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)
379
for (version_id, options, parents, size) in records:
380
index_entries.append((version_id, options, pos+offset,
382
if self._data._do_cache:
383
self._data._cache[version_id] = data[offset:offset+size]
385
self._index.add_versions(index_entries)
387
def enable_cache(self):
388
"""Start caching data for this knit"""
389
self._data.enable_cache()
391
def clear_cache(self):
392
"""Clear the data cache only."""
393
self._data.clear_cache()
395
def copy_to(self, name, transport):
396
"""See VersionedFile.copy_to()."""
397
# copy the current index to a temp index to avoid racing with local
399
transport.put(name + INDEX_SUFFIX + '.tmp', self.transport.get(self._index._filename),)
401
f = self._data._open_file()
403
transport.put(name + DATA_SUFFIX, f)
406
# move the copied index into place
407
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
409
def create_empty(self, name, transport, mode=None):
410
return KnitVersionedFile(name, transport, factory=self.factory, delta=self.delta, create=True)
412
def _fix_parents(self, version, new_parents):
413
"""Fix the parents list for version.
415
This is done by appending a new version to the index
416
with identical data except for the parents list.
417
the parents list must be a superset of the current
420
current_values = self._index._cache[version]
421
assert set(current_values[4]).difference(set(new_parents)) == set()
422
self._index.add_version(version,
428
def get_delta(self, version_id):
429
"""Get a delta for constructing version from some other version."""
430
if not self.has_version(version_id):
431
raise RevisionNotPresent(version_id, self.filename)
433
parents = self.get_parents(version_id)
438
data_pos, data_size = self._index.get_position(version_id)
439
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
440
version_idx = self._index.lookup(version_id)
441
noeol = 'no-eol' in self._index.get_options(version_id)
442
if 'fulltext' == self._index.get_method(version_id):
443
new_content = self.factory.parse_fulltext(data, version_idx)
444
if parent is not None:
445
reference_content = self._get_content(parent)
446
old_texts = reference_content.text()
449
new_texts = new_content.text()
450
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
451
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
453
delta = self.factory.parse_line_delta(data, version_idx)
454
return parent, sha1, noeol, delta
456
def get_graph_with_ghosts(self):
457
"""See VersionedFile.get_graph_with_ghosts()."""
458
graph_items = self._index.get_graph()
459
return dict(graph_items)
461
def get_sha1(self, version_id):
462
"""See VersionedFile.get_sha1()."""
463
record_map = self._get_record_map([version_id])
464
method, content, digest, next = record_map[version_id]
469
"""See VersionedFile.get_suffixes()."""
470
return [DATA_SUFFIX, INDEX_SUFFIX]
472
def has_ghost(self, version_id):
473
"""True if there is a ghost reference in the file to version_id."""
475
if self.has_version(version_id):
477
# optimisable if needed by memoising the _ghosts set.
478
items = self._index.get_graph()
479
for node, parents in items:
480
for parent in parents:
481
if parent not in self._index._cache:
482
if parent == version_id:
487
"""See VersionedFile.versions."""
488
return self._index.get_versions()
490
def has_version(self, version_id):
491
"""See VersionedFile.has_version."""
492
return self._index.has_version(version_id)
494
__contains__ = has_version
496
def _merge_annotations(self, content, parents, parent_texts={},
497
delta=None, annotated=None):
498
"""Merge annotations for content. This is done by comparing
499
the annotations based on changed to the text.
503
for parent_id in parents:
504
merge_content = self._get_content(parent_id, parent_texts)
505
seq = KnitSequenceMatcher(None, merge_content.text(), content.text())
506
if delta_seq is None:
507
# setup a delta seq to reuse.
509
for i, j, n in seq.get_matching_blocks():
512
# this appears to copy (origin, text) pairs across to the new
513
# content for any line that matches the last-checked parent.
514
# FIXME: save the sequence control data for delta compression
515
# against the most relevant parent rather than rediffing.
516
content._lines[j:j+n] = merge_content._lines[i:i+n]
519
reference_content = self._get_content(parents[0], parent_texts)
520
new_texts = content.text()
521
old_texts = reference_content.text()
522
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
523
return self._make_line_delta(delta_seq, content)
525
def _make_line_delta(self, delta_seq, new_content):
526
"""Generate a line delta from delta_seq and new_content."""
528
for op in delta_seq.get_opcodes():
531
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
534
def _get_components_positions(self, version_ids):
535
"""Produce a map of position data for the components of versions.
537
This data is intended to be used for retrieving the knit records.
539
A dict of version_id to (method, data_pos, data_size, next) is
541
method is the way referenced data should be applied.
542
data_pos is the position of the data in the knit.
543
data_size is the size of the data in the knit.
544
next is the build-parent of the version, or None for fulltexts.
547
for version_id in version_ids:
550
while cursor is not None and cursor not in component_data:
551
method = self._index.get_method(cursor)
552
if method == 'fulltext':
555
next = self.get_parents(cursor)[0]
556
data_pos, data_size = self._index.get_position(cursor)
557
component_data[cursor] = (method, data_pos, data_size, next)
559
return component_data
561
def _get_content(self, version_id, parent_texts={}):
562
"""Returns a content object that makes up the specified
564
if not self.has_version(version_id):
565
raise RevisionNotPresent(version_id, self.filename)
567
cached_version = parent_texts.get(version_id, None)
568
if cached_version is not None:
569
return cached_version
571
text_map, contents_map = self._get_content_maps([version_id])
572
return contents_map[version_id]
574
def _check_versions_present(self, version_ids):
575
"""Check that all specified versions are present."""
