155
273
internal representation is of the format:
156
274
(revid, plaintext)
160
origin, text = line.split(' ', 1)
161
lines.append((origin.decode('utf-8'), text))
162
return KnitContent(lines)
276
# TODO: jam 20070209 The tests expect this to be returned as tuples,
277
# but the code itself doesn't really depend on that.
278
# Figure out a way to not require the overhead of turning the
279
# list back into tuples.
280
lines = [tuple(line.split(' ', 1)) for line in content]
281
return AnnotatedKnitContent(lines)
164
283
def parse_line_delta_iter(self, lines):
165
for result_item in self.parse_line_delta[lines]:
284
return iter(self.parse_line_delta(lines))
168
def parse_line_delta(self, lines, version):
286
def parse_line_delta(self, lines, version_id, plain=False):
169
287
"""Convert a line based delta into internal representation.
171
289
line delta is in the form of:
172
290
intstart intend intcount
174
292
revid(utf8) newline\n
175
internal represnetation is
293
internal representation is
176
294
(start, end, count, [1..count tuples (revid, newline)])
296
:param plain: If True, the lines are returned as a plain
297
list without annotations, not as a list of (origin, content) tuples, i.e.
298
(start, end, count, [1..count newline])
179
301
lines = iter(lines)
180
302
next = lines.next
305
def cache_and_return(line):
306
origin, text = line.split(' ', 1)
307
return cache.setdefault(origin, origin), text
181
309
# walk through the lines parsing.
310
# Note that the plain test is explicitly pulled out of the
311
# loop to minimise any performance impact
314
start, end, count = [int(n) for n in header.split(',')]
315
contents = [next().split(' ', 1)[1] for i in xrange(count)]
316
result.append((start, end, count, contents))
319
start, end, count = [int(n) for n in header.split(',')]
320
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
321
result.append((start, end, count, contents))
324
def get_fulltext_content(self, lines):
325
"""Extract just the content lines from a fulltext."""
326
return (line.split(' ', 1)[1] for line in lines)
328
def get_linedelta_content(self, lines):
329
"""Extract just the content from a line delta.
331
This doesn't return all of the extra information stored in a delta.
332
Only the actual content lines.
182
336
for header in lines:
183
start, end, count = [int(n) for n in header.split(',')]
337
header = header.split(',')
338
count = int(header[2])
339
for i in xrange(count):
187
340
origin, text = next().split(' ', 1)
189
contents.append((origin.decode('utf-8'), text))
190
result.append((start, end, count, contents))
193
343
def lower_fulltext(self, content):
194
344
"""convert a fulltext content record into a serializable form.
196
346
see parse_fulltext which this inverts.
198
return ['%s %s' % (o.encode('utf-8'), t) for o, t in content._lines]
348
# TODO: jam 20070209 We only do the caching thing to make sure that
349
# the origin is a valid utf-8 line, eventually we could remove it
350
return ['%s %s' % (o, t) for o, t in content._lines]
200
352
def lower_line_delta(self, delta):
201
353
"""convert a delta into a serializable form.
203
355
See parse_line_delta which this inverts.
357
# TODO: jam 20070209 We only do the caching thing to make sure that
358
# the origin is a valid utf-8 line, eventually we could remove it
206
360
for start, end, c, lines in delta:
207
361
out.append('%d,%d,%d\n' % (start, end, c))
208
for origin, text in lines:
209
out.append('%s %s' % (origin.encode('utf-8'), text))
362
out.extend(origin + ' ' + text
363
for origin, text in lines)
213
class KnitPlainFactory(_KnitFactory):
366
def annotate_iter(self, knit, version_id):
367
content = knit._get_content(version_id)
368
return content.annotate_iter()
371
class KnitPlainFactory(object):
214
372
"""Factory for creating plain Content objects."""
216
374
annotated = False
218
def parse_fulltext(self, content, version):
376
def make(self, lines, version_id):
377
return PlainKnitContent(lines, version_id)
379
def parse_fulltext(self, content, version_id):
219
380
"""This parses an unannotated fulltext.
221
382
Note that this is not a noop - the internal representation
222
383
has (versionid, line) - its just a constant versionid.
224
return self.make(content, version)
385
return self.make(content, version_id)
226
def parse_line_delta_iter(self, lines, version):
228
header = lines.pop(0)
387
def parse_line_delta_iter(self, lines, version_id):
389
num_lines = len(lines)
390
while cur < num_lines:
229
393
start, end, c = [int(n) for n in header.split(',')]
230
yield start, end, c, zip([version] * c, lines[:c])
233
def parse_line_delta(self, lines, version):
234
return list(self.parse_line_delta_iter(lines, version))
394
yield start, end, c, lines[cur:cur+c]
397
def parse_line_delta(self, lines, version_id):
398
return list(self.parse_line_delta_iter(lines, version_id))
400
def get_fulltext_content(self, lines):
401
"""Extract just the content lines from a fulltext."""
404
def get_linedelta_content(self, lines):
405
"""Extract just the content from a line delta.
407
This doesn't return all of the extra information stored in a delta.
408
Only the actual content lines.
413
header = header.split(',')
414
count = int(header[2])
415
for i in xrange(count):
236
418
def lower_fulltext(self, content):
237
419
return content.text()
265
449
stored and retrieved.
268
def __init__(self, relpath, transport, file_mode=None, access_mode=None, factory=None,
269
basis_knit=None, delta=True, create=False):
452
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
453
factory=None, delta=True, create=False, create_parent_dir=False,
454
delay_create=False, dir_mode=None, index=None, access_method=None):
270
455
"""Construct a knit at location specified by relpath.
272
457
:param create: If not True, only open an existing knit.
458
:param create_parent_dir: If True, create the parent directory if
459
creating the file fails. (This is used for stores with
460
hash-prefixes that may not exist yet)
461
:param delay_create: The calling code is aware that the knit won't
462
actually be created until the first data is stored.
463
:param index: An index to use for the knit.
274
465
if access_mode is None:
275
466
access_mode = 'w'
276
467
super(KnitVersionedFile, self).__init__(access_mode)
277
468
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
278
assert not basis_knit or isinstance(basis_knit, KnitVersionedFile), \
281
469
self.transport = transport
282
470
self.filename = relpath
283
self.basis_knit = basis_knit
284
471
self.factory = factory or KnitAnnotateFactory()
285
472
self.writable = (access_mode == 'w')
286
473
self.delta = delta
288
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
289
access_mode, create=create, file_mode=file_mode)
290
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
291
access_mode, create=create and not len(self), file_mode=file_mode)
475
self._max_delta_chain = 200
478
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
479
access_mode, create=create, file_mode=file_mode,
480
create_parent_dir=create_parent_dir, delay_create=delay_create,
484
if access_method is None:
485
_access = _KnitAccess(transport, relpath + DATA_SUFFIX, file_mode, dir_mode,
486
((create and not len(self)) and delay_create), create_parent_dir)
488
_access = access_method
489
if create and not len(self) and not delay_create:
491
self._data = _KnitData(_access)
293
493
def __repr__(self):
294
return '%s(%s)' % (self.__class__.__name__,
494
return '%s(%s)' % (self.__class__.__name__,
295
495
self.transport.abspath(self.filename))
297
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
298
"""See VersionedFile._add_delta()."""
299
self._check_add(version_id, []) # should we check the lines ?
300
self._check_versions_present(parents)
304
for parent in parents:
305
if not self.has_version(parent):
306
ghosts.append(parent)
308
present_parents.append(parent)
310
if delta_parent is None:
311
# reconstitute as full text.
312
assert len(delta) == 1 or len(delta) == 0
314
assert delta[0][0] == 0
315
assert delta[0][1] == 0, delta[0][1]
316
return super(KnitVersionedFile, self)._add_delta(version_id,
327
options.append('no-eol')
329
if delta_parent is not None:
330
# determine the current delta chain length.
331
# To speed the extract of texts the delta chain is limited
332
# to a fixed number of deltas. This should minimize both
333
# I/O and the time spend applying deltas.
335
delta_parents = [delta_parent]
337
parent = delta_parents[0]
338
method = self._index.get_method(parent)
339
if method == 'fulltext':
341
delta_parents = self._index.get_parents(parent)
343
if method == 'line-delta':
344
# did not find a fulltext in the delta limit.
345
# just do a normal insertion.
346
return super(KnitVersionedFile, self)._add_delta(version_id,
353
options.append('line-delta')
354
store_lines = self.factory.lower_line_delta(delta)
356
where, size = self._data.add_record(version_id, digest, store_lines)
357
self._index.add_version(version_id, options, where, size, parents)
497
def _check_should_delta(self, first_parents):
498
"""Iterate back through the parent listing, looking for a fulltext.
500
This is used when we want to decide whether to add a delta or a new
501
fulltext. It searches for _max_delta_chain parents. When it finds a
502
fulltext parent, it sees if the total size of the deltas leading up to
503
it is large enough to indicate that we want a new full text anyway.
505
Return True if we should create a new delta, False if we should use a
510
delta_parents = first_parents
511
for count in xrange(self._max_delta_chain):
512
parent = delta_parents[0]
513
method = self._index.get_method(parent)
514
index, pos, size = self._index.get_position(parent)
515
if method == 'fulltext':
519
delta_parents = self._index.get_parents(parent)
521
# We couldn't find a fulltext, so we must create a new one
524
return fulltext_size > delta_size
359
526
def _add_raw_records(self, records, data):
360
527
"""Add all the records 'records' with data pre-joined in 'data'.
380
556
"""See VersionedFile.copy_to()."""
