306
751
for start, end, c, lines in delta:
307
752
out.append('%d,%d,%d\n' % (start, end, c))
308
out.extend([text for origin, text in lines])
312
def make_empty_knit(transport, relpath):
313
"""Construct a empty knit at the specified location."""
314
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
318
class KnitVersionedFile(VersionedFile):
319
"""Weave-like structure with faster random access.
321
A knit stores a number of texts and a summary of the relationships
322
between them. Texts are identified by a string version-id. Texts
323
are normally stored and retrieved as a series of lines, but can
324
also be passed as single strings.
326
Lines are stored with the trailing newline (if any) included, to
327
avoid special cases for files with no final newline. Lines are
328
composed of 8-bit characters, not unicode. The combination of
329
these approaches should mean any 'binary' file can be safely
330
stored and retrieved.
333
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
334
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
335
create=False, create_parent_dir=False, delay_create=False,
337
"""Construct a knit at location specified by relpath.
339
:param create: If not True, only open an existing knit.
340
:param create_parent_dir: If True, create the parent directory if
341
creating the file fails. (This is used for stores with
342
hash-prefixes that may not exist yet)
343
:param delay_create: The calling code is aware that the knit won't
344
actually be created until the first data is stored.
756
def annotate(self, knit, key):
757
annotator = _KnitAnnotator(knit)
758
return annotator.annotate_flat(key)
762
def make_file_factory(annotated, mapper):
763
"""Create a factory for creating a file based KnitVersionedFiles.
765
This is only functional enough to run interface tests, it doesn't try to
766
provide a full pack environment.
768
:param annotated: knit annotations are wanted.
769
:param mapper: The mapper from keys to paths.
771
def factory(transport):
772
index = _KndxIndex(transport, mapper, lambda:None, lambda:True, lambda:True)
773
access = _KnitKeyAccess(transport, mapper)
774
return KnitVersionedFiles(index, access, annotated=annotated)
778
def make_pack_factory(graph, delta, keylength):
779
"""Create a factory for creating a pack based VersionedFiles.
781
This is only functional enough to run interface tests, it doesn't try to
782
provide a full pack environment.
784
:param graph: Store a graph.
785
:param delta: Delta compress contents.
786
:param keylength: How long should keys be.
788
def factory(transport):
789
parents = graph or delta
795
max_delta_chain = 200
798
graph_index = _mod_index.InMemoryGraphIndex(reference_lists=ref_length,
799
key_elements=keylength)
800
stream = transport.open_write_stream('newpack')
801
writer = pack.ContainerWriter(stream.write)
803
index = _KnitGraphIndex(graph_index, lambda:True, parents=parents,
804
deltas=delta, add_callback=graph_index.add_nodes)
805
access = pack_repo._DirectPackAccess({})
806
access.set_writer(writer, graph_index, (transport, 'newpack'))
807
result = KnitVersionedFiles(index, access,
808
max_delta_chain=max_delta_chain)
809
result.stream = stream
810
result.writer = writer
815
def cleanup_pack_knit(versioned_files):
816
versioned_files.stream.close()
817
versioned_files.writer.end()
820
def _get_total_build_size(self, keys, positions):
821
"""Determine the total bytes to build these keys.
823
(helper function because _KnitGraphIndex and _KndxIndex work the same, but
824
don't inherit from a common base.)
826
:param keys: Keys that we want to build
827
:param positions: dict of {key, (info, index_memo, comp_parent)} (such
828
as returned by _get_components_positions)
829
:return: Number of bytes to build those keys
831
all_build_index_memos = {}
835
for key in build_keys:
836
# This is mostly for the 'stacked' case
837
# Where we will be getting the data from a fallback
838
if key not in positions:
840
_, index_memo, compression_parent = positions[key]
841
all_build_index_memos[key] = index_memo
842
if compression_parent not in all_build_index_memos:
843
next_keys.add(compression_parent)
844
build_keys = next_keys
845
return sum([index_memo[2] for index_memo
846
in all_build_index_memos.itervalues()])
849
class KnitVersionedFiles(VersionedFilesWithFallbacks):
850
"""Storage for many versioned files using knit compression.
852
Backend storage is managed by indices and data objects.
854
:ivar _index: A _KnitGraphIndex or similar that can describe the
855
parents, graph, compression and data location of entries in this
856
KnitVersionedFiles. Note that this is only the index for
857
*this* vfs; if there are fallbacks they must be queried separately.
860
def __init__(self, index, data_access, max_delta_chain=200,
861
annotated=False, reload_func=None):
862
"""Create a KnitVersionedFiles with index and data_access.
864
:param index: The index for the knit data.
865
:param data_access: The access object to store and retrieve knit
867
:param max_delta_chain: The maximum number of deltas to permit during
868
insertion. Set to 0 to prohibit the use of deltas.
869
:param annotated: Set to True to cause annotations to be calculated and
870
stored during insertion.
871
:param reload_func: An function that can be called if we think we need
872
to reload the pack listing and try again. See
873
'bzrlib.repofmt.pack_repo.AggregateIndex' for the signature.
346
if deprecated_passed(basis_knit):
347
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
348
" deprecated as of bzr 0.9.",
349
DeprecationWarning, stacklevel=2)
350
if access_mode is None:
352
super(KnitVersionedFile, self).__init__(access_mode)
353
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
354
self.transport = transport
355
self.filename = relpath
356
self.factory = factory or KnitAnnotateFactory()
357
self.writable = (access_mode == 'w')
360
self._max_delta_chain = 200
362
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
363
access_mode, create=create, file_mode=file_mode,
364
create_parent_dir=create_parent_dir, delay_create=delay_create,
366
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
367
access_mode, create=create and not len(self), file_mode=file_mode,
368
create_parent_dir=create_parent_dir, delay_create=delay_create,
876
self._access = data_access
877
self._max_delta_chain = max_delta_chain
879
self._factory = KnitAnnotateFactory()
881
self._factory = KnitPlainFactory()
882
self._immediate_fallback_vfs = []
883
self._reload_func = reload_func
371
885
def __repr__(self):
372
return '%s(%s)' % (self.__class__.__name__,
373
self.transport.abspath(self.filename))
375
def _check_should_delta(self, first_parents):
886
return "%s(%r, %r)" % (
887
self.__class__.__name__,
891
def without_fallbacks(self):
892
"""Return a clone of this object without any fallbacks configured."""
893
return KnitVersionedFiles(self._index, self._access,
894
self._max_delta_chain, self._factory.annotated,
897
def add_fallback_versioned_files(self, a_versioned_files):
898
"""Add a source of texts for texts not present in this knit.
900
:param a_versioned_files: A VersionedFiles object.
902
self._immediate_fallback_vfs.append(a_versioned_files)
904
def add_lines(self, key, parents, lines, parent_texts=None,
905
left_matching_blocks=None, nostore_sha=None, random_id=False,
907
"""See VersionedFiles.add_lines()."""
908
self._index._check_write_ok()
909
self._check_add(key, lines, random_id, check_content)
911
# The caller might pass None if there is no graph data, but kndx
912
# indexes can't directly store that, so we give them
913
# an empty tuple instead.
915
line_bytes = ''.join(lines)
916
return self._add(key, lines, parents,
917
parent_texts, left_matching_blocks, nostore_sha, random_id,
918
line_bytes=line_bytes)
920
def _add_text(self, key, parents, text, nostore_sha=None, random_id=False):
921
"""See VersionedFiles._add_text()."""
922
self._index._check_write_ok()
923
self._check_add(key, None, random_id, check_content=False)
924
if text.__class__ is not str:
925
raise errors.BzrBadParameterUnicode("text")
927
# The caller might pass None if there is no graph data, but kndx
928
# indexes can't directly store that, so we give them
929
# an empty tuple instead.
931
return self._add(key, None, parents,
932
None, None, nostore_sha, random_id,
935
def _add(self, key, lines, parents, parent_texts,
936
left_matching_blocks, nostore_sha, random_id,
938
"""Add a set of lines on top of version specified by parents.
940
Any versions not present will be converted into ghosts.
942
:param lines: A list of strings where each one is a single line (has a
943
single newline at the end of the string) This is now optional
944
(callers can pass None). It is left in its location for backwards
945
compatibility. It should ''.join(lines) must == line_bytes
946
:param line_bytes: A single string containing the content
948
We pass both lines and line_bytes because different routes bring the
949
values to this function. And for memory efficiency, we don't want to
950
have to split/join on-demand.
952
# first thing, if the content is something we don't need to store, find
954
digest = sha_string(line_bytes)
955
if nostore_sha == digest:
956
raise errors.ExistingContent
959
if parent_texts is None:
961
# Do a single query to ascertain parent presence; we only compress
962
# against parents in the same kvf.
963
present_parent_map = self._index.get_parent_map(parents)
964
for parent in parents:
965
if parent in present_parent_map:
966
present_parents.append(parent)
968
# Currently we can only compress against the left most present parent.
969
if (len(present_parents) == 0 or
970
present_parents[0] != parents[0]):
973
# To speed the extract of texts the delta chain is limited
974
# to a fixed number of deltas. This should minimize both
975
# I/O and the time spend applying deltas.
976
delta = self._check_should_delta(present_parents[0])
978
text_length = len(line_bytes)
981
# Note: line_bytes is not modified to add a newline, that is tracked
982
# via the no_eol flag. 'lines' *is* modified, because that is the
983
# general values needed by the Content code.
984
if line_bytes and line_bytes[-1] != '\n':
985
options.append('no-eol')
987
# Copy the existing list, or create a new one
989
lines = osutils.split_lines(line_bytes)
992
# Replace the last line with one that ends in a final newline
993
lines[-1] = lines[-1] + '\n'
995
lines = osutils.split_lines(line_bytes)
997
for element in key[:-1]:
998
if type(element) is not str:
999
raise TypeError("key contains non-strings: %r" % (key,))
1001
key = key[:-1] + ('sha1:' + digest,)
1002
elif type(key[-1]) is not str:
1003
raise TypeError("key contains non-strings: %r" % (key,))
1004
# Knit hunks are still last-element only
1005
version_id = key[-1]
1006
content = self._factory.make(lines, version_id)
1008
# Hint to the content object that its text() call should strip the
1010
content._should_strip_eol = True
1011
if delta or (self._factory.annotated and len(present_parents) > 0):
1012
# Merge annotations from parent texts if needed.
1013
delta_hunks = self._merge_annotations(content, present_parents,
1014
parent_texts, delta, self._factory.annotated,
1015
left_matching_blocks)
1018
options.append('line-delta')
1019
store_lines = self._factory.lower_line_delta(delta_hunks)
1020
size, bytes = self._record_to_data(key, digest,
1023
options.append('fulltext')
1024
# isinstance is slower and we have no hierarchy.
1025
if self._factory.__class__ is KnitPlainFactory:
1026
# Use the already joined bytes saving iteration time in
1028
dense_lines = [line_bytes]
1030
dense_lines.append('\n')
1031
size, bytes = self._record_to_data(key, digest,
1034
# get mixed annotation + content and feed it into the
1036
store_lines = self._factory.lower_fulltext(content)
1037
size, bytes = self._record_to_data(key, digest,
1040
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
1041
self._index.add_records(
1042
((key, options, access_memo, parents),),
1043
random_id=random_id)
1044
return digest, text_length, content
1046
def annotate(self, key):
1047
"""See VersionedFiles.annotate."""
1048
return self._factory.annotate(self, key)
1050
def get_annotator(self):
1051
return _KnitAnnotator(self)
1053
def check(self, progress_bar=None, keys=None):
1054
"""See VersionedFiles.check()."""
1056
return self._logical_check()
1058
# At the moment, check does not extra work over get_record_stream
1059
return self.get_record_stream(keys, 'unordered', True)
1061
def _logical_check(self):
1062
# This doesn't actually test extraction of everything, but that will
1063
# impact 'bzr check' substantially, and needs to be integrated with
1064
# care. However, it does check for the obvious problem of a delta with
1066
keys = self._index.keys()
1067
parent_map = self.get_parent_map(keys)
1069
if self._index.get_method(key) != 'fulltext':
1070
compression_parent = parent_map[key][0]
1071
if compression_parent not in parent_map:
1072
raise errors.KnitCorrupt(self,
1073
"Missing basis parent %s for %s" % (
1074
compression_parent, key))
1075
for fallback_vfs in self._immediate_fallback_vfs:
1076
fallback_vfs.check()
1078
def _check_add(self, key, lines, random_id, check_content):
1079
"""check that version_id and lines are safe to add."""
1080
version_id = key[-1]
1081
if version_id is not None:
1082
if contains_whitespace(version_id):
1083
raise InvalidRevisionId(version_id, self)
1084
self.check_not_reserved_id(version_id)
1085
# TODO: If random_id==False and the key is already present, we should
1086
# probably check that the existing content is identical to what is
1087
# being inserted, and otherwise raise an exception. This would make
1088
# the bundle code simpler.
1090
self._check_lines_not_unicode(lines)
1091
self._check_lines_are_lines(lines)
1093
def _check_header(self, key, line):
1094
rec = self._split_header(line)
1095
self._check_header_version(rec, key[-1])
1098
def _check_header_version(self, rec, version_id):
1099
"""Checks the header version on original format knit records.
1101
These have the last component of the key embedded in the record.
1103
if rec[1] != version_id:
1104
raise KnitCorrupt(self,
1105
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
1107
def _check_should_delta(self, parent):
376
1108
"""Iterate back through the parent listing, looking for a fulltext.
378
1110
This is used when we want to decide whether to add a delta or a new
387
1119
fulltext_size = None
388
delta_parents = first_parents
389
1120
for count in xrange(self._max_delta_chain):
390
parent = delta_parents[0]
391
method = self._index.get_method(parent)
392
pos, size = self._index.get_position(parent)
393
if method == 'fulltext':
1122
# Note that this only looks in the index of this particular
1123
# KnitVersionedFiles, not in the fallbacks. This ensures that
1124
# we won't store a delta spanning physical repository
1126
build_details = self._index.get_build_details([parent])
1127
parent_details = build_details[parent]
1128
except (RevisionNotPresent, KeyError), e:
1129
# Some basis is not locally present: always fulltext
1131
index_memo, compression_parent, _, _ = parent_details
1132
_, _, size = index_memo
1133
if compression_parent is None:
394
1134
fulltext_size = size
396
1136
delta_size += size
397
delta_parents = self._index.get_parents(parent)
1137
# We don't explicitly check for presence because this is in an
1138
# inner loop, and if it's missing it'll fail anyhow.
1139
parent = compression_parent
399
1141
# We couldn't find a fulltext, so we must create a new one
1143
# Simple heuristic - if the total I/O wold be greater as a delta than
1144
# the originally installed fulltext, we create a new fulltext.
402
1145
return fulltext_size > delta_size
404
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
405
"""See VersionedFile._add_delta()."""
406
self._check_add(version_id, []) # should we check the lines ?
407
self._check_versions_present(parents)
411
for parent in parents:
412
if not self.has_version(parent):
413
ghosts.append(parent)
415
present_parents.append(parent)
417
if delta_parent is None:
418
# reconstitute as full text.
419
assert len(delta) == 1 or len(delta) == 0
421
assert delta[0][0] == 0
422
assert delta[0][1] == 0, delta[0][1]
423
return super(KnitVersionedFile, self)._add_delta(version_id,
434
options.append('no-eol')
436
if delta_parent is not None:
437
# determine the current delta chain length.
438
# To speed the extract of texts the delta chain is limited
439
# to a fixed number of deltas. This should minimize both
440
# I/O and the time spend applying deltas.
441
# The window was changed to a maximum of 200 deltas, but also added
442
# was a check that the total compressed size of the deltas is
443
# smaller than the compressed size of the fulltext.
444
if not self._check_should_delta([delta_parent]):
445
# We don't want a delta here, just do a normal insertion.
446
return super(KnitVersionedFile, self)._add_delta(version_id,
453
options.append('line-delta')
454
store_lines = self.factory.lower_line_delta(delta)
456
where, size = self._data.add_record(version_id, digest, store_lines)
457
self._index.add_version(version_id, options, where, size, parents)
459
def _add_raw_records(self, records, data):
460
"""Add all the records 'records' with data pre-joined in 'data'.
462
:param records: A list of tuples(version_id, options, parents, size).
463
:param data: The data for the records. When it is written, the records
464
are adjusted to have pos pointing into data by the sum of
465
the preceding records sizes.