576
version_ids = set(version_ids)
577
for r in list(version_ids):
578
if self._index.has_version(r):
579
version_ids.remove(r)
581
raise RevisionNotPresent(list(version_ids)[0], self.filename)
583
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
584
"""See VersionedFile.add_lines_with_ghosts()."""
585
self._check_add(version_id, lines)
586
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
588
def _add_lines(self, version_id, parents, lines, parent_texts):
589
"""See VersionedFile.add_lines."""
590
self._check_add(version_id, lines)
591
self._check_versions_present(parents)
592
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
594
def _check_add(self, version_id, lines):
595
"""check that version_id and lines are safe to add."""
596
assert self.writable, "knit is not opened for write"
597
### FIXME escape. RBC 20060228
598
if contains_whitespace(version_id):
599
raise InvalidRevisionId(version_id, self.filename)
600
if self.has_version(version_id):
601
raise RevisionAlreadyPresent(version_id, self.filename)
602
self._check_lines_not_unicode(lines)
603
self._check_lines_are_lines(lines)
605
def _add(self, version_id, lines, parents, delta, parent_texts):
606
"""Add a set of lines on top of version specified by parents.
608
If delta is true, compress the text as a line-delta against
611
Any versions not present will be converted into ghosts.
613
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
614
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
615
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
616
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
617
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
618
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
619
# +1383 0 8.0370 8.0370 +<len>
620
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
621
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
622
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
623
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
627
if parent_texts is None:
629
for parent in parents:
630
if not self.has_version(parent):
631
ghosts.append(parent)
633
present_parents.append(parent)
635
if delta and not len(present_parents):
638
digest = sha_strings(lines)
641
if lines[-1][-1] != '\n':
642
options.append('no-eol')
643
lines[-1] = lines[-1] + '\n'
645
if len(present_parents) and delta:
646
# To speed the extract of texts the delta chain is limited
647
# to a fixed number of deltas. This should minimize both
648
# I/O and the time spend applying deltas.
650
delta_parents = present_parents
652
parent = delta_parents[0]
653
method = self._index.get_method(parent)
654
if method == 'fulltext':
656
delta_parents = self._index.get_parents(parent)
658
if method == 'line-delta':
661
lines = self.factory.make(lines, version_id)
662
if delta or (self.factory.annotated and len(present_parents) > 0):
663
# Merge annotations from parent texts if so is needed.
664
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
665
delta, self.factory.annotated)
668
options.append('line-delta')
669
store_lines = self.factory.lower_line_delta(delta_hunks)
671
options.append('fulltext')
672
store_lines = self.factory.lower_fulltext(lines)
674
where, size = self._data.add_record(version_id, digest, store_lines)
675
self._index.add_version(version_id, options, where, size, parents)
678
def check(self, progress_bar=None):
679
"""See VersionedFile.check()."""
681
def _clone_text(self, new_version_id, old_version_id, parents):
682
"""See VersionedFile.clone_text()."""
683
# FIXME RBC 20060228 make fast by only inserting an index with null
685
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
687
def get_lines(self, version_id):
688
"""See VersionedFile.get_lines()."""
689
return self.get_line_list([version_id])[0]
691
def _get_record_map(self, version_ids):
692
"""Produce a dictionary of knit records.
694
The keys are version_ids, the values are tuples of (method, content,
696
method is the way the content should be applied.
697
content is a KnitContent object.
698
digest is the SHA1 digest of this version id after all steps are done
699
next is the build-parent of the version, i.e. the leftmost ancestor.
700
If the method is fulltext, next will be None.
702
position_map = self._get_components_positions(version_ids)
703
# c = component_id, m = method, p = position, s = size, n = next
704
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
706
for component_id, content, digest in \
707
self._data.read_records_iter(records):
708
method, position, size, next = position_map[component_id]
709
record_map[component_id] = method, content, digest, next
713
def get_text(self, version_id):
714
"""See VersionedFile.get_text"""
715
return self.get_texts([version_id])[0]
717
def get_texts(self, version_ids):
718
return [''.join(l) for l in self.get_line_list(version_ids)]
720
def get_line_list(self, version_ids):
721
"""Return the texts of listed versions as a list of strings."""
722
text_map, content_map = self._get_content_maps(version_ids)
723
return [text_map[v] for v in version_ids]
725
def _get_content_maps(self, version_ids):
726
"""Produce maps of text and KnitContents
728
:return: (text_map, content_map) where text_map contains the texts for
729
the requested versions and content_map contains the KnitContents.
730
Both dicts take version_ids as their keys.
