1
# Copyright (C) 2005, 2006 by Canonical Ltd
2
# Written by Martin Pool.
3
# Modified by Johan Rydberg <jrydberg@gnu.org>
4
# Modified by Robert Collins <robert.collins@canonical.com>
5
# Modified by Aaron Bentley <aaron.bentley@utoronto.ca>
7
# This program is free software; you can redistribute it and/or modify
8
# it under the terms of the GNU General Public License as published by
9
# the Free Software Foundation; either version 2 of the License, or
10
# (at your option) any later version.
12
# This program is distributed in the hope that it will be useful,
13
# but WITHOUT ANY WARRANTY; without even the implied warranty of
14
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15
# GNU General Public License for more details.
17
# You should have received a copy of the GNU General Public License
18
# along with this program; if not, write to the Free Software
19
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21
"""Knit versionedfile implementation.
23
A knit is a versioned file implementation that supports efficient append only
27
lifeless: the data file is made up of "delta records". each delta record has a delta header
28
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
29
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
30
end-marker; simply "end VERSION"
32
delta can be line or full contents.a
33
... the 8's there are the index number of the annotation.
34
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
38
8 e.set('executable', 'yes')
40
8 if elt.get('executable') == 'yes':
41
8 ie.executable = True
42
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
46
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
47
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
48
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
49
09:33 < lifeless> right
50
09:33 < jrydberg> lifeless: the position and size is the range in the data file
53
so the index sequence is the dictionary compressed sequence number used
54
in the deltas to provide line annotation
59
# 10:16 < lifeless> make partial index writes safe
60
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
61
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
63
# move sha1 out of the content so that join is faster at verifying parents
64
# record content length ?
68
from cStringIO import StringIO
70
from itertools import izip, chain
77
import bzrlib.errors as errors
78
from bzrlib.errors import FileExists, NoSuchFile, KnitError, \
79
InvalidRevisionId, KnitCorrupt, KnitHeaderError, \
80
RevisionNotPresent, RevisionAlreadyPresent
81
from bzrlib.tuned_gzip import GzipFile
82
from bzrlib.trace import mutter
83
from bzrlib.osutils import contains_whitespace, contains_linebreaks, \
85
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
86
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
87
from bzrlib.tsort import topo_sort
91
# TODO: Split out code specific to this format into an associated object.
93
# TODO: Can we put in some kind of value to check that the index and data
94
# files belong together?
96
# TODO: accommodate binaries, perhaps by storing a byte count
98
# TODO: function to check whole file
100
# TODO: atomically append data, then measure backwards from the cursor
101
# position after writing to work out where it was located. we may need to
102
# bypass python file buffering.
104
DATA_SUFFIX = '.knit'
105
INDEX_SUFFIX = '.kndx'
108
class KnitContent(object):
109
"""Content of a knit version to which deltas can be applied."""
111
def __init__(self, lines):
114
def annotate_iter(self):
115
"""Yield tuples of (origin, text) for each content line."""
116
for origin, text in self._lines:
120
"""Return a list of (origin, text) tuples."""
121
return list(self.annotate_iter())
123
def line_delta_iter(self, new_lines):
124
"""Generate line-based delta from this content to new_lines."""
125
new_texts = [text for origin, text in new_lines._lines]
126
old_texts = [text for origin, text in self._lines]
127
s = KnitSequenceMatcher(None, old_texts, new_texts)
128
for op in s.get_opcodes():
131
# ofrom oto length data
132
yield (op[1], op[2], op[4]-op[3], new_lines._lines[op[3]:op[4]])
134
def line_delta(self, new_lines):
135
return list(self.line_delta_iter(new_lines))
138
return [text for origin, text in self._lines]
141
return KnitContent(self._lines[:])
144
class _KnitFactory(object):
145
"""Base factory for creating content objects."""
147
def make(self, lines, version):
148
num_lines = len(lines)
149
return KnitContent(zip([version] * num_lines, lines))
152
class KnitAnnotateFactory(_KnitFactory):
153
"""Factory for creating annotated Content objects."""
157
def parse_fulltext(self, content, version):
158
"""Convert fulltext to internal representation
160
fulltext content is of the format
161
revid(utf8) plaintext\n
162
internal representation is of the format:
167
origin, text = line.split(' ', 1)
168
lines.append((origin.decode('utf-8'), text))
169
return KnitContent(lines)
171
def parse_line_delta_iter(self, lines):
172
for result_item in self.parse_line_delta[lines]:
175
def parse_line_delta(self, lines, version):
176
"""Convert a line based delta into internal representation.
178
line delta is in the form of:
179
intstart intend intcount
181
revid(utf8) newline\n
182
internal representation is
183
(start, end, count, [1..count tuples (revid, newline)])
188
# walk through the lines parsing.
190
start, end, count = [int(n) for n in header.split(',')]
194
origin, text = next().split(' ', 1)
196
contents.append((origin.decode('utf-8'), text))
197
result.append((start, end, count, contents))
200
def lower_fulltext(self, content):
201
"""convert a fulltext content record into a serializable form.
203
see parse_fulltext which this inverts.
205
return ['%s %s' % (o.encode('utf-8'), t) for o, t in content._lines]
207
def lower_line_delta(self, delta):
208
"""convert a delta into a serializable form.
210
See parse_line_delta which this inverts.
213
for start, end, c, lines in delta:
214
out.append('%d,%d,%d\n' % (start, end, c))
215
for origin, text in lines:
216
out.append('%s %s' % (origin.encode('utf-8'), text))
220
class KnitPlainFactory(_KnitFactory):
221
"""Factory for creating plain Content objects."""
225
def parse_fulltext(self, content, version):
226
"""This parses an unannotated fulltext.
228
Note that this is not a noop - the internal representation
229
has (versionid, line) - its just a constant versionid.
231
return self.make(content, version)
233
def parse_line_delta_iter(self, lines, version):
235
header = lines.pop(0)
236
start, end, c = [int(n) for n in header.split(',')]
237
yield start, end, c, zip([version] * c, lines[:c])
240
def parse_line_delta(self, lines, version):
241
return list(self.parse_line_delta_iter(lines, version))
243
def lower_fulltext(self, content):
244
return content.text()
246
def lower_line_delta(self, delta):
248
for start, end, c, lines in delta:
249
out.append('%d,%d,%d\n' % (start, end, c))
250
out.extend([text for origin, text in lines])
254
def make_empty_knit(transport, relpath):
255
"""Construct a empty knit at the specified location."""
