1
# Copyright (C) 2005, 2006, 2007 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
"""Knit versionedfile implementation.
19
A knit is a versioned file implementation that supports efficient append only
23
lifeless: the data file is made up of "delta records". each delta record has a delta header
24
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
25
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
26
end-marker; simply "end VERSION"
28
delta can be line or full contents.a
29
... the 8's there are the index number of the annotation.
30
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
34
8 e.set('executable', 'yes')
36
8 if elt.get('executable') == 'yes':
37
8 ie.executable = True
38
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
42
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
43
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
44
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
45
09:33 < lifeless> right
46
09:33 < jrydberg> lifeless: the position and size is the range in the data file
49
so the index sequence is the dictionary compressed sequence number used
50
in the deltas to provide line annotation
55
# 10:16 < lifeless> make partial index writes safe
56
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
57
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
59
# move sha1 out of the content so that join is faster at verifying parents
60
# record content length ?
64
from cStringIO import StringIO
66
from itertools import izip, chain
73
from bzrlib.lazy_import import lazy_import
74
lazy_import(globals(), """
88
from bzrlib.errors import (
96
RevisionAlreadyPresent,
98
from bzrlib.tuned_gzip import GzipFile
99
from bzrlib.trace import mutter
100
from bzrlib.osutils import (
105
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
106
from bzrlib.tsort import topo_sort
109
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
112
# TODO: Split out code specific to this format into an associated object.
114
# TODO: Can we put in some kind of value to check that the index and data
115
# files belong together?
117
# TODO: accommodate binaries, perhaps by storing a byte count
119
# TODO: function to check whole file
121
# TODO: atomically append data, then measure backwards from the cursor
122
# position after writing to work out where it was located. we may need to
123
# bypass python file buffering.
125
DATA_SUFFIX = '.knit'
126
INDEX_SUFFIX = '.kndx'
129
class KnitContent(object):
130
"""Content of a knit version to which deltas can be applied."""
132
def __init__(self, lines):
135
def annotate_iter(self):
136
"""Yield tuples of (origin, text) for each content line."""
137
return iter(self._lines)
140
"""Return a list of (origin, text) tuples."""
141
return list(self.annotate_iter())
143
def line_delta_iter(self, new_lines):
144
"""Generate line-based delta from this content to new_lines."""
145
new_texts = new_lines.text()
146
old_texts = self.text()
147
s = KnitSequenceMatcher(None, old_texts, new_texts)
148
for tag, i1, i2, j1, j2 in s.get_opcodes():
151
# ofrom, oto, length, data
152
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
154
def line_delta(self, new_lines):
155
return list(self.line_delta_iter(new_lines))
158
return [text for origin, text in self._lines]
161
return KnitContent(self._lines[:])
164
def get_line_delta_blocks(knit_delta, source, target):
165
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
166
target_len = len(target)
169
for s_begin, s_end, t_len, new_text in knit_delta:
170
true_n = s_begin - s_pos
173
# knit deltas do not provide reliable info about whether the
174
# last line of a file matches, due to eol handling.
175
if source[s_pos + n -1] != target[t_pos + n -1]:
178
yield s_pos, t_pos, n
179
t_pos += t_len + true_n
181
n = target_len - t_pos
183
if source[s_pos + n -1] != target[t_pos + n -1]:
186
yield s_pos, t_pos, n
187
yield s_pos + (target_len - t_pos), target_len, 0
190
class _KnitFactory(object):
191
"""Base factory for creating content objects."""
193
def make(self, lines, version_id):
194
num_lines = len(lines)
195
return KnitContent(zip([version_id] * num_lines, lines))
198
class KnitAnnotateFactory(_KnitFactory):
199
"""Factory for creating annotated Content objects."""
203
def parse_fulltext(self, content, version_id):
204
"""Convert fulltext to internal representation
206
fulltext content is of the format
207
revid(utf8) plaintext\n
208
internal representation is of the format:
211
# TODO: jam 20070209 The tests expect this to be returned as tuples,
212
# but the code itself doesn't really depend on that.
213
# Figure out a way to not require the overhead of turning the
214
# list back into tuples.
215
lines = [tuple(line.split(' ', 1)) for line in content]
216
return KnitContent(lines)
218
def parse_line_delta_iter(self, lines):
219
return iter(self.parse_line_delta(lines))
221
def parse_line_delta(self, lines, version_id):
222
"""Convert a line based delta into internal representation.
224
line delta is in the form of:
225
intstart intend intcount
227
revid(utf8) newline\n
228
internal representation is
229
(start, end, count, [1..count tuples (revid, newline)])
236
def cache_and_return(line):
237
origin, text = line.split(' ', 1)
238
return cache.setdefault(origin, origin), text
240
# walk through the lines parsing.
242
start, end, count = [int(n) for n in header.split(',')]
243
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
244
result.append((start, end, count, contents))
247
def get_fulltext_content(self, lines):
248
"""Extract just the content lines from a fulltext."""
249
return (line.split(' ', 1)[1] for line in lines)
251
def get_linedelta_content(self, lines):
252
"""Extract just the content from a line delta.
254
This doesn't return all of the extra information stored in a delta.
255
Only the actual content lines.
260
header = header.split(',')
261
count = int(header[2])
262
for i in xrange(count):
263
origin, text = next().split(' ', 1)
266
def lower_fulltext(self, content):
267
"""convert a fulltext content record into a serializable form.
269
see parse_fulltext which this inverts.
271
# TODO: jam 20070209 We only do the caching thing to make sure that
272
# the origin is a valid utf-8 line, eventually we could remove it
273
return ['%s %s' % (o, t) for o, t in content._lines]
275
def lower_line_delta(self, delta):
276
"""convert a delta into a serializable form.
278
See parse_line_delta which this inverts.
280
# TODO: jam 20070209 We only do the caching thing to make sure that
281
# the origin is a valid utf-8 line, eventually we could remove it
283
for start, end, c, lines in delta:
284
out.append('%d,%d,%d\n' % (start, end, c))
285
out.extend(origin + ' ' + text
286
for origin, text in lines)
290
class KnitPlainFactory(_KnitFactory):
291
"""Factory for creating plain Content objects."""
295
def parse_fulltext(self, content, version_id):
296
"""This parses an unannotated fulltext.
298
Note that this is not a noop - the internal representation
299
has (versionid, line) - its just a constant versionid.
301
return self.make(content, version_id)
303
def parse_line_delta_iter(self, lines, version_id):
305
num_lines = len(lines)
306
while cur < num_lines:
309
start, end, c = [int(n) for n in header.split(',')]
310
yield start, end, c, zip([version_id] * c, lines[cur:cur+c])
313
def parse_line_delta(self, lines, version_id):
314
return list(self.parse_line_delta_iter(lines, version_id))
316
def get_fulltext_content(self, lines):
317
"""Extract just the content lines from a fulltext."""
320
def get_linedelta_content(self, lines):
321
"""Extract just the content from a line delta.
323
This doesn't return all of the extra information stored in a delta.
324
Only the actual content lines.
329
header = header.split(',')
330
count = int(header[2])
331
for i in xrange(count):
334
def lower_fulltext(self, content):
335
return content.text()
337
def lower_line_delta(self, delta):
339
for start, end, c, lines in delta:
340
out.append('%d,%d,%d\n' % (start, end, c))
341
out.extend([text for origin, text in lines])
345
def make_empty_knit(transport, relpath):
346
"""Construct a empty knit at the specified location."""
347
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
350
class KnitVersionedFile(VersionedFile):
351
"""Weave-like structure with faster random access.
353
A knit stores a number of texts and a summary of the relationships
354
between them. Texts are identified by a string version-id. Texts
355
are normally stored and retrieved as a series of lines, but can
356
also be passed as single strings.
358
Lines are stored with the trailing newline (if any) included, to
359
avoid special cases for files with no final newline. Lines are
360
composed of 8-bit characters, not unicode. The combination of
361
these approaches should mean any 'binary' file can be safely
362
stored and retrieved.
365
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
366
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
367
create=False, create_parent_dir=False, delay_create=False,
368
dir_mode=None, index=None, access_method=None):
369
"""Construct a knit at location specified by relpath.
371
:param create: If not True, only open an existing knit.
372
:param create_parent_dir: If True, create the parent directory if
373
creating the file fails. (This is used for stores with
374
hash-prefixes that may not exist yet)
375
:param delay_create: The calling code is aware that the knit won't
376
actually be created until the first data is stored.
377
:param index: An index to use for the knit.