732
for version_id in version_ids:
733
if not self.has_version(version_id):
734
raise RevisionNotPresent(version_id, self.filename)
735
record_map = self._get_record_map(version_ids)
740
for version_id in version_ids:
743
while cursor is not None:
744
method, data, digest, next = record_map[cursor]
745
components.append((cursor, method, data, digest))
746
if cursor in content_map:
751
for component_id, method, data, digest in reversed(components):
752
if component_id in content_map:
753
content = content_map[component_id]
755
version_idx = self._index.lookup(component_id)
756
if method == 'fulltext':
757
assert content is None
758
content = self.factory.parse_fulltext(data, version_idx)
759
elif method == 'line-delta':
760
delta = self.factory.parse_line_delta(data[:],
762
content = content.copy()
763
content._lines = self._apply_delta(content._lines,
765
content_map[component_id] = content
767
if 'no-eol' in self._index.get_options(version_id):
768
content = content.copy()
769
line = content._lines[-1][1].rstrip('\n')
770
content._lines[-1] = (content._lines[-1][0], line)
771
final_content[version_id] = content
773
# digest here is the digest from the last applied component.
774
text = content.text()
775
if sha_strings(text) != digest:
776
raise KnitCorrupt(self.filename,
777
'sha-1 does not match %s' % version_id)
779
text_map[version_id] = text
780
return text_map, final_content
782
def iter_lines_added_or_present_in_versions(self, version_ids=None):
783
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
784
if version_ids is None:
785
version_ids = self.versions()
786
# we don't care about inclusions, the caller cares.
787
# but we need to setup a list of records to visit.
788
# we need version_id, position, length
789
version_id_records = []
790
requested_versions = list(version_ids)
791
# filter for available versions
792
for version_id in requested_versions:
793
if not self.has_version(version_id):
794
raise RevisionNotPresent(version_id, self.filename)
795
# get a in-component-order queue:
797
for version_id in self.versions():
798
if version_id in requested_versions:
799
version_ids.append(version_id)
800
data_pos, length = self._index.get_position(version_id)
801
version_id_records.append((version_id, data_pos, length))
803
pb = bzrlib.ui.ui_factory.nested_progress_bar()
805
total = len(version_id_records)
807
pb.update('Walking content.', count, total)
808
for version_id, data, sha_value in \
809
self._data.read_records_iter(version_id_records):
810
pb.update('Walking content.', count, total)
811
method = self._index.get_method(version_id)
812
version_idx = self._index.lookup(version_id)
813
assert method in ('fulltext', 'line-delta')
814
if method == 'fulltext':
815
content = self.factory.parse_fulltext(data, version_idx)
816
for line in content.text():
819
delta = self.factory.parse_line_delta(data, version_idx)
820
for start, end, count, lines in delta:
821
for origin, line in lines:
824
pb.update('Walking content.', total, total)
827
pb.update('Walking content.', total, total)
831
def num_versions(self):
832
"""See VersionedFile.num_versions()."""
833
return self._index.num_versions()
835
__len__ = num_versions
837
def annotate_iter(self, version_id):
838
"""See VersionedFile.annotate_iter."""
839
content = self._get_content(version_id)
840
for origin, text in content.annotate_iter():
843
def get_parents(self, version_id):
844
"""See VersionedFile.get_parents."""
847
# 52554 calls in 1264 872 internal down from 3674
849
return self._index.get_parents(version_id)
851
raise RevisionNotPresent(version_id, self.filename)
853
def get_parents_with_ghosts(self, version_id):
854
"""See VersionedFile.get_parents."""
856
return self._index.get_parents_with_ghosts(version_id)
858
raise RevisionNotPresent(version_id, self.filename)
860
def get_ancestry(self, versions):
861
"""See VersionedFile.get_ancestry."""
862
if isinstance(versions, basestring):
863
versions = [versions]
866
self._check_versions_present(versions)
867
return self._index.get_ancestry(versions)
869
def get_ancestry_with_ghosts(self, versions):
870
"""See VersionedFile.get_ancestry_with_ghosts."""
871
if isinstance(versions, basestring):
872
versions = [versions]
875
self._check_versions_present(versions)
876
return self._index.get_ancestry_with_ghosts(versions)
878
#@deprecated_method(zero_eight)
879
def walk(self, version_ids):
880
"""See VersionedFile.walk."""
881
# We take the short path here, and extract all relevant texts
882
# and put them in a weave and let that do all the work. Far
883
# from optimal, but is much simpler.
884
# FIXME RB 20060228 this really is inefficient!
885
from bzrlib.weave import Weave
887
w = Weave(self.filename)
888
ancestry = self.get_ancestry(version_ids)
889
sorted_graph = topo_sort(self._index.get_graph())
890
version_list = [vid for vid in sorted_graph if vid in ancestry]
892
for version_id in version_list:
893
lines = self.get_lines(version_id)
894
w.add_lines(version_id, self.get_parents(version_id), lines)
896
for lineno, insert_id, dset, line in w.walk(version_ids):
897
yield lineno, insert_id, dset, line
899
def plan_merge(self, ver_a, ver_b):
900
"""See VersionedFile.plan_merge."""