256
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
260
class KnitVersionedFile(VersionedFile):
261
"""Weave-like structure with faster random access.
263
A knit stores a number of texts and a summary of the relationships
264
between them. Texts are identified by a string version-id. Texts
265
are normally stored and retrieved as a series of lines, but can
266
also be passed as single strings.
268
Lines are stored with the trailing newline (if any) included, to
269
avoid special cases for files with no final newline. Lines are
270
composed of 8-bit characters, not unicode. The combination of
271
these approaches should mean any 'binary' file can be safely
272
stored and retrieved.
275
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
276
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
278
"""Construct a knit at location specified by relpath.
280
:param create: If not True, only open an existing knit.
282
if deprecated_passed(basis_knit):
283
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
284
" deprecated as of bzr 0.9.",
285
DeprecationWarning, stacklevel=2)
286
if access_mode is None:
288
super(KnitVersionedFile, self).__init__(access_mode)
289
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
290
self.transport = transport
291
self.filename = relpath
292
self.factory = factory or KnitAnnotateFactory()
293
self.writable = (access_mode == 'w')
296
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
297
access_mode, create=create, file_mode=file_mode)
298
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
299
access_mode, create=create and not len(self), file_mode=file_mode)
302
return '%s(%s)' % (self.__class__.__name__,
303
self.transport.abspath(self.filename))
305
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
306
"""See VersionedFile._add_delta()."""
307
self._check_add(version_id, []) # should we check the lines ?
308
self._check_versions_present(parents)
312
for parent in parents:
313
if not self.has_version(parent):
314
ghosts.append(parent)
316
present_parents.append(parent)
318
if delta_parent is None:
319
# reconstitute as full text.
320
assert len(delta) == 1 or len(delta) == 0
322
assert delta[0][0] == 0
323
assert delta[0][1] == 0, delta[0][1]
324
return super(KnitVersionedFile, self)._add_delta(version_id,
335
options.append('no-eol')
337
if delta_parent is not None:
338
# determine the current delta chain length.
339
# To speed the extract of texts the delta chain is limited
340
# to a fixed number of deltas. This should minimize both
341
# I/O and the time spend applying deltas.
343
delta_parents = [delta_parent]
345
parent = delta_parents[0]
346
method = self._index.get_method(parent)
347
if method == 'fulltext':
349
delta_parents = self._index.get_parents(parent)
351
if method == 'line-delta':
352
# did not find a fulltext in the delta limit.
353
# just do a normal insertion.
354
return super(KnitVersionedFile, self)._add_delta(version_id,
361
options.append('line-delta')
362
store_lines = self.factory.lower_line_delta(delta)
364
where, size = self._data.add_record(version_id, digest, store_lines)
365
self._index.add_version(version_id, options, where, size, parents)
367
def _add_raw_records(self, records, data):
368
"""Add all the records 'records' with data pre-joined in 'data'.
370
:param records: A list of tuples(version_id, options, parents, size).
371
:param data: The data for the records. When it is written, the records
372
are adjusted to have pos pointing into data by the sum of
373
the preceding records sizes.
376
pos = self._data.add_raw_record(data)
378
for (version_id, options, parents, size) in records:
379
index_entries.append((version_id, options, pos, size, parents))
381
self._index.add_versions(index_entries)
383
def clear_cache(self):
384
"""Clear the data cache only."""
385
self._data.clear_cache()
387
def copy_to(self, name, transport):
388
"""See VersionedFile.copy_to()."""
389
# copy the current index to a temp index to avoid racing with local
391
transport.put(name + INDEX_SUFFIX + '.tmp', self.transport.get(self._index._filename),)
393
f = self._data._open_file()
395
transport.put(name + DATA_SUFFIX, f)
398
# move the copied index into place
399
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
401
def create_empty(self, name, transport, mode=None):
402
return KnitVersionedFile(name, transport, factory=self.factory, delta=self.delta, create=True)
404
def _fix_parents(self, version, new_parents):
405
"""Fix the parents list for version.
407
This is done by appending a new version to the index
408
with identical data except for the parents list.
409
the parents list must be a superset of the current
412
current_values = self._index._cache[version]
413
assert set(current_values[4]).difference(set(new_parents)) == set()
414
self._index.add_version(version,
420
def get_delta(self, version_id):
421
"""Get a delta for constructing version from some other version."""
422
if not self.has_version(version_id):
423
raise RevisionNotPresent(version_id, self.filename)
425
parents = self.get_parents(version_id)
430
data_pos, data_size = self._index.get_position(version_id)
431
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
432
version_idx = self._index.lookup(version_id)
433
noeol = 'no-eol' in self._index.get_options(version_id)
434
if 'fulltext' == self._index.get_method(version_id):
435
new_content = self.factory.parse_fulltext(data, version_idx)
436
if parent is not None:
437
reference_content = self._get_content(parent)
438
old_texts = reference_content.text()
441
new_texts = new_content.text()
442
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
443
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
445
delta = self.factory.parse_line_delta(data, version_idx)
446
return parent, sha1, noeol, delta
448
def get_graph_with_ghosts(self):
449
"""See VersionedFile.get_graph_with_ghosts()."""
450
graph_items = self._index.get_graph()
451
return dict(graph_items)
453
def get_sha1(self, version_id):
454
"""See VersionedFile.get_sha1()."""
455
record_map = self._get_record_map([version_id])
456
method, content, digest, next = record_map[version_id]
461
"""See VersionedFile.get_suffixes()."""
462
return [DATA_SUFFIX, INDEX_SUFFIX]
464
def has_ghost(self, version_id):
465
"""True if there is a ghost reference in the file to version_id."""
467
if self.has_version(version_id):
469
# optimisable if needed by memoising the _ghosts set.
470
items = self._index.get_graph()
471
for node, parents in items:
472
for parent in parents:
473
if parent not in self._index._cache:
474
if parent == version_id:
479
"""See VersionedFile.versions."""
480
return self._index.get_versions()
482
def has_version(self, version_id):
483
"""See VersionedFile.has_version."""