379
if deprecated_passed(basis_knit):
380
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
381
" deprecated as of bzr 0.9.",
382
DeprecationWarning, stacklevel=2)
383
if access_mode is None:
385
super(KnitVersionedFile, self).__init__(access_mode)
386
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
387
self.transport = transport
388
self.filename = relpath
389
self.factory = factory or KnitAnnotateFactory()
390
self.writable = (access_mode == 'w')
393
self._max_delta_chain = 200
396
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
397
access_mode, create=create, file_mode=file_mode,
398
create_parent_dir=create_parent_dir, delay_create=delay_create,
402
if access_method is None:
403
_access = _KnitAccess(transport, relpath + DATA_SUFFIX, file_mode, dir_mode,
404
((create and not len(self)) and delay_create), create_parent_dir)
406
_access = access_method
407
if create and not len(self) and not delay_create:
409
self._data = _KnitData(_access)
412
return '%s(%s)' % (self.__class__.__name__,
413
self.transport.abspath(self.filename))
415
def _check_should_delta(self, first_parents):
416
"""Iterate back through the parent listing, looking for a fulltext.
418
This is used when we want to decide whether to add a delta or a new
419
fulltext. It searches for _max_delta_chain parents. When it finds a
420
fulltext parent, it sees if the total size of the deltas leading up to
421
it is large enough to indicate that we want a new full text anyway.
423
Return True if we should create a new delta, False if we should use a
428
delta_parents = first_parents
429
for count in xrange(self._max_delta_chain):
430
parent = delta_parents[0]
431
method = self._index.get_method(parent)
432
index, pos, size = self._index.get_position(parent)
433
if method == 'fulltext':
437
delta_parents = self._index.get_parents(parent)
439
# We couldn't find a fulltext, so we must create a new one
442
return fulltext_size > delta_size
444
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
445
"""See VersionedFile._add_delta()."""
446
self._check_add(version_id, []) # should we check the lines ?
447
self._check_versions_present(parents)
451
for parent in parents:
452
if not self.has_version(parent):
453
ghosts.append(parent)
455
present_parents.append(parent)
457
if delta_parent is None:
458
# reconstitute as full text.
459
assert len(delta) == 1 or len(delta) == 0
461
assert delta[0][0] == 0
462
assert delta[0][1] == 0, delta[0][1]
463
return super(KnitVersionedFile, self)._add_delta(version_id,
474
options.append('no-eol')
476
if delta_parent is not None:
477
# determine the current delta chain length.
478
# To speed the extract of texts the delta chain is limited
479
# to a fixed number of deltas. This should minimize both
480
# I/O and the time spend applying deltas.
481
# The window was changed to a maximum of 200 deltas, but also added
482
# was a check that the total compressed size of the deltas is
483
# smaller than the compressed size of the fulltext.
484
if not self._check_should_delta([delta_parent]):
485
# We don't want a delta here, just do a normal insertion.
486
return super(KnitVersionedFile, self)._add_delta(version_id,
493
options.append('line-delta')
494
store_lines = self.factory.lower_line_delta(delta)
496
access_memo = self._data.add_record(version_id, digest, store_lines)
497
self._index.add_version(version_id, options, access_memo, parents)
499
def _add_raw_records(self, records, data):
500
"""Add all the records 'records' with data pre-joined in 'data'.
502
:param records: A list of tuples(version_id, options, parents, size).
503
:param data: The data for the records. When it is written, the records
504
are adjusted to have pos pointing into data by the sum of
505
the preceding records sizes.
508
raw_record_sizes = [record[3] for record in records]
509
positions = self._data.add_raw_records(raw_record_sizes, data)
512
for (version_id, options, parents, size), access_memo in zip(
514
index_entries.append((version_id, options, access_memo, parents))
515
if self._data._do_cache:
516
self._data._cache[version_id] = data[offset:offset+size]
518
self._index.add_versions(index_entries)
520
def enable_cache(self):
521
"""Start caching data for this knit"""
522
self._data.enable_cache()
524
def clear_cache(self):
525
"""Clear the data cache only."""
526
self._data.clear_cache()
528
def copy_to(self, name, transport):
529
"""See VersionedFile.copy_to()."""
530
# copy the current index to a temp index to avoid racing with local
532
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
533
self.transport.get(self._index._filename))
535
f = self._data._open_file()
537
transport.put_file(name + DATA_SUFFIX, f)
540
# move the copied index into place
541
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
543
def create_empty(self, name, transport, mode=None):
544
return KnitVersionedFile(name, transport, factory=self.factory,
545
delta=self.delta, create=True)
547
def _fix_parents(self, version_id, new_parents):
548
"""Fix the parents list for version.
550
This is done by appending a new version to the index
551
with identical data except for the parents list.
552
the parents list must be a superset of the current
555
current_values = self._index._cache[version_id]
556
assert set(current_values[4]).difference(set(new_parents)) == set()
557
self._index.add_version(version_id,
559
(None, current_values[2], current_values[3]),
562
def _extract_blocks(self, version_id, source, target):
563
if self._index.get_method(version_id) != 'line-delta':
565
parent, sha1, noeol, delta = self.get_delta(version_id)
566
return KnitContent.get_line_delta_blocks(delta, source, target)
568
def get_delta(self, version_id):
569
"""Get a delta for constructing version from some other version."""
570
version_id = osutils.safe_revision_id(version_id)
571
self.check_not_reserved_id(version_id)
572
if not self.has_version(version_id):
573
raise RevisionNotPresent(version_id, self.filename)
575
parents = self.get_parents(version_id)
580
index_memo = self._index.get_position(version_id)
581
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
582
noeol = 'no-eol' in self._index.get_options(version_id)
583
if 'fulltext' == self._index.get_method(version_id):
584
new_content = self.factory.parse_fulltext(data, version_id)
585
if parent is not None:
586
reference_content = self._get_content(parent)
587
old_texts = reference_content.text()
590
new_texts = new_content.text()
591
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
592
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
594
delta = self.factory.parse_line_delta(data, version_id)
595
return parent, sha1, noeol, delta
597
def get_graph_with_ghosts(self):
598
"""See VersionedFile.get_graph_with_ghosts()."""
599
graph_items = self._index.get_graph()
600
return dict(graph_items)
602
def get_sha1(self, version_id):
603
return self.get_sha1s([version_id])[0]
605
def get_sha1s(self, version_ids):
606
"""See VersionedFile.get_sha1()."""
607
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
608
record_map = self._get_record_map(version_ids)
609
# record entry 2 is the 'digest'.
610
return [record_map[v][2] for v in version_ids]
614
"""See VersionedFile.get_suffixes()."""
615
return [DATA_SUFFIX, INDEX_SUFFIX]
617
def has_ghost(self, version_id):
618
"""True if there is a ghost reference in the file to version_id."""
619
version_id = osutils.safe_revision_id(version_id)
621
if self.has_version(version_id):
623
# optimisable if needed by memoising the _ghosts set.
624
items = self._index.get_graph()
625
for node, parents in items:
626
for parent in parents:
627
if parent not in self._index._cache:
628
if parent == version_id:
633
"""See VersionedFile.versions."""
634
return self._index.get_versions()
636
def has_version(self, version_id):
637
"""See VersionedFile.has_version."""
638
version_id = osutils.safe_revision_id(version_id)
639
return self._index.has_version(version_id)
641
__contains__ = has_version
643
def _merge_annotations(self, content, parents, parent_texts={},
644
delta=None, annotated=None):
645
"""Merge annotations for content. This is done by comparing
646
the annotations based on changed to the text.
650
for parent_id in parents:
651
merge_content = self._get_content(parent_id, parent_texts)
652
seq = patiencediff.PatienceSequenceMatcher(
653
None, merge_content.text(), content.text())
654
if delta_seq is None:
655
# setup a delta seq to reuse.
657
for i, j, n in seq.get_matching_blocks():
660
# this appears to copy (origin, text) pairs across to the new
661
# content for any line that matches the last-checked parent.
662
# FIXME: save the sequence control data for delta compression
663
# against the most relevant parent rather than rediffing.
664
content._lines[j:j+n] = merge_content._lines[i:i+n]
667
reference_content = self._get_content(parents[0], parent_texts)
668
new_texts = content.text()
669
old_texts = reference_content.text()
670
delta_seq = patiencediff.PatienceSequenceMatcher(
671
None, old_texts, new_texts)
672
return self._make_line_delta(delta_seq, content)
674
def _make_line_delta(self, delta_seq, new_content):
675
"""Generate a line delta from delta_seq and new_content."""
677
for op in delta_seq.get_opcodes():
680
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
683
def _get_components_positions(self, version_ids):
684
"""Produce a map of position data for the components of versions.