468
pos = self._data.add_raw_record(data)
471
for (version_id, options, parents, size) in records:
472
index_entries.append((version_id, options, pos+offset,
474
if self._data._do_cache:
475
self._data._cache[version_id] = data[offset:offset+size]
477
self._index.add_versions(index_entries)
479
def enable_cache(self):
480
"""Start caching data for this knit"""
481
self._data.enable_cache()
483
def clear_cache(self):
484
"""Clear the data cache only."""
485
self._data.clear_cache()
487
def copy_to(self, name, transport):
488
"""See VersionedFile.copy_to()."""
489
# copy the current index to a temp index to avoid racing with local
491
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
492
self.transport.get(self._index._filename))
494
f = self._data._open_file()
496
transport.put_file(name + DATA_SUFFIX, f)
499
# move the copied index into place
500
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
502
def create_empty(self, name, transport, mode=None):
503
return KnitVersionedFile(name, transport, factory=self.factory,
504
delta=self.delta, create=True)
506
def _fix_parents(self, version_id, new_parents):
507
"""Fix the parents list for version.
509
This is done by appending a new version to the index
510
with identical data except for the parents list.
511
the parents list must be a superset of the current
514
current_values = self._index._cache[version_id]
515
assert set(current_values[4]).difference(set(new_parents)) == set()
516
self._index.add_version(version_id,
522
def get_delta(self, version_id):
523
"""Get a delta for constructing version from some other version."""
524
version_id = osutils.safe_revision_id(version_id)
525
self.check_not_reserved_id(version_id)
526
if not self.has_version(version_id):
527
raise RevisionNotPresent(version_id, self.filename)
529
parents = self.get_parents(version_id)
534
data_pos, data_size = self._index.get_position(version_id)
535
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
536
noeol = 'no-eol' in self._index.get_options(version_id)
537
if 'fulltext' == self._index.get_method(version_id):
538
new_content = self.factory.parse_fulltext(data, version_id)
539
if parent is not None:
540
reference_content = self._get_content(parent)
541
old_texts = reference_content.text()
544
new_texts = new_content.text()
545
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
546
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
548
delta = self.factory.parse_line_delta(data, version_id)
549
return parent, sha1, noeol, delta
551
def get_graph_with_ghosts(self):
552
"""See VersionedFile.get_graph_with_ghosts()."""
553
graph_items = self._index.get_graph()
554
return dict(graph_items)
556
def get_sha1(self, version_id):
557
"""See VersionedFile.get_sha1()."""
558
version_id = osutils.safe_revision_id(version_id)
559
record_map = self._get_record_map([version_id])
560
method, content, digest, next = record_map[version_id]
565
"""See VersionedFile.get_suffixes()."""
566
return [DATA_SUFFIX, INDEX_SUFFIX]
568
def has_ghost(self, version_id):
569
"""True if there is a ghost reference in the file to version_id."""
570
version_id = osutils.safe_revision_id(version_id)
572
if self.has_version(version_id):
574
# optimisable if needed by memoising the _ghosts set.
575
items = self._index.get_graph()
576
for node, parents in items:
577
for parent in parents:
578
if parent not in self._index._cache:
579
if parent == version_id:
584
"""See VersionedFile.versions."""
585
return self._index.get_versions()
587
def has_version(self, version_id):
588
"""See VersionedFile.has_version."""
589
version_id = osutils.safe_revision_id(version_id)
590
return self._index.has_version(version_id)
592
__contains__ = has_version
1147
def _build_details_to_components(self, build_details):
1148
"""Convert a build_details tuple to a position tuple."""
1149
# record_details, access_memo, compression_parent
1150
return build_details[3], build_details[0], build_details[1]
1152
def _get_components_positions(self, keys, allow_missing=False):
1153
"""Produce a map of position data for the components of keys.
1155
This data is intended to be used for retrieving the knit records.
1157
A dict of key to (record_details, index_memo, next, parents) is
1160
* method is the way referenced data should be applied.
1161
* index_memo is the handle to pass to the data access to actually get
1163
* next is the build-parent of the version, or None for fulltexts.
1164
* parents is the version_ids of the parents of this version
1166
:param allow_missing: If True do not raise an error on a missing
1167
component, just ignore it.
1170
pending_components = keys
1171
while pending_components:
1172
build_details = self._index.get_build_details(pending_components)
1173
current_components = set(pending_components)
1174
pending_components = set()
1175
for key, details in build_details.iteritems():
1176
(index_memo, compression_parent, parents,
1177
record_details) = details
1178
method = record_details[0]
1179
if compression_parent is not None:
1180
pending_components.add(compression_parent)
1181
component_data[key] = self._build_details_to_components(details)
1182
missing = current_components.difference(build_details)
1183
if missing and not allow_missing:
1184
raise errors.RevisionNotPresent(missing.pop(), self)
1185
return component_data
1187
def _get_content(self, key, parent_texts={}):
1188
"""Returns a content object that makes up the specified
1190
cached_version = parent_texts.get(key, None)
1191
if cached_version is not None:
1192
# Ensure the cache dict is valid.
1193
if not self.get_parent_map([key]):
1194
raise RevisionNotPresent(key, self)
1195
return cached_version
1196
generator = _VFContentMapGenerator(self, [key])
1197
return generator._get_content(key)
1199
def get_parent_map(self, keys):
1200
"""Get a map of the graph parents of keys.
1202
:param keys: The keys to look up parents for.
1203
:return: A mapping from keys to parents. Absent keys are absent from
1206
return self._get_parent_map_with_sources(keys)[0]
1208
def _get_parent_map_with_sources(self, keys):
1209
"""Get a map of the parents of keys.
1211
:param keys: The keys to look up parents for.
1212
:return: A tuple. The first element is a mapping from keys to parents.
1213
Absent keys are absent from the mapping. The second element is a
1214
list with the locations each key was found in. The first element
1215
is the in-this-knit parents, the second the first fallback source,
1219
sources = [self._index] + self._immediate_fallback_vfs
1222
for source in sources:
1225
new_result = source.get_parent_map(missing)
1226
source_results.append(new_result)
1227
result.update(new_result)
1228
missing.difference_update(set(new_result))
1229
return result, source_results
1231
def _get_record_map(self, keys, allow_missing=False):
1232
"""Produce a dictionary of knit records.
1234
:return: {key:(record, record_details, digest, next)}
1236
* record: data returned from read_records (a KnitContentobject)
1237
* record_details: opaque information to pass to parse_record
1238
* digest: SHA1 digest of the full text after all steps are done
1239
* next: build-parent of the version, i.e. the leftmost ancestor.
1240
Will be None if the record is not a delta.
1242
:param keys: The keys to build a map for
1243
:param allow_missing: If some records are missing, rather than
1244
error, just return the data that could be generated.
1246
raw_map = self._get_record_map_unparsed(keys,
1247
allow_missing=allow_missing)
1248
return self._raw_map_to_record_map(raw_map)
1250
def _raw_map_to_record_map(self, raw_map):
1251
"""Parse the contents of _get_record_map_unparsed.
1253
:return: see _get_record_map.
1257
data, record_details, next = raw_map[key]
1258
content, digest = self._parse_record(key[-1], data)
1259
result[key] = content, record_details, digest, next
1262
def _get_record_map_unparsed(self, keys, allow_missing=False):
1263
"""Get the raw data for reconstructing keys without parsing it.
1265
:return: A dict suitable for parsing via _raw_map_to_record_map.
1266
key-> raw_bytes, (method, noeol), compression_parent
1268
# This retries the whole request if anything fails. Potentially we
1269
# could be a bit more selective. We could track the keys whose records
1270
# we have successfully found, and then only request the new records
1271
# from there. However, _get_components_positions grabs the whole build
1272
# chain, which means we'll likely try to grab the same records again
1273
# anyway. Also, can the build chains change as part of a pack
1274
# operation? We wouldn't want to end up with a broken chain.
1277
position_map = self._get_components_positions(keys,
1278
allow_missing=allow_missing)
1279
# key = component_id, r = record_details, i_m = index_memo,
1281
records = [(key, i_m) for key, (r, i_m, n)
1282
in position_map.iteritems()]
1283
# Sort by the index memo, so that we request records from the
1284
# same pack file together, and in forward-sorted order
1285
records.sort(key=operator.itemgetter(1))
1287
for key, data in self._read_records_iter_unchecked(records):
1288
(record_details, index_memo, next) = position_map[key]
1289
raw_record_map[key] = data, record_details, next
1290
return raw_record_map
1291
except errors.RetryWithNewPacks, e:
1292
self._access.reload_or_raise(e)
1295
def _split_by_prefix(cls, keys):
1296
"""For the given keys, split them up based on their prefix.
1298
To keep memory pressure somewhat under control, split the
1299
requests back into per-file-id requests, otherwise "bzr co"
1300
extracts the full tree into memory before writing it to disk.
1301
This should be revisited if _get_content_maps() can ever cross
1304
The keys for a given file_id are kept in the same relative order.
1305
Ordering between file_ids is not, though prefix_order will return the
1306
order that the key was first seen.
1308
:param keys: An iterable of key tuples
1309
:return: (split_map, prefix_order)
1310
split_map A dictionary mapping prefix => keys
1311
prefix_order The order that we saw the various prefixes
1313
split_by_prefix = {}
1321
if prefix in split_by_prefix:
1322
split_by_prefix[prefix].append(key)
1324
split_by_prefix[prefix] = [key]
1325
prefix_order.append(prefix)
1326
return split_by_prefix, prefix_order
1328
def _group_keys_for_io(self, keys, non_local_keys, positions,
1329
_min_buffer_size=_STREAM_MIN_BUFFER_SIZE):
1330
"""For the given keys, group them into 'best-sized' requests.
1332
The idea is to avoid making 1 request per file, but to never try to
1333
unpack an entire 1.5GB source tree in a single pass. Also when
1334
possible, we should try to group requests to the same pack file
1337
:return: list of (keys, non_local) tuples that indicate what keys
1338
should be fetched next.
1340
# TODO: Ideally we would group on 2 factors. We want to extract texts
1341
# from the same pack file together, and we want to extract all
1342
# the texts for a given build-chain together. Ultimately it
1343
# probably needs a better global view.
1344
total_keys = len(keys)
1345
prefix_split_keys, prefix_order = self._split_by_prefix(keys)
1346
prefix_split_non_local_keys, _ = self._split_by_prefix(non_local_keys)
1348
cur_non_local = set()
1352
for prefix in prefix_order:
1353
keys = prefix_split_keys[prefix]
1354
non_local = prefix_split_non_local_keys.get(prefix, [])
1356
this_size = self._index._get_total_build_size(keys, positions)
1357
cur_size += this_size
1358
cur_keys.extend(keys)
1359
cur_non_local.update(non_local)
1360
if cur_size > _min_buffer_size:
1361
result.append((cur_keys, cur_non_local))
1362
sizes.append(cur_size)
1364
cur_non_local = set()
1367
result.append((cur_keys, cur_non_local))
1368
sizes.append(cur_size)
1371
def get_record_stream(self, keys, ordering, include_delta_closure):
1372
"""Get a stream of records for keys.
1374
:param keys: The keys to include.
1375
:param ordering: Either 'unordered' or 'topological'. A topologically
1376
sorted stream has compression parents strictly before their
1378
:param include_delta_closure: If True then the closure across any
1379
compression parents will be included (in the opaque data).
1380
:return: An iterator of ContentFactory objects, each of which is only
1381
valid until the iterator is advanced.
1383
# keys might be a generator
1387
if not self._index.has_graph:
1388
# Cannot sort when no graph has been stored.
1389
ordering = 'unordered'
1391
remaining_keys = keys
1394
keys = set(remaining_keys)
1395
for content_factory in self._get_remaining_record_stream(keys,
1396
ordering, include_delta_closure):
1397
remaining_keys.discard(content_factory.key)
1398
yield content_factory
1400
except errors.RetryWithNewPacks, e:
1401
self._access.reload_or_raise(e)
1403
def _get_remaining_record_stream(self, keys, ordering,
1404
include_delta_closure):
1405
"""This function is the 'retry' portion for get_record_stream."""
1406
if include_delta_closure:
1407
positions = self._get_components_positions(keys, allow_missing=True)
1409
build_details = self._index.get_build_details(keys)
1411
# (record_details, access_memo, compression_parent_key)
1412
positions = dict((key, self._build_details_to_components(details))
1413
for key, details in build_details.iteritems())
1414
absent_keys = keys.difference(set(positions))
1415
# There may be more absent keys : if we're missing the basis component
1416
# and are trying to include the delta closure.
1417
# XXX: We should not ever need to examine remote sources because we do
1418
# not permit deltas across versioned files boundaries.
1419
if include_delta_closure:
1420
needed_from_fallback = set()
1421
# Build up reconstructable_keys dict. key:True in this dict means
1422
# the key can be reconstructed.
1423
reconstructable_keys = {}
1427
chain = [key, positions[key][2]]
1429
needed_from_fallback.add(key)
1432
while chain[-1] is not None:
1433
if chain[-1] in reconstructable_keys:
1434
result = reconstructable_keys[chain[-1]]
1438
chain.append(positions[chain[-1]][2])
1440
# missing basis component
1441
needed_from_fallback.add(chain[-1])
1444
for chain_key in chain[:-1]:
1445
reconstructable_keys[chain_key] = result
1447
needed_from_fallback.add(key)
1448
# Double index lookups here : need a unified api ?
1449
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1450
if ordering in ('topological', 'groupcompress'):
1451
if ordering == 'topological':
1452
# Global topological sort
1453
present_keys = tsort.topo_sort(global_map)
1455
present_keys = sort_groupcompress(global_map)
1456
# Now group by source:
1458
current_source = None
1459
for key in present_keys:
1460
for parent_map in parent_maps:
1461
if key in parent_map:
1462
key_source = parent_map
1464
if current_source is not key_source:
1465
source_keys.append((key_source, []))
1466
current_source = key_source
1467
source_keys[-1][1].append(key)
1469
if ordering != 'unordered':
1470
raise AssertionError('valid values for ordering are:'
1471
' "unordered", "groupcompress" or "topological" not: %r'
1473
# Just group by source; remote sources first.
1476
for parent_map in reversed(parent_maps):
1477
source_keys.append((parent_map, []))
1478
for key in parent_map:
1479
present_keys.append(key)
1480
source_keys[-1][1].append(key)
1481
# We have been requested to return these records in an order that
1482
# suits us. So we ask the index to give us an optimally sorted
1484
for source, sub_keys in source_keys:
1485
if source is parent_maps[0]:
1486
# Only sort the keys for this VF
1487
self._index._sort_keys_by_io(sub_keys, positions)
1488
absent_keys = keys - set(global_map)
1489
for key in absent_keys:
1490
yield AbsentContentFactory(key)
1491
# restrict our view to the keys we can answer.
1492
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1493
# XXX: At that point we need to consider the impact of double reads by
1494
# utilising components multiple times.
1495
if include_delta_closure:
1496
# XXX: get_content_maps performs its own index queries; allow state
1498
non_local_keys = needed_from_fallback - absent_keys
1499
for keys, non_local_keys in self._group_keys_for_io(present_keys,
1502
generator = _VFContentMapGenerator(self, keys, non_local_keys,
1505
for record in generator.get_record_stream():
1508
for source, keys in source_keys:
1509
if source is parent_maps[0]:
1510
# this KnitVersionedFiles
1511
records = [(key, positions[key][1]) for key in keys]
1512
for key, raw_data in self._read_records_iter_unchecked(records):
1513
(record_details, index_memo, _) = positions[key]
1514
yield KnitContentFactory(key, global_map[key],
1515
record_details, None, raw_data, self._factory.annotated, None)
1517
vf = self._immediate_fallback_vfs[parent_maps.index(source) - 1]
1518
for record in vf.get_record_stream(keys, ordering,
1519
include_delta_closure):
1522
def get_sha1s(self, keys):
1523
"""See VersionedFiles.get_sha1s()."""
1525
record_map = self._get_record_map(missing, allow_missing=True)
1527
for key, details in record_map.iteritems():
1528
if key not in missing:
1530
# record entry 2 is the 'digest'.
1531
result[key] = details[2]
1532
missing.difference_update(set(result))
1533
for source in self._immediate_fallback_vfs:
1536
new_result = source.get_sha1s(missing)
1537
result.update(new_result)
1538
missing.difference_update(set(new_result))
1541
def insert_record_stream(self, stream):
1542
"""Insert a record stream into this container.