484
return self._index.has_version(version_id)
486
__contains__ = has_version
488
def _merge_annotations(self, content, parents, parent_texts={},
489
delta=None, annotated=None):
490
"""Merge annotations for content. This is done by comparing
491
the annotations based on changed to the text.
495
for parent_id in parents:
496
merge_content = self._get_content(parent_id, parent_texts)
497
seq = KnitSequenceMatcher(None, merge_content.text(), content.text())
498
if delta_seq is None:
499
# setup a delta seq to reuse.
501
for i, j, n in seq.get_matching_blocks():
504
# this appears to copy (origin, text) pairs across to the new
505
# content for any line that matches the last-checked parent.
506
# FIXME: save the sequence control data for delta compression
507
# against the most relevant parent rather than rediffing.
508
content._lines[j:j+n] = merge_content._lines[i:i+n]
511
reference_content = self._get_content(parents[0], parent_texts)
512
new_texts = content.text()
513
old_texts = reference_content.text()
514
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
515
return self._make_line_delta(delta_seq, content)
517
def _make_line_delta(self, delta_seq, new_content):
518
"""Generate a line delta from delta_seq and new_content."""
520
for op in delta_seq.get_opcodes():
523
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
526
def _get_components_positions(self, version_ids):
527
"""Produce a map of position data for the components of versions.
529
This data is intended to be used for retrieving the knit records.
531
A dict of version_id to (method, data_pos, data_size, next) is
533
method is the way referenced data should be applied.
534
data_pos is the position of the data in the knit.
535
data_size is the size of the data in the knit.
536
next is the build-parent of the version, or None for fulltexts.
539
for version_id in version_ids:
542
while cursor is not None and cursor not in component_data:
543
method = self._index.get_method(cursor)
544
if method == 'fulltext':
547
next = self.get_parents(cursor)[0]
548
data_pos, data_size = self._index.get_position(cursor)
549
component_data[cursor] = (method, data_pos, data_size, next)
551
return component_data
553
def _get_content(self, version_id, parent_texts={}):
554
"""Returns a content object that makes up the specified
556
if not self.has_version(version_id):
557
raise RevisionNotPresent(version_id, self.filename)
559
cached_version = parent_texts.get(version_id, None)
560
if cached_version is not None:
561
return cached_version
563
text_map, contents_map = self._get_content_maps([version_id])
564
return contents_map[version_id]
566
def _check_versions_present(self, version_ids):
567
"""Check that all specified versions are present."""
568
version_ids = set(version_ids)
569
for r in list(version_ids):
570
if self._index.has_version(r):
571
version_ids.remove(r)
573
raise RevisionNotPresent(list(version_ids)[0], self.filename)
575
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
576
"""See VersionedFile.add_lines_with_ghosts()."""
577
self._check_add(version_id, lines)
578
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
580
def _add_lines(self, version_id, parents, lines, parent_texts):
581
"""See VersionedFile.add_lines."""
582
self._check_add(version_id, lines)
583
self._check_versions_present(parents)
584
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
586
def _check_add(self, version_id, lines):
587
"""check that version_id and lines are safe to add."""
588
assert self.writable, "knit is not opened for write"
589
### FIXME escape. RBC 20060228
590
if contains_whitespace(version_id):
591
raise InvalidRevisionId(version_id, self.filename)
592
if self.has_version(version_id):
593
raise RevisionAlreadyPresent(version_id, self.filename)
594
self._check_lines_not_unicode(lines)
595
self._check_lines_are_lines(lines)
597
def _add(self, version_id, lines, parents, delta, parent_texts):
598
"""Add a set of lines on top of version specified by parents.
600
If delta is true, compress the text as a line-delta against
603
Any versions not present will be converted into ghosts.
605
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
606
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
607
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
608
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
609
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
610
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
611
# +1383 0 8.0370 8.0370 +<len>
612
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
613
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
614
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
615
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
619
if parent_texts is None:
621
for parent in parents:
622
if not self.has_version(parent):
623
ghosts.append(parent)
625
present_parents.append(parent)
627
if delta and not len(present_parents):
630
digest = sha_strings(lines)
633
if lines[-1][-1] != '\n':
634
options.append('no-eol')
635
lines[-1] = lines[-1] + '\n'
637
if len(present_parents) and delta:
638
# To speed the extract of texts the delta chain is limited
639
# to a fixed number of deltas. This should minimize both
640
# I/O and the time spend applying deltas.
642
delta_parents = present_parents
644
parent = delta_parents[0]
645
method = self._index.get_method(parent)
646
if method == 'fulltext':
648
delta_parents = self._index.get_parents(parent)
650
if method == 'line-delta':
653
lines = self.factory.make(lines, version_id)
654
if delta or (self.factory.annotated and len(present_parents) > 0):
655
# Merge annotations from parent texts if so is needed.
656
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
657
delta, self.factory.annotated)
660
options.append('line-delta')
661
store_lines = self.factory.lower_line_delta(delta_hunks)
663
options.append('fulltext')
664
store_lines = self.factory.lower_fulltext(lines)
666
where, size = self._data.add_record(version_id, digest, store_lines)
667
self._index.add_version(version_id, options, where, size, parents)
670
def check(self, progress_bar=None):
671
"""See VersionedFile.check()."""
673
def _clone_text(self, new_version_id, old_version_id, parents):
674
"""See VersionedFile.clone_text()."""
675
# FIXME RBC 20060228 make fast by only inserting an index with null
677
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
679
def get_lines(self, version_id):
680
"""See VersionedFile.get_lines()."""
681
return self.get_line_list([version_id])[0]
683
def _get_record_map(self, version_ids):
684
"""Produce a dictionary of knit records.
686
The keys are version_ids, the values are tuples of (method, content,
688
method is the way the content should be applied.
689
content is a KnitContent object.
690
digest is the SHA1 digest of this version id after all steps are done
691
next is the build-parent of the version, i.e. the leftmost ancestor.
692
If the method is fulltext, next will be None.
694
position_map = self._get_components_positions(version_ids)
695
# c = component_id, m = method, p = position, s = size, n = next
696
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
698
for component_id, content, digest in\
699
self._data.read_records_iter(records):
700
method, position, size, next = position_map[component_id]
701
record_map[component_id] = method, content, digest, next
705
def get_text(self, version_id):
706
"""See VersionedFile.get_text"""
707
return self.get_texts([version_id])[0]
709
def get_texts(self, version_ids):
710
return [''.join(l) for l in self.get_line_list(version_ids)]
712
def get_line_list(self, version_ids):
713
"""Return the texts of listed versions as a list of strings."""