381
557
# copy the current index to a temp index to avoid racing with local
383
transport.put(name + INDEX_SUFFIX + '.tmp', self.transport.get(self._index._filename),)
559
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
560
self.transport.get(self._index._filename))
384
561
# copy the data file
385
transport.put(name + DATA_SUFFIX, self._data._open_file())
386
# rename the copied index into place
387
transport.rename(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
562
f = self._data._open_file()
564
transport.put_file(name + DATA_SUFFIX, f)
567
# move the copied index into place
568
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
389
570
def create_empty(self, name, transport, mode=None):
390
return KnitVersionedFile(name, transport, factory=self.factory, delta=self.delta, create=True)
571
return KnitVersionedFile(name, transport, factory=self.factory,
572
delta=self.delta, create=True)
392
def _fix_parents(self, version, new_parents):
393
"""Fix the parents list for version.
574
def get_data_stream(self, required_versions):
575
"""Get a data stream for the specified versions.
577
Versions may be returned in any order, not necessarily the order
578
specified. They are returned in a partial order by compression
579
parent, so that the deltas can be applied as the data stream is
580
inserted; however note that compression parents will not be sent
581
unless they were specifically requested, as the client may already
584
:param required_versions: The exact set of versions to be extracted.
585
Unlike some other knit methods, this is not used to generate a
586
transitive closure, rather it is used precisely as given.
395
This is done by appending a new version to the index
396
with identical data except for the parents list.
397
the parents list must be a superset of the current
588
:returns: format_signature, list of (version, options, length, parents),
400
current_values = self._index._cache[version]
401
assert set(current_values[4]).difference(set(new_parents)) == set()
402
self._index.add_version(version,
591
required_version_set = frozenset(required_versions)
593
# list of revisions that can just be sent without waiting for their
596
# map from revision to the children based on it
598
# first, read all relevant index data, enough to sort into the right
600
for version_id in required_versions:
601
options = self._index.get_options(version_id)
602
parents = self._index.get_parents_with_ghosts(version_id)
603
index_memo = self._index.get_position(version_id)
604
version_index[version_id] = (index_memo, options, parents)
605
if ('line-delta' in options
606
and parents[0] in required_version_set):
607
# must wait until the parent has been sent
608
deferred.setdefault(parents[0], []). \
611
# either a fulltext, or a delta whose parent the client did
612
# not ask for and presumably already has
613
ready_to_send.append(version_id)
614
# build a list of results to return, plus instructions for data to
616
copy_queue_records = []
617
temp_version_list = []
619
# XXX: pushing and popping lists may be a bit inefficient
620
version_id = ready_to_send.pop(0)
621
(index_memo, options, parents) = version_index[version_id]
622
copy_queue_records.append((version_id, index_memo))
623
none, data_pos, data_size = index_memo
624
temp_version_list.append((version_id, options, data_size,
626
if version_id in deferred:
627
# now we can send all the children of this revision - we could
628
# put them in anywhere, but we hope that sending them soon
629
# after the fulltext will give good locality in the receiver
630
ready_to_send[:0] = deferred.pop(version_id)
631
assert len(deferred) == 0, \
632
"Still have compressed child versions waiting to be sent"
633
# XXX: The stream format is such that we cannot stream it - we have to
634
# know the length of all the data a-priori.
636
result_version_list = []
637
for (version_id, raw_data), \
638
(version_id2, options, _, parents) in \
639
izip(self._data.read_records_iter_raw(copy_queue_records),
641
assert version_id == version_id2, \
642
'logic error, inconsistent results'
643
raw_datum.append(raw_data)
644
result_version_list.append(
645
(version_id, options, len(raw_data), parents))
646
# provide a callback to get data incrementally.
647
pseudo_file = StringIO(''.join(raw_datum))
650
return pseudo_file.read()
652
return pseudo_file.read(length)
653
return (self.get_format_signature(), result_version_list, read)
655
def _extract_blocks(self, version_id, source, target):
656
if self._index.get_method(version_id) != 'line-delta':
658
parent, sha1, noeol, delta = self.get_delta(version_id)
659
return KnitContent.get_line_delta_blocks(delta, source, target)
408
661
def get_delta(self, version_id):
409
662
"""Get a delta for constructing version from some other version."""
410
if not self.has_version(version_id):
411
raise RevisionNotPresent(version_id, self.filename)
663
self.check_not_reserved_id(version_id)
413
664
parents = self.get_parents(version_id)
415
666
parent = parents[0]
418
data_pos, data_size = self._index.get_position(version_id)
419
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
420
version_idx = self._index.lookup(version_id)
669
index_memo = self._index.get_position(version_id)
670
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
421
671
noeol = 'no-eol' in self._index.get_options(version_id)
422
672
if 'fulltext' == self._index.get_method(version_id):
423
new_content = self.factory.parse_fulltext(data, version_idx)
673
new_content = self.factory.parse_fulltext(data, version_id)
424
674
if parent is not None:
425
675
reference_content = self._get_content(parent)
426
676
old_texts = reference_content.text()
429
679
new_texts = new_content.text()
430
delta_seq = SequenceMatcher(None, old_texts, new_texts)
680
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
431
682
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
433
delta = self.factory.parse_line_delta(data, version_idx)
684
delta = self.factory.parse_line_delta(data, version_id)
434
685
return parent, sha1, noeol, delta
687
def get_format_signature(self):
688
"""See VersionedFile.get_format_signature()."""
689
if self.factory.annotated:
690
annotated_part = "annotated"
692
annotated_part = "plain"
693
return "knit-%s" % (annotated_part,)
436
695
def get_graph_with_ghosts(self):
437
696
"""See VersionedFile.get_graph_with_ghosts()."""
728
def insert_data_stream(self, (format, data_list, reader_callable)):
729
"""Insert knit records from a data stream into this knit.
731
If a version in the stream is already present in this knit, it will not
732
be inserted a second time. It will be checked for consistency with the
733
stored version however, and may cause a KnitCorrupt error to be raised
734
if the data in the stream disagrees with the already stored data.
736
:seealso: get_data_stream
738
if format != self.get_format_signature():
739
if 'knit' in debug.debug_flags:
741
'incompatible format signature inserting to %r', self)
742
source = self._knit_from_datastream(
743
(format, data_list, reader_callable))
747
for version_id, options, length, parents in data_list:
748
if self.has_version(version_id):
749
# First check: the list of parents.
750
my_parents = self.get_parents_with_ghosts(version_id)
751
if my_parents != parents:
752
# XXX: KnitCorrupt is not quite the right exception here.
755
'parents list %r from data stream does not match '
756
'already recorded parents %r for %s'
757
% (parents, my_parents, version_id))
759
# Also check the SHA-1 of the fulltext this content will
761
raw_data = reader_callable(length)
762
my_fulltext_sha1 = self.get_sha1(version_id)
763
df, rec = self._data._parse_record_header(version_id, raw_data)
764
stream_fulltext_sha1 = rec[3]
765
if my_fulltext_sha1 != stream_fulltext_sha1:
766
# Actually, we don't know if it's this knit that's corrupt,
767
# or the data stream we're trying to insert.
769
self.filename, 'sha-1 does not match %s' % version_id)
771
if 'line-delta' in options:
772
# Make sure that this knit record is actually useful: a
773
# line-delta is no use unless we have its parent.
774
# Fetching from a broken repository with this problem
775
# shouldn't break the target repository.
777
# See https://bugs.launchpad.net/bzr/+bug/164443
778
if not self._index.has_version(parents[0]):
781
'line-delta from stream '
784
'missing parent %s\n'
785
'Try running "bzr check" '
786
'on the source repository, and "bzr reconcile" '
788
(version_id, parents[0]))
789
self._add_raw_records(
790
[(version_id, options, parents, length)],
791
reader_callable(length))
793
def _knit_from_datastream(self, (format, data_list, reader_callable)):
794
"""Create a knit object from a data stream.
796
This method exists to allow conversion of data streams that do not
797
match the signature of this knit. Generally it will be slower and use
798
more memory to use this method to insert data, but it will work.
800
:seealso: get_data_stream for details on datastreams.
801
:return: A knit versioned file which can be used to join the datastream
804
if format == "knit-plain":
805
factory = KnitPlainFactory()
806
elif format == "knit-annotated":
807
factory = KnitAnnotateFactory()
809
raise errors.KnitDataStreamUnknown(format)
810
index = _StreamIndex(data_list)
811
access = _StreamAccess(reader_callable, index, self, factory)
812
return KnitVersionedFile(self.filename, self.transport,
813
factory=factory, index=index, access_method=access)
465
815
def versions(self):
466
816
"""See VersionedFile.versions."""
817
if 'evil' in debug.debug_flags:
818
trace.mutter_callsite(2, "versions scales with size of history")
467
819
return self._index.get_versions()
469
821
def has_version(self, version_id):
470
822
"""See VersionedFile.has_version."""
823
if 'evil' in debug.debug_flags:
824
trace.mutter_callsite(2, "has_version is a LBYL scenario")
471
825
return self._index.has_version(version_id)
473
827
__contains__ = has_version
475
829
def _merge_annotations(self, content, parents, parent_texts={},
476
delta=None, annotated=None):
830
delta=None, annotated=None,
831
left_matching_blocks=None):
477
832
"""Merge annotations for content. This is done by comparing
478
833
the annotations based on changed to the text.