1544
:param stream: A stream of records to insert.
1546
:seealso VersionedFiles.get_record_stream:
1548
def get_adapter(adapter_key):
1550
return adapters[adapter_key]
1552
adapter_factory = adapter_registry.get(adapter_key)
1553
adapter = adapter_factory(self)
1554
adapters[adapter_key] = adapter
1557
if self._factory.annotated:
1558
# self is annotated, we need annotated knits to use directly.
1559
annotated = "annotated-"
1562
# self is not annotated, but we can strip annotations cheaply.
1564
convertibles = set(["knit-annotated-ft-gz"])
1565
if self._max_delta_chain:
1566
delta_types.add("knit-annotated-delta-gz")
1567
convertibles.add("knit-annotated-delta-gz")
1568
# The set of types we can cheaply adapt without needing basis texts.
1569
native_types = set()
1570
if self._max_delta_chain:
1571
native_types.add("knit-%sdelta-gz" % annotated)
1572
delta_types.add("knit-%sdelta-gz" % annotated)
1573
native_types.add("knit-%sft-gz" % annotated)
1574
knit_types = native_types.union(convertibles)
1576
# Buffer all index entries that we can't add immediately because their
1577
# basis parent is missing. We don't buffer all because generating
1578
# annotations may require access to some of the new records. However we
1579
# can't generate annotations from new deltas until their basis parent
1580
# is present anyway, so we get away with not needing an index that
1581
# includes the new keys.
1583
# See <http://launchpad.net/bugs/300177> about ordering of compression
1584
# parents in the records - to be conservative, we insist that all
1585
# parents must be present to avoid expanding to a fulltext.
1587
# key = basis_parent, value = index entry to add
1588
buffered_index_entries = {}
1589
for record in stream:
1590
kind = record.storage_kind
1591
if kind.startswith('knit-') and kind.endswith('-gz'):
1592
# Check that the ID in the header of the raw knit bytes matches
1593
# the record metadata.
1594
raw_data = record._raw_record
1595
df, rec = self._parse_record_header(record.key, raw_data)
1598
parents = record.parents
1599
if record.storage_kind in delta_types:
1600
# TODO: eventually the record itself should track
1601
# compression_parent
1602
compression_parent = parents[0]
1604
compression_parent = None
1605
# Raise an error when a record is missing.
1606
if record.storage_kind == 'absent':
1607
raise RevisionNotPresent([record.key], self)
1608
elif ((record.storage_kind in knit_types)
1609
and (compression_parent is None
1610
or not self._immediate_fallback_vfs
1611
or self._index.has_key(compression_parent)
1612
or not self.has_key(compression_parent))):
1613
# we can insert the knit record literally if either it has no
1614
# compression parent OR we already have its basis in this kvf
1615
# OR the basis is not present even in the fallbacks. In the
1616
# last case it will either turn up later in the stream and all
1617
# will be well, or it won't turn up at all and we'll raise an
1620
# TODO: self.has_key is somewhat redundant with
1621
# self._index.has_key; we really want something that directly
1622
# asks if it's only present in the fallbacks. -- mbp 20081119
1623
if record.storage_kind not in native_types:
1625
adapter_key = (record.storage_kind, "knit-delta-gz")
1626
adapter = get_adapter(adapter_key)
1628
adapter_key = (record.storage_kind, "knit-ft-gz")
1629
adapter = get_adapter(adapter_key)
1630
bytes = adapter.get_bytes(record)
1632
# It's a knit record, it has a _raw_record field (even if
1633
# it was reconstituted from a network stream).
1634
bytes = record._raw_record
1635
options = [record._build_details[0]]
1636
if record._build_details[1]:
1637
options.append('no-eol')
1638
# Just blat it across.
1639
# Note: This does end up adding data on duplicate keys. As
1640
# modern repositories use atomic insertions this should not
1641
# lead to excessive growth in the event of interrupted fetches.
1642
# 'knit' repositories may suffer excessive growth, but as a
1643
# deprecated format this is tolerable. It can be fixed if
1644
# needed by in the kndx index support raising on a duplicate
1645
# add with identical parents and options.
1646
access_memo = self._access.add_raw_records(
1647
[(record.key, len(bytes))], bytes)[0]
1648
index_entry = (record.key, options, access_memo, parents)
1649
if 'fulltext' not in options:
1650
# Not a fulltext, so we need to make sure the compression
1651
# parent will also be present.
1652
# Note that pack backed knits don't need to buffer here
1653
# because they buffer all writes to the transaction level,
1654
# but we don't expose that difference at the index level. If
1655
# the query here has sufficient cost to show up in
1656
# profiling we should do that.
1658
# They're required to be physically in this
1659
# KnitVersionedFiles, not in a fallback.
1660
if not self._index.has_key(compression_parent):
1661
pending = buffered_index_entries.setdefault(
1662
compression_parent, [])
1663
pending.append(index_entry)
1666
self._index.add_records([index_entry])
1667
elif record.storage_kind == 'chunked':
1668
self.add_lines(record.key, parents,
1669
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1671
# Not suitable for direct insertion as a
1672
# delta, either because it's not the right format, or this
1673
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1674
# 0) or because it depends on a base only present in the
1676
self._access.flush()
1678
# Try getting a fulltext directly from the record.
1679
bytes = record.get_bytes_as('fulltext')
1680
except errors.UnavailableRepresentation:
1681
adapter_key = record.storage_kind, 'fulltext'
1682
adapter = get_adapter(adapter_key)
1683
bytes = adapter.get_bytes(record)
1684
lines = split_lines(bytes)
1686
self.add_lines(record.key, parents, lines)
1687
except errors.RevisionAlreadyPresent:
1689
# Add any records whose basis parent is now available.
1691
added_keys = [record.key]
1693
key = added_keys.pop(0)
1694
if key in buffered_index_entries:
1695
index_entries = buffered_index_entries[key]
1696
self._index.add_records(index_entries)
1698
[index_entry[0] for index_entry in index_entries])
1699
del buffered_index_entries[key]
1700
if buffered_index_entries:
1701
# There were index entries buffered at the end of the stream,
1702
# So these need to be added (if the index supports holding such
1703
# entries for later insertion)
1705
for key in buffered_index_entries:
1706
index_entries = buffered_index_entries[key]
1707
all_entries.extend(index_entries)
1708
self._index.add_records(
1709
all_entries, missing_compression_parents=True)
1711
def get_missing_compression_parent_keys(self):
1712
"""Return an iterable of keys of missing compression parents.
1714
Check this after calling insert_record_stream to find out if there are
1715
any missing compression parents. If there are, the records that
1716
depend on them are not able to be inserted safely. For atomic
1717
KnitVersionedFiles built on packs, the transaction should be aborted or
1718
suspended - commit will fail at this point. Nonatomic knits will error
1719
earlier because they have no staging area to put pending entries into.
1721
return self._index.get_missing_compression_parents()
1723
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1724
"""Iterate over the lines in the versioned files from keys.
1726
This may return lines from other keys. Each item the returned
1727
iterator yields is a tuple of a line and a text version that that line
1728
is present in (not introduced in).
1730
Ordering of results is in whatever order is most suitable for the
1731
underlying storage format.
1733
If a progress bar is supplied, it may be used to indicate progress.
1734
The caller is responsible for cleaning up progress bars (because this
1738
* Lines are normalised by the underlying store: they will all have \\n
1740
* Lines are returned in arbitrary order.
1741
* If a requested key did not change any lines (or didn't have any
1742
lines), it may not be mentioned at all in the result.
1744
:param pb: Progress bar supplied by caller.
1745
:return: An iterator over (line, key).
1748
pb = ui.ui_factory.nested_progress_bar()
1754
# we don't care about inclusions, the caller cares.
1755
# but we need to setup a list of records to visit.
1756
# we need key, position, length
1758
build_details = self._index.get_build_details(keys)
1759
for key, details in build_details.iteritems():
1761
key_records.append((key, details[0]))
1762
records_iter = enumerate(self._read_records_iter(key_records))
1763
for (key_idx, (key, data, sha_value)) in records_iter:
1764
pb.update(gettext('Walking content'), key_idx, total)
1765
compression_parent = build_details[key][1]
1766
if compression_parent is None:
1768
line_iterator = self._factory.get_fulltext_content(data)
1771
line_iterator = self._factory.get_linedelta_content(data)
1772
# Now that we are yielding the data for this key, remove it
1775
# XXX: It might be more efficient to yield (key,
1776
# line_iterator) in the future. However for now, this is a
1777
# simpler change to integrate into the rest of the
1778
# codebase. RBC 20071110
1779
for line in line_iterator:
1782
except errors.RetryWithNewPacks, e:
1783
self._access.reload_or_raise(e)
1784
# If there are still keys we've not yet found, we look in the fallback
1785
# vfs, and hope to find them there. Note that if the keys are found
1786
# but had no changes or no content, the fallback may not return
1788
if keys and not self._immediate_fallback_vfs:
1789
# XXX: strictly the second parameter is meant to be the file id
1790
# but it's not easily accessible here.
1791
raise RevisionNotPresent(keys, repr(self))
1792
for source in self._immediate_fallback_vfs:
1796
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1797
source_keys.add(key)
1799
keys.difference_update(source_keys)
1800
pb.update(gettext('Walking content'), total, total)
1802
def _make_line_delta(self, delta_seq, new_content):
1803
"""Generate a line delta from delta_seq and new_content."""
1805
for op in delta_seq.get_opcodes():
1806
if op[0] == 'equal':
1808
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
594
1811
def _merge_annotations(self, content, parents, parent_texts={},
595
delta=None, annotated=None):
596
"""Merge annotations for content. This is done by comparing
597
the annotations based on changed to the text.
1812
delta=None, annotated=None,
1813
left_matching_blocks=None):
1814
"""Merge annotations for content and generate deltas.
1816
This is done by comparing the annotations based on changes to the text
1817
and generating a delta on the resulting full texts. If annotations are
1818
not being created then a simple delta is created.
1820
if left_matching_blocks is not None:
1821
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
600
1823
delta_seq = None
601
for parent_id in parents:
602
merge_content = self._get_content(parent_id, parent_texts)
603
seq = patiencediff.PatienceSequenceMatcher(
604
None, merge_content.text(), content.text())
605
if delta_seq is None:
606
# setup a delta seq to reuse.
1825
for parent_key in parents:
1826
merge_content = self._get_content(parent_key, parent_texts)
1827
if (parent_key == parents[0] and delta_seq is not None):
1830
seq = patiencediff.PatienceSequenceMatcher(
1831
None, merge_content.text(), content.text())
608
1832
for i, j, n in seq.get_matching_blocks():
611
# this appears to copy (origin, text) pairs across to the new
612
# content for any line that matches the last-checked parent.
613
# FIXME: save the sequence control data for delta compression
614
# against the most relevant parent rather than rediffing.
1835
# this copies (origin, text) pairs across to the new
1836
# content for any line that matches the last-checked
615
1838
content._lines[j:j+n] = merge_content._lines[i:i+n]
1839
# XXX: Robert says the following block is a workaround for a
1840
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1841
if content._lines and content._lines[-1][1][-1] != '\n':
1842
# The copied annotation was from a line without a trailing EOL,
1843
# reinstate one for the content object, to ensure correct
1845
line = content._lines[-1][1] + '\n'
1846
content._lines[-1] = (content._lines[-1][0], line)
1848
if delta_seq is None:
618
1849
reference_content = self._get_content(parents[0], parent_texts)
619
1850
new_texts = content.text()
620
1851
old_texts = reference_content.text()
622
1853
None, old_texts, new_texts)
623
1854
return self._make_line_delta(delta_seq, content)
625
def _make_line_delta(self, delta_seq, new_content):
626
"""Generate a line delta from delta_seq and new_content."""
628
for op in delta_seq.get_opcodes():
631
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
634
def _get_components_positions(self, version_ids):
635
"""Produce a map of position data for the components of versions.
637
This data is intended to be used for retrieving the knit records.
639
A dict of version_id to (method, data_pos, data_size, next) is
641
method is the way referenced data should be applied.
642
data_pos is the position of the data in the knit.
643
data_size is the size of the data in the knit.
644
next is the build-parent of the version, or None for fulltexts.
647
for version_id in version_ids:
650
while cursor is not None and cursor not in component_data:
651
method = self._index.get_method(cursor)
652
if method == 'fulltext':
655
next = self.get_parents(cursor)[0]
656
data_pos, data_size = self._index.get_position(cursor)
657
component_data[cursor] = (method, data_pos, data_size, next)
659
return component_data
661
def _get_content(self, version_id, parent_texts={}):
662
"""Returns a content object that makes up the specified
664
if not self.has_version(version_id):
665
raise RevisionNotPresent(version_id, self.filename)
667
cached_version = parent_texts.get(version_id, None)
668
if cached_version is not None:
669
return cached_version
671
text_map, contents_map = self._get_content_maps([version_id])
672
return contents_map[version_id]
674
def _check_versions_present(self, version_ids):
675
"""Check that all specified versions are present."""
676
self._index.check_versions_present(version_ids)
678
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
679
"""See VersionedFile.add_lines_with_ghosts()."""
680
self._check_add(version_id, lines)
681
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
683
def _add_lines(self, version_id, parents, lines, parent_texts):
684
"""See VersionedFile.add_lines."""
685
self._check_add(version_id, lines)
686
self._check_versions_present(parents)
687
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
689
def _check_add(self, version_id, lines):
690
"""check that version_id and lines are safe to add."""
691
assert self.writable, "knit is not opened for write"
692
### FIXME escape. RBC 20060228
693
if contains_whitespace(version_id):
694
raise InvalidRevisionId(version_id, self.filename)
695
self.check_not_reserved_id(version_id)
696
if self.has_version(version_id):
697
raise RevisionAlreadyPresent(version_id, self.filename)
698
self._check_lines_not_unicode(lines)
699
self._check_lines_are_lines(lines)
701
def _add(self, version_id, lines, parents, delta, parent_texts):
702
"""Add a set of lines on top of version specified by parents.
704
If delta is true, compress the text as a line-delta against
707
Any versions not present will be converted into ghosts.
709
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
710
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
711
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
712
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
713
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
714
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
715
# +1383 0 8.0370 8.0370 +<len>
716
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
717
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
718
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
719
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
723
if parent_texts is None:
725
for parent in parents:
726
if not self.has_version(parent):
727
ghosts.append(parent)
729
present_parents.append(parent)
731
if delta and not len(present_parents):
734
digest = sha_strings(lines)
737
if lines[-1][-1] != '\n':
738
options.append('no-eol')
739
lines[-1] = lines[-1] + '\n'
741
if len(present_parents) and delta:
742
# To speed the extract of texts the delta chain is limited
743
# to a fixed number of deltas. This should minimize both
744
# I/O and the time spend applying deltas.
745
delta = self._check_should_delta(present_parents)
747
assert isinstance(version_id, str)
748
lines = self.factory.make(lines, version_id)
749
if delta or (self.factory.annotated and len(present_parents) > 0):
750
# Merge annotations from parent texts if so is needed.
751
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
752
delta, self.factory.annotated)
755
options.append('line-delta')
756
store_lines = self.factory.lower_line_delta(delta_hunks)
1856
def _parse_record(self, version_id, data):
1857
"""Parse an original format knit record.
1859
These have the last element of the key only present in the stored data.
1861
rec, record_contents = self._parse_record_unchecked(data)
1862
self._check_header_version(rec, version_id)
1863
return record_contents, rec[3]
1865
def _parse_record_header(self, key, raw_data):
1866
"""Parse a record header for consistency.
1868
:return: the header and the decompressor stream.