714
text_map, content_map = self._get_content_maps(version_ids)
715
return [text_map[v] for v in version_ids]
717
def _get_content_maps(self, version_ids):
718
"""Produce maps of text and KnitContents
720
:return: (text_map, content_map) where text_map contains the texts for
721
the requested versions and content_map contains the KnitContents.
722
Both dicts take version_ids as their keys.
724
for version_id in version_ids:
725
if not self.has_version(version_id):
726
raise RevisionNotPresent(version_id, self.filename)
727
record_map = self._get_record_map(version_ids)
732
for version_id in version_ids:
735
while cursor is not None:
736
method, data, digest, next = record_map[cursor]
737
components.append((cursor, method, data, digest))
738
if cursor in content_map:
743
for component_id, method, data, digest in reversed(components):
744
if component_id in content_map:
745
content = content_map[component_id]
747
version_idx = self._index.lookup(component_id)
748
if method == 'fulltext':
749
assert content is None
750
content = self.factory.parse_fulltext(data, version_idx)
751
elif method == 'line-delta':
752
delta = self.factory.parse_line_delta(data[:],
754
content = content.copy()
755
content._lines = self._apply_delta(content._lines,
757
content_map[component_id] = content
759
if 'no-eol' in self._index.get_options(version_id):
760
content = content.copy()
761
line = content._lines[-1][1].rstrip('\n')
762
content._lines[-1] = (content._lines[-1][0], line)
763
final_content[version_id] = content
765
# digest here is the digest from the last applied component.
766
text = content.text()
767
if sha_strings(text) != digest:
768
raise KnitCorrupt(self.filename,
769
'sha-1 does not match %s' % version_id)
771
text_map[version_id] = text
772
return text_map, final_content
774
def iter_lines_added_or_present_in_versions(self, version_ids=None):
775
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
776
if version_ids is None:
777
version_ids = self.versions()
778
# we don't care about inclusions, the caller cares.
779
# but we need to setup a list of records to visit.
780
# we need version_id, position, length
781
version_id_records = []
782
requested_versions = list(version_ids)
783
# filter for available versions
784
for version_id in requested_versions:
785
if not self.has_version(version_id):
786
raise RevisionNotPresent(version_id, self.filename)
787
# get a in-component-order queue:
789
for version_id in self.versions():
790
if version_id in requested_versions:
791
version_ids.append(version_id)
792
data_pos, length = self._index.get_position(version_id)
793
version_id_records.append((version_id, data_pos, length))
795
pb = bzrlib.ui.ui_factory.nested_progress_bar()
797
total = len(version_id_records)
799
pb.update('Walking content.', count, total)
800
for version_id, data, sha_value in \
801
self._data.read_records_iter(version_id_records):
802
pb.update('Walking content.', count, total)
803
method = self._index.get_method(version_id)
804
version_idx = self._index.lookup(version_id)
805
assert method in ('fulltext', 'line-delta')
806
if method == 'fulltext':
807
content = self.factory.parse_fulltext(data, version_idx)
808
for line in content.text():
811
delta = self.factory.parse_line_delta(data, version_idx)
812
for start, end, count, lines in delta:
813
for origin, line in lines:
816
pb.update('Walking content.', total, total)
819
pb.update('Walking content.', total, total)
823
def num_versions(self):
824
"""See VersionedFile.num_versions()."""
825
return self._index.num_versions()
827
__len__ = num_versions
829
def annotate_iter(self, version_id):
830
"""See VersionedFile.annotate_iter."""
831
content = self._get_content(version_id)
832
for origin, text in content.annotate_iter():
835
def get_parents(self, version_id):
836
"""See VersionedFile.get_parents."""
839
# 52554 calls in 1264 872 internal down from 3674
841
return self._index.get_parents(version_id)
843
raise RevisionNotPresent(version_id, self.filename)
845
def get_parents_with_ghosts(self, version_id):
846
"""See VersionedFile.get_parents."""
848
return self._index.get_parents_with_ghosts(version_id)
850
raise RevisionNotPresent(version_id, self.filename)
852
def get_ancestry(self, versions):
853
"""See VersionedFile.get_ancestry."""
854
if isinstance(versions, basestring):
855
versions = [versions]
858
self._check_versions_present(versions)
859
return self._index.get_ancestry(versions)
861
def get_ancestry_with_ghosts(self, versions):
862
"""See VersionedFile.get_ancestry_with_ghosts."""
863
if isinstance(versions, basestring):
864
versions = [versions]
867
self._check_versions_present(versions)
868
return self._index.get_ancestry_with_ghosts(versions)
870
#@deprecated_method(zero_eight)
871
def walk(self, version_ids):
872
"""See VersionedFile.walk."""
873
# We take the short path here, and extract all relevant texts
874
# and put them in a weave and let that do all the work. Far
875
# from optimal, but is much simpler.
876
# FIXME RB 20060228 this really is inefficient!
877
from bzrlib.weave import Weave
879
w = Weave(self.filename)
880
ancestry = self.get_ancestry(version_ids)
881
sorted_graph = topo_sort(self._index.get_graph())
882
version_list = [vid for vid in sorted_graph if vid in ancestry]
884
for version_id in version_list:
885
lines = self.get_lines(version_id)
886
w.add_lines(version_id, self.get_parents(version_id), lines)
888
for lineno, insert_id, dset, line in w.walk(version_ids):
889
yield lineno, insert_id, dset, line
891
def plan_merge(self, ver_a, ver_b):
892
"""See VersionedFile.plan_merge."""
