1587
# FIXME TEST this check for cycles being introduced works
1588
# the logic is we have a cycle if in our graph we are an
1589
# ancestor of any of the n2 revisions.
1595
parent_ancestors = self.source.get_ancestry(parent)
1596
if version in parent_ancestors:
1597
raise errors.GraphCycleError([parent, version])
1598
# ensure this parent will be available later.
1599
new_parents = n2.difference(n1)
1600
needed_versions.update(new_parents.difference(this_versions))
1601
mismatched_versions.add(version)
1603
if not needed_versions and not mismatched_versions:
1605
full_list = topo_sort(self.source.get_graph())
1607
version_list = [i for i in full_list if (not self.target.has_version(i)
1608
and i in needed_versions)]
1612
total = len(version_list)
1613
for version_id in version_list:
1614
pb.update("Converting to knit", count, total)
1615
parents = self.source.get_parents(version_id)
1616
# check that its will be a consistent copy:
1617
for parent in parents:
1618
# if source has the parent, we must already have it
1619
assert (self.target.has_version(parent))
1620
self.target.add_lines(
1621
version_id, parents, self.source.get_lines(version_id))
1624
for version in mismatched_versions:
1625
# FIXME RBC 20060309 is this needed?
1626
n1 = set(self.target.get_parents_with_ghosts(version))
1627
n2 = set(self.source.get_parents(version))
1628
# write a combined record to our history preserving the current
1629
# parents as first in the list
1630
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1631
self.target.fix_parents(version, new_parents)
1637
InterVersionedFile.register_optimiser(WeaveToKnit)
1640
class KnitSequenceMatcher(difflib.SequenceMatcher):
1641
"""Knit tuned sequence matcher.
1643
This is based on profiling of difflib which indicated some improvements
1644
for our usage pattern.
1647
def find_longest_match(self, alo, ahi, blo, bhi):
1648
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1650
If isjunk is not defined:
1652
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1653
alo <= i <= i+k <= ahi
1654
blo <= j <= j+k <= bhi
1655
and for all (i',j',k') meeting those conditions,
1658
and if i == i', j <= j'
1660
In other words, of all maximal matching blocks, return one that
1661
starts earliest in a, and of all those maximal matching blocks that
1662
start earliest in a, return the one that starts earliest in b.
1664
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1665
>>> s.find_longest_match(0, 5, 0, 9)
1668
If isjunk is defined, first the longest matching block is
1669
determined as above, but with the additional restriction that no
1670
junk element appears in the block. Then that block is extended as
1671
far as possible by matching (only) junk elements on both sides. So
1672
the resulting block never matches on junk except as identical junk
1673
happens to be adjacent to an "interesting" match.
1675
Here's the same example as before, but considering blanks to be
1676
junk. That prevents " abcd" from matching the " abcd" at the tail
1677
end of the second sequence directly. Instead only the "abcd" can
1678
match, and matches the leftmost "abcd" in the second sequence:
1680
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1681
>>> s.find_longest_match(0, 5, 0, 9)
1684
If no blocks match, return (alo, blo, 0).
1686
>>> s = SequenceMatcher(None, "ab", "c")
1687
>>> s.find_longest_match(0, 2, 0, 1)