1869
as (stream, header_record)
1871
df = gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1874
rec = self._check_header(key, df.readline())
1875
except Exception, e:
1876
raise KnitCorrupt(self,
1877
"While reading {%s} got %s(%s)"
1878
% (key, e.__class__.__name__, str(e)))
1881
def _parse_record_unchecked(self, data):
1883
# 4168 calls in 2880 217 internal
1884
# 4168 calls to _parse_record_header in 2121
1885
# 4168 calls to readlines in 330
1886
df = gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1888
record_contents = df.readlines()
1889
except Exception, e:
1890
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1891
(data, e.__class__.__name__, str(e)))
1892
header = record_contents.pop(0)
1893
rec = self._split_header(header)
1894
last_line = record_contents.pop()
1895
if len(record_contents) != int(rec[2]):
1896
raise KnitCorrupt(self,
1897
'incorrect number of lines %s != %s'
1898
' for version {%s} %s'
1899
% (len(record_contents), int(rec[2]),
1900
rec[1], record_contents))
1901
if last_line != 'end %s\n' % rec[1]:
1902
raise KnitCorrupt(self,
1903
'unexpected version end line %r, wanted %r'
1904
% (last_line, rec[1]))
1906
return rec, record_contents
1908
def _read_records_iter(self, records):
1909
"""Read text records from data file and yield result.
1911
The result will be returned in whatever is the fastest to read.
1912
Not by the order requested. Also, multiple requests for the same
1913
record will only yield 1 response.
1915
:param records: A list of (key, access_memo) entries
1916
:return: Yields (key, contents, digest) in the order
1917
read, not the order requested
1922
# XXX: This smells wrong, IO may not be getting ordered right.
1923
needed_records = sorted(set(records), key=operator.itemgetter(1))
1924
if not needed_records:
1927
# The transport optimizes the fetching as well
1928
# (ie, reads continuous ranges.)
1929
raw_data = self._access.get_raw_records(
1930
[index_memo for key, index_memo in needed_records])
1932
for (key, index_memo), data in \
1933
izip(iter(needed_records), raw_data):
1934
content, digest = self._parse_record(key[-1], data)
1935
yield key, content, digest
1937
def _read_records_iter_raw(self, records):
1938
"""Read text records from data file and yield raw data.
1940
This unpacks enough of the text record to validate the id is
1941
as expected but thats all.
1943
Each item the iterator yields is (key, bytes,
1944
expected_sha1_of_full_text).
1946
for key, data in self._read_records_iter_unchecked(records):
1947
# validate the header (note that we can only use the suffix in
1948
# current knit records).
1949
df, rec = self._parse_record_header(key, data)
1951
yield key, data, rec[3]
1953
def _read_records_iter_unchecked(self, records):
1954
"""Read text records from data file and yield raw data.
1956
No validation is done.
1958
Yields tuples of (key, data).
1960
# setup an iterator of the external records:
1961
# uses readv so nice and fast we hope.
1963
# grab the disk data needed.
1964
needed_offsets = [index_memo for key, index_memo
1966
raw_records = self._access.get_raw_records(needed_offsets)
1968
for key, index_memo in records:
1969
data = raw_records.next()
1972
def _record_to_data(self, key, digest, lines, dense_lines=None):
1973
"""Convert key, digest, lines into a raw data block.
1975
:param key: The key of the record. Currently keys are always serialised
1976
using just the trailing component.
1977
:param dense_lines: The bytes of lines but in a denser form. For
1978
instance, if lines is a list of 1000 bytestrings each ending in
1979
\\n, dense_lines may be a list with one line in it, containing all
1980
the 1000's lines and their \\n's. Using dense_lines if it is
1981
already known is a win because the string join to create bytes in
1982
this function spends less time resizing the final string.
1983
:return: (len, a StringIO instance with the raw data ready to read.)
1985
chunks = ["version %s %d %s\n" % (key[-1], len(lines), digest)]
1986
chunks.extend(dense_lines or lines)
1987
chunks.append("end %s\n" % key[-1])
1988
for chunk in chunks:
1989
if type(chunk) is not str:
1990
raise AssertionError(
1991
'data must be plain bytes was %s' % type(chunk))
1992
if lines and lines[-1][-1] != '\n':
1993
raise ValueError('corrupt lines value %r' % lines)
1994
compressed_bytes = tuned_gzip.chunks_to_gzip(chunks)
1995
return len(compressed_bytes), compressed_bytes
1997
def _split_header(self, line):
2000
raise KnitCorrupt(self,
2001
'unexpected number of elements in record header')
2005
"""See VersionedFiles.keys."""
2006
if 'evil' in debug.debug_flags:
2007
trace.mutter_callsite(2, "keys scales with size of history")
2008
sources = [self._index] + self._immediate_fallback_vfs
2010
for source in sources:
2011
result.update(source.keys())
2015
class _ContentMapGenerator(object):
2016
"""Generate texts or expose raw deltas for a set of texts."""
2018
def __init__(self, ordering='unordered'):
2019
self._ordering = ordering
2021
def _get_content(self, key):
2022
"""Get the content object for key."""
2023
# Note that _get_content is only called when the _ContentMapGenerator
2024
# has been constructed with just one key requested for reconstruction.
2025
if key in self.nonlocal_keys:
2026
record = self.get_record_stream().next()
2027
# Create a content object on the fly
2028
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
2029
return PlainKnitContent(lines, record.key)
758
options.append('fulltext')
759
store_lines = self.factory.lower_fulltext(lines)
761
where, size = self._data.add_record(version_id, digest, store_lines)
762
self._index.add_version(version_id, options, where, size, parents)
765
def check(self, progress_bar=None):
766
"""See VersionedFile.check()."""
768
def _clone_text(self, new_version_id, old_version_id, parents):
769
"""See VersionedFile.clone_text()."""
770
# FIXME RBC 20060228 make fast by only inserting an index with null
772
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
774
def get_lines(self, version_id):
775
"""See VersionedFile.get_lines()."""
776
return self.get_line_list([version_id])[0]
778
def _get_record_map(self, version_ids):
779
"""Produce a dictionary of knit records.
781
The keys are version_ids, the values are tuples of (method, content,
783
method is the way the content should be applied.
784
content is a KnitContent object.
785
digest is the SHA1 digest of this version id after all steps are done
786
next is the build-parent of the version, i.e. the leftmost ancestor.
787
If the method is fulltext, next will be None.
789
position_map = self._get_components_positions(version_ids)
790
# c = component_id, m = method, p = position, s = size, n = next
791
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
793
for component_id, content, digest in \
794
self._data.read_records_iter(records):
795
method, position, size, next = position_map[component_id]
796
record_map[component_id] = method, content, digest, next
800
def get_text(self, version_id):
801
"""See VersionedFile.get_text"""
802
return self.get_texts([version_id])[0]
804
def get_texts(self, version_ids):
805
return [''.join(l) for l in self.get_line_list(version_ids)]
807
def get_line_list(self, version_ids):
808
"""Return the texts of listed versions as a list of strings."""
809
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
810
for version_id in version_ids:
811
self.check_not_reserved_id(version_id)
812
text_map, content_map = self._get_content_maps(version_ids)
813
return [text_map[v] for v in version_ids]
815
def _get_content_maps(self, version_ids):
816
"""Produce maps of text and KnitContents
2031
# local keys we can ask for directly
2032
return self._get_one_work(key)
2034
def get_record_stream(self):
2035
"""Get a record stream for the keys requested during __init__."""
2036
for record in self._work():
2040
"""Produce maps of text and KnitContents as dicts.
818
2042
:return: (text_map, content_map) where text_map contains the texts for
819
the requested versions and content_map contains the KnitContents.
820
Both dicts take version_ids as their keys.
2043
the requested versions and content_map contains the KnitContents.
822
for version_id in version_ids:
823
if not self.has_version(version_id):
824
raise RevisionNotPresent(version_id, self.filename)
825
record_map = self._get_record_map(version_ids)
830
for version_id in version_ids:
2045
# NB: By definition we never need to read remote sources unless texts
2046
# are requested from them: we don't delta across stores - and we
2047
# explicitly do not want to to prevent data loss situations.
2048
if self.global_map is None:
2049
self.global_map = self.vf.get_parent_map(self.keys)
2050
nonlocal_keys = self.nonlocal_keys
2052
missing_keys = set(nonlocal_keys)
2053
# Read from remote versioned file instances and provide to our caller.
2054
for source in self.vf._immediate_fallback_vfs:
2055
if not missing_keys:
2057
# Loop over fallback repositories asking them for texts - ignore
2058
# any missing from a particular fallback.
2059
for record in source.get_record_stream(missing_keys,
2060
self._ordering, True):
2061
if record.storage_kind == 'absent':
2062
# Not in thie particular stream, may be in one of the
2063
# other fallback vfs objects.
2065
missing_keys.remove(record.key)
2068
if self._raw_record_map is None:
2069
raise AssertionError('_raw_record_map should have been filled')
2071
for key in self.keys:
2072
if key in self.nonlocal_keys:
2074
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2077
def _get_one_work(self, requested_key):
2078
# Now, if we have calculated everything already, just return the
2080
if requested_key in self._contents_map:
2081
return self._contents_map[requested_key]
2082
# To simplify things, parse everything at once - code that wants one text
2083
# probably wants them all.
2084
# FUTURE: This function could be improved for the 'extract many' case
2085
# by tracking each component and only doing the copy when the number of
2086
# children than need to apply delta's to it is > 1 or it is part of the
2088
multiple_versions = len(self.keys) != 1
2089
if self._record_map is None:
2090
self._record_map = self.vf._raw_map_to_record_map(
2091
self._raw_record_map)
2092
record_map = self._record_map
2093
# raw_record_map is key:
2094
# Have read and parsed records at this point.
2095
for key in self.keys:
2096
if key in self.nonlocal_keys:
833
2101
while cursor is not None:
834
method, data, digest, next = record_map[cursor]
835
components.append((cursor, method, data, digest))
836
if cursor in content_map:
2103
record, record_details, digest, next = record_map[cursor]
2105
raise RevisionNotPresent(cursor, self)
2106
components.append((cursor, record, record_details, digest))
2108
if cursor in self._contents_map:
2109
# no need to plan further back
2110
components.append((cursor, None, None, None))
841
for component_id, method, data, digest in reversed(components):
842
if component_id in content_map:
843
content = content_map[component_id]
2114
for (component_id, record, record_details,
2115
digest) in reversed(components):
2116
if component_id in self._contents_map:
2117
content = self._contents_map[component_id]
845
if method == 'fulltext':
846
assert content is None
847
content = self.factory.parse_fulltext(data, version_id)
848
elif method == 'line-delta':
849
delta = self.factory.parse_line_delta(data, version_id)
850
content = content.copy()
851
content._lines = self._apply_delta(content._lines,
853
content_map[component_id] = content
855
if 'no-eol' in self._index.get_options(version_id):
856
content = content.copy()
857
line = content._lines[-1][1].rstrip('\n')
858
content._lines[-1] = (content._lines[-1][0], line)
859
final_content[version_id] = content
2119
content, delta = self._factory.parse_record(key[-1],
2120
record, record_details, content,
2121
copy_base_content=multiple_versions)
2122
if multiple_versions:
2123
self._contents_map[component_id] = content
861
2125
# digest here is the digest from the last applied component.
862
2126
text = content.text()
863
if sha_strings(text) != digest:
864
raise KnitCorrupt(self.filename,
865
'sha-1 does not match %s' % version_id)
867
text_map[version_id] = text
868
return text_map, final_content
870
def iter_lines_added_or_present_in_versions(self, version_ids=None,
872
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
873
if version_ids is None:
874
version_ids = self.versions()
876
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
878
pb = progress.DummyProgress()
879
# we don't care about inclusions, the caller cares.
880
# but we need to setup a list of records to visit.
881
# we need version_id, position, length
882
version_id_records = []
883
requested_versions = set(version_ids)
884
# filter for available versions
885
for version_id in requested_versions:
886
if not self.has_version(version_id):
887
raise RevisionNotPresent(version_id, self.filename)
888
# get a in-component-order queue:
889
for version_id in self.versions():
890
if version_id in requested_versions:
891
data_pos, length = self._index.get_position(version_id)
892
version_id_records.append((version_id, data_pos, length))
894
total = len(version_id_records)
895
for version_idx, (version_id, data, sha_value) in \
896
enumerate(self._data.read_records_iter(version_id_records)):
897
pb.update('Walking content.', version_idx, total)
898
method = self._index.get_method(version_id)
900
assert method in ('fulltext', 'line-delta')
901
if method == 'fulltext':
902
line_iterator = self.factory.get_fulltext_content(data)
904
line_iterator = self.factory.get_linedelta_content(data)
905
for line in line_iterator:
908
pb.update('Walking content.', total, total)
910
def num_versions(self):
911
"""See VersionedFile.num_versions()."""
912
return self._index.num_versions()
914
__len__ = num_versions
916
def annotate_iter(self, version_id):
917
"""See VersionedFile.annotate_iter."""
918
version_id = osutils.safe_revision_id(version_id)
919
content = self._get_content(version_id)
920
for origin, text in content.annotate_iter():
923
def get_parents(self, version_id):
924
"""See VersionedFile.get_parents."""
927
# 52554 calls in 1264 872 internal down from 3674
928
version_id = osutils.safe_revision_id(version_id)
930
return self._index.get_parents(version_id)
932
raise RevisionNotPresent(version_id, self.filename)
934
def get_parents_with_ghosts(self, version_id):
935
"""See VersionedFile.get_parents."""
936
version_id = osutils.safe_revision_id(version_id)
938
return self._index.get_parents_with_ghosts(version_id)
940
raise RevisionNotPresent(version_id, self.filename)
942
def get_ancestry(self, versions, topo_sorted=True):
943
"""See VersionedFile.get_ancestry."""
944
if isinstance(versions, basestring):
945
versions = [versions]
948
versions = [osutils.safe_revision_id(v) for v in versions]
949
return self._index.get_ancestry(versions, topo_sorted)
951
def get_ancestry_with_ghosts(self, versions):
952
"""See VersionedFile.get_ancestry_with_ghosts."""
953
if isinstance(versions, basestring):
954
versions = [versions]
957
versions = [osutils.safe_revision_id(v) for v in versions]
958
return self._index.get_ancestry_with_ghosts(versions)
960
#@deprecated_method(zero_eight)
961
def walk(self, version_ids):
962
"""See VersionedFile.walk."""
963
# We take the short path here, and extract all relevant texts
964
# and put them in a weave and let that do all the work. Far
965
# from optimal, but is much simpler.
966
# FIXME RB 20060228 this really is inefficient!
967
from bzrlib.weave import Weave
969
w = Weave(self.filename)
970
ancestry = set(self.get_ancestry(version_ids, topo_sorted=False))
971
sorted_graph = topo_sort(self._index.get_graph())
972
version_list = [vid for vid in sorted_graph if vid in ancestry]
974
for version_id in version_list:
975
lines = self.get_lines(version_id)
976
w.add_lines(version_id, self.get_parents(version_id), lines)
978
for lineno, insert_id, dset, line in w.walk(version_ids):
979
yield lineno, insert_id, dset, line
981
def plan_merge(self, ver_a, ver_b):
982
"""See VersionedFile.plan_merge."""
983
ver_a = osutils.safe_revision_id(ver_a)
984
ver_b = osutils.safe_revision_id(ver_b)
985
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
986
def status_a(revision, text):
987
if revision in ancestors_b:
988
return 'killed-b', text
992
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
993
def status_b(revision, text):
994
if revision in ancestors_a:
995
return 'killed-a', text
999
annotated_a = self.annotate(ver_a)
1000
annotated_b = self.annotate(ver_b)
1001
plain_a = [t for (a, t) in annotated_a]
1002
plain_b = [t for (a, t) in annotated_b]
1003
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
1006
for ai, bi, l in blocks:
1007
# process all mismatched sections
1008
# (last mismatched section is handled because blocks always
1009
# includes a 0-length last block)
1010
for revision, text in annotated_a[a_cur:ai]:
1011
yield status_a(revision, text)
1012
for revision, text in annotated_b[b_cur:bi]:
1013
yield status_b(revision, text)
1015
# and now the matched section
1018
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
1019
assert text_a == text_b
1020
yield "unchanged", text_a
1023
class _KnitComponentFile(object):
1024
"""One of the files used to implement a knit database"""
1026
def __init__(self, transport, filename, mode, file_mode=None,
1027
create_parent_dir=False, dir_mode=None):
1028
self._transport = transport
1029
self._filename = filename
1031
self._file_mode = file_mode
1032
self._dir_mode = dir_mode
1033
self._create_parent_dir = create_parent_dir
1034
self._need_to_create = False
1036
def _full_path(self):
1037
"""Return the full path to this file."""