893
ancestors_b = set(self.get_ancestry(ver_b))
894
def status_a(revision, text):
895
if revision in ancestors_b:
896
return 'killed-b', text
900
ancestors_a = set(self.get_ancestry(ver_a))
901
def status_b(revision, text):
902
if revision in ancestors_a:
903
return 'killed-a', text
907
annotated_a = self.annotate(ver_a)
908
annotated_b = self.annotate(ver_b)
909
plain_a = [t for (a, t) in annotated_a]
910
plain_b = [t for (a, t) in annotated_b]
911
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
914
for ai, bi, l in blocks:
915
# process all mismatched sections
916
# (last mismatched section is handled because blocks always
917
# includes a 0-length last block)
918
for revision, text in annotated_a[a_cur:ai]:
919
yield status_a(revision, text)
920
for revision, text in annotated_b[b_cur:bi]:
921
yield status_b(revision, text)
923
# and now the matched section
926
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
927
assert text_a == text_b
928
yield "unchanged", text_a
931
class _KnitComponentFile(object):
932
"""One of the files used to implement a knit database"""
934
def __init__(self, transport, filename, mode, file_mode=None):
935
self._transport = transport
936
self._filename = filename
938
self._file_mode=file_mode
940
def write_header(self):
941
if self._transport.append(self._filename, StringIO(self.HEADER),
942
mode=self._file_mode):
943
raise KnitCorrupt(self._filename, 'misaligned after writing header')
945
def check_header(self, fp):
947
if line != self.HEADER:
948
raise KnitHeaderError(badline=line)
951
"""Commit is a nop."""
954
return '%s(%s)' % (self.__class__.__name__, self._filename)
957
class _KnitIndex(_KnitComponentFile):
958
"""Manages knit index file.
960
The index is already kept in memory and read on startup, to enable
961
fast lookups of revision information. The cursor of the index
962
file is always pointing to the end, making it easy to append
965
_cache is a cache for fast mapping from version id to a Index
968
_history is a cache for fast mapping from indexes to version ids.
970
The index data format is dictionary compressed when it comes to
971
parent references; a index entry may only have parents that with a
972
lover index number. As a result, the index is topological sorted.
974
Duplicate entries may be written to the index for a single version id
975
if this is done then the latter one completely replaces the former:
976
this allows updates to correct version and parent information.
977
Note that the two entries may share the delta, and that successive
978
annotations and references MUST point to the first entry.
980
The index file on disc contains a header, followed by one line per knit
981
record. The same revision can be present in an index file more than once.
982
The first occurrence gets assigned a sequence number starting from 0.
984
The format of a single line is
985
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
986
REVISION_ID is a utf8-encoded revision id
987
FLAGS is a comma separated list of flags about the record. Values include
988
no-eol, line-delta, fulltext.
989
BYTE_OFFSET is the ascii representation of the byte offset in the data file
990
that the the compressed data starts at.
991
LENGTH is the ascii representation of the length of the data file.
992
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
994
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
995
revision id already in the knit that is a parent of REVISION_ID.
996
The ' :' marker is the end of record marker.
999
when a write is interrupted to the index file, it will result in a line that
1000
does not end in ' :'. If the ' :' is not present at the end of a line, or at
1001
the end of the file, then the record that is missing it will be ignored by
1004
When writing new records to the index file, the data is preceded by '\n'
1005
to ensure that records always start on new lines even if the last write was
1006
interrupted. As a result its normal for the last line in the index to be
1007
missing a trailing newline. One can be added with no harmful effects.
1010
HEADER = "# bzr knit index 8\n"
1012
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1013
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1015
def _cache_version(self, version_id, options, pos, size, parents):
1016
"""Cache a version record in the history array and index cache.
1018
This is inlined into __init__ for performance. KEEP IN SYNC.
1019
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1022
# only want the _history index to reference the 1st index entry
1024
if version_id not in self._cache:
1025
index = len(self._history)
1026
self._history.append(version_id)
1028
index = self._cache[version_id][5]
1029
self._cache[version_id] = (version_id,
1036
def __init__(self, transport, filename, mode, create=False, file_mode=None):
1037
_KnitComponentFile.__init__(self, transport, filename, mode, file_mode)
1039
# position in _history is the 'official' index for a revision
1040
# but the values may have come from a newer entry.
1041
# so - wc -l of a knit index is != the number of unique names
1044
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1049
pb.update('read knit index', count, total)
1050
fp = self._transport.get(self._filename)
1052
self.check_header(fp)
1053
# readlines reads the whole file at once:
1054
# bad for transports like http, good for local disk
1055
# we save 60 ms doing this one change (
1056
# from calling readline each time to calling
1058
# probably what we want for nice behaviour on
1059
# http is a incremental readlines that yields, or
1060
# a check for local vs non local indexes,
1061
for l in fp.readlines():
1063
if len(rec) < 5 or rec[-1] != ':':
1065
# FIXME: in the future we should determine if its a
1066
# short write - and ignore it
1067
# or a different failure, and raise. RBC 20060407
1071
#pb.update('read knit index', count, total)
1072
# See self._parse_parents
1074
for value in rec[4:-1]:
1076
# uncompressed reference
1077
parents.append(value[1:])
1079
# this is 15/4000ms faster than isinstance,
1081
# this function is called thousands of times a
1082
# second so small variations add up.
1083
assert value.__class__ is str
1084
parents.append(self._history[int(value)])
1085
# end self._parse_parents
1086
# self._cache_version(rec[0],
1087
# rec[1].split(','),
1091
# --- self._cache_version
1092
# only want the _history index to reference the 1st
1093
# index entry for version_id
1095
if version_id not in self._cache:
1096
index = len(self._history)
1097
self._history.append(version_id)
1099
index = self._cache[version_id][5]
1100
self._cache[version_id] = (version_id,
1106
# --- self._cache_version
1109
except NoSuchFile, e:
1110
if mode != 'w' or not create:
1114
pb.update('read knit index', total, total)
1117
def _parse_parents(self, compressed_parents):
1118
"""convert a list of string parent values into version ids.
1120
ints are looked up in the index.
1121
.FOO values are ghosts and converted in to FOO.
1123
NOTE: the function is retained here for clarity, and for possible
1124
use in partial index reads. However bulk processing now has
1125
it inlined in __init__ for inner-loop optimisation.
1128
for value in compressed_parents:
1129
if value[-1] == '.':
1130
# uncompressed reference
1131
result.append(value[1:])
1133
# this is 15/4000ms faster than isinstance,
1134
# this function is called thousands of times a
1135
# second so small variations add up.
1136
assert value.__class__ is str
1137
result.append(self._history[int(value)])
1140
def get_graph(self):
1142
for version_id, index in self._cache.iteritems():
1143
graph.append((version_id, index[4]))
1146
def get_ancestry(self, versions):
1147
"""See VersionedFile.get_ancestry."""