686
This data is intended to be used for retrieving the knit records.
688
A dict of version_id to (method, data_pos, data_size, next) is
690
method is the way referenced data should be applied.
691
data_pos is the position of the data in the knit.
692
data_size is the size of the data in the knit.
693
next is the build-parent of the version, or None for fulltexts.
696
for version_id in version_ids:
699
while cursor is not None and cursor not in component_data:
700
method = self._index.get_method(cursor)
701
if method == 'fulltext':
704
next = self.get_parents(cursor)[0]
705
index_memo = self._index.get_position(cursor)
706
component_data[cursor] = (method, index_memo, next)
708
return component_data
710
def _get_content(self, version_id, parent_texts={}):
711
"""Returns a content object that makes up the specified
713
if not self.has_version(version_id):
714
raise RevisionNotPresent(version_id, self.filename)
716
cached_version = parent_texts.get(version_id, None)
717
if cached_version is not None:
718
return cached_version
720
text_map, contents_map = self._get_content_maps([version_id])
721
return contents_map[version_id]
723
def _check_versions_present(self, version_ids):
724
"""Check that all specified versions are present."""
725
self._index.check_versions_present(version_ids)
727
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
728
"""See VersionedFile.add_lines_with_ghosts()."""
729
self._check_add(version_id, lines)
730
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
732
def _add_lines(self, version_id, parents, lines, parent_texts):
733
"""See VersionedFile.add_lines."""
734
self._check_add(version_id, lines)
735
self._check_versions_present(parents)
736
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
738
def _check_add(self, version_id, lines):
739
"""check that version_id and lines are safe to add."""
740
assert self.writable, "knit is not opened for write"
741
### FIXME escape. RBC 20060228
742
if contains_whitespace(version_id):
743
raise InvalidRevisionId(version_id, self.filename)
744
self.check_not_reserved_id(version_id)
745
if self.has_version(version_id):
746
raise RevisionAlreadyPresent(version_id, self.filename)
747
self._check_lines_not_unicode(lines)
748
self._check_lines_are_lines(lines)
750
def _add(self, version_id, lines, parents, delta, parent_texts):
751
"""Add a set of lines on top of version specified by parents.
753
If delta is true, compress the text as a line-delta against
756
Any versions not present will be converted into ghosts.
758
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
759
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
760
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
761
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
762
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
763
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
764
# +1383 0 8.0370 8.0370 +<len>
765
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
766
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
767
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
768
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
772
if parent_texts is None:
774
for parent in parents:
775
if not self.has_version(parent):
776
ghosts.append(parent)
778
present_parents.append(parent)
780
if delta and not len(present_parents):
783
digest = sha_strings(lines)
786
if lines[-1][-1] != '\n':
787
options.append('no-eol')
788
lines[-1] = lines[-1] + '\n'
790
if len(present_parents) and delta:
791
# To speed the extract of texts the delta chain is limited
792
# to a fixed number of deltas. This should minimize both
793
# I/O and the time spend applying deltas.
794
delta = self._check_should_delta(present_parents)
796
assert isinstance(version_id, str)
797
lines = self.factory.make(lines, version_id)
798
if delta or (self.factory.annotated and len(present_parents) > 0):
799
# Merge annotations from parent texts if so is needed.
800
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
801
delta, self.factory.annotated)
804
options.append('line-delta')
805
store_lines = self.factory.lower_line_delta(delta_hunks)
807
options.append('fulltext')
808
store_lines = self.factory.lower_fulltext(lines)
810
access_memo = self._data.add_record(version_id, digest, store_lines)
811
self._index.add_version(version_id, options, access_memo, parents)
814
def check(self, progress_bar=None):
815
"""See VersionedFile.check()."""
817
def _clone_text(self, new_version_id, old_version_id, parents):
818
"""See VersionedFile.clone_text()."""
819
# FIXME RBC 20060228 make fast by only inserting an index with null
821
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
823
def get_lines(self, version_id):
824
"""See VersionedFile.get_lines()."""
825
return self.get_line_list([version_id])[0]
827
def _get_record_map(self, version_ids):
828
"""Produce a dictionary of knit records.
830
The keys are version_ids, the values are tuples of (method, content,
832
method is the way the content should be applied.
833
content is a KnitContent object.
834
digest is the SHA1 digest of this version id after all steps are done
835
next is the build-parent of the version, i.e. the leftmost ancestor.
836
If the method is fulltext, next will be None.
838
position_map = self._get_components_positions(version_ids)
839
# c = component_id, m = method, i_m = index_memo, n = next
840
records = [(c, i_m) for c, (m, i_m, n) in position_map.iteritems()]
842
for component_id, content, digest in \
843
self._data.read_records_iter(records):
844
method, index_memo, next = position_map[component_id]
845
record_map[component_id] = method, content, digest, next
849
def get_text(self, version_id):
850
"""See VersionedFile.get_text"""
851
return self.get_texts([version_id])[0]
853
def get_texts(self, version_ids):
854
return [''.join(l) for l in self.get_line_list(version_ids)]
856
def get_line_list(self, version_ids):
857
"""Return the texts of listed versions as a list of strings."""
858
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
859
for version_id in version_ids:
860
self.check_not_reserved_id(version_id)
861
text_map, content_map = self._get_content_maps(version_ids)
862
return [text_map[v] for v in version_ids]
864
_get_lf_split_line_list = get_line_list
866
def _get_content_maps(self, version_ids):
867
"""Produce maps of text and KnitContents
869
:return: (text_map, content_map) where text_map contains the texts for
870
the requested versions and content_map contains the KnitContents.
871
Both dicts take version_ids as their keys.
873
for version_id in version_ids:
874
if not self.has_version(version_id):
875
raise RevisionNotPresent(version_id, self.filename)
876
record_map = self._get_record_map(version_ids)
881
for version_id in version_ids:
884
while cursor is not None:
885
method, data, digest, next = record_map[cursor]
886
components.append((cursor, method, data, digest))
887
if cursor in content_map:
892
for component_id, method, data, digest in reversed(components):
893
if component_id in content_map:
894
content = content_map[component_id]
896
if method == 'fulltext':
897
assert content is None
898
content = self.factory.parse_fulltext(data, version_id)
899
elif method == 'line-delta':
900
delta = self.factory.parse_line_delta(data, version_id)
901
content = content.copy()
902
content._lines = self._apply_delta(content._lines,
904
content_map[component_id] = content
906
if 'no-eol' in self._index.get_options(version_id):
907
content = content.copy()
908
line = content._lines[-1][1].rstrip('\n')
909
content._lines[-1] = (content._lines[-1][0], line)
910
final_content[version_id] = content
912
# digest here is the digest from the last applied component.
913
text = content.text()
914
if sha_strings(text) != digest:
915
raise KnitCorrupt(self.filename,
916
'sha-1 does not match %s' % version_id)
918
text_map[version_id] = text
919
return text_map, final_content
921
def iter_lines_added_or_present_in_versions(self, version_ids=None,
923
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
924
if version_ids is None:
925
version_ids = self.versions()
927
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
929
pb = progress.DummyProgress()
930
# we don't care about inclusions, the caller cares.
931
# but we need to setup a list of records to visit.
932
# we need version_id, position, length
933
version_id_records = []
934
requested_versions = set(version_ids)
935
# filter for available versions
936
for version_id in requested_versions:
937
if not self.has_version(version_id):
938
raise RevisionNotPresent(version_id, self.filename)
939
# get a in-component-order queue:
940
for version_id in self.versions():
941
if version_id in requested_versions:
942
index_memo = self._index.get_position(version_id)
943
version_id_records.append((version_id, index_memo))
945
total = len(version_id_records)
946
for version_idx, (version_id, data, sha_value) in \
947
enumerate(self._data.read_records_iter(version_id_records)):
948
pb.update('Walking content.', version_idx, total)
949
method = self._index.get_method(version_id)
951
assert method in ('fulltext', 'line-delta')
952
if method == 'fulltext':
953
line_iterator = self.factory.get_fulltext_content(data)
955
line_iterator = self.factory.get_linedelta_content(data)
956
for line in line_iterator:
959
pb.update('Walking content.', total, total)
961
def iter_parents(self, version_ids):
962
"""Iterate through the parents for many version ids.
964
:param version_ids: An iterable yielding version_ids.
965
:return: An iterator that yields (version_id, parents). Requested
966
version_ids not present in the versioned file are simply skipped.
967
The order is undefined, allowing for different optimisations in
968
the underlying implementation.