901
ancestors_b = set(self.get_ancestry(ver_b))
902
def status_a(revision, text):
903
if revision in ancestors_b:
904
return 'killed-b', text
908
ancestors_a = set(self.get_ancestry(ver_a))
909
def status_b(revision, text):
910
if revision in ancestors_a:
911
return 'killed-a', text
915
annotated_a = self.annotate(ver_a)
916
annotated_b = self.annotate(ver_b)
917
plain_a = [t for (a, t) in annotated_a]
918
plain_b = [t for (a, t) in annotated_b]
919
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
922
for ai, bi, l in blocks:
923
# process all mismatched sections
924
# (last mismatched section is handled because blocks always
925
# includes a 0-length last block)
926
for revision, text in annotated_a[a_cur:ai]:
927
yield status_a(revision, text)
928
for revision, text in annotated_b[b_cur:bi]:
929
yield status_b(revision, text)
931
# and now the matched section
934
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
935
assert text_a == text_b
936
yield "unchanged", text_a
939
class _KnitComponentFile(object):
940
"""One of the files used to implement a knit database"""
942
def __init__(self, transport, filename, mode, file_mode=None):
943
self._transport = transport
944
self._filename = filename
946
self._file_mode=file_mode
948
def write_header(self):
949
if self._transport.append(self._filename, StringIO(self.HEADER),
950
mode=self._file_mode):
951
raise KnitCorrupt(self._filename, 'misaligned after writing header')
953
def check_header(self, fp):
955
if line != self.HEADER:
956
raise KnitHeaderError(badline=line)
959
"""Commit is a nop."""
962
return '%s(%s)' % (self.__class__.__name__, self._filename)
965
class _KnitIndex(_KnitComponentFile):
966
"""Manages knit index file.
968
The index is already kept in memory and read on startup, to enable
969
fast lookups of revision information. The cursor of the index
970
file is always pointing to the end, making it easy to append
973
_cache is a cache for fast mapping from version id to a Index
976
_history is a cache for fast mapping from indexes to version ids.
978
The index data format is dictionary compressed when it comes to
979
parent references; a index entry may only have parents that with a
980
lover index number. As a result, the index is topological sorted.
982
Duplicate entries may be written to the index for a single version id
983
if this is done then the latter one completely replaces the former:
984
this allows updates to correct version and parent information.
985
Note that the two entries may share the delta, and that successive
986
annotations and references MUST point to the first entry.
988
The index file on disc contains a header, followed by one line per knit
989
record. The same revision can be present in an index file more than once.
990
The first occurrence gets assigned a sequence number starting from 0.
992
The format of a single line is
993
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
994
REVISION_ID is a utf8-encoded revision id
995
FLAGS is a comma separated list of flags about the record. Values include
996
no-eol, line-delta, fulltext.
997
BYTE_OFFSET is the ascii representation of the byte offset in the data file
998
that the the compressed data starts at.
999
LENGTH is the ascii representation of the length of the data file.
1000
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1002
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1003
revision id already in the knit that is a parent of REVISION_ID.
1004
The ' :' marker is the end of record marker.
1007
when a write is interrupted to the index file, it will result in a line that
1008
does not end in ' :'. If the ' :' is not present at the end of a line, or at
1009
the end of the file, then the record that is missing it will be ignored by
1012
When writing new records to the index file, the data is preceded by '\n'
1013
to ensure that records always start on new lines even if the last write was
1014
interrupted. As a result its normal for the last line in the index to be
1015
missing a trailing newline. One can be added with no harmful effects.
1018
HEADER = "# bzr knit index 8\n"
1020
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1021
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1023
def _cache_version(self, version_id, options, pos, size, parents):
1024
"""Cache a version record in the history array and index cache.
1026
This is inlined into __init__ for performance. KEEP IN SYNC.
1027
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1030
# only want the _history index to reference the 1st index entry
1032
if version_id not in self._cache:
1033
index = len(self._history)
1034
self._history.append(version_id)
1036
index = self._cache[version_id][5]
1037
self._cache[version_id] = (version_id,
1044
def __init__(self, transport, filename, mode, create=False, file_mode=None):
1045
_KnitComponentFile.__init__(self, transport, filename, mode, file_mode)
1047
# position in _history is the 'official' index for a revision
1048
# but the values may have come from a newer entry.
1049
# so - wc -l of a knit index is != the number of unique names
1052
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1057
pb.update('read knit index', count, total)
1058
fp = self._transport.get(self._filename)
1060
self.check_header(fp)
1061
# readlines reads the whole file at once:
1062
# bad for transports like http, good for local disk
1063
# we save 60 ms doing this one change (
1064
# from calling readline each time to calling
1066
# probably what we want for nice behaviour on
1067
# http is a incremental readlines that yields, or
1068
# a check for local vs non local indexes,
1069
for l in fp.readlines():
1071
if len(rec) < 5 or rec[-1] != ':':
1073
# FIXME: in the future we should determine if its a
1074
# short write - and ignore it
1075
# or a different failure, and raise. RBC 20060407
1079
#pb.update('read knit index', count, total)
1080
# See self._parse_parents
1082
for value in rec[4:-1]:
1084
# uncompressed reference
1085
parents.append(value[1:])
1087
# this is 15/4000ms faster than isinstance,
1089
# this function is called thousands of times a
1090
# second so small variations add up.
1091
assert value.__class__ is str
1092
parents.append(self._history[int(value)])
1093
# end self._parse_parents
1094
# self._cache_version(rec[0],
1095
# rec[1].split(','),
1099
# --- self._cache_version
1100
# only want the _history index to reference the 1st
1101
# index entry for version_id
1103
if version_id not in self._cache:
1104
index = len(self._history)
1105
self._history.append(version_id)
1107
index = self._cache[version_id][5]
1108
self._cache[version_id] = (version_id,
1114
# --- self._cache_version
1117
except NoSuchFile, e:
1118
if mode != 'w' or not create:
1122
pb.update('read knit index', total, total)
1125
def _parse_parents(self, compressed_parents):
1126
"""convert a list of string parent values into version ids.