1148
# get a graph of all the mentioned versions:
1150
pending = set(versions)
1152
version = pending.pop()
1153
parents = self._cache[version][4]
1154
# got the parents ok
1156
parents = [parent for parent in parents if parent in self._cache]
1157
for parent in parents:
1158
# if not completed and not a ghost
1159
if parent not in graph:
1161
graph[version] = parents
1162
return topo_sort(graph.items())
1164
def get_ancestry_with_ghosts(self, versions):
1165
"""See VersionedFile.get_ancestry_with_ghosts."""
1166
# get a graph of all the mentioned versions:
1168
pending = set(versions)
1170
version = pending.pop()
1172
parents = self._cache[version][4]
1178
# got the parents ok
1179
for parent in parents:
1180
if parent not in graph:
1182
graph[version] = parents
1183
return topo_sort(graph.items())
1185
def num_versions(self):
1186
return len(self._history)
1188
__len__ = num_versions
1190
def get_versions(self):
1191
return self._history
1193
def idx_to_name(self, idx):
1194
return self._history[idx]
1196
def lookup(self, version_id):
1197
assert version_id in self._cache
1198
return self._cache[version_id][5]
1200
def _version_list_to_index(self, versions):
1202
for version in versions:
1203
if version in self._cache:
1204
# -- inlined lookup() --
1205
result_list.append(str(self._cache[version][5]))
1206
# -- end lookup () --
1208
result_list.append('.' + version.encode('utf-8'))
1209
return ' '.join(result_list)
1211
def add_version(self, version_id, options, pos, size, parents):
1212
"""Add a version record to the index."""
1213
self.add_versions(((version_id, options, pos, size, parents),))
1215
def add_versions(self, versions):
1216
"""Add multiple versions to the index.
1218
:param versions: a list of tuples:
1219
(version_id, options, pos, size, parents).
1222
for version_id, options, pos, size, parents in versions:
1223
line = "\n%s %s %s %s %s :" % (version_id.encode('utf-8'),
1227
self._version_list_to_index(parents))
1228
assert isinstance(line, str), \
1229
'content must be utf-8 encoded: %r' % (line,)
1231
self._transport.append(self._filename, StringIO(''.join(lines)))
1232
# cache after writing, so that a failed write leads to missing cache
1233
# entries not extra ones. XXX TODO: RBC 20060502 in the event of a
1234
# failure, reload the index or flush it or some such, to prevent
1235
# writing records that did complete twice.
1236
for version_id, options, pos, size, parents in versions:
1237
self._cache_version(version_id, options, pos, size, parents)
1239
def has_version(self, version_id):
1240
"""True if the version is in the index."""
1241
return self._cache.has_key(version_id)
1243
def get_position(self, version_id):
1244
"""Return data position and size of specified version."""
1245
return (self._cache[version_id][2], \
1246
self._cache[version_id][3])
1248
def get_method(self, version_id):
1249
"""Return compression method of specified version."""
1250
options = self._cache[version_id][1]
1251
if 'fulltext' in options:
1254
assert 'line-delta' in options
1257
def get_options(self, version_id):
1258
return self._cache[version_id][1]
1260
def get_parents(self, version_id):
1261
"""Return parents of specified version ignoring ghosts."""
1262
return [parent for parent in self._cache[version_id][4]
1263
if parent in self._cache]
1265
def get_parents_with_ghosts(self, version_id):
1266
"""Return parents of specified version with ghosts."""
1267
return self._cache[version_id][4]
1269
def check_versions_present(self, version_ids):
1270
"""Check that all specified versions are present."""
1271
version_ids = set(version_ids)
1272
for version_id in list(version_ids):
1273
if version_id in self._cache:
1274
version_ids.remove(version_id)
1276
raise RevisionNotPresent(list(version_ids)[0], self.filename)
1279
class _KnitData(_KnitComponentFile):
1280
"""Contents of the knit data file"""
1282
HEADER = "# bzr knit data 8\n"
1284
def __init__(self, transport, filename, mode, create=False, file_mode=None):
1285
_KnitComponentFile.__init__(self, transport, filename, mode)
1286
self._checked = False
1288
self._transport.put(self._filename, StringIO(''), mode=file_mode)
1290
def clear_cache(self):
1291
"""Clear the record cache."""
1294
def _open_file(self):
1296
return self._transport.get(self._filename)
1301
def _record_to_data(self, version_id, digest, lines):
1302
"""Convert version_id, digest, lines into a raw data block.
1304
:return: (len, a StringIO instance with the raw data ready to read.)
1307
data_file = GzipFile(None, mode='wb', fileobj=sio)
1308
data_file.writelines(chain(
1309
["version %s %d %s\n" % (version_id.encode('utf-8'),
1313
["end %s\n" % version_id.encode('utf-8')]))
1320
def add_raw_record(self, raw_data):
1321
"""Append a prepared record to the data file.
1323
:return: the offset in the data file raw_data was written.
1325
assert isinstance(raw_data, str), 'data must be plain bytes'
1326
return self._transport.append(self._filename, StringIO(raw_data))
1328
def add_record(self, version_id, digest, lines):
1329
"""Write new text record to disk. Returns the position in the
1330
file where it was written."""
1331
size, sio = self._record_to_data(version_id, digest, lines)
1333
start_pos = self._transport.append(self._filename, sio)
1334
return start_pos, size
1336
def _parse_record_header(self, version_id, raw_data):
1337
"""Parse a record header for consistency.
1339
:return: the header and the decompressor stream.
1340
as (stream, header_record)
1342
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1343
rec = df.readline().split()
1345
raise KnitCorrupt(self._filename, 'unexpected number of elements in record header')
1346
if rec[1].decode('utf-8')!= version_id:
1347
raise KnitCorrupt(self._filename,
1348
'unexpected version, wanted %r, got %r' % (
1349
version_id, rec[1]))
1352
def _parse_record(self, version_id, data):
1354
# 4168 calls in 2880 217 internal
1355
# 4168 calls to _parse_record_header in 2121
1356
# 4168 calls to readlines in 330
1357
df, rec = self._parse_record_header(version_id, data)
1358
record_contents = df.readlines()
1359
l = record_contents.pop()
1360
assert len(record_contents) == int(rec[2])
1361
if l.decode('utf-8') != 'end %s\n' % version_id:
1362
raise KnitCorrupt(self._filename, 'unexpected version end line %r, wanted %r'
1365
return record_contents, rec[3]
1367
def read_records_iter_raw(self, records):
1368
"""Read text records from data file and yield raw data.