835
if left_matching_blocks is not None:
836
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
482
840
for parent_id in parents:
483
841
merge_content = self._get_content(parent_id, parent_texts)
484
seq = SequenceMatcher(None, merge_content.text(), content.text())
485
if delta_seq is None:
486
# setup a delta seq to reuse.
842
if (parent_id == parents[0] and delta_seq is not None):
845
seq = patiencediff.PatienceSequenceMatcher(
846
None, merge_content.text(), content.text())
488
847
for i, j, n in seq.get_matching_blocks():
491
# this appears to copy (origin, text) pairs across to the new
492
# content for any line that matches the last-checked parent.
493
# FIXME: save the sequence control data for delta compression
494
# against the most relevant parent rather than rediffing.
850
# this appears to copy (origin, text) pairs across to the
851
# new content for any line that matches the last-checked
495
853
content._lines[j:j+n] = merge_content._lines[i:i+n]
855
if delta_seq is None:
498
856
reference_content = self._get_content(parents[0], parent_texts)
499
857
new_texts = content.text()
500
858
old_texts = reference_content.text()
501
delta_seq = SequenceMatcher(None, old_texts, new_texts)
859
delta_seq = patiencediff.PatienceSequenceMatcher(
860
None, old_texts, new_texts)
502
861
return self._make_line_delta(delta_seq, content)
504
863
def _make_line_delta(self, delta_seq, new_content):
510
869
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
511
870
return diff_hunks
513
def _get_components(self, version_id):
514
"""Return a list of (version_id, method, data) tuples that
515
makes up version specified by version_id of the knit.
517
The components should be applied in the order of the returned
520
The basis knit will be used to the largest extent possible
521
since it is assumed that accesses to it is faster.
872
def _get_components_positions(self, version_ids):
873
"""Produce a map of position data for the components of versions.
875
This data is intended to be used for retrieving the knit records.
877
A dict of version_id to (method, data_pos, data_size, next) is
879
method is the way referenced data should be applied.
880
data_pos is the position of the data in the knit.
881
data_size is the size of the data in the knit.
882
next is the build-parent of the version, or None for fulltexts.
524
# 4168 calls in 14912, 2289 internal
525
# 4168 in 9711 to read_records
526
# 52554 in 1250 to get_parents
527
# 170166 in 865 to list.append
529
# needed_revisions holds a list of (method, version_id) of
530
# versions that is needed to be fetched to construct the final
531
# version of the file.
533
# basis_revisions is a list of versions that needs to be
534
# fetched but exists in the basis knit.
536
basis = self.basis_knit
543
if basis and basis._index.has_version(cursor):
545
basis_versions.append(cursor)
546
method = picked_knit._index.get_method(cursor)
547
needed_versions.append((method, cursor))
548
if method == 'fulltext':
550
cursor = picked_knit.get_parents(cursor)[0]
555
for comp_id in basis_versions:
556
data_pos, data_size = basis._index.get_data_position(comp_id)
557
records.append((piece_id, data_pos, data_size))
558
components.update(basis._data.read_records(records))
561
for comp_id in [vid for method, vid in needed_versions
562
if vid not in basis_versions]:
563
data_pos, data_size = self._index.get_position(comp_id)
564
records.append((comp_id, data_pos, data_size))
565
components.update(self._data.read_records(records))
567
# get_data_records returns a mapping with the version id as
568
# index and the value as data. The order the components need
569
# to be applied is held by needed_versions (reversed).
571
for method, comp_id in reversed(needed_versions):
572
out.append((comp_id, method, components[comp_id]))
885
for version_id in version_ids:
888
while cursor is not None and cursor not in component_data:
889
method = self._index.get_method(cursor)
890
if method == 'fulltext':
893
next = self.get_parents_with_ghosts(cursor)[0]
894
index_memo = self._index.get_position(cursor)
895
component_data[cursor] = (method, index_memo, next)
897
return component_data
576
899
def _get_content(self, version_id, parent_texts={}):
577
900
"""Returns a content object that makes up the specified
579
if not self.has_version(version_id):
580
raise RevisionNotPresent(version_id, self.filename)
582
902
cached_version = parent_texts.get(version_id, None)
583
903
if cached_version is not None:
904
if not self.has_version(version_id):
905
raise RevisionNotPresent(version_id, self.filename)
584
906
return cached_version
586
if self.basis_knit and version_id in self.basis_knit:
587
return self.basis_knit._get_content(version_id)
590
components = self._get_components(version_id)
591
for component_id, method, (data, digest) in components:
592
version_idx = self._index.lookup(component_id)
593
if method == 'fulltext':
594
assert content is None
595
content = self.factory.parse_fulltext(data, version_idx)
596
elif method == 'line-delta':
597
delta = self.factory.parse_line_delta(data, version_idx)
598
content._lines = self._apply_delta(content._lines, delta)
600
if 'no-eol' in self._index.get_options(version_id):
601
line = content._lines[-1][1].rstrip('\n')
602
content._lines[-1] = (content._lines[-1][0], line)
604
# digest here is the digest from the last applied component.
605
if sha_strings(content.text()) != digest:
606
import pdb;pdb.set_trace()
607
raise KnitCorrupt(self.filename, 'sha-1 does not match %s' % version_id)
908
text_map, contents_map = self._get_content_maps([version_id])
909
return contents_map[version_id]
611
911
def _check_versions_present(self, version_ids):
612
912
"""Check that all specified versions are present."""
613
version_ids = set(version_ids)
614
for r in list(version_ids):
615
if self._index.has_version(r):
616
version_ids.remove(r)
618
raise RevisionNotPresent(list(version_ids)[0], self.filename)
913
self._index.check_versions_present(version_ids)
620
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
915
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
916
nostore_sha, random_id, check_content):
621
917
"""See VersionedFile.add_lines_with_ghosts()."""
622
self._check_add(version_id, lines)
623
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
918
self._check_add(version_id, lines, random_id, check_content)
919
return self._add(version_id, lines, parents, self.delta,
920
parent_texts, None, nostore_sha, random_id)
625
def _add_lines(self, version_id, parents, lines, parent_texts):
922
def _add_lines(self, version_id, parents, lines, parent_texts,
923
left_matching_blocks, nostore_sha, random_id, check_content):
626
924
"""See VersionedFile.add_lines."""
627
self._check_add(version_id, lines)
925
self._check_add(version_id, lines, random_id, check_content)
628
926
self._check_versions_present(parents)
629
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
927
return self._add(version_id, lines[:], parents, self.delta,
928
parent_texts, left_matching_blocks, nostore_sha, random_id)
631
def _check_add(self, version_id, lines):
930
def _check_add(self, version_id, lines, random_id, check_content):
632
931
"""check that version_id and lines are safe to add."""
633
assert self.writable, "knit is not opened for write"
634
### FIXME escape. RBC 20060228
635
932
if contains_whitespace(version_id):
636
933
raise InvalidRevisionId(version_id, self.filename)
637
if self.has_version(version_id):
934
self.check_not_reserved_id(version_id)
935
# Technically this could be avoided if we are happy to allow duplicate
936
# id insertion when other things than bzr core insert texts, but it
937
# seems useful for folk using the knit api directly to have some safety
938
# blanket that we can disable.
939
if not random_id and self.has_version(version_id):
638
940
raise RevisionAlreadyPresent(version_id, self.filename)
639
self._check_lines_not_unicode(lines)
640
self._check_lines_are_lines(lines)
942
self._check_lines_not_unicode(lines)
943
self._check_lines_are_lines(lines)
642
def _add(self, version_id, lines, parents, delta, parent_texts):
945
def _add(self, version_id, lines, parents, delta, parent_texts,
946
left_matching_blocks, nostore_sha, random_id):
643
947
"""Add a set of lines on top of version specified by parents.
645
949
If delta is true, compress the text as a line-delta against
648
952
Any versions not present will be converted into ghosts.
650
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
651
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
652
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
653
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
654
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
655
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
656
# +1383 0 8.0370 8.0370 +<len>
657
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
658
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
659
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
660
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
954
# first thing, if the content is something we don't need to store, find
956
line_bytes = ''.join(lines)
957
digest = sha_string(line_bytes)
958
if nostore_sha == digest:
959
raise errors.ExistingContent
662
961
present_parents = []
664
962
if parent_texts is None:
665
963
parent_texts = {}
666
964
for parent in parents:
667
if not self.has_version(parent):
668
ghosts.append(parent)
965
if self.has_version(parent):
670
966
present_parents.append(parent)
672
if delta and not len(present_parents):
968
# can only compress against the left most present parent.
970
(len(present_parents) == 0 or
971
present_parents[0] != parents[0])):
675
digest = sha_strings(lines)
974
text_length = len(line_bytes)
678
977
if lines[-1][-1] != '\n':
978
# copy the contents of lines.
679
980
options.append('no-eol')
680
981
lines[-1] = lines[-1] + '\n'
682
if len(present_parents) and delta:
683
985
# To speed the extract of texts the delta chain is limited
684
986
# to a fixed number of deltas. This should minimize both
685
987
# I/O and the time spend applying deltas.
687
delta_parents = present_parents
689
parent = delta_parents[0]
690
method = self._index.get_method(parent)
691
if method == 'fulltext':
693
delta_parents = self._index.get_parents(parent)
695
if method == 'line-delta':
988
delta = self._check_should_delta(present_parents)
698
lines = self.factory.make(lines, version_id)
990
assert isinstance(version_id, str)
991
content = self.factory.make(lines, version_id)
699
992
if delta or (self.factory.annotated and len(present_parents) > 0):
700
# Merge annotations from parent texts if so is needed.
701
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
702
delta, self.factory.annotated)
993
# Merge annotations from parent texts if needed.
994
delta_hunks = self._merge_annotations(content, present_parents,
995
parent_texts, delta, self.factory.annotated,
996
left_matching_blocks)
705
999
options.append('line-delta')
706
1000
store_lines = self.factory.lower_line_delta(delta_hunks)
1001
size, bytes = self._data._record_to_data(version_id, digest,
708
1004
options.append('fulltext')
709
store_lines = self.factory.lower_fulltext(lines)
1005
# isinstance is slower and we have no hierarchy.
1006
if self.factory.__class__ == KnitPlainFactory:
1007
# Use the already joined bytes saving iteration time in
1009
size, bytes = self._data._record_to_data(version_id, digest,
1010
lines, [line_bytes])
1012
# get mixed annotation + content and feed it into the
1014
store_lines = self.factory.lower_fulltext(content)
1015
size, bytes = self._data._record_to_data(version_id, digest,
711
where, size = self._data.add_record(version_id, digest, store_lines)
712
self._index.add_version(version_id, options, where, size, parents)
1018
access_memo = self._data.add_raw_records([size], bytes)[0]
1019
self._index.add_versions(
1020
((version_id, options, access_memo, parents),),
1021
random_id=random_id)
1022
return digest, text_length, content
715
1024
def check(self, progress_bar=None):
716
1025
"""See VersionedFile.check()."""