1128
ints are looked up in the index.
1129
.FOO values are ghosts and converted in to FOO.
1131
NOTE: the function is retained here for clarity, and for possible
1132
use in partial index reads. However bulk processing now has
1133
it inlined in __init__ for inner-loop optimisation.
1136
for value in compressed_parents:
1137
if value[-1] == '.':
1138
# uncompressed reference
1139
result.append(value[1:])
1141
# this is 15/4000ms faster than isinstance,
1142
# this function is called thousands of times a
1143
# second so small variations add up.
1144
assert value.__class__ is str
1145
result.append(self._history[int(value)])
1148
def get_graph(self):
1150
for version_id, index in self._cache.iteritems():
1151
graph.append((version_id, index[4]))
1154
def get_ancestry(self, versions):
1155
"""See VersionedFile.get_ancestry."""
1156
# get a graph of all the mentioned versions:
1158
pending = set(versions)
1160
version = pending.pop()
1161
parents = self._cache[version][4]
1162
# got the parents ok
1164
parents = [parent for parent in parents if parent in self._cache]
1165
for parent in parents:
1166
# if not completed and not a ghost
1167
if parent not in graph:
1169
graph[version] = parents
1170
return topo_sort(graph.items())
1172
def get_ancestry_with_ghosts(self, versions):
1173
"""See VersionedFile.get_ancestry_with_ghosts."""
1174
# get a graph of all the mentioned versions:
1176
pending = set(versions)
1178
version = pending.pop()
1180
parents = self._cache[version][4]
1186
# got the parents ok
1187
for parent in parents:
1188
if parent not in graph:
1190
graph[version] = parents
1191
return topo_sort(graph.items())
1193
def num_versions(self):
1194
return len(self._history)
1196
__len__ = num_versions
1198
def get_versions(self):
1199
return self._history
1201
def idx_to_name(self, idx):
1202
return self._history[idx]
1204
def lookup(self, version_id):
1205
assert version_id in self._cache
1206
return self._cache[version_id][5]
1208
def _version_list_to_index(self, versions):
1210
for version in versions:
1211
if version in self._cache:
1212
# -- inlined lookup() --
1213
result_list.append(str(self._cache[version][5]))
1214
# -- end lookup () --
1216
result_list.append('.' + version.encode('utf-8'))
1217
return ' '.join(result_list)
1219
def add_version(self, version_id, options, pos, size, parents):
1220
"""Add a version record to the index."""
1221
self.add_versions(((version_id, options, pos, size, parents),))
1223
def add_versions(self, versions):
1224
"""Add multiple versions to the index.
1226
:param versions: a list of tuples:
1227
(version_id, options, pos, size, parents).
1230
for version_id, options, pos, size, parents in versions:
1231
line = "\n%s %s %s %s %s :" % (version_id.encode('utf-8'),
1235
self._version_list_to_index(parents))
1236
assert isinstance(line, str), \
1237
'content must be utf-8 encoded: %r' % (line,)
1239
self._transport.append(self._filename, StringIO(''.join(lines)))
1240
# cache after writing, so that a failed write leads to missing cache
1241
# entries not extra ones. XXX TODO: RBC 20060502 in the event of a
1242
# failure, reload the index or flush it or some such, to prevent
1243
# writing records that did complete twice.
1244
for version_id, options, pos, size, parents in versions:
1245
self._cache_version(version_id, options, pos, size, parents)
1247
def has_version(self, version_id):
1248
"""True if the version is in the index."""
1249
return self._cache.has_key(version_id)
1251
def get_position(self, version_id):
1252
"""Return data position and size of specified version."""
1253
return (self._cache[version_id][2], \
1254
self._cache[version_id][3])
1256
def get_method(self, version_id):
1257
"""Return compression method of specified version."""
1258
options = self._cache[version_id][1]
1259
if 'fulltext' in options:
1262
assert 'line-delta' in options
1265
def get_options(self, version_id):
1266
return self._cache[version_id][1]
1268
def get_parents(self, version_id):
1269
"""Return parents of specified version ignoring ghosts."""
1270
return [parent for parent in self._cache[version_id][4]
1271
if parent in self._cache]
1273
def get_parents_with_ghosts(self, version_id):
1274
"""Return parents of specified version with ghosts."""
1275
return self._cache[version_id][4]
1277
def check_versions_present(self, version_ids):
1278
"""Check that all specified versions are present."""