1370
This unpacks enough of the text record to validate the id is
1371
as expected but thats all.
1373
It will actively recompress currently cached records on the
1374
basis that that is cheaper than I/O activity.
1376
# setup an iterator of the external records:
1377
# uses readv so nice and fast we hope.
1379
# grab the disk data needed.
1380
raw_records = self._transport.readv(self._filename,
1381
[(pos, size) for version_id, pos, size in records])
1383
for version_id, pos, size in records:
1384
pos, data = raw_records.next()
1385
# validate the header
1386
df, rec = self._parse_record_header(version_id, data)
1388
yield version_id, data
1390
def read_records_iter(self, records):
1391
"""Read text records from data file and yield result.
1393
Each passed record is a tuple of (version_id, pos, len) and
1394
will be read in the given order. Yields (version_id,
1397
if len(records) == 0:
1400
# 60890 calls for 4168 extractions in 5045, 683 internal.
1401
# 4168 calls to readv in 1411
1402
# 4168 calls to parse_record in 2880
1404
# Get unique records, sorted by position
1405
needed_records = sorted(set(records), key=operator.itemgetter(1))
1407
# We take it that the transport optimizes the fetching as good
1408
# as possible (ie, reads continuous ranges.)
1409
response = self._transport.readv(self._filename,
1410
[(pos, size) for version_id, pos, size in needed_records])
1413
for (record_id, pos, size), (pos, data) in \
1414
izip(iter(needed_records), response):
1415
content, digest = self._parse_record(record_id, data)
1416
record_map[record_id] = (digest, content)
1418
for version_id, pos, size in records:
1419
digest, content = record_map[version_id]
1420
yield version_id, content, digest
1422
def read_records(self, records):
1423
"""Read records into a dictionary."""
1425
for record_id, content, digest in self.read_records_iter(records):
1426
components[record_id] = (content, digest)
1430
class InterKnit(InterVersionedFile):
1431
"""Optimised code paths for knit to knit operations."""
1433
_matching_file_from_factory = KnitVersionedFile
1434
_matching_file_to_factory = KnitVersionedFile
1437
def is_compatible(source, target):
1438
"""Be compatible with knits. """
1440
return (isinstance(source, KnitVersionedFile) and
1441
isinstance(target, KnitVersionedFile))
1442
except AttributeError:
1445
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1446
"""See InterVersionedFile.join."""
1447
assert isinstance(self.source, KnitVersionedFile)
1448
assert isinstance(self.target, KnitVersionedFile)
1450
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1455
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1457
version_ids = list(version_ids)
1458
if None in version_ids:
1459
version_ids.remove(None)
1461
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1462
this_versions = set(self.target._index.get_versions())
1463
needed_versions = self.source_ancestry - this_versions
1464
cross_check_versions = self.source_ancestry.intersection(this_versions)
1465
mismatched_versions = set()
1466
for version in cross_check_versions:
1467
# scan to include needed parents.
1468
n1 = set(self.target.get_parents_with_ghosts(version))
1469
n2 = set(self.source.get_parents_with_ghosts(version))
1471
# FIXME TEST this check for cycles being introduced works
1472
# the logic is we have a cycle if in our graph we are an
1473
# ancestor of any of the n2 revisions.
1479
parent_ancestors = self.source.get_ancestry(parent)
1480
if version in parent_ancestors:
1481
raise errors.GraphCycleError([parent, version])
1482
# ensure this parent will be available later.
1483
new_parents = n2.difference(n1)
1484
needed_versions.update(new_parents.difference(this_versions))
1485
mismatched_versions.add(version)
1487
if not needed_versions and not mismatched_versions:
1489
full_list = topo_sort(self.source.get_graph())
1491
version_list = [i for i in full_list if (not self.target.has_version(i)
1492
and i in needed_versions)]
1496
copy_queue_records = []
1498
for version_id in version_list:
1499
options = self.source._index.get_options(version_id)
1500
parents = self.source._index.get_parents_with_ghosts(version_id)
1501
# check that its will be a consistent copy:
1502
for parent in parents:
1503
# if source has the parent, we must :
1504
# * already have it or
1505
# * have it scheduled already
1506
# otherwise we don't care
1507
assert (self.target.has_version(parent) or
1508
parent in copy_set or
1509
not self.source.has_version(parent))
1510
data_pos, data_size = self.source._index.get_position(version_id)
1511
copy_queue_records.append((version_id, data_pos, data_size))
1512
copy_queue.append((version_id, options, parents))
1513
copy_set.add(version_id)
1515
# data suck the join:
1517
total = len(version_list)
1520
for (version_id, raw_data), \
1521
(version_id2, options, parents) in \
1522
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1524
assert version_id == version_id2, 'logic error, inconsistent results'
1526
pb.update("Joining knit", count, total)
1527
raw_records.append((version_id, options, parents, len(raw_data)))
1528
raw_datum.append(raw_data)
1529
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1531
for version in mismatched_versions:
1532
# FIXME RBC 20060309 is this needed?
1533
n1 = set(self.target.get_parents_with_ghosts(version))
1534
n2 = set(self.source.get_parents_with_ghosts(version))
1535
# write a combined record to our history preserving the current
1536
# parents as first in the list
1537
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1538
self.target.fix_parents(version, new_parents)
1544
InterVersionedFile.register_optimiser(InterKnit)
1547
class WeaveToKnit(InterVersionedFile):
1548
"""Optimised code paths for weave to knit operations."""
1550
_matching_file_from_factory = bzrlib.weave.WeaveFile
1551
_matching_file_to_factory = KnitVersionedFile
1554
def is_compatible(source, target):
1555
"""Be compatible with weaves to knits."""
1557
return (isinstance(source, bzrlib.weave.Weave) and
1558
isinstance(target, KnitVersionedFile))
1559
except AttributeError:
1562
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1563
"""See InterVersionedFile.join."""