718
1027
def _clone_text(self, new_version_id, old_version_id, parents):
719
1028
"""See VersionedFile.clone_text()."""
720
# FIXME RBC 20060228 make fast by only inserting an index with null delta.
1029
# FIXME RBC 20060228 make fast by only inserting an index with null
721
1031
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
723
1033
def get_lines(self, version_id):
724
1034
"""See VersionedFile.get_lines()."""
725
return self._get_content(version_id).text()
727
def iter_lines_added_or_present_in_versions(self, version_ids=None):
1035
return self.get_line_list([version_id])[0]
1037
def _get_record_map(self, version_ids):
1038
"""Produce a dictionary of knit records.
1040
The keys are version_ids, the values are tuples of (method, content,
1042
method is the way the content should be applied.
1043
content is a KnitContent object.
1044
digest is the SHA1 digest of this version id after all steps are done
1045
next is the build-parent of the version, i.e. the leftmost ancestor.
1046
If the method is fulltext, next will be None.
1048
position_map = self._get_components_positions(version_ids)
1049
# c = component_id, m = method, i_m = index_memo, n = next
1050
records = [(c, i_m) for c, (m, i_m, n) in position_map.iteritems()]
1052
for component_id, content, digest in \
1053
self._data.read_records_iter(records):
1054
method, index_memo, next = position_map[component_id]
1055
record_map[component_id] = method, content, digest, next
1059
def get_text(self, version_id):
1060
"""See VersionedFile.get_text"""
1061
return self.get_texts([version_id])[0]
1063
def get_texts(self, version_ids):
1064
return [''.join(l) for l in self.get_line_list(version_ids)]
1066
def get_line_list(self, version_ids):
1067
"""Return the texts of listed versions as a list of strings."""
1068
for version_id in version_ids:
1069
self.check_not_reserved_id(version_id)
1070
text_map, content_map = self._get_content_maps(version_ids)
1071
return [text_map[v] for v in version_ids]
1073
_get_lf_split_line_list = get_line_list
1075
def _get_content_maps(self, version_ids):
1076
"""Produce maps of text and KnitContents
1078
:return: (text_map, content_map) where text_map contains the texts for
1079
the requested versions and content_map contains the KnitContents.
1080
Both dicts take version_ids as their keys.
1082
# FUTURE: This function could be improved for the 'extract many' case
1083
# by tracking each component and only doing the copy when the number of
1084
# children than need to apply delta's to it is > 1 or it is part of the
1086
version_ids = list(version_ids)
1087
multiple_versions = len(version_ids) != 1
1088
record_map = self._get_record_map(version_ids)
1093
for version_id in version_ids:
1096
while cursor is not None:
1097
method, data, digest, next = record_map[cursor]
1098
components.append((cursor, method, data, digest))
1099
if cursor in content_map:
1104
for component_id, method, data, digest in reversed(components):
1105
if component_id in content_map:
1106
content = content_map[component_id]
1108
if method == 'fulltext':
1109
assert content is None
1110
content = self.factory.parse_fulltext(data, version_id)
1111
elif method == 'line-delta':
1112
delta = self.factory.parse_line_delta(data, version_id)
1113
if multiple_versions:
1114
# only doing this when we want multiple versions
1115
# output avoids list copies - which reference and
1116
# dereference many strings.
1117
content = content.copy()
1118
content.apply_delta(delta, version_id)
1119
if multiple_versions:
1120
content_map[component_id] = content
1122
if 'no-eol' in self._index.get_options(version_id):
1123
if multiple_versions:
1124
content = content.copy()
1125
content.strip_last_line_newline()
1126
final_content[version_id] = content
1128
# digest here is the digest from the last applied component.
1129
text = content.text()
1130
actual_sha = sha_strings(text)
1131
if actual_sha != digest:
1132
raise KnitCorrupt(self.filename,
1134
'\n of reconstructed text does not match'
1136
'\n for version %s' %
1137
(actual_sha, digest, version_id))
1138
text_map[version_id] = text
1139
return text_map, final_content
1141
def iter_lines_added_or_present_in_versions(self, version_ids=None,
728
1143
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
729
1144
if version_ids is None:
730
1145
version_ids = self.versions()
731
# we dont care about inclusions, the caller cares.
1147
pb = progress.DummyProgress()
1148
# we don't care about inclusions, the caller cares.
732
1149
# but we need to setup a list of records to visit.
733
1150
# we need version_id, position, length
734
1151
version_id_records = []
735
requested_versions = list(version_ids)
1152
requested_versions = set(version_ids)
736
1153
# filter for available versions
737
1154
for version_id in requested_versions:
738
1155
if not self.has_version(version_id):
739
1156
raise RevisionNotPresent(version_id, self.filename)
740
1157
# get a in-component-order queue:
742
1158
for version_id in self.versions():
743
1159
if version_id in requested_versions:
744
version_ids.append(version_id)
745
data_pos, length = self._index.get_position(version_id)
746
version_id_records.append((version_id, data_pos, length))
1160
index_memo = self._index.get_position(version_id)
1161
version_id_records.append((version_id, index_memo))
748
pb = bzrlib.ui.ui_factory.nested_progress_bar()
750
1163
total = len(version_id_records)
752
pb.update('Walking content.', count, total)
753
for version_id, data, sha_value in \
754
self._data.read_records_iter(version_id_records):
755
pb.update('Walking content.', count, total)
756
method = self._index.get_method(version_id)
757
version_idx = self._index.lookup(version_id)
758
assert method in ('fulltext', 'line-delta')
759
if method == 'fulltext':
760
content = self.factory.parse_fulltext(data, version_idx)
761
for line in content.text():
764
delta = self.factory.parse_line_delta(data, version_idx)
765
for start, end, count, lines in delta:
766
for origin, line in lines:
769
pb.update('Walking content.', total, total)
772
pb.update('Walking content.', total, total)
1164
for version_idx, (version_id, data, sha_value) in \
1165
enumerate(self._data.read_records_iter(version_id_records)):
1166
pb.update('Walking content.', version_idx, total)
1167
method = self._index.get_method(version_id)
1169
assert method in ('fulltext', 'line-delta')
1170
if method == 'fulltext':
1171
line_iterator = self.factory.get_fulltext_content(data)
1173
line_iterator = self.factory.get_linedelta_content(data)
1174
# XXX: It might be more efficient to yield (version_id,
1175
# line_iterator) in the future. However for now, this is a simpler
1176
# change to integrate into the rest of the codebase. RBC 20071110
1177
for line in line_iterator:
1178
yield line, version_id
1180
pb.update('Walking content.', total, total)
1182
def iter_parents(self, version_ids):
1183
"""Iterate through the parents for many version ids.
1185
:param version_ids: An iterable yielding version_ids.
1186
:return: An iterator that yields (version_id, parents). Requested
1187
version_ids not present in the versioned file are simply skipped.
1188
The order is undefined, allowing for different optimisations in
1189
the underlying implementation.
1191
return self._index.iter_parents(version_ids)
776
1193
def num_versions(self):
777
1194
"""See VersionedFile.num_versions()."""
778
1195
return self._index.num_versions()
979
1346
self._history.append(version_id)
981
1348
index = self._cache[version_id][5]
982
self._cache[version_id] = (version_id,
1349
self._cache[version_id] = (version_id,
989
def __init__(self, transport, filename, mode, create=False, file_mode=None):
990
_KnitComponentFile.__init__(self, transport, filename, mode, file_mode)
1356
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1357
create_parent_dir=False, delay_create=False, dir_mode=None):
1358
_KnitComponentFile.__init__(self, transport, filename, mode,
1359
file_mode=file_mode,
1360
create_parent_dir=create_parent_dir,
991
1362
self._cache = {}
992
1363
# position in _history is the 'official' index for a revision
993
1364
# but the values may have come from a newer entry.
994
# so - wc -l of a knit index is != the number of uniqe names
1365
# so - wc -l of a knit index is != the number of unique names
996
1367
self._history = []
997
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1369
fp = self._transport.get(self._filename)
1002
pb.update('read knit index', count, total)
1003
fp = self._transport.get(self._filename)
1004
self.check_header(fp)
1005
# readlines reads the whole file at once:
1006
# bad for transports like http, good for local disk
1007
# we save 60 ms doing this one change (
1008
# from calling readline each time to calling
1010
# probably what we want for nice behaviour on
1011
# http is a incremental readlines that yields, or
1012
# a check for local vs non local indexes,
1013
for l in fp.readlines():
1015
if len(rec) < 5 or rec[-1] != ':':
1017
# FIXME: in the future we should determine if its a
1018
# short write - and ignore it
1019
# or a different failure, and raise. RBC 20060407
1023
#pb.update('read knit index', count, total)
1024
# See self._parse_parents
1026
for value in rec[4:-1]:
1028
# uncompressed reference
1029
parents.append(value[1:])
1031
# this is 15/4000ms faster than isinstance,
1033
# this function is called thousands of times a
1034
# second so small variations add up.
1035
assert value.__class__ is str
1036
parents.append(self._history[int(value)])
1037
# end self._parse_parents
1038
# self._cache_version(rec[0],
1039
# rec[1].split(','),
1043
# --- self._cache_version
1044
# only want the _history index to reference the 1st
1045
# index entry for version_id
1047
if version_id not in self._cache:
1048
index = len(self._history)
1049
self._history.append(version_id)
1051
index = self._cache[version_id][5]
1052
self._cache[version_id] = (version_id,
1058
# --- self._cache_version
1059
except NoSuchFile, e:
1060
if mode != 'w' or not create:
1064
pb.update('read knit index', total, total)
1067
def _parse_parents(self, compressed_parents):
1068
"""convert a list of string parent values into version ids.