1691
# CAUTION: stripping common prefix or suffix would be incorrect.
1695
# Longest matching block is "ab", but if common prefix is
1696
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1697
# strip, so ends up claiming that ab is changed to acab by
1698
# inserting "ca" in the middle. That's minimal but unintuitive:
1699
# "it's obvious" that someone inserted "ac" at the front.
1700
# Windiff ends up at the same place as diff, but by pairing up
1701
# the unique 'b's and then matching the first two 'a's.
1703
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1704
besti, bestj, bestsize = alo, blo, 0
1705
# find longest junk-free match
1706
# during an iteration of the loop, j2len[j] = length of longest
1707
# junk-free match ending with a[i-1] and b[j]
1711
for i in xrange(alo, ahi):
1712
# look at all instances of a[i] in b; note that because
1713
# b2j has no junk keys, the loop is skipped if a[i] is junk
1714
j2lenget = j2len.get
1717
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1718
# following improvement
1719
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1720
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1721
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1723
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1724
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1725
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
3326
records, ann_keys = self._get_build_graph(key)
3327
for idx, (sub_key, text, num_lines) in enumerate(
3328
self._extract_texts(records)):
3330
pb.update(gettext('annotating'), idx, len(records))
3331
yield sub_key, text, num_lines
3332
for sub_key in ann_keys:
3333
text = self._text_cache[sub_key]
3334
num_lines = len(text) # bad assumption
3335
yield sub_key, text, num_lines
3337
except errors.RetryWithNewPacks, e:
3338
self._vf._access.reload_or_raise(e)
3339
# The cached build_details are no longer valid
3340
self._all_build_details.clear()
3342
def _cache_delta_blocks(self, key, compression_parent, delta, lines):
3343
parent_lines = self._text_cache[compression_parent]
3344
blocks = list(KnitContent.get_line_delta_blocks(delta, parent_lines, lines))
3345
self._matching_blocks[(key, compression_parent)] = blocks
3347
def _expand_record(self, key, parent_keys, compression_parent, record,
3350
if compression_parent:
3351
if compression_parent not in self._content_objects:
3352
# Waiting for the parent
3353
self._pending_deltas.setdefault(compression_parent, []).append(
3354
(key, parent_keys, record, record_details))
3356
# We have the basis parent, so expand the delta
3357
num = self._num_compression_children[compression_parent]
3360
base_content = self._content_objects.pop(compression_parent)
3361
self._num_compression_children.pop(compression_parent)
1737
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1739
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1742
# Extend the best by non-junk elements on each end. In particular,
1743
# "popular" non-junk elements aren't in b2j, which greatly speeds
1744
# the inner loop above, but also means "the best" match so far
1745
# doesn't contain any junk *or* popular non-junk elements.
1746
while besti > alo and bestj > blo and \
1747
not isbjunk(b[bestj-1]) and \
1748
a[besti-1] == b[bestj-1]:
1749
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1750
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1751
not isbjunk(b[bestj+bestsize]) and \
1752
a[besti+bestsize] == b[bestj+bestsize]:
1755
# Now that we have a wholly interesting match (albeit possibly
1756
# empty!), we may as well suck up the matching junk on each
1757
# side of it too. Can't think of a good reason not to, and it
1758
# saves post-processing the (possibly considerable) expense of
1759
# figuring out what to do with it. In the case of an empty
1760
# interesting match, this is clearly the right thing to do,
1761
# because no other kind of match is possible in the regions.
1762
while besti > alo and bestj > blo and \
1763
isbjunk(b[bestj-1]) and \
1764
a[besti-1] == b[bestj-1]:
1765
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1766
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1767
isbjunk(b[bestj+bestsize]) and \
1768
a[besti+bestsize] == b[bestj+bestsize]:
1769
bestsize = bestsize + 1
1771
return besti, bestj, bestsize
3363
self._num_compression_children[compression_parent] = num
3364
base_content = self._content_objects[compression_parent]
3365
# It is tempting to want to copy_base_content=False for the last
3366
# child object. However, whenever noeol=False,
3367
# self._text_cache[parent_key] is content._lines. So mutating it
3368
# gives very bad results.
3369
# The alternative is to copy the lines into text cache, but then we
3370
# are copying anyway, so just do it here.
3371
content, delta = self._vf._factory.parse_record(
3372
key, record, record_details, base_content,
3373
copy_base_content=True)
3376
content, _ = self._vf._factory.parse_record(
3377
key, record, record_details, None)
3378
if self._num_compression_children.get(key, 0) > 0:
3379
self._content_objects[key] = content
3380
lines = content.text()
3381
self._text_cache[key] = lines
3382
if delta is not None:
3383
self._cache_delta_blocks(key, compression_parent, delta, lines)
3386
def _get_parent_annotations_and_matches(self, key, text, parent_key):
3387
"""Get the list of annotations for the parent, and the matching lines.