1038
return self._transport.base + self._filename
1040
def check_header(self, fp):
1041
line = fp.readline()
1043
# An empty file can actually be treated as though the file doesn't
1045
raise errors.NoSuchFile(self._full_path())
1046
if line != self.HEADER:
1047
raise KnitHeaderError(badline=line,
1048
filename=self._transport.abspath(self._filename))
1051
"""Commit is a nop."""
1054
return '%s(%s)' % (self.__class__.__name__, self._filename)
1057
class _KnitIndex(_KnitComponentFile):
1058
"""Manages knit index file.
1060
The index is already kept in memory and read on startup, to enable
2127
actual_sha = sha_strings(text)
2128
if actual_sha != digest:
2129
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
2130
if multiple_versions:
2131
return self._contents_map[requested_key]
2135
def _wire_bytes(self):
2136
"""Get the bytes to put on the wire for 'key'.
2138
The first collection of bytes asked for returns the serialised
2139
raw_record_map and the additional details (key, parent) for key.
2140
Subsequent calls return just the additional details (key, parent).
2141
The wire storage_kind given for the first key is 'knit-delta-closure',
2142
For subsequent keys it is 'knit-delta-closure-ref'.
2144
:param key: A key from the content generator.
2145
:return: Bytes to put on the wire.
2148
# kind marker for dispatch on the far side,
2149
lines.append('knit-delta-closure')
2151
if self.vf._factory.annotated:
2152
lines.append('annotated')
2155
# then the list of keys
2156
lines.append('\t'.join(['\x00'.join(key) for key in self.keys
2157
if key not in self.nonlocal_keys]))
2158
# then the _raw_record_map in serialised form:
2160
# for each item in the map:
2162
# 1 line with parents if the key is to be yielded (None: for None, '' for ())
2163
# one line with method
2164
# one line with noeol
2165
# one line with next ('' for None)
2166
# one line with byte count of the record bytes
2168
for key, (record_bytes, (method, noeol), next) in \
2169
self._raw_record_map.iteritems():
2170
key_bytes = '\x00'.join(key)
2171
parents = self.global_map.get(key, None)
2173
parent_bytes = 'None:'
2175
parent_bytes = '\t'.join('\x00'.join(key) for key in parents)
2176
method_bytes = method
2182
next_bytes = '\x00'.join(next)
2185
map_byte_list.append('%s\n%s\n%s\n%s\n%s\n%d\n%s' % (
2186
key_bytes, parent_bytes, method_bytes, noeol_bytes, next_bytes,
2187
len(record_bytes), record_bytes))
2188
map_bytes = ''.join(map_byte_list)
2189
lines.append(map_bytes)
2190
bytes = '\n'.join(lines)
2194
class _VFContentMapGenerator(_ContentMapGenerator):
2195
"""Content map generator reading from a VersionedFiles object."""
2197
def __init__(self, versioned_files, keys, nonlocal_keys=None,
2198
global_map=None, raw_record_map=None, ordering='unordered'):
2199
"""Create a _ContentMapGenerator.
2201
:param versioned_files: The versioned files that the texts are being
2203
:param keys: The keys to produce content maps for.
2204
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
2205
which are known to not be in this knit, but rather in one of the
2207
:param global_map: The result of get_parent_map(keys) (or a supermap).
2208
This is required if get_record_stream() is to be used.
2209
:param raw_record_map: A unparsed raw record map to use for answering
2212
_ContentMapGenerator.__init__(self, ordering=ordering)
2213
# The vf to source data from
2214
self.vf = versioned_files
2216
self.keys = list(keys)
2217
# Keys known to be in fallback vfs objects
2218
if nonlocal_keys is None:
2219
self.nonlocal_keys = set()
2221
self.nonlocal_keys = frozenset(nonlocal_keys)
2222
# Parents data for keys to be returned in get_record_stream
2223
self.global_map = global_map
2224
# The chunked lists for self.keys in text form
2226
# A cache of KnitContent objects used in extracting texts.
2227
self._contents_map = {}
2228
# All the knit records needed to assemble the requested keys as full
2230
self._record_map = None
2231
if raw_record_map is None:
2232
self._raw_record_map = self.vf._get_record_map_unparsed(keys,
2235
self._raw_record_map = raw_record_map
2236
# the factory for parsing records
2237
self._factory = self.vf._factory
2240
class _NetworkContentMapGenerator(_ContentMapGenerator):
2241
"""Content map generator sourced from a network stream."""
2243
def __init__(self, bytes, line_end):
2244
"""Construct a _NetworkContentMapGenerator from a bytes block."""
2246
self.global_map = {}
2247
self._raw_record_map = {}
2248
self._contents_map = {}
2249
self._record_map = None
2250
self.nonlocal_keys = []
2251
# Get access to record parsing facilities
2252
self.vf = KnitVersionedFiles(None, None)
2255
line_end = bytes.find('\n', start)
2256
line = bytes[start:line_end]
2257
start = line_end + 1
2258
if line == 'annotated':
2259
self._factory = KnitAnnotateFactory()
2261
self._factory = KnitPlainFactory()
2262
# list of keys to emit in get_record_stream
2263
line_end = bytes.find('\n', start)
2264
line = bytes[start:line_end]
2265
start = line_end + 1
2267
tuple(segment.split('\x00')) for segment in line.split('\t')
2269
# now a loop until the end. XXX: It would be nice if this was just a
2270
# bunch of the same records as get_record_stream(..., False) gives, but
2271
# there is a decent sized gap stopping that at the moment.
2275
line_end = bytes.find('\n', start)
2276
key = tuple(bytes[start:line_end].split('\x00'))
2277
start = line_end + 1
2278
# 1 line with parents (None: for None, '' for ())
2279
line_end = bytes.find('\n', start)
2280
line = bytes[start:line_end]
2285
[tuple(segment.split('\x00')) for segment in line.split('\t')
2287
self.global_map[key] = parents
2288
start = line_end + 1
2289
# one line with method
2290
line_end = bytes.find('\n', start)
2291
line = bytes[start:line_end]
2293
start = line_end + 1
2294
# one line with noeol
2295
line_end = bytes.find('\n', start)
2296
line = bytes[start:line_end]
2298
start = line_end + 1
2299
# one line with next ('' for None)
2300
line_end = bytes.find('\n', start)
2301
line = bytes[start:line_end]
2305
next = tuple(bytes[start:line_end].split('\x00'))
2306
start = line_end + 1
2307
# one line with byte count of the record bytes
2308
line_end = bytes.find('\n', start)
2309
line = bytes[start:line_end]
2311
start = line_end + 1
2313
record_bytes = bytes[start:start+count]
2314
start = start + count
2316
self._raw_record_map[key] = (record_bytes, (method, noeol), next)
2318
def get_record_stream(self):
2319
"""Get a record stream for for keys requested by the bytestream."""
2321
for key in self.keys:
2322
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2325
def _wire_bytes(self):
2329
class _KndxIndex(object):
2330
"""Manages knit index files
2332
The index is kept in memory and read on startup, to enable
1061
2333
fast lookups of revision information. The cursor of the index
1062
2334
file is always pointing to the end, making it easy to append
1105
2377
to ensure that records always start on new lines even if the last write was
1106
2378
interrupted. As a result its normal for the last line in the index to be
1107
2379
missing a trailing newline. One can be added with no harmful effects.
2381
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
2382
where prefix is e.g. the (fileid,) for .texts instances or () for
2383
constant-mapped things like .revisions, and the old state is
2384
tuple(history_vector, cache_dict). This is used to prevent having an
2385
ABI change with the C extension that reads .kndx files.
1110
2388
HEADER = "# bzr knit index 8\n"
1112
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1113
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1115
def _cache_version(self, version_id, options, pos, size, parents):
2390
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
2391
"""Create a _KndxIndex on transport using mapper."""
2392
self._transport = transport
2393
self._mapper = mapper
2394
self._get_scope = get_scope
2395
self._allow_writes = allow_writes
2396
self._is_locked = is_locked
2398
self.has_graph = True
2400
def add_records(self, records, random_id=False, missing_compression_parents=False):
2401
"""Add multiple records to the index.
2403
:param records: a list of tuples:
2404
(key, options, access_memo, parents).
2405
:param random_id: If True the ids being added were randomly generated
2406
and no check for existence will be performed.
2407
:param missing_compression_parents: If True the records being added are
2408
only compressed against texts already in the index (or inside
2409
records). If False the records all refer to unavailable texts (or
2410
texts inside records) as compression parents.
2412
if missing_compression_parents:
2413
# It might be nice to get the edge of the records. But keys isn't
2415
keys = sorted(record[0] for record in records)
2416
raise errors.RevisionNotPresent(keys, self)
2418
for record in records:
2421
path = self._mapper.map(key) + '.kndx'
2422
path_keys = paths.setdefault(path, (prefix, []))
2423
path_keys[1].append(record)
2424
for path in sorted(paths):
2425
prefix, path_keys = paths[path]
2426
self._load_prefixes([prefix])
2428
orig_history = self._kndx_cache[prefix][1][:]
2429
orig_cache = self._kndx_cache[prefix][0].copy()
2432
for key, options, (_, pos, size), parents in path_keys:
2434
# kndx indices cannot be parentless.
2436
line = "\n%s %s %s %s %s :" % (
2437
key[-1], ','.join(options), pos, size,
2438
self._dictionary_compress(parents))
2439
if type(line) is not str:
2440
raise AssertionError(
2441
'data must be utf8 was %s' % type(line))
2443
self._cache_key(key, options, pos, size, parents)
2444
if len(orig_history):
2445
self._transport.append_bytes(path, ''.join(lines))
2447
self._init_index(path, lines)
2449
# If any problems happen, restore the original values and re-raise
2450
self._kndx_cache[prefix] = (orig_cache, orig_history)
2453
def scan_unvalidated_index(self, graph_index):
2454
"""See _KnitGraphIndex.scan_unvalidated_index."""
2455
# Because kndx files do not support atomic insertion via separate index
2456
# files, they do not support this method.
2457
raise NotImplementedError(self.scan_unvalidated_index)
2459
def get_missing_compression_parents(self):
2460
"""See _KnitGraphIndex.get_missing_compression_parents."""
2461
# Because kndx files do not support atomic insertion via separate index
2462
# files, they do not support this method.
2463
raise NotImplementedError(self.get_missing_compression_parents)
2465
def _cache_key(self, key, options, pos, size, parent_keys):
1116
2466
"""Cache a version record in the history array and index cache.
1118
2468
This is inlined into _load_data for performance. KEEP IN SYNC.
1119
2469
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
2473
version_id = key[-1]
2474
# last-element only for compatibilty with the C load_data.
2475
parents = tuple(parent[-1] for parent in parent_keys)
2476
for parent in parent_keys:
2477
if parent[:-1] != prefix:
2478
raise ValueError("mismatched prefixes for %r, %r" % (
2480
cache, history = self._kndx_cache[prefix]
1122
2481
# only want the _history index to reference the 1st index entry
1123
2482
# for version_id
1124
if version_id not in self._cache:
1125
index = len(self._history)
1126
self._history.append(version_id)
2483
if version_id not in cache:
2484
index = len(history)
2485
history.append(version_id)
1128
index = self._cache[version_id][5]
1129
self._cache[version_id] = (version_id,
2487
index = cache[version_id][5]
2488
cache[version_id] = (version_id,
1136
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1137
create_parent_dir=False, delay_create=False, dir_mode=None):
1138
_KnitComponentFile.__init__(self, transport, filename, mode,
1139
file_mode=file_mode,
1140
create_parent_dir=create_parent_dir,
1143
# position in _history is the 'official' index for a revision
1144
# but the values may have come from a newer entry.
1145
# so - wc -l of a knit index is != the number of unique names
2495
def check_header(self, fp):
2496
line = fp.readline()
2498
# An empty file can actually be treated as though the file doesn't
2500
raise errors.NoSuchFile(self)
2501
if line != self.HEADER:
2502
raise KnitHeaderError(badline=line, filename=self)
2504
def _check_read(self):
2505
if not self._is_locked():
2506
raise errors.ObjectNotLocked(self)
2507
if self._get_scope() != self._scope:
2510
def _check_write_ok(self):
2511
"""Assert if not writes are permitted."""
2512
if not self._is_locked():
2513
raise errors.ObjectNotLocked(self)
2514
if self._get_scope() != self._scope:
2516
if self._mode != 'w':
2517
raise errors.ReadOnlyObjectDirtiedError(self)
2519
def get_build_details(self, keys):
2520
"""Get the method, index_memo and compression parent for keys.
2522
Ghosts are omitted from the result.
2524
:param keys: An iterable of keys.
2525
:return: A dict of key:(index_memo, compression_parent, parents,
2528
opaque structure to pass to read_records to extract the raw
2531
Content that this record is built upon, may be None
2533
Logical parents of this node
2535
extra information about the content which needs to be passed to
2536
Factory.parse_record
2538
parent_map = self.get_parent_map(keys)
2541
if key not in parent_map:
2543
method = self.get_method(key)
2544
parents = parent_map[key]
2545
if method == 'fulltext':
2546
compression_parent = None
2548
compression_parent = parents[0]
2549
noeol = 'no-eol' in self.get_options(key)
2550
index_memo = self.get_position(key)
2551
result[key] = (index_memo, compression_parent,
2552
parents, (method, noeol))
2555
def get_method(self, key):
2556
"""Return compression method of specified key."""
2557
options = self.get_options(key)
2558
if 'fulltext' in options:
2560
elif 'line-delta' in options:
2563
raise errors.KnitIndexUnknownMethod(self, options)
2565
def get_options(self, key):
2566
"""Return a list representing options.
2570
prefix, suffix = self._split_key(key)
2571
self._load_prefixes([prefix])
1149
fp = self._transport.get(self._filename)
1151
# _load_data may raise NoSuchFile if the target knit is
1157
if mode != 'w' or not create:
1160
self._need_to_create = True
1162
self._transport.put_bytes_non_atomic(
1163
self._filename, self.HEADER, mode=self._file_mode)
1165
def _load_data(self, fp):
1167
history = self._history
1169
self.check_header(fp)
1170
# readlines reads the whole file at once:
1171
# bad for transports like http, good for local disk
1172
# we save 60 ms doing this one change (
1173
# from calling readline each time to calling
1175
# probably what we want for nice behaviour on
1176
# http is a incremental readlines that yields, or
1177
# a check for local vs non local indexes,
1178
history_top = len(history) - 1
1179
for line in fp.readlines():
1181
if len(rec) < 5 or rec[-1] != ':':
1183
# FIXME: in the future we should determine if its a
1184
# short write - and ignore it
1185
# or a different failure, and raise. RBC 20060407
2573
return self._kndx_cache[prefix][0][suffix][1]
2575
raise RevisionNotPresent(key, self)
2577
def find_ancestry(self, keys):
2578
"""See CombinedGraphIndex.find_ancestry()"""
2579
prefixes = set(key[:-1] for key in keys)
2580
self._load_prefixes(prefixes)
2583
missing_keys = set()
2584
pending_keys = list(keys)
2585
# This assumes that keys will not reference parents in a different
2586
# prefix, which is accurate so far.
2588
key = pending_keys.pop()
2589
if key in parent_map:
1190
for value in rec[4:-1]:
1192
# uncompressed reference
1193
parent_id = value[1:]
1195
parent_id = history[int(value)]
1196
parents.append(parent_id)
1197
except (IndexError, ValueError), e:
1198
# The parent could not be decoded to get its parent row. This
1199
# at a minimum will cause this row to have wrong parents, or
1200
# even to apply a delta to the wrong base and decode
1201
# incorrectly. its therefore not usable, and because we have
1202
# encountered a situation where a new knit index had this
1203
# corrupt we can't asssume that no other rows referring to the
1204
# index of this record actually mean the subsequent uncorrupt
1206
raise errors.KnitCorrupt(self._filename,
1207
"line %r: %s" % (rec, e))
1209
version_id, options, pos, size = rec[:4]
1210
version_id = version_id
1212
# See self._cache_version
1213
# only want the _history index to reference the 1st
1214
# index entry for version_id
1215
if version_id not in cache:
1218
history.append(version_id)
1220
index = cache[version_id][5]
1221
cache[version_id] = (version_id,
1227
# end self._cache_version
1229
def get_graph(self):
1230
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1232
def get_ancestry(self, versions, topo_sorted=True):
1233
"""See VersionedFile.get_ancestry."""