970
version_ids = [osutils.safe_revision_id(version_id) for
971
version_id in version_ids]
972
return self._index.iter_parents(version_ids)
974
def num_versions(self):
975
"""See VersionedFile.num_versions()."""
976
return self._index.num_versions()
978
__len__ = num_versions
980
def annotate_iter(self, version_id):
981
"""See VersionedFile.annotate_iter."""
982
version_id = osutils.safe_revision_id(version_id)
983
content = self._get_content(version_id)
984
for origin, text in content.annotate_iter():
987
def get_parents(self, version_id):
988
"""See VersionedFile.get_parents."""
991
# 52554 calls in 1264 872 internal down from 3674
992
version_id = osutils.safe_revision_id(version_id)
994
return self._index.get_parents(version_id)
996
raise RevisionNotPresent(version_id, self.filename)
998
def get_parents_with_ghosts(self, version_id):
999
"""See VersionedFile.get_parents."""
1000
version_id = osutils.safe_revision_id(version_id)
1002
return self._index.get_parents_with_ghosts(version_id)
1004
raise RevisionNotPresent(version_id, self.filename)
1006
def get_ancestry(self, versions, topo_sorted=True):
1007
"""See VersionedFile.get_ancestry."""
1008
if isinstance(versions, basestring):
1009
versions = [versions]
1012
versions = [osutils.safe_revision_id(v) for v in versions]
1013
return self._index.get_ancestry(versions, topo_sorted)
1015
def get_ancestry_with_ghosts(self, versions):
1016
"""See VersionedFile.get_ancestry_with_ghosts."""
1017
if isinstance(versions, basestring):
1018
versions = [versions]
1021
versions = [osutils.safe_revision_id(v) for v in versions]
1022
return self._index.get_ancestry_with_ghosts(versions)
1024
#@deprecated_method(zero_eight)
1025
def walk(self, version_ids):
1026
"""See VersionedFile.walk."""
1027
# We take the short path here, and extract all relevant texts
1028
# and put them in a weave and let that do all the work. Far
1029
# from optimal, but is much simpler.
1030
# FIXME RB 20060228 this really is inefficient!
1031
from bzrlib.weave import Weave
1033
w = Weave(self.filename)
1034
ancestry = set(self.get_ancestry(version_ids, topo_sorted=False))
1035
sorted_graph = topo_sort(self._index.get_graph())
1036
version_list = [vid for vid in sorted_graph if vid in ancestry]
1038
for version_id in version_list:
1039
lines = self.get_lines(version_id)
1040
w.add_lines(version_id, self.get_parents(version_id), lines)
1042
for lineno, insert_id, dset, line in w.walk(version_ids):
1043
yield lineno, insert_id, dset, line
1045
def plan_merge(self, ver_a, ver_b):
1046
"""See VersionedFile.plan_merge."""
1047
ver_a = osutils.safe_revision_id(ver_a)
1048
ver_b = osutils.safe_revision_id(ver_b)
1049
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1051
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1052
annotated_a = self.annotate(ver_a)
1053
annotated_b = self.annotate(ver_b)
1054
return merge._plan_annotate_merge(annotated_a, annotated_b,
1055
ancestors_a, ancestors_b)
1058
class _KnitComponentFile(object):
1059
"""One of the files used to implement a knit database"""
1061
def __init__(self, transport, filename, mode, file_mode=None,
1062
create_parent_dir=False, dir_mode=None):
1063
self._transport = transport
1064
self._filename = filename
1066
self._file_mode = file_mode
1067
self._dir_mode = dir_mode
1068
self._create_parent_dir = create_parent_dir
1069
self._need_to_create = False
1071
def _full_path(self):
1072
"""Return the full path to this file."""
1073
return self._transport.base + self._filename
1075
def check_header(self, fp):
1076
line = fp.readline()
1078
# An empty file can actually be treated as though the file doesn't
1080
raise errors.NoSuchFile(self._full_path())
1081
if line != self.HEADER:
1082
raise KnitHeaderError(badline=line,
1083
filename=self._transport.abspath(self._filename))
1086
return '%s(%s)' % (self.__class__.__name__, self._filename)
1089
class _KnitIndex(_KnitComponentFile):
1090
"""Manages knit index file.
1092
The index is already kept in memory and read on startup, to enable
1093
fast lookups of revision information. The cursor of the index
1094
file is always pointing to the end, making it easy to append
1097
_cache is a cache for fast mapping from version id to a Index
1100
_history is a cache for fast mapping from indexes to version ids.
1102
The index data format is dictionary compressed when it comes to
1103
parent references; a index entry may only have parents that with a
1104
lover index number. As a result, the index is topological sorted.
1106
Duplicate entries may be written to the index for a single version id
1107
if this is done then the latter one completely replaces the former:
1108
this allows updates to correct version and parent information.
1109
Note that the two entries may share the delta, and that successive
1110
annotations and references MUST point to the first entry.
1112
The index file on disc contains a header, followed by one line per knit
1113
record. The same revision can be present in an index file more than once.
1114
The first occurrence gets assigned a sequence number starting from 0.
1116
The format of a single line is
1117
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1118
REVISION_ID is a utf8-encoded revision id
1119
FLAGS is a comma separated list of flags about the record. Values include
1120
no-eol, line-delta, fulltext.
1121
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1122
that the the compressed data starts at.
1123
LENGTH is the ascii representation of the length of the data file.
1124
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1126
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1127
revision id already in the knit that is a parent of REVISION_ID.
1128
The ' :' marker is the end of record marker.
1131
when a write is interrupted to the index file, it will result in a line
1132
that does not end in ' :'. If the ' :' is not present at the end of a line,
1133
or at the end of the file, then the record that is missing it will be
1134
ignored by the parser.
1136
When writing new records to the index file, the data is preceded by '\n'
1137
to ensure that records always start on new lines even if the last write was
1138
interrupted. As a result its normal for the last line in the index to be
1139
missing a trailing newline. One can be added with no harmful effects.
1142
HEADER = "# bzr knit index 8\n"
1144
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1145
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1147
def _cache_version(self, version_id, options, pos, size, parents):
1148
"""Cache a version record in the history array and index cache.
1150
This is inlined into _load_data for performance. KEEP IN SYNC.
1151
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1154
# only want the _history index to reference the 1st index entry
1156
if version_id not in self._cache:
1157
index = len(self._history)
1158
self._history.append(version_id)
1160
index = self._cache[version_id][5]
1161
self._cache[version_id] = (version_id,
1168
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1169
create_parent_dir=False, delay_create=False, dir_mode=None):
1170
_KnitComponentFile.__init__(self, transport, filename, mode,
1171
file_mode=file_mode,
1172
create_parent_dir=create_parent_dir,
1175
# position in _history is the 'official' index for a revision
1176
# but the values may have come from a newer entry.
1177
# so - wc -l of a knit index is != the number of unique names
1181
fp = self._transport.get(self._filename)
1183
# _load_data may raise NoSuchFile if the target knit is
1185
_load_data(self, fp)
1189
if mode != 'w' or not create:
1192
self._need_to_create = True
1194
self._transport.put_bytes_non_atomic(
1195
self._filename, self.HEADER, mode=self._file_mode)
1197
def get_graph(self):
1198
"""Return a list of the node:parents lists from this knit index."""
1199
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1201
def get_ancestry(self, versions, topo_sorted=True):
1202
"""See VersionedFile.get_ancestry."""
1203
# get a graph of all the mentioned versions:
1205
pending = set(versions)
1208
version = pending.pop()
1211
parents = [p for p in cache[version][4] if p in cache]
1213
raise RevisionNotPresent(version, self._filename)
1214
# if not completed and not a ghost
1215
pending.update([p for p in parents if p not in graph])
1216
graph[version] = parents
1219
return topo_sort(graph.items())
1221
def get_ancestry_with_ghosts(self, versions):
1222
"""See VersionedFile.get_ancestry_with_ghosts."""
1223
# get a graph of all the mentioned versions:
1224
self.check_versions_present(versions)
1227
pending = set(versions)
1229
version = pending.pop()
1231
parents = cache[version][4]
1237
pending.update([p for p in parents if p not in graph])
1238
graph[version] = parents
1239
return topo_sort(graph.items())
1241
def iter_parents(self, version_ids):
1242
"""Iterate through the parents for many version ids.
1244
:param version_ids: An iterable yielding version_ids.
1245
:return: An iterator that yields (version_id, parents). Requested
1246
version_ids not present in the versioned file are simply skipped.
1247
The order is undefined, allowing for different optimisations in
1248
the underlying implementation.