1279
version_ids = set(version_ids)
1280
for version_id in list(version_ids):
1281
if version_id in self._cache:
1282
version_ids.remove(version_id)
1284
raise RevisionNotPresent(list(version_ids)[0], self.filename)
1287
class _KnitData(_KnitComponentFile):
1288
"""Contents of the knit data file"""
1290
HEADER = "# bzr knit data 8\n"
1292
def __init__(self, transport, filename, mode, create=False, file_mode=None):
1293
_KnitComponentFile.__init__(self, transport, filename, mode)
1294
self._checked = False
1295
# TODO: jam 20060713 conceptually, this could spill to disk
1296
# if the cached size gets larger than a certain amount
1297
# but it complicates the model a bit, so for now just use
1298
# a simple dictionary
1300
self._do_cache = False
1302
self._transport.put(self._filename, StringIO(''), mode=file_mode)
1304
def enable_cache(self):
1305
"""Enable caching of reads."""
1306
self._do_cache = True
1308
def clear_cache(self):
1309
"""Clear the record cache."""
1310
self._do_cache = False
1313
def _open_file(self):
1315
return self._transport.get(self._filename)
1320
def _record_to_data(self, version_id, digest, lines):
1321
"""Convert version_id, digest, lines into a raw data block.
1323
:return: (len, a StringIO instance with the raw data ready to read.)
1326
data_file = GzipFile(None, mode='wb', fileobj=sio)
1327
data_file.writelines(chain(
1328
["version %s %d %s\n" % (version_id.encode('utf-8'),
1332
["end %s\n" % version_id.encode('utf-8')]))
1339
def add_raw_record(self, raw_data):
1340
"""Append a prepared record to the data file.
1342
:return: the offset in the data file raw_data was written.
1344
assert isinstance(raw_data, str), 'data must be plain bytes'
1345
return self._transport.append(self._filename, StringIO(raw_data))
1347
def add_record(self, version_id, digest, lines):
1348
"""Write new text record to disk. Returns the position in the
1349
file where it was written."""
1350
size, sio = self._record_to_data(version_id, digest, lines)
1352
start_pos = self._transport.append(self._filename, sio)
1354
self._cache[version_id] = sio.getvalue()
1355
return start_pos, size
1357
def _parse_record_header(self, version_id, raw_data):
1358
"""Parse a record header for consistency.
1360
:return: the header and the decompressor stream.
1361
as (stream, header_record)
1363
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1364
rec = df.readline().split()
1366
raise KnitCorrupt(self._filename, 'unexpected number of elements in record header')
1367
if rec[1].decode('utf-8')!= version_id:
1368
raise KnitCorrupt(self._filename,
1369
'unexpected version, wanted %r, got %r' % (
1370
version_id, rec[1]))
1373
def _parse_record(self, version_id, data):
1375
# 4168 calls in 2880 217 internal
1376
# 4168 calls to _parse_record_header in 2121
1377
# 4168 calls to readlines in 330
1378
df, rec = self._parse_record_header(version_id, data)
1379
record_contents = df.readlines()
1380
l = record_contents.pop()
1381
assert len(record_contents) == int(rec[2])
1382
if l.decode('utf-8') != 'end %s\n' % version_id:
1383
raise KnitCorrupt(self._filename, 'unexpected version end line %r, wanted %r'
1386
return record_contents, rec[3]
1388
def read_records_iter_raw(self, records):
1389
"""Read text records from data file and yield raw data.
1391
This unpacks enough of the text record to validate the id is
1392
as expected but thats all.
1394
# setup an iterator of the external records:
1395
# uses readv so nice and fast we hope.
1397
# grab the disk data needed.
1399
# Don't check _cache if it is empty
1400
needed_offsets = [(pos, size) for version_id, pos, size
1402
if version_id not in self._cache]
1404
needed_offsets = [(pos, size) for version_id, pos, size
1407
raw_records = self._transport.readv(self._filename, needed_offsets)
1410
for version_id, pos, size in records:
1411
if version_id in self._cache:
1412
# This data has already been validated
1413
data = self._cache[version_id]
1415
pos, data = raw_records.next()
1417
self._cache[version_id] = data
1419
# validate the header
1420
df, rec = self._parse_record_header(version_id, data)
1422
yield version_id, data
1424
def read_records_iter(self, records):
1425
"""Read text records from data file and yield result.
1427
The result will be returned in whatever is the fastest to read.
1428
Not by the order requested. Also, multiple requests for the same
1429
record will only yield 1 response.
1430
:param records: A list of (version_id, pos, len) entries
1431
:return: Yields (version_id, contents, digest) in the order
1432
read, not the order requested
1438
# Skip records we have alread seen
1439
yielded_records = set()
1440
needed_records = set()
1441
for record in records:
1442
if record[0] in self._cache:
1443
if record[0] in yielded_records:
1445
yielded_records.add(record[0])
1446
data = self._cache[record[0]]
1447
content, digest = self._parse_record(record[0], data)
1448
yield (record[0], content, digest)
1450
needed_records.add(record)
1451
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1453
needed_records = sorted(set(records), key=operator.itemgetter(1))
1455
if not needed_records:
1458
# The transport optimizes the fetching as well
1459
# (ie, reads continuous ranges.)
1460
readv_response = self._transport.readv(self._filename,
1461
[(pos, size) for version_id, pos, size in needed_records])
1463
for (version_id, pos, size), (pos, data) in \
1464
izip(iter(needed_records), readv_response):
1465
content, digest = self._parse_record(version_id, data)
1467
self._cache[version_id] = data
1468
yield version_id, content, digest
1470
def read_records(self, records):
1471
"""Read records into a dictionary."""