3389
:param text: The opaque value given by _get_needed_texts
3390
:param parent_key: The key for the parent text
3391
:return: (parent_annotations, matching_blocks)
3392
parent_annotations is a list as long as the number of lines in
3394
matching_blocks is a list of (parent_idx, text_idx, len) tuples
3395
indicating which lines match between the two texts
3397
block_key = (key, parent_key)
3398
if block_key in self._matching_blocks:
3399
blocks = self._matching_blocks.pop(block_key)
3400
parent_annotations = self._annotations_cache[parent_key]
3401
return parent_annotations, blocks
3402
return annotate.Annotator._get_parent_annotations_and_matches(self,
3403
key, text, parent_key)
3405
def _process_pending(self, key):
3406
"""The content for 'key' was just processed.
3408
Determine if there is any more pending work to be processed.
3411
if key in self._pending_deltas:
3412
compression_parent = key
3413
children = self._pending_deltas.pop(key)
3414
for child_key, parent_keys, record, record_details in children:
3415
lines = self._expand_record(child_key, parent_keys,
3417
record, record_details)
3418
if self._check_ready_for_annotations(child_key, parent_keys):
3419
to_return.append(child_key)
3420
# Also check any children that are waiting for this parent to be
3422
if key in self._pending_annotation:
3423
children = self._pending_annotation.pop(key)
3424
to_return.extend([c for c, p_keys in children
3425
if self._check_ready_for_annotations(c, p_keys)])
3428
def _check_ready_for_annotations(self, key, parent_keys):
3429
"""return true if this text is ready to be yielded.
3431
Otherwise, this will return False, and queue the text into
3432
self._pending_annotation
3434
for parent_key in parent_keys:
3435
if parent_key not in self._annotations_cache:
3436
# still waiting on at least one parent text, so queue it up
3437
# Note that if there are multiple parents, we need to wait
3439
self._pending_annotation.setdefault(parent_key,
3440
[]).append((key, parent_keys))
3444
def _extract_texts(self, records):
3445
"""Extract the various texts needed based on records"""
3446
# We iterate in the order read, rather than a strict order requested
3447
# However, process what we can, and put off to the side things that
3448
# still need parents, cleaning them up when those parents are
3451
# 1) As 'records' are read, see if we can expand these records into
3452
# Content objects (and thus lines)
3453
# 2) If a given line-delta is waiting on its compression parent, it
3454
# gets queued up into self._pending_deltas, otherwise we expand
3455
# it, and put it into self._text_cache and self._content_objects
3456
# 3) If we expanded the text, we will then check to see if all
3457
# parents have also been processed. If so, this text gets yielded,
3458
# else this record gets set aside into pending_annotation
3459
# 4) Further, if we expanded the text in (2), we will then check to
3460
# see if there are any children in self._pending_deltas waiting to
3461
# also be processed. If so, we go back to (2) for those
3462
# 5) Further again, if we yielded the text, we can then check if that
3463
# 'unlocks' any of the texts in pending_annotations, which should
3464
# then get yielded as well
3465
# Note that both steps 4 and 5 are 'recursive' in that unlocking one
3466
# compression child could unlock yet another, and yielding a fulltext
3467
# will also 'unlock' the children that are waiting on that annotation.
3468
# (Though also, unlocking 1 parent's fulltext, does not unlock a child
3469
# if other parents are also waiting.)
3470
# We want to yield content before expanding child content objects, so
3471
# that we know when we can re-use the content lines, and the annotation
3472
# code can know when it can stop caching fulltexts, as well.
3474
# Children that are missing their compression parent
3476
for (key, record, digest) in self._vf._read_records_iter(records):
3478
details = self._all_build_details[key]
3479
(_, compression_parent, parent_keys, record_details) = details
3480
lines = self._expand_record(key, parent_keys, compression_parent,
3481
record, record_details)
3483
# Pending delta should be queued up
3485
# At this point, we may be able to yield this content, if all
3486
# parents are also finished
3487
yield_this_text = self._check_ready_for_annotations(key,
3490
# All parents present
3491
yield key, lines, len(lines)
3492
to_process = self._process_pending(key)
3494
this_process = to_process
3496
for key in this_process:
3497
lines = self._text_cache[key]
3498
yield key, lines, len(lines)
3499
to_process.extend(self._process_pending(key))
3502
from bzrlib._knit_load_data_pyx import _load_data_c as _load_data
3503
except ImportError, e:
3504
osutils.failed_to_load_extension(e)
3505
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