1250
for version_id in version_ids:
1252
yield version_id, tuple(self.get_parents(version_id))
1256
def num_versions(self):
1257
return len(self._history)
1259
__len__ = num_versions
1261
def get_versions(self):
1262
"""Get all the versions in the file. not topologically sorted."""
1263
return self._history
1265
def _version_list_to_index(self, versions):
1268
for version in versions:
1269
if version in cache:
1270
# -- inlined lookup() --
1271
result_list.append(str(cache[version][5]))
1272
# -- end lookup () --
1274
result_list.append('.' + version)
1275
return ' '.join(result_list)
1277
def add_version(self, version_id, options, index_memo, parents):
1278
"""Add a version record to the index."""
1279
self.add_versions(((version_id, options, index_memo, parents),))
1281
def add_versions(self, versions):
1282
"""Add multiple versions to the index.
1284
:param versions: a list of tuples:
1285
(version_id, options, pos, size, parents).
1288
orig_history = self._history[:]
1289
orig_cache = self._cache.copy()
1292
for version_id, options, (index, pos, size), parents in versions:
1293
line = "\n%s %s %s %s %s :" % (version_id,
1297
self._version_list_to_index(parents))
1298
assert isinstance(line, str), \
1299
'content must be utf-8 encoded: %r' % (line,)
1301
self._cache_version(version_id, options, pos, size, parents)
1302
if not self._need_to_create:
1303
self._transport.append_bytes(self._filename, ''.join(lines))
1306
sio.write(self.HEADER)
1307
sio.writelines(lines)
1309
self._transport.put_file_non_atomic(self._filename, sio,
1310
create_parent_dir=self._create_parent_dir,
1311
mode=self._file_mode,
1312
dir_mode=self._dir_mode)
1313
self._need_to_create = False
1315
# If any problems happen, restore the original values and re-raise
1316
self._history = orig_history
1317
self._cache = orig_cache
1320
def has_version(self, version_id):
1321
"""True if the version is in the index."""
1322
return version_id in self._cache
1324
def get_position(self, version_id):
1325
"""Return details needed to access the version.
1327
.kndx indices do not support split-out data, so return None for the
1330
:return: a tuple (None, data position, size) to hand to the access
1331
logic to get the record.
1333
entry = self._cache[version_id]
1334
return None, entry[2], entry[3]
1336
def get_method(self, version_id):
1337
"""Return compression method of specified version."""
1338
options = self._cache[version_id][1]
1339
if 'fulltext' in options:
1342
if 'line-delta' not in options:
1343
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1346
def get_options(self, version_id):
1347
"""Return a string represention options.
1351
return self._cache[version_id][1]
1353
def get_parents(self, version_id):
1354
"""Return parents of specified version ignoring ghosts."""
1355
return [parent for parent in self._cache[version_id][4]
1356
if parent in self._cache]
1358
def get_parents_with_ghosts(self, version_id):
1359
"""Return parents of specified version with ghosts."""
1360
return self._cache[version_id][4]
1362
def check_versions_present(self, version_ids):
1363
"""Check that all specified versions are present."""
1365
for version_id in version_ids:
1366
if version_id not in cache:
1367
raise RevisionNotPresent(version_id, self._filename)
1370
class KnitGraphIndex(object):
1371
"""A knit index that builds on GraphIndex."""
1373
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1374
"""Construct a KnitGraphIndex on a graph_index.
1376
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1377
:param deltas: Allow delta-compressed records.
1378
:param add_callback: If not None, allow additions to the index and call
1379
this callback with a list of added GraphIndex nodes:
1380
[(node, value, node_refs), ...]
1381
:param parents: If True, record knits parents, if not do not record
1384
self._graph_index = graph_index
1385
self._deltas = deltas
1386
self._add_callback = add_callback
1387
self._parents = parents
1388
if deltas and not parents:
1389
raise KnitCorrupt(self, "Cannot do delta compression without "
1392
def _get_entries(self, keys, check_present=False):
1393
"""Get the entries for keys.
1395
:param keys: An iterable of index keys, - 1-tuples.
1400
for node in self._graph_index.iter_entries(keys):
1402
found_keys.add(node[1])
1404
# adapt parentless index to the rest of the code.
1405
for node in self._graph_index.iter_entries(keys):
1406
yield node[0], node[1], node[2], ()
1407
found_keys.add(node[1])
1409
missing_keys = keys.difference(found_keys)
1411
raise RevisionNotPresent(missing_keys.pop(), self)
1413
def _present_keys(self, version_ids):
1415
node[1] for node in self._get_entries(version_ids)])
1417
def _parentless_ancestry(self, versions):
1418
"""Honour the get_ancestry API for parentless knit indices."""
1419
wanted_keys = self._version_ids_to_keys(versions)
1420
present_keys = self._present_keys(wanted_keys)
1421
missing = set(wanted_keys).difference(present_keys)
1423
raise RevisionNotPresent(missing.pop(), self)
1424
return list(self._keys_to_version_ids(present_keys))
1426
def get_ancestry(self, versions, topo_sorted=True):
1427
"""See VersionedFile.get_ancestry."""
1428
if not self._parents:
1429
return self._parentless_ancestry(versions)
1430
# XXX: This will do len(history) index calls - perhaps
1431
# it should be altered to be a index core feature?
1432
# get a graph of all the mentioned versions:
1435
versions = self._version_ids_to_keys(versions)
1436
pending = set(versions)
1438
# get all pending nodes
1439
this_iteration = pending
1440
new_nodes = self._get_entries(this_iteration)
1443
for (index, key, value, node_refs) in new_nodes:
1444
# dont ask for ghosties - otherwise
1445
# we we can end up looping with pending
1446
# being entirely ghosted.
1447
graph[key] = [parent for parent in node_refs[0]
1448
if parent not in ghosts]
1450
for parent in graph[key]:
1451
# dont examine known nodes again
1456
ghosts.update(this_iteration.difference(found))
1457
if versions.difference(graph):
1458
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1460
result_keys = topo_sort(graph.items())
1462
result_keys = graph.iterkeys()
1463
return [key[0] for key in result_keys]
1465
def get_ancestry_with_ghosts(self, versions):
1466
"""See VersionedFile.get_ancestry."""
1467
if not self._parents:
1468
return self._parentless_ancestry(versions)
1469
# XXX: This will do len(history) index calls - perhaps
1470
# it should be altered to be a index core feature?
1471
# get a graph of all the mentioned versions:
1473
versions = self._version_ids_to_keys(versions)
1474
pending = set(versions)
1476
# get all pending nodes
1477
this_iteration = pending
1478
new_nodes = self._get_entries(this_iteration)
1480
for (index, key, value, node_refs) in new_nodes:
1481
graph[key] = node_refs[0]
1483
for parent in graph[key]:
1484
# dont examine known nodes again
1488
missing_versions = this_iteration.difference(graph)
1489
missing_needed = versions.intersection(missing_versions)
1491
raise RevisionNotPresent(missing_needed.pop(), self)
1492
for missing_version in missing_versions:
1493
# add a key, no parents
1494
graph[missing_version] = []
1495
pending.discard(missing_version) # don't look for it
1496
result_keys = topo_sort(graph.items())
1497
return [key[0] for key in result_keys]
1499
def get_graph(self):
1500
"""Return a list of the node:parents lists from this knit index."""
1501
if not self._parents:
1502
return [(key, ()) for key in self.get_versions()]
1504
for index, key, value, refs in self._graph_index.iter_all_entries():
1505
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1508
def iter_parents(self, version_ids):
1509
"""Iterate through the parents for many version ids.
1511
:param version_ids: An iterable yielding version_ids.
1512
:return: An iterator that yields (version_id, parents). Requested
1513
version_ids not present in the versioned file are simply skipped.
1514
The order is undefined, allowing for different optimisations in
1515
the underlying implementation.
1518
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1520
present_parents = set()
1521
for node in all_nodes:
1522
all_parents.update(node[3][0])
1523
# any node we are querying must be present
1524
present_parents.add(node[1])
1525
unknown_parents = all_parents.difference(present_parents)
1526
present_parents.update(self._present_keys(unknown_parents))
1527
for node in all_nodes:
1529
for parent in node[3][0]:
1530
if parent in present_parents:
1531
parents.append(parent[0])
1532
yield node[1][0], tuple(parents)
1534
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1535
yield node[1][0], ()
1537
def num_versions(self):
1538
return len(list(self._graph_index.iter_all_entries()))
1540
__len__ = num_versions
1542
def get_versions(self):
1543
"""Get all the versions in the file. not topologically sorted."""