1070
ints are looked up in the index.
1071
.FOO values are ghosts and converted in to FOO.
1073
NOTE: the function is retained here for clarity, and for possible
1074
use in partial index reads. However bulk processing now has
1075
it inlined in __init__ for inner-loop optimisation.
1078
for value in compressed_parents:
1079
if value[-1] == '.':
1080
# uncompressed reference
1081
result.append(value[1:])
1371
# _load_data may raise NoSuchFile if the target knit is
1373
_load_data(self, fp)
1377
if mode != 'w' or not create:
1380
self._need_to_create = True
1083
# this is 15/4000ms faster than isinstance,
1084
# this function is called thousands of times a
1085
# second so small variations add up.
1086
assert value.__class__ is str
1087
result.append(self._history[int(value)])
1382
self._transport.put_bytes_non_atomic(
1383
self._filename, self.HEADER, mode=self._file_mode)
1090
1385
def get_graph(self):
1092
for version_id, index in self._cache.iteritems():
1093
graph.append((version_id, index[4]))
1386
"""Return a list of the node:parents lists from this knit index."""
1387
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1096
def get_ancestry(self, versions):
1389
def get_ancestry(self, versions, topo_sorted=True):
1097
1390
"""See VersionedFile.get_ancestry."""
1098
1391
# get a graph of all the mentioned versions:
1100
1393
pending = set(versions)
1102
1396
version = pending.pop()
1103
parents = self._cache[version][4]
1104
# got the parents ok
1106
parents = [parent for parent in parents if parent in self._cache]
1107
for parent in parents:
1108
# if not completed and not a ghost
1109
if parent not in graph:
1399
parents = [p for p in cache[version][4] if p in cache]
1401
raise RevisionNotPresent(version, self._filename)
1402
# if not completed and not a ghost
1403
pending.update([p for p in parents if p not in graph])
1111
1404
graph[version] = parents
1112
1407
return topo_sort(graph.items())
1114
1409
def get_ancestry_with_ghosts(self, versions):
1115
1410
"""See VersionedFile.get_ancestry_with_ghosts."""
1116
1411
# get a graph of all the mentioned versions:
1412
self.check_versions_present(versions)
1118
1415
pending = set(versions)
1120
1417
version = pending.pop()
1122
parents = self._cache[version][4]
1419
parents = cache[version][4]
1123
1420
except KeyError:
1124
1421
# ghost, fake it
1125
1422
graph[version] = []
1128
# got the parents ok
1129
for parent in parents:
1130
if parent not in graph:
1425
pending.update([p for p in parents if p not in graph])
1132
1426
graph[version] = parents
1133
1427
return topo_sort(graph.items())
1429
def iter_parents(self, version_ids):
1430
"""Iterate through the parents for many version ids.
1432
:param version_ids: An iterable yielding version_ids.
1433
:return: An iterator that yields (version_id, parents). Requested
1434
version_ids not present in the versioned file are simply skipped.
1435
The order is undefined, allowing for different optimisations in
1436
the underlying implementation.
1438
for version_id in version_ids:
1440
yield version_id, tuple(self.get_parents(version_id))
1135
1444
def num_versions(self):
1136
1445
return len(self._history)
1138
1447
__len__ = num_versions
1140
1449
def get_versions(self):
1450
"""Get all the versions in the file. not topologically sorted."""
1141
1451
return self._history
1143
def idx_to_name(self, idx):
1144
return self._history[idx]
1146
def lookup(self, version_id):
1147
assert version_id in self._cache
1148
return self._cache[version_id][5]
1150
1453
def _version_list_to_index(self, versions):
1151
1454
result_list = []
1152
1456
for version in versions:
1153
if version in self._cache:
1457
if version in cache:
1154
1458
# -- inlined lookup() --
1155
result_list.append(str(self._cache[version][5]))
1459
result_list.append(str(cache[version][5]))
1156
1460
# -- end lookup () --
1158
result_list.append('.' + version.encode('utf-8'))
1462
result_list.append('.' + version)
1159
1463
return ' '.join(result_list)
1161
def add_version(self, version_id, options, pos, size, parents):
1465
def add_version(self, version_id, options, index_memo, parents):
1162
1466
"""Add a version record to the index."""
1163
self.add_versions(((version_id, options, pos, size, parents),))
1467
self.add_versions(((version_id, options, index_memo, parents),))
1165
def add_versions(self, versions):
1469
def add_versions(self, versions, random_id=False):
1166
1470
"""Add multiple versions to the index.
1168
1472
:param versions: a list of tuples:
1169
1473
(version_id, options, pos, size, parents).
1474
:param random_id: If True the ids being added were randomly generated
1475
and no check for existence will be performed.
1172
for version_id, options, pos, size, parents in versions:
1173
line = "\n%s %s %s %s %s :" % (version_id.encode('utf-8'),
1177
self._version_list_to_index(parents))
1178
assert isinstance(line, str), \
1179
'content must be utf-8 encoded: %r' % (line,)
1181
self._transport.append(self._filename, StringIO(''.join(lines)))
1182
# cache after writing, so that a failed write leads to missing cache
1183
# entries not extra ones. XXX TODO: RBC 20060502 in the event of a
1184
# failure, reload the index or flush it or some such, to prevent
1185
# writing records that did complete twice.
1186
for version_id, options, pos, size, parents in versions:
1187
self._cache_version(version_id, options, pos, size, parents)
1478
orig_history = self._history[:]
1479
orig_cache = self._cache.copy()
1482
for version_id, options, (index, pos, size), parents in versions:
1483
line = "\n%s %s %s %s %s :" % (version_id,
1487
self._version_list_to_index(parents))
1488
assert isinstance(line, str), \
1489
'content must be utf-8 encoded: %r' % (line,)
1491
self._cache_version(version_id, options, pos, size, parents)
1492
if not self._need_to_create:
1493
self._transport.append_bytes(self._filename, ''.join(lines))
1496
sio.write(self.HEADER)
1497
sio.writelines(lines)
1499
self._transport.put_file_non_atomic(self._filename, sio,
1500
create_parent_dir=self._create_parent_dir,
1501
mode=self._file_mode,
1502
dir_mode=self._dir_mode)
1503
self._need_to_create = False
1505
# If any problems happen, restore the original values and re-raise
1506
self._history = orig_history
1507
self._cache = orig_cache
1189
1510
def has_version(self, version_id):
1190
1511
"""True if the version is in the index."""
1191
return self._cache.has_key(version_id)
1512
return version_id in self._cache
1193
1514
def get_position(self, version_id):
1194
"""Return data position and size of specified version."""
1195
return (self._cache[version_id][2], \
1196
self._cache[version_id][3])
1515
"""Return details needed to access the version.
1517
.kndx indices do not support split-out data, so return None for the
1520
:return: a tuple (None, data position, size) to hand to the access
1521
logic to get the record.
1523
entry = self._cache[version_id]
1524
return None, entry[2], entry[3]
1198
1526
def get_method(self, version_id):
1199
1527
"""Return compression method of specified version."""
1200
options = self._cache[version_id][1]
1529
options = self._cache[version_id][1]
1531
raise RevisionNotPresent(version_id, self._filename)
1201
1532
if 'fulltext' in options:
1202
1533
return 'fulltext'
1204
assert 'line-delta' in options
1535
if 'line-delta' not in options:
1536
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1205
1537
return 'line-delta'
1207
1539
def get_options(self, version_id):
1540
"""Return a list representing options.
1208
1544
return self._cache[version_id][1]
1210
1546
def get_parents(self, version_id):
1213
1549
if parent in self._cache]
1215
1551
def get_parents_with_ghosts(self, version_id):
1216
"""Return parents of specified version wth ghosts."""
1552
"""Return parents of specified version with ghosts."""
1217
1553
return self._cache[version_id][4]
1219
1555
def check_versions_present(self, version_ids):
1220
1556
"""Check that all specified versions are present."""
1221
version_ids = set(version_ids)
1222
for version_id in list(version_ids):
1223
if version_id in self._cache:
1224
version_ids.remove(version_id)
1226
raise RevisionNotPresent(list(version_ids)[0], self.filename)
1229
class _KnitData(_KnitComponentFile):
1230
"""Contents of the knit data file"""
1232
HEADER = "# bzr knit data 8\n"
1234
def __init__(self, transport, filename, mode, create=False, file_mode=None):
1235
_KnitComponentFile.__init__(self, transport, filename, mode)
1558
for version_id in version_ids:
1559
if version_id not in cache:
1560
raise RevisionNotPresent(version_id, self._filename)
1563
class KnitGraphIndex(object):
1564
"""A knit index that builds on GraphIndex."""