1234
# get a graph of all the mentioned versions:
1236
pending = set(versions)
1239
version = pending.pop()
1242
parents = [p for p in cache[version][4] if p in cache]
1244
raise RevisionNotPresent(version, self._filename)
1245
# if not completed and not a ghost
1246
pending.update([p for p in parents if p not in graph])
1247
graph[version] = parents
1250
return topo_sort(graph.items())
1252
def get_ancestry_with_ghosts(self, versions):
1253
"""See VersionedFile.get_ancestry_with_ghosts."""
1254
# get a graph of all the mentioned versions:
1255
self.check_versions_present(versions)
1258
pending = set(versions)
1260
version = pending.pop()
1262
parents = cache[version][4]
1268
pending.update([p for p in parents if p not in graph])
1269
graph[version] = parents
1270
return topo_sort(graph.items())
1272
def num_versions(self):
1273
return len(self._history)
1275
__len__ = num_versions
1277
def get_versions(self):
1278
return self._history
1280
def idx_to_name(self, idx):
1281
return self._history[idx]
1283
def lookup(self, version_id):
1284
assert version_id in self._cache
1285
return self._cache[version_id][5]
1287
def _version_list_to_index(self, versions):
2593
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2595
missing_keys.add(key)
2597
parent_keys = tuple([prefix + (suffix,)
2598
for suffix in suffix_parents])
2599
parent_map[key] = parent_keys
2600
pending_keys.extend([p for p in parent_keys
2601
if p not in parent_map])
2602
return parent_map, missing_keys
2604
def get_parent_map(self, keys):
2605
"""Get a map of the parents of keys.
2607
:param keys: The keys to look up parents for.
2608
:return: A mapping from keys to parents. Absent keys are absent from
2611
# Parse what we need to up front, this potentially trades off I/O
2612
# locality (.kndx and .knit in the same block group for the same file
2613
# id) for less checking in inner loops.
2614
prefixes = set(key[:-1] for key in keys)
2615
self._load_prefixes(prefixes)
2620
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2624
result[key] = tuple(prefix + (suffix,) for
2625
suffix in suffix_parents)
2628
def get_position(self, key):
2629
"""Return details needed to access the version.
2631
:return: a tuple (key, data position, size) to hand to the access
2632
logic to get the record.
2634
prefix, suffix = self._split_key(key)
2635
self._load_prefixes([prefix])
2636
entry = self._kndx_cache[prefix][0][suffix]
2637
return key, entry[2], entry[3]
2639
has_key = _mod_index._has_key_from_parent_map
2641
def _init_index(self, path, extra_lines=[]):
2642
"""Initialize an index."""
2644
sio.write(self.HEADER)
2645
sio.writelines(extra_lines)
2647
self._transport.put_file_non_atomic(path, sio,
2648
create_parent_dir=True)
2649
# self._create_parent_dir)
2650
# mode=self._file_mode,
2651
# dir_mode=self._dir_mode)
2654
"""Get all the keys in the collection.
2656
The keys are not ordered.
2659
# Identify all key prefixes.
2660
# XXX: A bit hacky, needs polish.
2661
if type(self._mapper) is ConstantMapper:
2665
for quoted_relpath in self._transport.iter_files_recursive():
2666
path, ext = os.path.splitext(quoted_relpath)
2668
prefixes = [self._mapper.unmap(path) for path in relpaths]
2669
self._load_prefixes(prefixes)
2670
for prefix in prefixes:
2671
for suffix in self._kndx_cache[prefix][1]:
2672
result.add(prefix + (suffix,))
2675
def _load_prefixes(self, prefixes):
2676
"""Load the indices for prefixes."""
2678
for prefix in prefixes:
2679
if prefix not in self._kndx_cache:
2680
# the load_data interface writes to these variables.
2683
self._filename = prefix
2685
path = self._mapper.map(prefix) + '.kndx'
2686
fp = self._transport.get(path)
2688
# _load_data may raise NoSuchFile if the target knit is
2690
_load_data(self, fp)
2693
self._kndx_cache[prefix] = (self._cache, self._history)
2698
self._kndx_cache[prefix] = ({}, [])
2699
if type(self._mapper) is ConstantMapper:
2700
# preserve behaviour for revisions.kndx etc.
2701
self._init_index(path)
2706
missing_keys = _mod_index._missing_keys_from_parent_map
2708
def _partition_keys(self, keys):
2709
"""Turn keys into a dict of prefix:suffix_list."""
2712
prefix_keys = result.setdefault(key[:-1], [])
2713
prefix_keys.append(key[-1])
2716
def _dictionary_compress(self, keys):
2717
"""Dictionary compress keys.
2719
:param keys: The keys to generate references to.
2720
:return: A string representation of keys. keys which are present are
2721
dictionary compressed, and others are emitted as fulltext with a
1288
2726
result_list = []
1290
for version in versions:
1291
if version in cache:
2727
prefix = keys[0][:-1]
2728
cache = self._kndx_cache[prefix][0]
2730
if key[:-1] != prefix:
2731
# kndx indices cannot refer across partitioned storage.
2732
raise ValueError("mismatched prefixes for %r" % keys)
2733
if key[-1] in cache:
1292
2734
# -- inlined lookup() --
1293
result_list.append(str(cache[version][5]))
2735
result_list.append(str(cache[key[-1]][5]))
1294
2736
# -- end lookup () --
1296
result_list.append('.' + version)
2738
result_list.append('.' + key[-1])
1297
2739
return ' '.join(result_list)
1299
def add_version(self, version_id, options, pos, size, parents):
1300
"""Add a version record to the index."""
1301
self.add_versions(((version_id, options, pos, size, parents),))
1303
def add_versions(self, versions):
1304
"""Add multiple versions to the index.
1306
:param versions: a list of tuples:
1307
(version_id, options, pos, size, parents).
1310
orig_history = self._history[:]
1311
orig_cache = self._cache.copy()
1314
for version_id, options, pos, size, parents in versions:
1315
line = "\n%s %s %s %s %s :" % (version_id,
1319
self._version_list_to_index(parents))
1320
assert isinstance(line, str), \
1321
'content must be utf-8 encoded: %r' % (line,)
1323
self._cache_version(version_id, options, pos, size, parents)
1324
if not self._need_to_create:
1325
self._transport.append_bytes(self._filename, ''.join(lines))
1328
sio.write(self.HEADER)
1329
sio.writelines(lines)
1331
self._transport.put_file_non_atomic(self._filename, sio,
1332
create_parent_dir=self._create_parent_dir,
1333
mode=self._file_mode,
1334
dir_mode=self._dir_mode)
1335
self._need_to_create = False
1337
# If any problems happen, restore the original values and re-raise
1338
self._history = orig_history
1339
self._cache = orig_cache
1342
def has_version(self, version_id):
1343
"""True if the version is in the index."""
1344
return version_id in self._cache
1346
def get_position(self, version_id):
1347
"""Return data position and size of specified version."""
1348
entry = self._cache[version_id]
1349
return entry[2], entry[3]
1351
def get_method(self, version_id):
1352
"""Return compression method of specified version."""
1353
options = self._cache[version_id][1]
1354
if 'fulltext' in options:
2741
def _reset_cache(self):
2742
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2743
# (cache_dict, history_vector) for parsed kndx files.
2744
self._kndx_cache = {}
2745
self._scope = self._get_scope()
2746
allow_writes = self._allow_writes()
2752
def _sort_keys_by_io(self, keys, positions):
2753
"""Figure out an optimal order to read the records for the given keys.
2755
Sort keys, grouped by index and sorted by position.
2757
:param keys: A list of keys whose records we want to read. This will be
2759
:param positions: A dict, such as the one returned by
2760
_get_components_positions()
2763
def get_sort_key(key):
2764
index_memo = positions[key][1]
2765
# Group by prefix and position. index_memo[0] is the key, so it is
2766
# (file_id, revision_id) and we don't want to sort on revision_id,
2767
# index_memo[1] is the position, and index_memo[2] is the size,
2768
# which doesn't matter for the sort
2769
return index_memo[0][:-1], index_memo[1]
2770
return keys.sort(key=get_sort_key)
2772
_get_total_build_size = _get_total_build_size
2774
def _split_key(self, key):
2775
"""Split key into a prefix and suffix."""
2776
return key[:-1], key[-1]
2779
class _KnitGraphIndex(object):
2780
"""A KnitVersionedFiles index layered on GraphIndex."""
2782
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2783
add_callback=None, track_external_parent_refs=False):
2784
"""Construct a KnitGraphIndex on a graph_index.
2786
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2787
:param is_locked: A callback to check whether the object should answer
2789
:param deltas: Allow delta-compressed records.
2790
:param parents: If True, record knits parents, if not do not record
2792
:param add_callback: If not None, allow additions to the index and call
2793
this callback with a list of added GraphIndex nodes:
2794
[(node, value, node_refs), ...]
2795
:param is_locked: A callback, returns True if the index is locked and
2797
:param track_external_parent_refs: If True, record all external parent
2798
references parents from added records. These can be retrieved
2799
later by calling get_missing_parents().
2801
self._add_callback = add_callback
2802
self._graph_index = graph_index
2803
self._deltas = deltas
2804
self._parents = parents
2805
if deltas and not parents:
2806
# XXX: TODO: Delta tree and parent graph should be conceptually
2808
raise KnitCorrupt(self, "Cannot do delta compression without "
2810
self.has_graph = parents
2811
self._is_locked = is_locked
2812
self._missing_compression_parents = set()
2813
if track_external_parent_refs:
2814
self._key_dependencies = _KeyRefs()
2816
self._key_dependencies = None
2819
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2821
def add_records(self, records, random_id=False,
2822
missing_compression_parents=False):
2823
"""Add multiple records to the index.
2825
This function does not insert data into the Immutable GraphIndex
2826
backing the KnitGraphIndex, instead it prepares data for insertion by
2827
the caller and checks that it is safe to insert then calls
2828
self._add_callback with the prepared GraphIndex nodes.
2830
:param records: a list of tuples:
2831
(key, options, access_memo, parents).
2832
:param random_id: If True the ids being added were randomly generated
2833
and no check for existence will be performed.
2834
:param missing_compression_parents: If True the records being added are
2835
only compressed against texts already in the index (or inside
2836
records). If False the records all refer to unavailable texts (or
2837
texts inside records) as compression parents.
2839
if not self._add_callback:
2840
raise errors.ReadOnlyError(self)
2841
# we hope there are no repositories with inconsistent parentage
2845
compression_parents = set()
2846
key_dependencies = self._key_dependencies
2847
for (key, options, access_memo, parents) in records:
2849
parents = tuple(parents)
2850
if key_dependencies is not None:
2851
key_dependencies.add_references(key, parents)
2852
index, pos, size = access_memo
2853
if 'no-eol' in options:
2857
value += "%d %d" % (pos, size)
2858
if not self._deltas:
2859
if 'line-delta' in options:
2860
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2863
if 'line-delta' in options:
2864
node_refs = (parents, (parents[0],))
2865
if missing_compression_parents:
2866
compression_parents.add(parents[0])
2868
node_refs = (parents, ())
2870
node_refs = (parents, )
2873
raise KnitCorrupt(self, "attempt to add node with parents "
2874
"in parentless index.")
2876
keys[key] = (value, node_refs)
2879
present_nodes = self._get_entries(keys)
2880
for (index, key, value, node_refs) in present_nodes:
2881
parents = node_refs[:1]
2882
# Sometimes these are passed as a list rather than a tuple
2883
passed = static_tuple.as_tuples(keys[key])
2884
passed_parents = passed[1][:1]
2885
if (value[0] != keys[key][0][0] or
2886
parents != passed_parents):
2887
node_refs = static_tuple.as_tuples(node_refs)
2888
raise KnitCorrupt(self, "inconsistent details in add_records"
2889
": %s %s" % ((value, node_refs), passed))
2893
for key, (value, node_refs) in keys.iteritems():
2894
result.append((key, value, node_refs))
2896
for key, (value, node_refs) in keys.iteritems():
2897
result.append((key, value))
2898
self._add_callback(result)
2899
if missing_compression_parents:
2900
# This may appear to be incorrect (it does not check for
2901
# compression parents that are in the existing graph index),
2902
# but such records won't have been buffered, so this is
2903
# actually correct: every entry when
2904
# missing_compression_parents==True either has a missing parent, or
2905
# a parent that is one of the keys in records.
2906
compression_parents.difference_update(keys)
2907
self._missing_compression_parents.update(compression_parents)
2908
# Adding records may have satisfied missing compression parents.
2909
self._missing_compression_parents.difference_update(keys)
2911
def scan_unvalidated_index(self, graph_index):
2912
"""Inform this _KnitGraphIndex that there is an unvalidated index.
2914
This allows this _KnitGraphIndex to keep track of any missing
2915
compression parents we may want to have filled in to make those
2918
:param graph_index: A GraphIndex
2921
new_missing = graph_index.external_references(ref_list_num=1)
2922
new_missing.difference_update(self.get_parent_map(new_missing))
2923
self._missing_compression_parents.update(new_missing)
2924
if self._key_dependencies is not None:
2925
# Add parent refs from graph_index (and discard parent refs that
2926
# the graph_index has).
2927
for node in graph_index.iter_all_entries():
2928
self._key_dependencies.add_references(node[1], node[3][0])
2930
def get_missing_compression_parents(self):
2931
"""Return the keys of missing compression parents.
2933
Missing compression parents occur when a record stream was missing
2934
basis texts, or a index was scanned that had missing basis texts.
2936
return frozenset(self._missing_compression_parents)
2938
def get_missing_parents(self):
2939
"""Return the keys of missing parents."""
2940
# If updating this, you should also update
2941
# groupcompress._GCGraphIndex.get_missing_parents
2942
# We may have false positives, so filter those out.
2943
self._key_dependencies.satisfy_refs_for_keys(
2944
self.get_parent_map(self._key_dependencies.get_unsatisfied_refs()))
2945
return frozenset(self._key_dependencies.get_unsatisfied_refs())
2947
def _check_read(self):
2948
"""raise if reads are not permitted."""
2949
if not self._is_locked():
2950
raise errors.ObjectNotLocked(self)
2952
def _check_write_ok(self):
2953
"""Assert if writes are not permitted."""
2954
if not self._is_locked():
2955
raise errors.ObjectNotLocked(self)
2957
def _compression_parent(self, an_entry):
2958
# return the key that an_entry is compressed against, or None
2959
# Grab the second parent list (as deltas implies parents currently)
2960
compression_parents = an_entry[3][1]
2961
if not compression_parents:
2963
if len(compression_parents) != 1:
2964
raise AssertionError(
2965
"Too many compression parents: %r" % compression_parents)
2966
return compression_parents[0]
2968
def get_build_details(self, keys):
2969
"""Get the method, index_memo and compression parent for version_ids.
2971
Ghosts are omitted from the result.
2973
:param keys: An iterable of keys.
2974
:return: A dict of key:
2975
(index_memo, compression_parent, parents, record_details).
2977
opaque structure to pass to read_records to extract the raw
2980
Content that this record is built upon, may be None
2982
Logical parents of this node
2984
extra information about the content which needs to be passed to
2985
Factory.parse_record
2989
entries = self._get_entries(keys, False)
2990
for entry in entries:
2992
if not self._parents:
2995
parents = entry[3][0]
2996
if not self._deltas:
2997
compression_parent_key = None
2999
compression_parent_key = self._compression_parent(entry)
3000
noeol = (entry[2][0] == 'N')
3001
if compression_parent_key:
3002
method = 'line-delta'
3005
result[key] = (self._node_to_position(entry),
3006
compression_parent_key, parents,
3010
def _get_entries(self, keys, check_present=False):
3011
"""Get the entries for keys.
3013
:param keys: An iterable of index key tuples.
3018
for node in self._graph_index.iter_entries(keys):
3020
found_keys.add(node[1])
3022
# adapt parentless index to the rest of the code.
3023
for node in self._graph_index.iter_entries(keys):
3024
yield node[0], node[1], node[2], ()
3025
found_keys.add(node[1])
3027
missing_keys = keys.difference(found_keys)
3029
raise RevisionNotPresent(missing_keys.pop(), self)
3031
def get_method(self, key):
3032
"""Return compression method of specified key."""