1544
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1546
def has_version(self, version_id):
1547
"""True if the version is in the index."""
1548
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1550
def _keys_to_version_ids(self, keys):
1551
return tuple(key[0] for key in keys)
1553
def get_position(self, version_id):
1554
"""Return details needed to access the version.
1556
:return: a tuple (index, data position, size) to hand to the access
1557
logic to get the record.
1559
node = self._get_node(version_id)
1560
bits = node[2][1:].split(' ')
1561
return node[0], int(bits[0]), int(bits[1])
1563
def get_method(self, version_id):
1564
"""Return compression method of specified version."""
1565
if not self._deltas:
1567
return self._parent_compression(self._get_node(version_id)[3][1])
1569
def _parent_compression(self, reference_list):
1570
# use the second reference list to decide if this is delta'd or not.
1571
if len(reference_list):
1576
def _get_node(self, version_id):
1577
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1579
def get_options(self, version_id):
1580
"""Return a string represention options.
1584
node = self._get_node(version_id)
1585
if not self._deltas:
1586
options = ['fulltext']
1588
options = [self._parent_compression(node[3][1])]
1589
if node[2][0] == 'N':
1590
options.append('no-eol')
1593
def get_parents(self, version_id):
1594
"""Return parents of specified version ignoring ghosts."""
1595
parents = list(self.iter_parents([version_id]))
1598
raise errors.RevisionNotPresent(version_id, self)
1599
return parents[0][1]
1601
def get_parents_with_ghosts(self, version_id):
1602
"""Return parents of specified version with ghosts."""
1603
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1604
check_present=True))
1605
if not self._parents:
1607
return self._keys_to_version_ids(nodes[0][3][0])
1609
def check_versions_present(self, version_ids):
1610
"""Check that all specified versions are present."""
1611
keys = self._version_ids_to_keys(version_ids)
1612
present = self._present_keys(keys)
1613
missing = keys.difference(present)
1615
raise RevisionNotPresent(missing.pop(), self)
1617
def add_version(self, version_id, options, access_memo, parents):
1618
"""Add a version record to the index."""
1619
return self.add_versions(((version_id, options, access_memo, parents),))
1621
def add_versions(self, versions):
1622
"""Add multiple versions to the index.
1624
This function does not insert data into the Immutable GraphIndex
1625
backing the KnitGraphIndex, instead it prepares data for insertion by
1626
the caller and checks that it is safe to insert then calls
1627
self._add_callback with the prepared GraphIndex nodes.
1629
:param versions: a list of tuples:
1630
(version_id, options, pos, size, parents).
1632
if not self._add_callback:
1633
raise errors.ReadOnlyError(self)
1634
# we hope there are no repositories with inconsistent parentage
1639
for (version_id, options, access_memo, parents) in versions:
1640
index, pos, size = access_memo
1641
key = (version_id, )
1642
parents = tuple((parent, ) for parent in parents)
1643
if 'no-eol' in options:
1647
value += "%d %d" % (pos, size)
1648
if not self._deltas:
1649
if 'line-delta' in options:
1650
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1653
if 'line-delta' in options:
1654
node_refs = (parents, (parents[0],))
1656
node_refs = (parents, ())
1658
node_refs = (parents, )
1661
raise KnitCorrupt(self, "attempt to add node with parents "
1662
"in parentless index.")
1664
keys[key] = (value, node_refs)
1665
present_nodes = self._get_entries(keys)
1666
for (index, key, value, node_refs) in present_nodes:
1667
if (value, node_refs) != keys[key]:
1668
raise KnitCorrupt(self, "inconsistent details in add_versions"
1669
": %s %s" % ((value, node_refs), keys[key]))
1673
for key, (value, node_refs) in keys.iteritems():
1674
result.append((key, value, node_refs))
1676
for key, (value, node_refs) in keys.iteritems():
1677
result.append((key, value))
1678
self._add_callback(result)
1680
def _version_ids_to_keys(self, version_ids):
1681
return set((version_id, ) for version_id in version_ids)
1684
class _KnitAccess(object):
1685
"""Access to knit records in a .knit file."""
1687
def __init__(self, transport, filename, _file_mode, _dir_mode,
1688
_need_to_create, _create_parent_dir):
1689
"""Create a _KnitAccess for accessing and inserting data.
1691
:param transport: The transport the .knit is located on.
1692
:param filename: The filename of the .knit.
1694
self._transport = transport
1695
self._filename = filename
1696
self._file_mode = _file_mode
1697
self._dir_mode = _dir_mode
1698
self._need_to_create = _need_to_create
1699
self._create_parent_dir = _create_parent_dir
1701
def add_raw_records(self, sizes, raw_data):
1702
"""Add raw knit bytes to a storage area.
1704
The data is spooled to whereever the access method is storing data.
1706
:param sizes: An iterable containing the size of each raw data segment.
1707
:param raw_data: A bytestring containing the data.
1708
:return: A list of memos to retrieve the record later. Each memo is a
1709
tuple - (index, pos, length), where the index field is always None
1710
for the .knit access method.
1712
assert type(raw_data) == str, \
1713
'data must be plain bytes was %s' % type(raw_data)
1714
if not self._need_to_create:
1715
base = self._transport.append_bytes(self._filename, raw_data)
1717
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1718
create_parent_dir=self._create_parent_dir,
1719
mode=self._file_mode,
1720
dir_mode=self._dir_mode)
1721
self._need_to_create = False
1725
result.append((None, base, size))
1730
"""IFF this data access has its own storage area, initialise it.
1734
self._transport.put_bytes_non_atomic(self._filename, '',
1735
mode=self._file_mode)
1737
def open_file(self):
1738
"""IFF this data access can be represented as a single file, open it.
1740
For knits that are not mapped to a single file on disk this will
1743
:return: None or a file handle.
1746
return self._transport.get(self._filename)
1751
def get_raw_records(self, memos_for_retrieval):
1752
"""Get the raw bytes for a records.
1754
:param memos_for_retrieval: An iterable containing the (index, pos,
1755
length) memo for retrieving the bytes. The .knit method ignores
1756
the index as there is always only a single file.
1757
:return: An iterator over the bytes of the records.
1759
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
1760
for pos, data in self._transport.readv(self._filename, read_vector):
1764
class _PackAccess(object):
1765
"""Access to knit records via a collection of packs."""
1767
def __init__(self, index_to_packs, writer=None):
1768
"""Create a _PackAccess object.
1770
:param index_to_packs: A dict mapping index objects to the transport
1771
and file names for obtaining data.
1772
:param writer: A tuple (pack.ContainerWriter, write_index) which
1773
contains the pack to write, and the index that reads from it will
1777
self.container_writer = writer[0]
1778
self.write_index = writer[1]
1780
self.container_writer = None
1781
self.write_index = None
1782
self.indices = index_to_packs
1784
def add_raw_records(self, sizes, raw_data):
1785
"""Add raw knit bytes to a storage area.
1787
The data is spooled to the container writer in one bytes-record per
1790
:param sizes: An iterable containing the size of each raw data segment.
1791
:param raw_data: A bytestring containing the data.
1792
:return: A list of memos to retrieve the record later. Each memo is a
1793
tuple - (index, pos, length), where the index field is the
1794
write_index object supplied to the PackAccess object.
1796
assert type(raw_data) == str, \
1797
'data must be plain bytes was %s' % type(raw_data)
1801
p_offset, p_length = self.container_writer.add_bytes_record(
1802
raw_data[offset:offset+size], [])
1804
result.append((self.write_index, p_offset, p_length))
1808
"""Pack based knits do not get individually created."""
1810
def get_raw_records(self, memos_for_retrieval):
1811
"""Get the raw bytes for a records.
1813
:param memos_for_retrieval: An iterable containing the (index, pos,
1814
length) memo for retrieving the bytes. The Pack access method
1815
looks up the pack to use for a given record in its index_to_pack
1817
:return: An iterator over the bytes of the records.
1819
# first pass, group into same-index requests
1821
current_index = None
1822
for (index, offset, length) in memos_for_retrieval:
1823
if current_index == index:
1824
current_list.append((offset, length))
1826
if current_index is not None:
1827
request_lists.append((current_index, current_list))
1828
current_index = index
1829
current_list = [(offset, length)]
1830
# handle the last entry
1831
if current_index is not None:
1832
request_lists.append((current_index, current_list))
1833
for index, offsets in request_lists:
1834
transport, path = self.indices[index]
1835
reader = pack.make_readv_reader(transport, path, offsets)
1836
for names, read_func in reader.iter_records():
1837
yield read_func(None)
1839
def open_file(self):
1840
"""Pack based knits have no single file."""