1566
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1567
"""Construct a KnitGraphIndex on a graph_index.
1569
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1570
:param deltas: Allow delta-compressed records.
1571
:param add_callback: If not None, allow additions to the index and call
1572
this callback with a list of added GraphIndex nodes:
1573
[(node, value, node_refs), ...]
1574
:param parents: If True, record knits parents, if not do not record
1577
self._graph_index = graph_index
1578
self._deltas = deltas
1579
self._add_callback = add_callback
1580
self._parents = parents
1581
if deltas and not parents:
1582
raise KnitCorrupt(self, "Cannot do delta compression without "
1585
def _get_entries(self, keys, check_present=False):
1586
"""Get the entries for keys.
1588
:param keys: An iterable of index keys, - 1-tuples.
1593
for node in self._graph_index.iter_entries(keys):
1595
found_keys.add(node[1])
1597
# adapt parentless index to the rest of the code.
1598
for node in self._graph_index.iter_entries(keys):
1599
yield node[0], node[1], node[2], ()
1600
found_keys.add(node[1])
1602
missing_keys = keys.difference(found_keys)
1604
raise RevisionNotPresent(missing_keys.pop(), self)
1606
def _present_keys(self, version_ids):
1608
node[1] for node in self._get_entries(version_ids)])
1610
def _parentless_ancestry(self, versions):
1611
"""Honour the get_ancestry API for parentless knit indices."""
1612
wanted_keys = self._version_ids_to_keys(versions)
1613
present_keys = self._present_keys(wanted_keys)
1614
missing = set(wanted_keys).difference(present_keys)
1616
raise RevisionNotPresent(missing.pop(), self)
1617
return list(self._keys_to_version_ids(present_keys))
1619
def get_ancestry(self, versions, topo_sorted=True):
1620
"""See VersionedFile.get_ancestry."""
1621
if not self._parents:
1622
return self._parentless_ancestry(versions)
1623
# XXX: This will do len(history) index calls - perhaps
1624
# it should be altered to be a index core feature?
1625
# get a graph of all the mentioned versions:
1628
versions = self._version_ids_to_keys(versions)
1629
pending = set(versions)
1631
# get all pending nodes
1632
this_iteration = pending
1633
new_nodes = self._get_entries(this_iteration)
1636
for (index, key, value, node_refs) in new_nodes:
1637
# dont ask for ghosties - otherwise
1638
# we we can end up looping with pending
1639
# being entirely ghosted.
1640
graph[key] = [parent for parent in node_refs[0]
1641
if parent not in ghosts]
1643
for parent in graph[key]:
1644
# dont examine known nodes again
1649
ghosts.update(this_iteration.difference(found))
1650
if versions.difference(graph):
1651
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1653
result_keys = topo_sort(graph.items())
1655
result_keys = graph.iterkeys()
1656
return [key[0] for key in result_keys]
1658
def get_ancestry_with_ghosts(self, versions):
1659
"""See VersionedFile.get_ancestry."""
1660
if not self._parents:
1661
return self._parentless_ancestry(versions)
1662
# XXX: This will do len(history) index calls - perhaps
1663
# it should be altered to be a index core feature?
1664
# get a graph of all the mentioned versions:
1666
versions = self._version_ids_to_keys(versions)
1667
pending = set(versions)
1669
# get all pending nodes
1670
this_iteration = pending
1671
new_nodes = self._get_entries(this_iteration)
1673
for (index, key, value, node_refs) in new_nodes:
1674
graph[key] = node_refs[0]
1676
for parent in graph[key]:
1677
# dont examine known nodes again
1681
missing_versions = this_iteration.difference(graph)
1682
missing_needed = versions.intersection(missing_versions)
1684
raise RevisionNotPresent(missing_needed.pop(), self)
1685
for missing_version in missing_versions:
1686
# add a key, no parents
1687
graph[missing_version] = []
1688
pending.discard(missing_version) # don't look for it
1689
result_keys = topo_sort(graph.items())
1690
return [key[0] for key in result_keys]
1692
def get_graph(self):
1693
"""Return a list of the node:parents lists from this knit index."""
1694
if not self._parents:
1695
return [(key, ()) for key in self.get_versions()]
1697
for index, key, value, refs in self._graph_index.iter_all_entries():
1698
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1701
def iter_parents(self, version_ids):
1702
"""Iterate through the parents for many version ids.
1704
:param version_ids: An iterable yielding version_ids.
1705
:return: An iterator that yields (version_id, parents). Requested
1706
version_ids not present in the versioned file are simply skipped.
1707
The order is undefined, allowing for different optimisations in
1708
the underlying implementation.
1711
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1713
present_parents = set()
1714
for node in all_nodes:
1715
all_parents.update(node[3][0])
1716
# any node we are querying must be present
1717
present_parents.add(node[1])
1718
unknown_parents = all_parents.difference(present_parents)
1719
present_parents.update(self._present_keys(unknown_parents))
1720
for node in all_nodes:
1722
for parent in node[3][0]:
1723
if parent in present_parents:
1724
parents.append(parent[0])
1725
yield node[1][0], tuple(parents)
1727
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1728
yield node[1][0], ()
1730
def num_versions(self):
1731
return len(list(self._graph_index.iter_all_entries()))
1733
__len__ = num_versions
1735
def get_versions(self):
1736
"""Get all the versions in the file. not topologically sorted."""
1737
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1739
def has_version(self, version_id):
1740
"""True if the version is in the index."""
1741
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1743
def _keys_to_version_ids(self, keys):
1744
return tuple(key[0] for key in keys)
1746
def get_position(self, version_id):
1747
"""Return details needed to access the version.
1749
:return: a tuple (index, data position, size) to hand to the access
1750
logic to get the record.
1752
node = self._get_node(version_id)
1753
bits = node[2][1:].split(' ')
1754
return node[0], int(bits[0]), int(bits[1])
1756
def get_method(self, version_id):
1757
"""Return compression method of specified version."""
1758
if not self._deltas:
1760
return self._parent_compression(self._get_node(version_id)[3][1])
1762
def _parent_compression(self, reference_list):
1763
# use the second reference list to decide if this is delta'd or not.
1764
if len(reference_list):
1769
def _get_node(self, version_id):
1771
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1773
raise RevisionNotPresent(version_id, self)
1775
def get_options(self, version_id):
1776
"""Return a list representing options.
1780
node = self._get_node(version_id)
1781
if not self._deltas:
1782
options = ['fulltext']
1784
options = [self._parent_compression(node[3][1])]
1785
if node[2][0] == 'N':
1786
options.append('no-eol')
1789
def get_parents(self, version_id):
1790
"""Return parents of specified version ignoring ghosts."""
1791
parents = list(self.iter_parents([version_id]))
1794
raise errors.RevisionNotPresent(version_id, self)
1795
return parents[0][1]
1797
def get_parents_with_ghosts(self, version_id):
1798
"""Return parents of specified version with ghosts."""
1799
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1800
check_present=True))
1801
if not self._parents:
1803
return self._keys_to_version_ids(nodes[0][3][0])
1805
def check_versions_present(self, version_ids):
1806
"""Check that all specified versions are present."""
1807
keys = self._version_ids_to_keys(version_ids)
1808
present = self._present_keys(keys)
1809
missing = keys.difference(present)
1811
raise RevisionNotPresent(missing.pop(), self)
1813
def add_version(self, version_id, options, access_memo, parents):
1814
"""Add a version record to the index."""
1815
return self.add_versions(((version_id, options, access_memo, parents),))
1817
def add_versions(self, versions, random_id=False):
1818
"""Add multiple versions to the index.
1820
This function does not insert data into the Immutable GraphIndex
1821
backing the KnitGraphIndex, instead it prepares data for insertion by
1822
the caller and checks that it is safe to insert then calls
1823
self._add_callback with the prepared GraphIndex nodes.
1825
:param versions: a list of tuples:
1826
(version_id, options, pos, size, parents).
1827
:param random_id: If True the ids being added were randomly generated
1828
and no check for existence will be performed.
1830
if not self._add_callback:
1831
raise errors.ReadOnlyError(self)
1832
# we hope there are no repositories with inconsistent parentage
1837
for (version_id, options, access_memo, parents) in versions:
1838
index, pos, size = access_memo
1839
key = (version_id, )
1840
parents = tuple((parent, ) for parent in parents)
1841
if 'no-eol' in options:
1845
value += "%d %d" % (pos, size)
1846
if not self._deltas:
1847
if 'line-delta' in options:
1848
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1851
if 'line-delta' in options:
1852
node_refs = (parents, (parents[0],))
1854
node_refs = (parents, ())
1856
node_refs = (parents, )
1859
raise KnitCorrupt(self, "attempt to add node with parents "
1860
"in parentless index.")
1862
keys[key] = (value, node_refs)
1864
present_nodes = self._get_entries(keys)
1865
for (index, key, value, node_refs) in present_nodes:
1866
if (value, node_refs) != keys[key]:
1867
raise KnitCorrupt(self, "inconsistent details in add_versions"
1868
": %s %s" % ((value, node_refs), keys[key]))
1872
for key, (value, node_refs) in keys.iteritems():
1873
result.append((key, value, node_refs))
1875
for key, (value, node_refs) in keys.iteritems():
1876
result.append((key, value))
1877
self._add_callback(result)
1879
def _version_ids_to_keys(self, version_ids):
1880
return set((version_id, ) for version_id in version_ids)
1883
class _KnitAccess(object):
1884
"""Access to knit records in a .knit file."""