3033
return self._get_method(self._get_node(key))
3035
def _get_method(self, node):
3036
if not self._deltas:
1355
3037
return 'fulltext'
1357
if 'line-delta' not in options:
1358
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
3038
if self._compression_parent(node):
1359
3039
return 'line-delta'
1361
def get_options(self, version_id):
1362
return self._cache[version_id][1]
1364
def get_parents(self, version_id):
1365
"""Return parents of specified version ignoring ghosts."""
1366
return [parent for parent in self._cache[version_id][4]
1367
if parent in self._cache]
1369
def get_parents_with_ghosts(self, version_id):
1370
"""Return parents of specified version with ghosts."""
1371
return self._cache[version_id][4]
1373
def check_versions_present(self, version_ids):
1374
"""Check that all specified versions are present."""
1376
for version_id in version_ids:
1377
if version_id not in cache:
1378
raise RevisionNotPresent(version_id, self._filename)
1381
class _KnitData(_KnitComponentFile):
1382
"""Contents of the knit data file"""
1384
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1385
create_parent_dir=False, delay_create=False,
1387
_KnitComponentFile.__init__(self, transport, filename, mode,
1388
file_mode=file_mode,
1389
create_parent_dir=create_parent_dir,
1391
self._checked = False
1392
# TODO: jam 20060713 conceptually, this could spill to disk
1393
# if the cached size gets larger than a certain amount
1394
# but it complicates the model a bit, so for now just use
1395
# a simple dictionary
1397
self._do_cache = False
1400
self._need_to_create = create
1402
self._transport.put_bytes_non_atomic(self._filename, '',
1403
mode=self._file_mode)
1405
def enable_cache(self):
1406
"""Enable caching of reads."""
1407
self._do_cache = True
1409
def clear_cache(self):
1410
"""Clear the record cache."""
1411
self._do_cache = False
1414
def _open_file(self):
1416
return self._transport.get(self._filename)
1421
def _record_to_data(self, version_id, digest, lines):
1422
"""Convert version_id, digest, lines into a raw data block.
1424
:return: (len, a StringIO instance with the raw data ready to read.)
1427
data_file = GzipFile(None, mode='wb', fileobj=sio)
1429
assert isinstance(version_id, str)
1430
data_file.writelines(chain(
1431
["version %s %d %s\n" % (version_id,
1435
["end %s\n" % version_id]))
1442
def add_raw_record(self, raw_data):
1443
"""Append a prepared record to the data file.
1445
:return: the offset in the data file raw_data was written.
1447
assert isinstance(raw_data, str), 'data must be plain bytes'
1448
if not self._need_to_create:
1449
return self._transport.append_bytes(self._filename, raw_data)
1451
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1452
create_parent_dir=self._create_parent_dir,
1453
mode=self._file_mode,
1454
dir_mode=self._dir_mode)
1455
self._need_to_create = False
1458
def add_record(self, version_id, digest, lines):
1459
"""Write new text record to disk. Returns the position in the
1460
file where it was written."""
1461
size, sio = self._record_to_data(version_id, digest, lines)
1463
if not self._need_to_create:
1464
start_pos = self._transport.append_file(self._filename, sio)
1466
self._transport.put_file_non_atomic(self._filename, sio,
1467
create_parent_dir=self._create_parent_dir,
1468
mode=self._file_mode,
1469
dir_mode=self._dir_mode)
1470
self._need_to_create = False
1473
self._cache[version_id] = sio.getvalue()
1474
return start_pos, size
1476
def _parse_record_header(self, version_id, raw_data):
1477
"""Parse a record header for consistency.
1479
:return: the header and the decompressor stream.
1480
as (stream, header_record)
1482
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1484
rec = self._check_header(version_id, df.readline())
1485
except Exception, e:
1486
raise KnitCorrupt(self._filename,
1487
"While reading {%s} got %s(%s)"
1488
% (version_id, e.__class__.__name__, str(e)))
1491
def _check_header(self, version_id, line):
1494
raise KnitCorrupt(self._filename,
1495
'unexpected number of elements in record header')
1496
if rec[1] != version_id:
1497
raise KnitCorrupt(self._filename,
1498
'unexpected version, wanted %r, got %r'
1499
% (version_id, rec[1]))
1502
def _parse_record(self, version_id, data):
1504
# 4168 calls in 2880 217 internal
1505
# 4168 calls to _parse_record_header in 2121
1506
# 4168 calls to readlines in 330
1507
df = GzipFile(mode='rb', fileobj=StringIO(data))
1510
record_contents = df.readlines()
1511
except Exception, e:
1512
raise KnitCorrupt(self._filename,
1513
"While reading {%s} got %s(%s)"
1514
% (version_id, e.__class__.__name__, str(e)))
1515
header = record_contents.pop(0)
1516
rec = self._check_header(version_id, header)
1518
last_line = record_contents.pop()
1519
if len(record_contents) != int(rec[2]):
1520
raise KnitCorrupt(self._filename,
1521
'incorrect number of lines %s != %s'
1523
% (len(record_contents), int(rec[2]),
1525
if last_line != 'end %s\n' % rec[1]:
1526
raise KnitCorrupt(self._filename,
1527
'unexpected version end line %r, wanted %r'
1528
% (last_line, version_id))
1530
return record_contents, rec[3]
1532
def read_records_iter_raw(self, records):
1533
"""Read text records from data file and yield raw data.
1535
This unpacks enough of the text record to validate the id is
1536
as expected but thats all.
1538
# setup an iterator of the external records:
1539
# uses readv so nice and fast we hope.
1541
# grab the disk data needed.
1543
# Don't check _cache if it is empty
1544
needed_offsets = [(pos, size) for version_id, pos, size
1546
if version_id not in self._cache]
1548
needed_offsets = [(pos, size) for version_id, pos, size
1551
raw_records = self._transport.readv(self._filename, needed_offsets)
1553
for version_id, pos, size in records:
1554
if version_id in self._cache:
1555
# This data has already been validated
1556
data = self._cache[version_id]
1558
pos, data = raw_records.next()
1560
self._cache[version_id] = data
1562
# validate the header
1563
df, rec = self._parse_record_header(version_id, data)
1565
yield version_id, data
1567
def read_records_iter(self, records):
1568
"""Read text records from data file and yield result.
1570
The result will be returned in whatever is the fastest to read.
1571
Not by the order requested. Also, multiple requests for the same
1572
record will only yield 1 response.
1573
:param records: A list of (version_id, pos, len) entries
1574
:return: Yields (version_id, contents, digest) in the order
1575
read, not the order requested
1581
# Skip records we have alread seen
1582
yielded_records = set()
1583
needed_records = set()
1584
for record in records:
1585
if record[0] in self._cache:
1586
if record[0] in yielded_records:
1588
yielded_records.add(record[0])
1589
data = self._cache[record[0]]
1590
content, digest = self._parse_record(record[0], data)
1591
yield (record[0], content, digest)
1593
needed_records.add(record)
1594
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1596
needed_records = sorted(set(records), key=operator.itemgetter(1))
1598
if not needed_records:
1601
# The transport optimizes the fetching as well
1602
# (ie, reads continuous ranges.)
1603
readv_response = self._transport.readv(self._filename,
1604
[(pos, size) for version_id, pos, size in needed_records])
1606
for (version_id, pos, size), (pos, data) in \
1607
izip(iter(needed_records), readv_response):
1608
content, digest = self._parse_record(version_id, data)
1610
self._cache[version_id] = data
1611
yield version_id, content, digest
1613
def read_records(self, records):
1614
"""Read records into a dictionary."""
1616
for record_id, content, digest in \
1617
self.read_records_iter(records):
1618
components[record_id] = (content, digest)
1622
class InterKnit(InterVersionedFile):
1623
"""Optimised code paths for knit to knit operations."""
1625
_matching_file_from_factory = KnitVersionedFile
1626
_matching_file_to_factory = KnitVersionedFile
1629
def is_compatible(source, target):
1630
"""Be compatible with knits. """
1632
return (isinstance(source, KnitVersionedFile) and
1633
isinstance(target, KnitVersionedFile))
1634
except AttributeError:
1637
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1638
"""See InterVersionedFile.join."""
1639
assert isinstance(self.source, KnitVersionedFile)
1640
assert isinstance(self.target, KnitVersionedFile)
1642
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1647
pb = ui.ui_factory.nested_progress_bar()
1649
version_ids = list(version_ids)
1650
if None in version_ids:
1651
version_ids.remove(None)
1653
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1654
this_versions = set(self.target._index.get_versions())
1655
needed_versions = self.source_ancestry - this_versions
1656
cross_check_versions = self.source_ancestry.intersection(this_versions)
1657
mismatched_versions = set()
1658
for version in cross_check_versions:
1659
# scan to include needed parents.
1660
n1 = set(self.target.get_parents_with_ghosts(version))
1661
n2 = set(self.source.get_parents_with_ghosts(version))
1663
# FIXME TEST this check for cycles being introduced works
1664
# the logic is we have a cycle if in our graph we are an
1665
# ancestor of any of the n2 revisions.
1671
parent_ancestors = self.source.get_ancestry(parent)
1672
if version in parent_ancestors:
1673
raise errors.GraphCycleError([parent, version])
1674
# ensure this parent will be available later.
1675
new_parents = n2.difference(n1)
1676
needed_versions.update(new_parents.difference(this_versions))
1677
mismatched_versions.add(version)
1679
if not needed_versions and not mismatched_versions:
1681
full_list = topo_sort(self.source.get_graph())
1683
version_list = [i for i in full_list if (not self.target.has_version(i)
1684
and i in needed_versions)]
1688
copy_queue_records = []
1690
for version_id in version_list:
1691
options = self.source._index.get_options(version_id)
1692
parents = self.source._index.get_parents_with_ghosts(version_id)
1693
# check that its will be a consistent copy:
1694
for parent in parents:
1695
# if source has the parent, we must :
1696
# * already have it or
1697
# * have it scheduled already
1698
# otherwise we don't care
1699
assert (self.target.has_version(parent) or
1700
parent in copy_set or
1701
not self.source.has_version(parent))
1702
data_pos, data_size = self.source._index.get_position(version_id)
1703
copy_queue_records.append((version_id, data_pos, data_size))
1704
copy_queue.append((version_id, options, parents))
1705
copy_set.add(version_id)
1707
# data suck the join:
1709
total = len(version_list)
1712
for (version_id, raw_data), \
1713
(version_id2, options, parents) in \
1714
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1716
assert version_id == version_id2, 'logic error, inconsistent results'
1718
pb.update("Joining knit", count, total)
1719
raw_records.append((version_id, options, parents, len(raw_data)))
1720
raw_datum.append(raw_data)
1721
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1723
for version in mismatched_versions:
1724
# FIXME RBC 20060309 is this needed?
1725
n1 = set(self.target.get_parents_with_ghosts(version))
1726
n2 = set(self.source.get_parents_with_ghosts(version))
1727
# write a combined record to our history preserving the current
1728
# parents as first in the list
1729
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1730
self.target.fix_parents(version, new_parents)
1736
InterVersionedFile.register_optimiser(InterKnit)
1739
class WeaveToKnit(InterVersionedFile):
1740
"""Optimised code paths for weave to knit operations."""
1742
_matching_file_from_factory = bzrlib.weave.WeaveFile
1743
_matching_file_to_factory = KnitVersionedFile
1746
def is_compatible(source, target):
1747
"""Be compatible with weaves to knits."""
1749
return (isinstance(source, bzrlib.weave.Weave) and
1750
isinstance(target, KnitVersionedFile))
1751
except AttributeError:
1754
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1755
"""See InterVersionedFile.join."""
1756
assert isinstance(self.source, bzrlib.weave.Weave)
1757
assert isinstance(self.target, KnitVersionedFile)
1759
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1764
pb = ui.ui_factory.nested_progress_bar()
1766
version_ids = list(version_ids)
1768
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1769
this_versions = set(self.target._index.get_versions())
1770
needed_versions = self.source_ancestry - this_versions
1771
cross_check_versions = self.source_ancestry.intersection(this_versions)
1772
mismatched_versions = set()
1773
for version in cross_check_versions:
1774
# scan to include needed parents.
1775
n1 = set(self.target.get_parents_with_ghosts(version))
1776
n2 = set(self.source.get_parents(version))
1777
# if all of n2's parents are in n1, then its fine.
1778
if n2.difference(n1):
1779
# FIXME TEST this check for cycles being introduced works
1780
# the logic is we have a cycle if in our graph we are an
1781
# ancestor of any of the n2 revisions.
1787
parent_ancestors = self.source.get_ancestry(parent)
1788
if version in parent_ancestors:
1789
raise errors.GraphCycleError([parent, version])
1790
# ensure this parent will be available later.
1791
new_parents = n2.difference(n1)
1792
needed_versions.update(new_parents.difference(this_versions))
1793
mismatched_versions.add(version)
1795
if not needed_versions and not mismatched_versions:
1797
full_list = topo_sort(self.source.get_graph())
1799
version_list = [i for i in full_list if (not self.target.has_version(i)
1800
and i in needed_versions)]
1804
total = len(version_list)
1805
for version_id in version_list:
1806
pb.update("Converting to knit", count, total)
1807
parents = self.source.get_parents(version_id)
1808
# check that its will be a consistent copy:
1809
for parent in parents:
1810
# if source has the parent, we must already have it
1811
assert (self.target.has_version(parent))
1812
self.target.add_lines(
1813
version_id, parents, self.source.get_lines(version_id))
1816
for version in mismatched_versions:
1817
# FIXME RBC 20060309 is this needed?
1818
n1 = set(self.target.get_parents_with_ghosts(version))
1819
n2 = set(self.source.get_parents(version))
1820
# write a combined record to our history preserving the current
1821
# parents as first in the list
1822
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1823
self.target.fix_parents(version, new_parents)
1829
InterVersionedFile.register_optimiser(WeaveToKnit)
1832
class KnitSequenceMatcher(difflib.SequenceMatcher):
1833
"""Knit tuned sequence matcher.
1835
This is based on profiling of difflib which indicated some improvements
1836
for our usage pattern.
3043
def _get_node(self, key):
3045
return list(self._get_entries([key]))[0]
3047
raise RevisionNotPresent(key, self)
3049
def get_options(self, key):
3050
"""Return a list representing options.
3054
node = self._get_node(key)
3055
options = [self._get_method(node)]
3056
if node[2][0] == 'N':
3057
options.append('no-eol')
3060
def find_ancestry(self, keys):
3061
"""See CombinedGraphIndex.find_ancestry()"""
3062
return self._graph_index.find_ancestry(keys, 0)
3064
def get_parent_map(self, keys):
3065
"""Get a map of the parents of keys.
3067
:param keys: The keys to look up parents for.
3068
:return: A mapping from keys to parents. Absent keys are absent from
3072
nodes = self._get_entries(keys)
3076
result[node[1]] = node[3][0]
3079
result[node[1]] = None
3082
def get_position(self, key):
3083
"""Return details needed to access the version.
3085
:return: a tuple (index, data position, size) to hand to the access
3086
logic to get the record.
3088
node = self._get_node(key)
3089
return self._node_to_position(node)
3091
has_key = _mod_index._has_key_from_parent_map
3094
"""Get all the keys in the collection.
3096
The keys are not ordered.
3099
return [node[1] for node in self._graph_index.iter_all_entries()]
3101
missing_keys = _mod_index._missing_keys_from_parent_map
3103
def _node_to_position(self, node):
3104
"""Convert an index value to position details."""
3105
bits = node[2][1:].split(' ')
3106
return node[0], int(bits[0]), int(bits[1])
3108
def _sort_keys_by_io(self, keys, positions):
3109
"""Figure out an optimal order to read the records for the given keys.
3111
Sort keys, grouped by index and sorted by position.
3113
:param keys: A list of keys whose records we want to read. This will be
3115
:param positions: A dict, such as the one returned by
3116
_get_components_positions()
3119
def get_index_memo(key):
3120
# index_memo is at offset [1]. It is made up of (GraphIndex,
3121
# position, size). GI is an object, which will be unique for each
3122
# pack file. This causes us to group by pack file, then sort by
3123
# position. Size doesn't matter, but it isn't worth breaking up the
3125
return positions[key][1]
3126
return keys.sort(key=get_index_memo)
3128
_get_total_build_size = _get_total_build_size
3131
class _KnitKeyAccess(object):
3132
"""Access to records in .knit files."""
3134
def __init__(self, transport, mapper):
3135
"""Create a _KnitKeyAccess with transport and mapper.