1843
def set_writer(self, writer, index, (transport, packname)):
1844
"""Set a writer to use for adding data."""
1845
self.indices[index] = (transport, packname)
1846
self.container_writer = writer
1847
self.write_index = index
1850
class _KnitData(object):
1851
"""Manage extraction of data from a KnitAccess, caching and decompressing.
1853
The KnitData class provides the logic for parsing and using knit records,
1854
making use of an access method for the low level read and write operations.
1857
def __init__(self, access):
1858
"""Create a KnitData object.
1860
:param access: The access method to use. Access methods such as
1861
_KnitAccess manage the insertion of raw records and the subsequent
1862
retrieval of the same.
1864
self._access = access
1865
self._checked = False
1866
# TODO: jam 20060713 conceptually, this could spill to disk
1867
# if the cached size gets larger than a certain amount
1868
# but it complicates the model a bit, so for now just use
1869
# a simple dictionary
1871
self._do_cache = False
1873
def enable_cache(self):
1874
"""Enable caching of reads."""
1875
self._do_cache = True
1877
def clear_cache(self):
1878
"""Clear the record cache."""
1879
self._do_cache = False
1882
def _open_file(self):
1883
return self._access.open_file()
1885
def _record_to_data(self, version_id, digest, lines):
1886
"""Convert version_id, digest, lines into a raw data block.
1888
:return: (len, a StringIO instance with the raw data ready to read.)
1891
data_file = GzipFile(None, mode='wb', fileobj=sio)
1893
assert isinstance(version_id, str)
1894
data_file.writelines(chain(
1895
["version %s %d %s\n" % (version_id,
1899
["end %s\n" % version_id]))
1906
def add_raw_records(self, sizes, raw_data):
1907
"""Append a prepared record to the data file.
1909
:param sizes: An iterable containing the size of each raw data segment.
1910
:param raw_data: A bytestring containing the data.
1911
:return: a list of index data for the way the data was stored.
1912
See the access method add_raw_records documentation for more
1915
return self._access.add_raw_records(sizes, raw_data)
1917
def add_record(self, version_id, digest, lines):
1918
"""Write new text record to disk.
1920
Returns index data for retrieving it later, as per add_raw_records.
1922
size, sio = self._record_to_data(version_id, digest, lines)
1923
result = self.add_raw_records([size], sio.getvalue())
1925
self._cache[version_id] = sio.getvalue()
1928
def _parse_record_header(self, version_id, raw_data):
1929
"""Parse a record header for consistency.
1931
:return: the header and the decompressor stream.
1932
as (stream, header_record)
1934
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1936
rec = self._check_header(version_id, df.readline())
1937
except Exception, e:
1938
raise KnitCorrupt(self._access,
1939
"While reading {%s} got %s(%s)"
1940
% (version_id, e.__class__.__name__, str(e)))
1943
def _check_header(self, version_id, line):
1946
raise KnitCorrupt(self._access,
1947
'unexpected number of elements in record header')
1948
if rec[1] != version_id:
1949
raise KnitCorrupt(self._access,
1950
'unexpected version, wanted %r, got %r'
1951
% (version_id, rec[1]))
1954
def _parse_record(self, version_id, data):
1956
# 4168 calls in 2880 217 internal
1957
# 4168 calls to _parse_record_header in 2121
1958
# 4168 calls to readlines in 330
1959
df = GzipFile(mode='rb', fileobj=StringIO(data))
1962
record_contents = df.readlines()
1963
except Exception, e:
1964
raise KnitCorrupt(self._access,
1965
"While reading {%s} got %s(%s)"
1966
% (version_id, e.__class__.__name__, str(e)))
1967
header = record_contents.pop(0)
1968
rec = self._check_header(version_id, header)
1970
last_line = record_contents.pop()
1971
if len(record_contents) != int(rec[2]):
1972
raise KnitCorrupt(self._access,
1973
'incorrect number of lines %s != %s'
1975
% (len(record_contents), int(rec[2]),
1977
if last_line != 'end %s\n' % rec[1]:
1978
raise KnitCorrupt(self._access,
1979
'unexpected version end line %r, wanted %r'
1980
% (last_line, version_id))
1982
return record_contents, rec[3]
1984
def read_records_iter_raw(self, records):
1985
"""Read text records from data file and yield raw data.
1987
This unpacks enough of the text record to validate the id is
1988
as expected but thats all.
1990
# setup an iterator of the external records:
1991
# uses readv so nice and fast we hope.
1993
# grab the disk data needed.
1995
# Don't check _cache if it is empty
1996
needed_offsets = [index_memo for version_id, index_memo
1998
if version_id not in self._cache]
2000
needed_offsets = [index_memo for version_id, index_memo
2003
raw_records = self._access.get_raw_records(needed_offsets)
2005
for version_id, index_memo in records:
2006
if version_id in self._cache:
2007
# This data has already been validated
2008
data = self._cache[version_id]
2010
data = raw_records.next()
2012
self._cache[version_id] = data
2014
# validate the header
2015
df, rec = self._parse_record_header(version_id, data)
2017
yield version_id, data
2019
def read_records_iter(self, records):
2020
"""Read text records from data file and yield result.
2022
The result will be returned in whatever is the fastest to read.
2023
Not by the order requested. Also, multiple requests for the same
2024
record will only yield 1 response.
2025
:param records: A list of (version_id, pos, len) entries
2026
:return: Yields (version_id, contents, digest) in the order
2027
read, not the order requested
2033
# Skip records we have alread seen
2034
yielded_records = set()
2035
needed_records = set()
2036
for record in records:
2037
if record[0] in self._cache:
2038
if record[0] in yielded_records:
2040
yielded_records.add(record[0])
2041
data = self._cache[record[0]]
2042
content, digest = self._parse_record(record[0], data)
2043
yield (record[0], content, digest)
2045
needed_records.add(record)
2046
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2048
needed_records = sorted(set(records), key=operator.itemgetter(1))
2050
if not needed_records:
2053
# The transport optimizes the fetching as well
2054
# (ie, reads continuous ranges.)
2055
raw_data = self._access.get_raw_records(
2056
[index_memo for version_id, index_memo in needed_records])
2058
for (version_id, index_memo), data in \
2059
izip(iter(needed_records), raw_data):
2060
content, digest = self._parse_record(version_id, data)
2062
self._cache[version_id] = data
2063
yield version_id, content, digest
2065
def read_records(self, records):
2066
"""Read records into a dictionary."""
2068
for record_id, content, digest in \
2069
self.read_records_iter(records):
2070
components[record_id] = (content, digest)
2074
class InterKnit(InterVersionedFile):
2075
"""Optimised code paths for knit to knit operations."""
2077
_matching_file_from_factory = KnitVersionedFile
2078
_matching_file_to_factory = KnitVersionedFile
2081
def is_compatible(source, target):
2082
"""Be compatible with knits. """
2084
return (isinstance(source, KnitVersionedFile) and
2085
isinstance(target, KnitVersionedFile))
2086
except AttributeError:
2089
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2090
"""See InterVersionedFile.join."""
2091
assert isinstance(self.source, KnitVersionedFile)
2092
assert isinstance(self.target, KnitVersionedFile)
2094
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2099
pb = ui.ui_factory.nested_progress_bar()
2101
version_ids = list(version_ids)
2102
if None in version_ids:
2103
version_ids.remove(None)
2105
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2106
this_versions = set(self.target._index.get_versions())
2107
needed_versions = self.source_ancestry - this_versions
2108
cross_check_versions = self.source_ancestry.intersection(this_versions)
2109
mismatched_versions = set()
2110
for version in cross_check_versions:
2111
# scan to include needed parents.
2112
n1 = set(self.target.get_parents_with_ghosts(version))
2113
n2 = set(self.source.get_parents_with_ghosts(version))
2115
# FIXME TEST this check for cycles being introduced works
2116
# the logic is we have a cycle if in our graph we are an
2117
# ancestor of any of the n2 revisions.
2123
parent_ancestors = self.source.get_ancestry(parent)
2124
if version in parent_ancestors:
2125
raise errors.GraphCycleError([parent, version])