1473
for record_id, content, digest in \
1474
self.read_records_iter(records):
1475
components[record_id] = (content, digest)
1479
class InterKnit(InterVersionedFile):
1480
"""Optimised code paths for knit to knit operations."""
1482
_matching_file_from_factory = KnitVersionedFile
1483
_matching_file_to_factory = KnitVersionedFile
1486
def is_compatible(source, target):
1487
"""Be compatible with knits. """
1489
return (isinstance(source, KnitVersionedFile) and
1490
isinstance(target, KnitVersionedFile))
1491
except AttributeError:
1494
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1495
"""See InterVersionedFile.join."""
1496
assert isinstance(self.source, KnitVersionedFile)
1497
assert isinstance(self.target, KnitVersionedFile)
1499
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1504
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1506
version_ids = list(version_ids)
1507
if None in version_ids:
1508
version_ids.remove(None)
1510
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1511
this_versions = set(self.target._index.get_versions())
1512
needed_versions = self.source_ancestry - this_versions
1513
cross_check_versions = self.source_ancestry.intersection(this_versions)
1514
mismatched_versions = set()
1515
for version in cross_check_versions:
1516
# scan to include needed parents.
1517
n1 = set(self.target.get_parents_with_ghosts(version))
1518
n2 = set(self.source.get_parents_with_ghosts(version))
1520
# FIXME TEST this check for cycles being introduced works
1521
# the logic is we have a cycle if in our graph we are an
1522
# ancestor of any of the n2 revisions.
1528
parent_ancestors = self.source.get_ancestry(parent)
1529
if version in parent_ancestors:
1530
raise errors.GraphCycleError([parent, version])
1531
# ensure this parent will be available later.
1532
new_parents = n2.difference(n1)
1533
needed_versions.update(new_parents.difference(this_versions))
1534
mismatched_versions.add(version)
1536
if not needed_versions and not mismatched_versions:
1538
full_list = topo_sort(self.source.get_graph())
1540
version_list = [i for i in full_list if (not self.target.has_version(i)
1541
and i in needed_versions)]
1545
copy_queue_records = []
1547
for version_id in version_list:
1548
options = self.source._index.get_options(version_id)
1549
parents = self.source._index.get_parents_with_ghosts(version_id)
1550
# check that its will be a consistent copy:
1551
for parent in parents:
1552
# if source has the parent, we must :
1553
# * already have it or
1554
# * have it scheduled already
1555
# otherwise we don't care
1556
assert (self.target.has_version(parent) or
1557
parent in copy_set or
1558
not self.source.has_version(parent))
1559
data_pos, data_size = self.source._index.get_position(version_id)
1560
copy_queue_records.append((version_id, data_pos, data_size))
1561
copy_queue.append((version_id, options, parents))
1562
copy_set.add(version_id)
1564
# data suck the join:
1566
total = len(version_list)
1569
for (version_id, raw_data), \
1570
(version_id2, options, parents) in \
1571
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1573
assert version_id == version_id2, 'logic error, inconsistent results'
1575
pb.update("Joining knit", count, total)
1576
raw_records.append((version_id, options, parents, len(raw_data)))
1577
raw_datum.append(raw_data)
1578
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1580
for version in mismatched_versions:
1581
# FIXME RBC 20060309 is this needed?
1582
n1 = set(self.target.get_parents_with_ghosts(version))
1583
n2 = set(self.source.get_parents_with_ghosts(version))
1584
# write a combined record to our history preserving the current
1585
# parents as first in the list
1586
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1587
self.target.fix_parents(version, new_parents)
1593
InterVersionedFile.register_optimiser(InterKnit)
1596
class WeaveToKnit(InterVersionedFile):
1597
"""Optimised code paths for weave to knit operations."""
1599
_matching_file_from_factory = bzrlib.weave.WeaveFile
1600
_matching_file_to_factory = KnitVersionedFile
1603
def is_compatible(source, target):
1604
"""Be compatible with weaves to knits."""
1606
return (isinstance(source, bzrlib.weave.Weave) and
1607
isinstance(target, KnitVersionedFile))
1608
except AttributeError:
1611
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1612
"""See InterVersionedFile.join."""
1613
assert isinstance(self.source, bzrlib.weave.Weave)
1614
assert isinstance(self.target, KnitVersionedFile)
1616
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1621
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1623
version_ids = list(version_ids)
1625
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1626
this_versions = set(self.target._index.get_versions())
1627
needed_versions = self.source_ancestry - this_versions
1628
cross_check_versions = self.source_ancestry.intersection(this_versions)
1629
mismatched_versions = set()
1630
for version in cross_check_versions:
1631
# scan to include needed parents.
1632
n1 = set(self.target.get_parents_with_ghosts(version))
1633
n2 = set(self.source.get_parents(version))
1634
# if all of n2's parents are in n1, then its fine.
1635
if n2.difference(n1):
1636
# FIXME TEST this check for cycles being introduced works
1637
# the logic is we have a cycle if in our graph we are an
1638
# ancestor of any of the n2 revisions.
1644
parent_ancestors = self.source.get_ancestry(parent)
1645
if version in parent_ancestors:
1646
raise errors.GraphCycleError([parent, version])