3137
:param transport: The transport the access object is rooted at.
3138
:param mapper: The mapper used to map keys to .knit files.
3140
self._transport = transport
3141
self._mapper = mapper
3143
def add_raw_records(self, key_sizes, raw_data):
3144
"""Add raw knit bytes to a storage area.
3146
The data is spooled to the container writer in one bytes-record per
3149
:param sizes: An iterable of tuples containing the key and size of each
3151
:param raw_data: A bytestring containing the data.
3152
:return: A list of memos to retrieve the record later. Each memo is an
3153
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
3154
length), where the key is the record key.
3156
if type(raw_data) is not str:
3157
raise AssertionError(
3158
'data must be plain bytes was %s' % type(raw_data))
3161
# TODO: This can be tuned for writing to sftp and other servers where
3162
# append() is relatively expensive by grouping the writes to each key
3164
for key, size in key_sizes:
3165
path = self._mapper.map(key)
3167
base = self._transport.append_bytes(path + '.knit',
3168
raw_data[offset:offset+size])
3169
except errors.NoSuchFile:
3170
self._transport.mkdir(osutils.dirname(path))
3171
base = self._transport.append_bytes(path + '.knit',
3172
raw_data[offset:offset+size])
3176
result.append((key, base, size))
3180
"""Flush pending writes on this access object.
3182
For .knit files this is a no-op.
3186
def get_raw_records(self, memos_for_retrieval):
3187
"""Get the raw bytes for a records.
3189
:param memos_for_retrieval: An iterable containing the access memo for
3190
retrieving the bytes.
3191
:return: An iterator over the bytes of the records.
3193
# first pass, group into same-index request to minimise readv's issued.
3195
current_prefix = None
3196
for (key, offset, length) in memos_for_retrieval:
3197
if current_prefix == key[:-1]:
3198
current_list.append((offset, length))
3200
if current_prefix is not None:
3201
request_lists.append((current_prefix, current_list))
3202
current_prefix = key[:-1]
3203
current_list = [(offset, length)]
3204
# handle the last entry
3205
if current_prefix is not None:
3206
request_lists.append((current_prefix, current_list))
3207
for prefix, read_vector in request_lists:
3208
path = self._mapper.map(prefix) + '.knit'
3209
for pos, data in self._transport.readv(path, read_vector):
3213
def annotate_knit(knit, revision_id):
3214
"""Annotate a knit with no cached annotations.
3216
This implementation is for knits with no cached annotations.
3217
It will work for knits with cached annotations, but this is not
1839
def find_longest_match(self, alo, ahi, blo, bhi):
1840
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1842
If isjunk is not defined:
1844
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1845
alo <= i <= i+k <= ahi
1846
blo <= j <= j+k <= bhi
1847
and for all (i',j',k') meeting those conditions,
1850
and if i == i', j <= j'
1852
In other words, of all maximal matching blocks, return one that
1853
starts earliest in a, and of all those maximal matching blocks that
1854
start earliest in a, return the one that starts earliest in b.
1856
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1857
>>> s.find_longest_match(0, 5, 0, 9)
1860
If isjunk is defined, first the longest matching block is
1861
determined as above, but with the additional restriction that no
1862
junk element appears in the block. Then that block is extended as
1863
far as possible by matching (only) junk elements on both sides. So
1864
the resulting block never matches on junk except as identical junk
1865
happens to be adjacent to an "interesting" match.
1867
Here's the same example as before, but considering blanks to be
1868
junk. That prevents " abcd" from matching the " abcd" at the tail
1869
end of the second sequence directly. Instead only the "abcd" can
1870
match, and matches the leftmost "abcd" in the second sequence:
1872
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1873
>>> s.find_longest_match(0, 5, 0, 9)
1876
If no blocks match, return (alo, blo, 0).
1878
>>> s = SequenceMatcher(None, "ab", "c")
1879
>>> s.find_longest_match(0, 2, 0, 1)
3220
annotator = _KnitAnnotator(knit)
3221
return iter(annotator.annotate_flat(revision_id))
3224
class _KnitAnnotator(annotate.Annotator):
3225
"""Build up the annotations for a text."""
3227
def __init__(self, vf):
3228
annotate.Annotator.__init__(self, vf)
3230
# TODO: handle Nodes which cannot be extracted
3231
# self._ghosts = set()
3233
# Map from (key, parent_key) => matching_blocks, should be 'use once'
3234
self._matching_blocks = {}
3236
# KnitContent objects
3237
self._content_objects = {}
3238
# The number of children that depend on this fulltext content object
3239
self._num_compression_children = {}
3240
# Delta records that need their compression parent before they can be
3242
self._pending_deltas = {}
3243
# Fulltext records that are waiting for their parents fulltexts before
3244
# they can be yielded for annotation
3245
self._pending_annotation = {}
3247
self._all_build_details = {}
3249
def _get_build_graph(self, key):
3250
"""Get the graphs for building texts and annotations.
3252
The data you need for creating a full text may be different than the
3253
data you need to annotate that text. (At a minimum, you need both
3254
parents to create an annotation, but only need 1 parent to generate the
3257
:return: A list of (key, index_memo) records, suitable for
3258
passing to read_records_iter to start reading in the raw data from
1883
# CAUTION: stripping common prefix or suffix would be incorrect.
1887
# Longest matching block is "ab", but if common prefix is
1888
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1889
# strip, so ends up claiming that ab is changed to acab by
1890
# inserting "ca" in the middle. That's minimal but unintuitive:
1891
# "it's obvious" that someone inserted "ac" at the front.
1892
# Windiff ends up at the same place as diff, but by pairing up
1893
# the unique 'b's and then matching the first two 'a's.
1895
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1896
besti, bestj, bestsize = alo, blo, 0
1897
# find longest junk-free match
1898
# during an iteration of the loop, j2len[j] = length of longest
1899
# junk-free match ending with a[i-1] and b[j]
1903
for i in xrange(alo, ahi):
1904
# look at all instances of a[i] in b; note that because
1905
# b2j has no junk keys, the loop is skipped if a[i] is junk
1906
j2lenget = j2len.get
1909
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1910
# following improvement
1911
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1912
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1913
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1915
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1916
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1917
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
3261
pending = set([key])
3264
self._num_needed_children[key] = 1
3266
# get all pending nodes
3267
this_iteration = pending
3268
build_details = self._vf._index.get_build_details(this_iteration)
3269
self._all_build_details.update(build_details)
3270
# new_nodes = self._vf._index._get_entries(this_iteration)
3272
for key, details in build_details.iteritems():
3273
(index_memo, compression_parent, parent_keys,
3274
record_details) = details
3275
self._parent_map[key] = parent_keys
3276
self._heads_provider = None
3277
records.append((key, index_memo))
3278
# Do we actually need to check _annotated_lines?
3279
pending.update([p for p in parent_keys
3280
if p not in self._all_build_details])
3282
for parent_key in parent_keys:
3283
if parent_key in self._num_needed_children:
3284
self._num_needed_children[parent_key] += 1
3286
self._num_needed_children[parent_key] = 1
3287
if compression_parent:
3288
if compression_parent in self._num_compression_children:
3289
self._num_compression_children[compression_parent] += 1
3291
self._num_compression_children[compression_parent] = 1
3293
missing_versions = this_iteration.difference(build_details.keys())
3294
if missing_versions:
3295
for key in missing_versions:
3296
if key in self._parent_map and key in self._text_cache:
3297
# We already have this text ready, we just need to
3298
# yield it later so we get it annotated
3300
parent_keys = self._parent_map[key]
3301
for parent_key in parent_keys:
3302
if parent_key in self._num_needed_children:
3303
self._num_needed_children[parent_key] += 1
3305
self._num_needed_children[parent_key] = 1
3306
pending.update([p for p in parent_keys
3307
if p not in self._all_build_details])
3309
raise errors.RevisionNotPresent(key, self._vf)
3310
# Generally we will want to read the records in reverse order, because
3311
# we find the parent nodes after the children
3313
return records, ann_keys
3315
def _get_needed_texts(self, key, pb=None):
3316
# if True or len(self._vf._immediate_fallback_vfs) > 0:
3317
if len(self._vf._immediate_fallback_vfs) > 0:
3318
# If we have fallbacks, go to the generic path
3319
for v in annotate.Annotator._get_needed_texts(self, key, pb=pb):
3324
records, ann_keys = self._get_build_graph(key)
3325
for idx, (sub_key, text, num_lines) in enumerate(
3326
self._extract_texts(records)):
3328
pb.update(gettext('annotating'), idx, len(records))
3329
yield sub_key, text, num_lines
3330
for sub_key in ann_keys:
3331
text = self._text_cache[sub_key]
3332
num_lines = len(text) # bad assumption
3333
yield sub_key, text, num_lines
3335
except errors.RetryWithNewPacks, e:
3336
self._vf._access.reload_or_raise(e)
3337
# The cached build_details are no longer valid
3338
self._all_build_details.clear()
3340
def _cache_delta_blocks(self, key, compression_parent, delta, lines):
3341
parent_lines = self._text_cache[compression_parent]
3342
blocks = list(KnitContent.get_line_delta_blocks(delta, parent_lines, lines))
3343
self._matching_blocks[(key, compression_parent)] = blocks
3345
def _expand_record(self, key, parent_keys, compression_parent, record,
3348
if compression_parent:
3349
if compression_parent not in self._content_objects:
3350
# Waiting for the parent
3351
self._pending_deltas.setdefault(compression_parent, []).append(
3352
(key, parent_keys, record, record_details))
3354
# We have the basis parent, so expand the delta
3355
num = self._num_compression_children[compression_parent]
3358
base_content = self._content_objects.pop(compression_parent)
3359
self._num_compression_children.pop(compression_parent)
1929
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1931
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1934
# Extend the best by non-junk elements on each end. In particular,
1935
# "popular" non-junk elements aren't in b2j, which greatly speeds
1936
# the inner loop above, but also means "the best" match so far
1937
# doesn't contain any junk *or* popular non-junk elements.
1938
while besti > alo and bestj > blo and \
1939
not isbjunk(b[bestj-1]) and \
1940
a[besti-1] == b[bestj-1]:
1941
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1942
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1943
not isbjunk(b[bestj+bestsize]) and \
1944
a[besti+bestsize] == b[bestj+bestsize]:
1947
# Now that we have a wholly interesting match (albeit possibly
1948
# empty!), we may as well suck up the matching junk on each
1949
# side of it too. Can't think of a good reason not to, and it
1950
# saves post-processing the (possibly considerable) expense of
1951
# figuring out what to do with it. In the case of an empty
1952
# interesting match, this is clearly the right thing to do,
1953
# because no other kind of match is possible in the regions.
1954
while besti > alo and bestj > blo and \
1955
isbjunk(b[bestj-1]) and \
1956
a[besti-1] == b[bestj-1]:
1957
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1958
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1959
isbjunk(b[bestj+bestsize]) and \
1960
a[besti+bestsize] == b[bestj+bestsize]:
1961
bestsize = bestsize + 1
1963
return besti, bestj, bestsize
3361
self._num_compression_children[compression_parent] = num
3362
base_content = self._content_objects[compression_parent]
3363
# It is tempting to want to copy_base_content=False for the last
3364
# child object. However, whenever noeol=False,
3365
# self._text_cache[parent_key] is content._lines. So mutating it
3366
# gives very bad results.
3367
# The alternative is to copy the lines into text cache, but then we
3368
# are copying anyway, so just do it here.
3369
content, delta = self._vf._factory.parse_record(
3370
key, record, record_details, base_content,
3371
copy_base_content=True)
3374
content, _ = self._vf._factory.parse_record(
3375
key, record, record_details, None)
3376
if self._num_compression_children.get(key, 0) > 0:
3377
self._content_objects[key] = content
3378
lines = content.text()
3379
self._text_cache[key] = lines
3380
if delta is not None:
3381
self._cache_delta_blocks(key, compression_parent, delta, lines)
3384
def _get_parent_annotations_and_matches(self, key, text, parent_key):
3385
"""Get the list of annotations for the parent, and the matching lines.
3387
:param text: The opaque value given by _get_needed_texts
3388
:param parent_key: The key for the parent text
3389
:return: (parent_annotations, matching_blocks)
3390
parent_annotations is a list as long as the number of lines in
3392
matching_blocks is a list of (parent_idx, text_idx, len) tuples
3393
indicating which lines match between the two texts
3395
block_key = (key, parent_key)
3396
if block_key in self._matching_blocks:
3397
blocks = self._matching_blocks.pop(block_key)
3398
parent_annotations = self._annotations_cache[parent_key]
3399
return parent_annotations, blocks
3400
return annotate.Annotator._get_parent_annotations_and_matches(self,
3401
key, text, parent_key)
3403
def _process_pending(self, key):
3404
"""The content for 'key' was just processed.
3406
Determine if there is any more pending work to be processed.
3409
if key in self._pending_deltas:
3410
compression_parent = key
3411
children = self._pending_deltas.pop(key)
3412
for child_key, parent_keys, record, record_details in children:
3413
lines = self._expand_record(child_key, parent_keys,
3415
record, record_details)
3416
if self._check_ready_for_annotations(child_key, parent_keys):
3417
to_return.append(child_key)
3418
# Also check any children that are waiting for this parent to be
3420
if key in self._pending_annotation:
3421
children = self._pending_annotation.pop(key)
3422
to_return.extend([c for c, p_keys in children
3423
if self._check_ready_for_annotations(c, p_keys)])
3426
def _check_ready_for_annotations(self, key, parent_keys):
3427
"""return true if this text is ready to be yielded.
3429
Otherwise, this will return False, and queue the text into
3430
self._pending_annotation
3432
for parent_key in parent_keys:
3433
if parent_key not in self._annotations_cache:
3434
# still waiting on at least one parent text, so queue it up
3435
# Note that if there are multiple parents, we need to wait
3437
self._pending_annotation.setdefault(parent_key,
3438
[]).append((key, parent_keys))
3442
def _extract_texts(self, records):
3443
"""Extract the various texts needed based on records"""
3444
# We iterate in the order read, rather than a strict order requested
3445
# However, process what we can, and put off to the side things that
3446
# still need parents, cleaning them up when those parents are
3449
# 1) As 'records' are read, see if we can expand these records into
3450
# Content objects (and thus lines)
3451
# 2) If a given line-delta is waiting on its compression parent, it
3452
# gets queued up into self._pending_deltas, otherwise we expand
3453
# it, and put it into self._text_cache and self._content_objects
3454
# 3) If we expanded the text, we will then check to see if all
3455
# parents have also been processed. If so, this text gets yielded,
3456
# else this record gets set aside into pending_annotation
3457
# 4) Further, if we expanded the text in (2), we will then check to
3458
# see if there are any children in self._pending_deltas waiting to
3459
# also be processed. If so, we go back to (2) for those
3460
# 5) Further again, if we yielded the text, we can then check if that
3461
# 'unlocks' any of the texts in pending_annotations, which should
3462
# then get yielded as well
3463
# Note that both steps 4 and 5 are 'recursive' in that unlocking one
3464
# compression child could unlock yet another, and yielding a fulltext
3465
# will also 'unlock' the children that are waiting on that annotation.
3466
# (Though also, unlocking 1 parent's fulltext, does not unlock a child
3467
# if other parents are also waiting.)
3468
# We want to yield content before expanding child content objects, so
3469
# that we know when we can re-use the content lines, and the annotation
3470
# code can know when it can stop caching fulltexts, as well.
3472
# Children that are missing their compression parent
3474
for (key, record, digest) in self._vf._read_records_iter(records):
3476
details = self._all_build_details[key]
3477
(_, compression_parent, parent_keys, record_details) = details
3478
lines = self._expand_record(key, parent_keys, compression_parent,
3479
record, record_details)
3481
# Pending delta should be queued up
3483
# At this point, we may be able to yield this content, if all
3484
# parents are also finished
3485
yield_this_text = self._check_ready_for_annotations(key,
3488
# All parents present
3489
yield key, lines, len(lines)
3490
to_process = self._process_pending(key)
3492
this_process = to_process
3494
for key in this_process:
3495
lines = self._text_cache[key]
3496
yield key, lines, len(lines)
3497
to_process.extend(self._process_pending(key))
3500
from bzrlib._knit_load_data_pyx import _load_data_c as _load_data
3501
except ImportError, e:
3502
osutils.failed_to_load_extension(e)
3503
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