2126
# ensure this parent will be available later.
2127
new_parents = n2.difference(n1)
2128
needed_versions.update(new_parents.difference(this_versions))
2129
mismatched_versions.add(version)
2131
if not needed_versions and not mismatched_versions:
2133
full_list = topo_sort(self.source.get_graph())
2135
version_list = [i for i in full_list if (not self.target.has_version(i)
2136
and i in needed_versions)]
2140
copy_queue_records = []
2142
for version_id in version_list:
2143
options = self.source._index.get_options(version_id)
2144
parents = self.source._index.get_parents_with_ghosts(version_id)
2145
# check that its will be a consistent copy:
2146
for parent in parents:
2147
# if source has the parent, we must :
2148
# * already have it or
2149
# * have it scheduled already
2150
# otherwise we don't care
2151
assert (self.target.has_version(parent) or
2152
parent in copy_set or
2153
not self.source.has_version(parent))
2154
index_memo = self.source._index.get_position(version_id)
2155
copy_queue_records.append((version_id, index_memo))
2156
copy_queue.append((version_id, options, parents))
2157
copy_set.add(version_id)
2159
# data suck the join:
2161
total = len(version_list)
2164
for (version_id, raw_data), \
2165
(version_id2, options, parents) in \
2166
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2168
assert version_id == version_id2, 'logic error, inconsistent results'
2170
pb.update("Joining knit", count, total)
2171
raw_records.append((version_id, options, parents, len(raw_data)))
2172
raw_datum.append(raw_data)
2173
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2175
for version in mismatched_versions:
2176
# FIXME RBC 20060309 is this needed?
2177
n1 = set(self.target.get_parents_with_ghosts(version))
2178
n2 = set(self.source.get_parents_with_ghosts(version))
2179
# write a combined record to our history preserving the current
2180
# parents as first in the list
2181
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2182
self.target.fix_parents(version, new_parents)
2188
InterVersionedFile.register_optimiser(InterKnit)
2191
class WeaveToKnit(InterVersionedFile):
2192
"""Optimised code paths for weave to knit operations."""
2194
_matching_file_from_factory = bzrlib.weave.WeaveFile
2195
_matching_file_to_factory = KnitVersionedFile
2198
def is_compatible(source, target):
2199
"""Be compatible with weaves to knits."""
2201
return (isinstance(source, bzrlib.weave.Weave) and
2202
isinstance(target, KnitVersionedFile))
2203
except AttributeError:
2206
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2207
"""See InterVersionedFile.join."""
2208
assert isinstance(self.source, bzrlib.weave.Weave)
2209
assert isinstance(self.target, KnitVersionedFile)
2211
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2216
pb = ui.ui_factory.nested_progress_bar()
2218
version_ids = list(version_ids)
2220
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2221
this_versions = set(self.target._index.get_versions())
2222
needed_versions = self.source_ancestry - this_versions
2223
cross_check_versions = self.source_ancestry.intersection(this_versions)
2224
mismatched_versions = set()
2225
for version in cross_check_versions:
2226
# scan to include needed parents.
2227
n1 = set(self.target.get_parents_with_ghosts(version))
2228
n2 = set(self.source.get_parents(version))
2229
# if all of n2's parents are in n1, then its fine.
2230
if n2.difference(n1):
2231
# FIXME TEST this check for cycles being introduced works
2232
# the logic is we have a cycle if in our graph we are an
2233
# ancestor of any of the n2 revisions.
2239
parent_ancestors = self.source.get_ancestry(parent)
2240
if version in parent_ancestors:
2241
raise errors.GraphCycleError([parent, version])
2242
# ensure this parent will be available later.
2243
new_parents = n2.difference(n1)
2244
needed_versions.update(new_parents.difference(this_versions))
2245
mismatched_versions.add(version)
2247
if not needed_versions and not mismatched_versions:
2249
full_list = topo_sort(self.source.get_graph())
2251
version_list = [i for i in full_list if (not self.target.has_version(i)
2252
and i in needed_versions)]
2256
total = len(version_list)
2257
for version_id in version_list:
2258
pb.update("Converting to knit", count, total)
2259
parents = self.source.get_parents(version_id)
2260
# check that its will be a consistent copy:
2261
for parent in parents:
2262
# if source has the parent, we must already have it
2263
assert (self.target.has_version(parent))
2264
self.target.add_lines(
2265
version_id, parents, self.source.get_lines(version_id))
2268
for version in mismatched_versions:
2269
# FIXME RBC 20060309 is this needed?
2270
n1 = set(self.target.get_parents_with_ghosts(version))
2271
n2 = set(self.source.get_parents(version))
2272
# write a combined record to our history preserving the current
2273
# parents as first in the list
2274
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2275
self.target.fix_parents(version, new_parents)
2281
InterVersionedFile.register_optimiser(WeaveToKnit)
2284
class KnitSequenceMatcher(difflib.SequenceMatcher):
2285
"""Knit tuned sequence matcher.
2287
This is based on profiling of difflib which indicated some improvements
2288
for our usage pattern.
2291
def find_longest_match(self, alo, ahi, blo, bhi):
2292
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
2294
If isjunk is not defined:
2296
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
2297
alo <= i <= i+k <= ahi
2298
blo <= j <= j+k <= bhi
2299
and for all (i',j',k') meeting those conditions,
2302
and if i == i', j <= j'
2304
In other words, of all maximal matching blocks, return one that
2305
starts earliest in a, and of all those maximal matching blocks that
2306
start earliest in a, return the one that starts earliest in b.
2308
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
2309
>>> s.find_longest_match(0, 5, 0, 9)
2312
If isjunk is defined, first the longest matching block is
2313
determined as above, but with the additional restriction that no
2314
junk element appears in the block. Then that block is extended as
2315
far as possible by matching (only) junk elements on both sides. So
2316
the resulting block never matches on junk except as identical junk
2317
happens to be adjacent to an "interesting" match.
2319
Here's the same example as before, but considering blanks to be
2320
junk. That prevents " abcd" from matching the " abcd" at the tail
2321
end of the second sequence directly. Instead only the "abcd" can
2322
match, and matches the leftmost "abcd" in the second sequence:
2324
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
2325
>>> s.find_longest_match(0, 5, 0, 9)
2328
If no blocks match, return (alo, blo, 0).
2330
>>> s = SequenceMatcher(None, "ab", "c")
2331
>>> s.find_longest_match(0, 2, 0, 1)
2335
# CAUTION: stripping common prefix or suffix would be incorrect.
2339
# Longest matching block is "ab", but if common prefix is
2340
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
2341
# strip, so ends up claiming that ab is changed to acab by
2342
# inserting "ca" in the middle. That's minimal but unintuitive:
2343
# "it's obvious" that someone inserted "ac" at the front.
2344
# Windiff ends up at the same place as diff, but by pairing up
2345
# the unique 'b's and then matching the first two 'a's.
2347
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
2348
besti, bestj, bestsize = alo, blo, 0
2349
# find longest junk-free match
2350
# during an iteration of the loop, j2len[j] = length of longest
2351
# junk-free match ending with a[i-1] and b[j]
2355
for i in xrange(alo, ahi):
2356
# look at all instances of a[i] in b; note that because
2357
# b2j has no junk keys, the loop is skipped if a[i] is junk
2358
j2lenget = j2len.get
2361
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
2362
# following improvement
2363
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
2364
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
2365
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
2367
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
2368
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
2369
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
2381
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
2383
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
2386
# Extend the best by non-junk elements on each end. In particular,
2387
# "popular" non-junk elements aren't in b2j, which greatly speeds
2388
# the inner loop above, but also means "the best" match so far
2389
# doesn't contain any junk *or* popular non-junk elements.
2390
while besti > alo and bestj > blo and \
2391
not isbjunk(b[bestj-1]) and \
2392
a[besti-1] == b[bestj-1]:
2393
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
2394
while besti+bestsize < ahi and bestj+bestsize < bhi and \
2395
not isbjunk(b[bestj+bestsize]) and \
2396
a[besti+bestsize] == b[bestj+bestsize]:
2399
# Now that we have a wholly interesting match (albeit possibly
2400
# empty!), we may as well suck up the matching junk on each
2401
# side of it too. Can't think of a good reason not to, and it
2402
# saves post-processing the (possibly considerable) expense of
2403
# figuring out what to do with it. In the case of an empty
2404
# interesting match, this is clearly the right thing to do,
2405
# because no other kind of match is possible in the regions.
2406
while besti > alo and bestj > blo and \
2407
isbjunk(b[bestj-1]) and \
2408
a[besti-1] == b[bestj-1]:
2409
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
2410
while besti+bestsize < ahi and bestj+bestsize < bhi and \
2411
isbjunk(b[bestj+bestsize]) and \
2412
a[besti+bestsize] == b[bestj+bestsize]:
2413
bestsize = bestsize + 1
2415
return besti, bestj, bestsize
2419
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2421
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