1886
def __init__(self, transport, filename, _file_mode, _dir_mode,
1887
_need_to_create, _create_parent_dir):
1888
"""Create a _KnitAccess for accessing and inserting data.
1890
:param transport: The transport the .knit is located on.
1891
:param filename: The filename of the .knit.
1893
self._transport = transport
1894
self._filename = filename
1895
self._file_mode = _file_mode
1896
self._dir_mode = _dir_mode
1897
self._need_to_create = _need_to_create
1898
self._create_parent_dir = _create_parent_dir
1900
def add_raw_records(self, sizes, raw_data):
1901
"""Add raw knit bytes to a storage area.
1903
The data is spooled to whereever the access method is storing data.
1905
:param sizes: An iterable containing the size of each raw data segment.
1906
:param raw_data: A bytestring containing the data.
1907
:return: A list of memos to retrieve the record later. Each memo is a
1908
tuple - (index, pos, length), where the index field is always None
1909
for the .knit access method.
1911
assert type(raw_data) == str, \
1912
'data must be plain bytes was %s' % type(raw_data)
1913
if not self._need_to_create:
1914
base = self._transport.append_bytes(self._filename, raw_data)
1916
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1917
create_parent_dir=self._create_parent_dir,
1918
mode=self._file_mode,
1919
dir_mode=self._dir_mode)
1920
self._need_to_create = False
1924
result.append((None, base, size))
1929
"""IFF this data access has its own storage area, initialise it.
1933
self._transport.put_bytes_non_atomic(self._filename, '',
1934
mode=self._file_mode)
1936
def open_file(self):
1937
"""IFF this data access can be represented as a single file, open it.
1939
For knits that are not mapped to a single file on disk this will
1942
:return: None or a file handle.
1945
return self._transport.get(self._filename)
1950
def get_raw_records(self, memos_for_retrieval):
1951
"""Get the raw bytes for a records.
1953
:param memos_for_retrieval: An iterable containing the (index, pos,
1954
length) memo for retrieving the bytes. The .knit method ignores
1955
the index as there is always only a single file.
1956
:return: An iterator over the bytes of the records.
1958
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
1959
for pos, data in self._transport.readv(self._filename, read_vector):
1963
class _PackAccess(object):
1964
"""Access to knit records via a collection of packs."""
1966
def __init__(self, index_to_packs, writer=None):
1967
"""Create a _PackAccess object.
1969
:param index_to_packs: A dict mapping index objects to the transport
1970
and file names for obtaining data.
1971
:param writer: A tuple (pack.ContainerWriter, write_index) which
1972
contains the pack to write, and the index that reads from it will
1976
self.container_writer = writer[0]
1977
self.write_index = writer[1]
1979
self.container_writer = None
1980
self.write_index = None
1981
self.indices = index_to_packs
1983
def add_raw_records(self, sizes, raw_data):
1984
"""Add raw knit bytes to a storage area.
1986
The data is spooled to the container writer in one bytes-record per
1989
:param sizes: An iterable containing the size of each raw data segment.
1990
:param raw_data: A bytestring containing the data.
1991
:return: A list of memos to retrieve the record later. Each memo is a
1992
tuple - (index, pos, length), where the index field is the
1993
write_index object supplied to the PackAccess object.
1995
assert type(raw_data) == str, \
1996
'data must be plain bytes was %s' % type(raw_data)
2000
p_offset, p_length = self.container_writer.add_bytes_record(
2001
raw_data[offset:offset+size], [])
2003
result.append((self.write_index, p_offset, p_length))
2007
"""Pack based knits do not get individually created."""
2009
def get_raw_records(self, memos_for_retrieval):
2010
"""Get the raw bytes for a records.
2012
:param memos_for_retrieval: An iterable containing the (index, pos,
2013
length) memo for retrieving the bytes. The Pack access method
2014
looks up the pack to use for a given record in its index_to_pack
2016
:return: An iterator over the bytes of the records.
2018
# first pass, group into same-index requests
2020
current_index = None
2021
for (index, offset, length) in memos_for_retrieval:
2022
if current_index == index:
2023
current_list.append((offset, length))
2025
if current_index is not None:
2026
request_lists.append((current_index, current_list))
2027
current_index = index
2028
current_list = [(offset, length)]
2029
# handle the last entry
2030
if current_index is not None:
2031
request_lists.append((current_index, current_list))
2032
for index, offsets in request_lists:
2033
transport, path = self.indices[index]
2034
reader = pack.make_readv_reader(transport, path, offsets)
2035
for names, read_func in reader.iter_records():
2036
yield read_func(None)
2038
def open_file(self):
2039
"""Pack based knits have no single file."""
2042
def set_writer(self, writer, index, (transport, packname)):
2043
"""Set a writer to use for adding data."""
2044
if index is not None:
2045
self.indices[index] = (transport, packname)
2046
self.container_writer = writer
2047
self.write_index = index
2050
class _StreamAccess(object):
2051
"""A Knit Access object that provides data from a datastream.
2053
It also provides a fallback to present as unannotated data, annotated data
2054
from a *backing* access object.
2056
This is triggered by a index_memo which is pointing to a different index
2057
than this was constructed with, and is used to allow extracting full
2058
unannotated texts for insertion into annotated knits.
2061
def __init__(self, reader_callable, stream_index, backing_knit,
2063
"""Create a _StreamAccess object.
2065
:param reader_callable: The reader_callable from the datastream.
2066
This is called to buffer all the data immediately, for
2068
:param stream_index: The index the data stream this provides access to
2069
which will be present in native index_memo's.
2070
:param backing_knit: The knit object that will provide access to
2071
annotated texts which are not available in the stream, so as to
2072
create unannotated texts.
2073
:param orig_factory: The original content factory used to generate the
2074
stream. This is used for checking whether the thunk code for
2075
supporting _copy_texts will generate the correct form of data.
2077
self.data = reader_callable(None)
2078
self.stream_index = stream_index
2079
self.backing_knit = backing_knit
2080
self.orig_factory = orig_factory
2082
def get_raw_records(self, memos_for_retrieval):
2083
"""Get the raw bytes for a records.
2085
:param memos_for_retrieval: An iterable containing the (thunk_flag,
2086
index, start, end) memo for retrieving the bytes.
2087
:return: An iterator over the bytes of the records.
2089
# use a generator for memory friendliness
2090
for thunk_flag, version_id, start, end in memos_for_retrieval:
2091
if version_id is self.stream_index:
2092
yield self.data[start:end]
2094
# we have been asked to thunk. This thunking only occurs when
2095
# we are obtaining plain texts from an annotated backing knit
2096
# so that _copy_texts will work.
2097
# We could improve performance here by scanning for where we need
2098
# to do this and using get_line_list, then interleaving the output
2099
# as desired. However, for now, this is sufficient.
2100
if self.orig_factory.__class__ != KnitPlainFactory:
2101
raise errors.KnitCorrupt(
2102
self, 'Bad thunk request %r' % version_id)
2103
lines = self.backing_knit.get_lines(version_id)
2104
line_bytes = ''.join(lines)
2105
digest = sha_string(line_bytes)
2107
if lines[-1][-1] != '\n':
2108
lines[-1] = lines[-1] + '\n'
2110
orig_options = list(self.backing_knit._index.get_options(version_id))
2111
if 'fulltext' not in orig_options:
2112
if 'line-delta' not in orig_options:
2113
raise errors.KnitCorrupt(self,
2114
'Unknown compression method %r' % orig_options)
2115
orig_options.remove('line-delta')
2116
orig_options.append('fulltext')
2117
# We want plain data, because we expect to thunk only to allow text
2119
size, bytes = self.backing_knit._data._record_to_data(version_id,
2120
digest, lines, line_bytes)
2124
class _StreamIndex(object):
2125
"""A Knit Index object that uses the data map from a datastream."""
2127
def __init__(self, data_list):
2128
"""Create a _StreamIndex object.
2130
:param data_list: The data_list from the datastream.
2132
self.data_list = data_list
2133
self._by_version = {}
2135
for key, options, length, parents in data_list:
2136
self._by_version[key] = options, (pos, pos + length), parents
2139
def get_ancestry(self, versions, topo_sorted):
2140
"""Get an ancestry list for versions."""
2142
# Not needed for basic joins
2143
raise NotImplementedError(self.get_ancestry)
2144
# get a graph of all the mentioned versions:
2145
# Little ugly - basically copied from KnitIndex, but don't want to
2146
# accidentally incorporate too much of that index's code.
2148
pending = set(versions)
2149
cache = self._by_version
2151
version = pending.pop()
2154
parents = [p for p in cache[version][2] if p in cache]
2156
raise RevisionNotPresent(version, self)
2157
# if not completed and not a ghost
2158
pending.update([p for p in parents if p not in ancestry])
2159
ancestry.add(version)
2160
return list(ancestry)
2162
def get_method(self, version_id):
2163
"""Return compression method of specified version."""
2165
options = self._by_version[version_id][0]
2167
# Strictly speaking this should check in the backing knit, but
2168
# until we have a test to discriminate, this will do.
2170
if 'fulltext' in options:
2172
elif 'line-delta' in options:
2175
raise errors.KnitIndexUnknownMethod(self, options)
2177
def get_options(self, version_id):
2178
"""Return a list representing options.
2182
return self._by_version[version_id][0]
2184
def get_parents_with_ghosts(self, version_id):
2185
"""Return parents of specified version with ghosts."""
2186
return self._by_version[version_id][2]
2188
def get_position(self, version_id):
2189
"""Return details needed to access the version.