1647
# ensure this parent will be available later.
1648
new_parents = n2.difference(n1)
1649
needed_versions.update(new_parents.difference(this_versions))
1650
mismatched_versions.add(version)
1652
if not needed_versions and not mismatched_versions:
1654
full_list = topo_sort(self.source.get_graph())
1656
version_list = [i for i in full_list if (not self.target.has_version(i)
1657
and i in needed_versions)]
1661
total = len(version_list)
1662
for version_id in version_list:
1663
pb.update("Converting to knit", count, total)
1664
parents = self.source.get_parents(version_id)
1665
# check that its will be a consistent copy:
1666
for parent in parents:
1667
# if source has the parent, we must already have it
1668
assert (self.target.has_version(parent))
1669
self.target.add_lines(
1670
version_id, parents, self.source.get_lines(version_id))
1673
for version in mismatched_versions:
1674
# FIXME RBC 20060309 is this needed?
1675
n1 = set(self.target.get_parents_with_ghosts(version))
1676
n2 = set(self.source.get_parents(version))
1677
# write a combined record to our history preserving the current
1678
# parents as first in the list
1679
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1680
self.target.fix_parents(version, new_parents)
1686
InterVersionedFile.register_optimiser(WeaveToKnit)
1689
class KnitSequenceMatcher(difflib.SequenceMatcher):
1690
"""Knit tuned sequence matcher.
1692
This is based on profiling of difflib which indicated some improvements
1693
for our usage pattern.
1696
def find_longest_match(self, alo, ahi, blo, bhi):
1697
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1699
If isjunk is not defined:
1701
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1702
alo <= i <= i+k <= ahi
1703
blo <= j <= j+k <= bhi
1704
and for all (i',j',k') meeting those conditions,
1707
and if i == i', j <= j'
1709
In other words, of all maximal matching blocks, return one that
1710
starts earliest in a, and of all those maximal matching blocks that
1711
start earliest in a, return the one that starts earliest in b.
1713
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1714
>>> s.find_longest_match(0, 5, 0, 9)
1717
If isjunk is defined, first the longest matching block is
1718
determined as above, but with the additional restriction that no
1719
junk element appears in the block. Then that block is extended as
1720
far as possible by matching (only) junk elements on both sides. So
1721
the resulting block never matches on junk except as identical junk
1722
happens to be adjacent to an "interesting" match.
1724
Here's the same example as before, but considering blanks to be
1725
junk. That prevents " abcd" from matching the " abcd" at the tail
1726
end of the second sequence directly. Instead only the "abcd" can
1727
match, and matches the leftmost "abcd" in the second sequence:
1729
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1730
>>> s.find_longest_match(0, 5, 0, 9)
1733
If no blocks match, return (alo, blo, 0).
1735
>>> s = SequenceMatcher(None, "ab", "c")
1736
>>> s.find_longest_match(0, 2, 0, 1)
1740
# CAUTION: stripping common prefix or suffix would be incorrect.
1744
# Longest matching block is "ab", but if common prefix is
1745
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1746
# strip, so ends up claiming that ab is changed to acab by
1747
# inserting "ca" in the middle. That's minimal but unintuitive:
1748
# "it's obvious" that someone inserted "ac" at the front.
1749
# Windiff ends up at the same place as diff, but by pairing up
1750
# the unique 'b's and then matching the first two 'a's.
1752
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1753
besti, bestj, bestsize = alo, blo, 0
1754
# find longest junk-free match
1755
# during an iteration of the loop, j2len[j] = length of longest
1756
# junk-free match ending with a[i-1] and b[j]
1760
for i in xrange(alo, ahi):
1761
# look at all instances of a[i] in b; note that because
1762
# b2j has no junk keys, the loop is skipped if a[i] is junk
1763
j2lenget = j2len.get
1766
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1767
# following improvement
1768
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1769
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1770
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1772
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1773
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1774
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
1786
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1788
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1791
# Extend the best by non-junk elements on each end. In particular,
1792
# "popular" non-junk elements aren't in b2j, which greatly speeds
1793
# the inner loop above, but also means "the best" match so far
1794
# doesn't contain any junk *or* popular non-junk elements.
1795
while besti > alo and bestj > blo and \
1796
not isbjunk(b[bestj-1]) and \
1797
a[besti-1] == b[bestj-1]:
1798
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1799
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1800
not isbjunk(b[bestj+bestsize]) and \
1801
a[besti+bestsize] == b[bestj+bestsize]:
1804
# Now that we have a wholly interesting match (albeit possibly
1805
# empty!), we may as well suck up the matching junk on each
1806
# side of it too. Can't think of a good reason not to, and it
1807
# saves post-processing the (possibly considerable) expense of
1808
# figuring out what to do with it. In the case of an empty
1809
# interesting match, this is clearly the right thing to do,
1810
# because no other kind of match is possible in the regions.
1811
while besti > alo and bestj > blo and \
1812
isbjunk(b[bestj-1]) and \
1813
a[besti-1] == b[bestj-1]:
1814
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1815
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1816
isbjunk(b[bestj+bestsize]) and \
1817
a[besti+bestsize] == b[bestj+bestsize]:
1818
bestsize = bestsize + 1
1820
return besti, bestj, bestsize
added
removed
bzrlib/knit.py