1564
assert isinstance(self.source, bzrlib.weave.Weave)
1565
assert isinstance(self.target, KnitVersionedFile)
1567
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1572
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1574
version_ids = list(version_ids)
1576
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1577
this_versions = set(self.target._index.get_versions())
1578
needed_versions = self.source_ancestry - this_versions
1579
cross_check_versions = self.source_ancestry.intersection(this_versions)
1580
mismatched_versions = set()
1581
for version in cross_check_versions:
1582
# scan to include needed parents.
1583
n1 = set(self.target.get_parents_with_ghosts(version))
1584
n2 = set(self.source.get_parents(version))
1585
# if all of n2's parents are in n1, then its fine.
1586
if n2.difference(n1):
1587
# FIXME TEST this check for cycles being introduced works
1588
# the logic is we have a cycle if in our graph we are an
1589
# ancestor of any of the n2 revisions.
1595
parent_ancestors = self.source.get_ancestry(parent)
1596
if version in parent_ancestors:
1597
raise errors.GraphCycleError([parent, version])
1598
# ensure this parent will be available later.
1599
new_parents = n2.difference(n1)
1600
needed_versions.update(new_parents.difference(this_versions))
1601
mismatched_versions.add(version)
1603
if not needed_versions and not mismatched_versions:
1605
full_list = topo_sort(self.source.get_graph())
1607
version_list = [i for i in full_list if (not self.target.has_version(i)
1608
and i in needed_versions)]
1612
total = len(version_list)
1613
for version_id in version_list:
1614
pb.update("Converting to knit", count, total)
1615
parents = self.source.get_parents(version_id)
1616
# check that its will be a consistent copy:
1617
for parent in parents:
1618
# if source has the parent, we must already have it
1619
assert (self.target.has_version(parent))
1620
self.target.add_lines(
1621
version_id, parents, self.source.get_lines(version_id))
1624
for version in mismatched_versions:
1625
# FIXME RBC 20060309 is this needed?
1626
n1 = set(self.target.get_parents_with_ghosts(version))
1627
n2 = set(self.source.get_parents(version))
1628
# write a combined record to our history preserving the current
1629
# parents as first in the list
1630
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1631
self.target.fix_parents(version, new_parents)
1637
InterVersionedFile.register_optimiser(WeaveToKnit)
1640
class KnitSequenceMatcher(difflib.SequenceMatcher):
1641
"""Knit tuned sequence matcher.
1643
This is based on profiling of difflib which indicated some improvements
1644
for our usage pattern.
1647
def find_longest_match(self, alo, ahi, blo, bhi):
1648
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1650
If isjunk is not defined:
1652
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1653
alo <= i <= i+k <= ahi
1654
blo <= j <= j+k <= bhi
1655
and for all (i',j',k') meeting those conditions,
1658
and if i == i', j <= j'
1660
In other words, of all maximal matching blocks, return one that
1661
starts earliest in a, and of all those maximal matching blocks that
1662
start earliest in a, return the one that starts earliest in b.
1664
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1665
>>> s.find_longest_match(0, 5, 0, 9)
1668
If isjunk is defined, first the longest matching block is
1669
determined as above, but with the additional restriction that no
1670
junk element appears in the block. Then that block is extended as
1671
far as possible by matching (only) junk elements on both sides. So
1672
the resulting block never matches on junk except as identical junk
1673
happens to be adjacent to an "interesting" match.
1675
Here's the same example as before, but considering blanks to be
1676
junk. That prevents " abcd" from matching the " abcd" at the tail
1677
end of the second sequence directly. Instead only the "abcd" can
1678
match, and matches the leftmost "abcd" in the second sequence:
1680
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1681
>>> s.find_longest_match(0, 5, 0, 9)
1684
If no blocks match, return (alo, blo, 0).
1686
>>> s = SequenceMatcher(None, "ab", "c")
1687
>>> s.find_longest_match(0, 2, 0, 1)
1691
# CAUTION: stripping common prefix or suffix would be incorrect.
1695
# Longest matching block is "ab", but if common prefix is
1696
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1697
# strip, so ends up claiming that ab is changed to acab by
1698
# inserting "ca" in the middle. That's minimal but unintuitive:
1699
# "it's obvious" that someone inserted "ac" at the front.
1700
# Windiff ends up at the same place as diff, but by pairing up
1701
# the unique 'b's and then matching the first two 'a's.
1703
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1704
besti, bestj, bestsize = alo, blo, 0
1705
# find longest junk-free match
1706
# during an iteration of the loop, j2len[j] = length of longest
1707
# junk-free match ending with a[i-1] and b[j]
1711
for i in xrange(alo, ahi):
1712
# look at all instances of a[i] in b; note that because
1713
# b2j has no junk keys, the loop is skipped if a[i] is junk
1714
j2lenget = j2len.get
1717
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1718
# following improvement
1719
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1720
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1721
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1723
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1724
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1725
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
1737
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1739
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1742
# Extend the best by non-junk elements on each end. In particular,
1743
# "popular" non-junk elements aren't in b2j, which greatly speeds
1744
# the inner loop above, but also means "the best" match so far
1745
# doesn't contain any junk *or* popular non-junk elements.
1746
while besti > alo and bestj > blo and \
1747
not isbjunk(b[bestj-1]) and \
1748
a[besti-1] == b[bestj-1]:
1749
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1750
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1751
not isbjunk(b[bestj+bestsize]) and \
1752
a[besti+bestsize] == b[bestj+bestsize]:
1755
# Now that we have a wholly interesting match (albeit possibly
1756
# empty!), we may as well suck up the matching junk on each
1757
# side of it too. Can't think of a good reason not to, and it
1758
# saves post-processing the (possibly considerable) expense of
1759
# figuring out what to do with it. In the case of an empty
1760
# interesting match, this is clearly the right thing to do,
1761
# because no other kind of match is possible in the regions.
1762
while besti > alo and bestj > blo and \
1763
isbjunk(b[bestj-1]) and \
1764
a[besti-1] == b[bestj-1]:
1765
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1766
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1767
isbjunk(b[bestj+bestsize]) and \
1768
a[besti+bestsize] == b[bestj+bestsize]:
1769
bestsize = bestsize + 1
1771
return besti, bestj, bestsize
added
removed
bzrlib/knit.py