2191
_StreamAccess has the data as a big array, so we return slice
2192
coordinates into that (as index_memo's are opaque outside the
2193
index and matching access class).
2195
:return: a tuple (thunk_flag, index, start, end). If thunk_flag is
2196
False, index will be self, otherwise it will be a version id.
2199
start, end = self._by_version[version_id][1]
2200
return False, self, start, end
2202
# Signal to the access object to handle this from the backing knit.
2203
return (True, version_id, None, None)
2205
def get_versions(self):
2206
"""Get all the versions in the stream."""
2207
return self._by_version.keys()
2209
def iter_parents(self, version_ids):
2210
"""Iterate through the parents for many version ids.
2212
:param version_ids: An iterable yielding version_ids.
2213
:return: An iterator that yields (version_id, parents). Requested
2214
version_ids not present in the versioned file are simply skipped.
2215
The order is undefined, allowing for different optimisations in
2216
the underlying implementation.
2219
for version in version_ids:
2221
result.append((version, self._by_version[version][2]))
2227
class _KnitData(object):
2228
"""Manage extraction of data from a KnitAccess, caching and decompressing.
2230
The KnitData class provides the logic for parsing and using knit records,
2231
making use of an access method for the low level read and write operations.
2234
def __init__(self, access):
2235
"""Create a KnitData object.
2237
:param access: The access method to use. Access methods such as
2238
_KnitAccess manage the insertion of raw records and the subsequent
2239
retrieval of the same.
2241
self._access = access
1237
2242
self._checked = False
1239
self._transport.put(self._filename, StringIO(''), mode=file_mode)
2243
# TODO: jam 20060713 conceptually, this could spill to disk
2244
# if the cached size gets larger than a certain amount
2245
# but it complicates the model a bit, so for now just use
2246
# a simple dictionary
2248
self._do_cache = False
2250
def enable_cache(self):
2251
"""Enable caching of reads."""
2252
self._do_cache = True
1242
2254
def clear_cache(self):
1243
2255
"""Clear the record cache."""
2256
self._do_cache = False
1246
2259
def _open_file(self):
1247
if self._file is None:
1249
self._file = self._transport.get(self._filename)
2260
return self._access.open_file()
1254
def _record_to_data(self, version_id, digest, lines):
2262
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
1255
2263
"""Convert version_id, digest, lines into a raw data block.
2265
:param dense_lines: The bytes of lines but in a denser form. For
2266
instance, if lines is a list of 1000 bytestrings each ending in \n,
2267
dense_lines may be a list with one line in it, containing all the
2268
1000's lines and their \n's. Using dense_lines if it is already
2269
known is a win because the string join to create bytes in this
2270
function spends less time resizing the final string.
1257
2271
:return: (len, a StringIO instance with the raw data ready to read.)
1260
data_file = GzipFile(None, mode='wb', fileobj=sio)
1261
data_file.writelines(chain(
1262
["version %s %d %s\n" % (version_id.encode('utf-8'),
2273
# Note: using a string copy here increases memory pressure with e.g.
2274
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2275
# when doing the initial commit of a mozilla tree. RBC 20070921
2276
bytes = ''.join(chain(
2277
["version %s %d %s\n" % (version_id,
1266
["end %s\n" % version_id.encode('utf-8')]))
1273
def add_raw_record(self, raw_data):
2280
dense_lines or lines,
2281
["end %s\n" % version_id]))
2282
assert bytes.__class__ == str
2283
compressed_bytes = bytes_to_gzip(bytes)
2284
return len(compressed_bytes), compressed_bytes
2286
def add_raw_records(self, sizes, raw_data):
1274
2287
"""Append a prepared record to the data file.
1276
:return: the offset in the data file raw_data was written.
2289
:param sizes: An iterable containing the size of each raw data segment.
2290
:param raw_data: A bytestring containing the data.
2291
:return: a list of index data for the way the data was stored.
2292
See the access method add_raw_records documentation for more
1278
assert isinstance(raw_data, str), 'data must be plain bytes'
1279
return self._transport.append(self._filename, StringIO(raw_data))
2295
return self._access.add_raw_records(sizes, raw_data)
1281
def add_record(self, version_id, digest, lines):
1282
"""Write new text record to disk. Returns the position in the
1283
file where it was written."""
1284
size, sio = self._record_to_data(version_id, digest, lines)
1286
self._records[version_id] = (digest, lines)
1288
start_pos = self._transport.append(self._filename, sio)
1289
return start_pos, size
1291
2297
def _parse_record_header(self, version_id, raw_data):
1292
2298
"""Parse a record header for consistency.
1603
2657
InterVersionedFile.register_optimiser(WeaveToKnit)
1606
class SequenceMatcher(difflib.SequenceMatcher):
1607
"""Knit tuned sequence matcher.
1609
This is based on profiling of difflib which indicated some improvements
1610
for our usage pattern.
2660
# Deprecated, use PatienceSequenceMatcher instead
2661
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2664
def annotate_knit(knit, revision_id):
2665
"""Annotate a knit with no cached annotations.
2667
This implementation is for knits with no cached annotations.
2668
It will work for knits with cached annotations, but this is not
1613
def find_longest_match(self, alo, ahi, blo, bhi):
1614
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1616
If isjunk is not defined:
1618
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1619
alo <= i <= i+k <= ahi
1620
blo <= j <= j+k <= bhi
1621
and for all (i',j',k') meeting those conditions,
1624
and if i == i', j <= j'
1626
In other words, of all maximal matching blocks, return one that
1627
starts earliest in a, and of all those maximal matching blocks that
1628
start earliest in a, return the one that starts earliest in b.
1630
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1631
>>> s.find_longest_match(0, 5, 0, 9)
1634
If isjunk is defined, first the longest matching block is
1635
determined as above, but with the additional restriction that no
1636
junk element appears in the block. Then that block is extended as
1637
far as possible by matching (only) junk elements on both sides. So
1638
the resulting block never matches on junk except as identical junk
1639
happens to be adjacent to an "interesting" match.
1641
Here's the same example as before, but considering blanks to be
1642
junk. That prevents " abcd" from matching the " abcd" at the tail
1643
end of the second sequence directly. Instead only the "abcd" can
1644
match, and matches the leftmost "abcd" in the second sequence:
1646
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1647
>>> s.find_longest_match(0, 5, 0, 9)
1650
If no blocks match, return (alo, blo, 0).
1652
>>> s = SequenceMatcher(None, "ab", "c")
1653
>>> s.find_longest_match(0, 2, 0, 1)
1657
# CAUTION: stripping common prefix or suffix would be incorrect.
1661
# Longest matching block is "ab", but if common prefix is
1662
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1663
# strip, so ends up claiming that ab is changed to acab by
1664
# inserting "ca" in the middle. That's minimal but unintuitive:
1665
# "it's obvious" that someone inserted "ac" at the front.
1666
# Windiff ends up at the same place as diff, but by pairing up
1667
# the unique 'b's and then matching the first two 'a's.
1669
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1670
besti, bestj, bestsize = alo, blo, 0
1671
# find longest junk-free match
1672
# during an iteration of the loop, j2len[j] = length of longest
1673
# junk-free match ending with a[i-1] and b[j]
1677
for i in xrange(alo, ahi):
1678
# look at all instances of a[i] in b; note that because
1679
# b2j has no junk keys, the loop is skipped if a[i] is junk
1680
j2lenget = j2len.get
1683
# changing b2j.get(a[i], nothing) to a try:Keyerror pair produced the
1684
# following improvement
1685
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1686
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1687
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1689
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1690
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1691
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
1703
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1705
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1708
# Extend the best by non-junk elements on each end. In particular,
1709
# "popular" non-junk elements aren't in b2j, which greatly speeds
1710
# the inner loop above, but also means "the best" match so far
1711
# doesn't contain any junk *or* popular non-junk elements.
1712
while besti > alo and bestj > blo and \
1713
not isbjunk(b[bestj-1]) and \
1714
a[besti-1] == b[bestj-1]:
1715
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1716
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1717
not isbjunk(b[bestj+bestsize]) and \
1718
a[besti+bestsize] == b[bestj+bestsize]:
1721
# Now that we have a wholly interesting match (albeit possibly
1722
# empty!), we may as well suck up the matching junk on each
1723
# side of it too. Can't think of a good reason not to, and it
1724
# saves post-processing the (possibly considerable) expense of
1725
# figuring out what to do with it. In the case of an empty
1726
# interesting match, this is clearly the right thing to do,
1727
# because no other kind of match is possible in the regions.
1728
while besti > alo and bestj > blo and \
1729
isbjunk(b[bestj-1]) and \
1730
a[besti-1] == b[bestj-1]:
1731
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1732
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1733
isbjunk(b[bestj+bestsize]) and \
1734
a[besti+bestsize] == b[bestj+bestsize]:
1735
bestsize = bestsize + 1
1737
return besti, bestj, bestsize
2671
ancestry = knit.get_ancestry(revision_id)
2672
fulltext = dict(zip(ancestry, knit.get_line_list(ancestry)))
2674
for candidate in ancestry:
2675
if candidate in annotations:
2677
parents = knit.get_parents(candidate)
2678
if len(parents) == 0:
2680
elif knit._index.get_method(candidate) != 'line-delta':
2683
parent, sha1, noeol, delta = knit.get_delta(candidate)
2684
blocks = KnitContent.get_line_delta_blocks(delta,
2685
fulltext[parents[0]], fulltext[candidate])
2686
annotations[candidate] = list(annotate.reannotate([annotations[p]
2687
for p in parents], fulltext[candidate], candidate, blocks))
2688
return iter(annotations[revision_id])
2692
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2694
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
