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
81
from bzrlib.errors import FileExists, NoSuchFile, KnitError, \
82
InvalidRevisionId, KnitCorrupt, KnitHeaderError, \
83
RevisionNotPresent, RevisionAlreadyPresent
84
from bzrlib.tuned_gzip import GzipFile
85
from bzrlib.trace import mutter
86
from bzrlib.osutils import contains_whitespace, contains_linebreaks, \
88
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
89
from bzrlib.tsort import topo_sort
91
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
94
# TODO: Split out code specific to this format into an associated object.
96
# TODO: Can we put in some kind of value to check that the index and data
97
# files belong together?
99
# TODO: accommodate binaries, perhaps by storing a byte count
101
# TODO: function to check whole file
103
# TODO: atomically append data, then measure backwards from the cursor
104
# position after writing to work out where it was located. we may need to
105
# bypass python file buffering.
107
DATA_SUFFIX = '.knit'
108
INDEX_SUFFIX = '.kndx'
111
class KnitContent(object):
112
"""Content of a knit version to which deltas can be applied."""
114
def __init__(self, lines):
117
def annotate_iter(self):
118
"""Yield tuples of (origin, text) for each content line."""
119
for origin, text in self._lines:
123
"""Return a list of (origin, text) tuples."""
124
return list(self.annotate_iter())
126
def line_delta_iter(self, new_lines):
127
"""Generate line-based delta from this content to new_lines."""
128
new_texts = [text for origin, text in new_lines._lines]
129
old_texts = [text for origin, text in self._lines]
130
s = KnitSequenceMatcher(None, old_texts, new_texts)
131
for op in s.get_opcodes():
134
# ofrom oto length data
135
yield (op[1], op[2], op[4]-op[3], new_lines._lines[op[3]:op[4]])
137
def line_delta(self, new_lines):
138
return list(self.line_delta_iter(new_lines))
141
return [text for origin, text in self._lines]
144
return KnitContent(self._lines[:])
147
class _KnitFactory(object):
148
"""Base factory for creating content objects."""
150
def make(self, lines, version):
151
num_lines = len(lines)
152
return KnitContent(zip([version] * num_lines, lines))
155
class KnitAnnotateFactory(_KnitFactory):
156
"""Factory for creating annotated Content objects."""
160
def parse_fulltext(self, content, version):
161
"""Convert fulltext to internal representation
163
fulltext content is of the format
164
revid(utf8) plaintext\n
165
internal representation is of the format:
168
decode_utf8 = cache_utf8.decode
171
origin, text = line.split(' ', 1)
172
lines.append((decode_utf8(origin), text))
173
return KnitContent(lines)
175
def parse_line_delta_iter(self, lines):
176
for result_item in self.parse_line_delta[lines]:
179
def parse_line_delta(self, lines, version):
180
"""Convert a line based delta into internal representation.
182
line delta is in the form of:
183
intstart intend intcount
185
revid(utf8) newline\n
186
internal representation is
187
(start, end, count, [1..count tuples (revid, newline)])
189
decode_utf8 = cache_utf8.decode
193
# walk through the lines parsing.
195
start, end, count = [int(n) for n in header.split(',')]
199
origin, text = next().split(' ', 1)
201
contents.append((decode_utf8(origin), text))
202
result.append((start, end, count, contents))
205
def lower_fulltext(self, content):
206
"""convert a fulltext content record into a serializable form.
208
see parse_fulltext which this inverts.
210
encode_utf8 = cache_utf8.encode
211
return ['%s %s' % (encode_utf8(o), t) for o, t in content._lines]
213
def lower_line_delta(self, delta):
214
"""convert a delta into a serializable form.
216
See parse_line_delta which this inverts.
218
encode_utf8 = cache_utf8.encode
220
for start, end, c, lines in delta:
221
out.append('%d,%d,%d\n' % (start, end, c))
222
out.extend(encode_utf8(origin) + ' ' + text
223
for origin, text in lines)
227
class KnitPlainFactory(_KnitFactory):
228
"""Factory for creating plain Content objects."""
232
def parse_fulltext(self, content, version):
233
"""This parses an unannotated fulltext.
235
Note that this is not a noop - the internal representation
236
has (versionid, line) - its just a constant versionid.
238
return self.make(content, version)
240
def parse_line_delta_iter(self, lines, version):
242
header = lines.pop(0)
243
start, end, c = [int(n) for n in header.split(',')]
244
yield start, end, c, zip([version] * c, lines[:c])
247
def parse_line_delta(self, lines, version):
248
return list(self.parse_line_delta_iter(lines, version))
250
def lower_fulltext(self, content):
251
return content.text()
253
def lower_line_delta(self, delta):
255
for start, end, c, lines in delta:
256
out.append('%d,%d,%d\n' % (start, end, c))
257
out.extend([text for origin, text in lines])
261
def make_empty_knit(transport, relpath):
262
"""Construct a empty knit at the specified location."""
263
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
267
class KnitVersionedFile(VersionedFile):
268
"""Weave-like structure with faster random access.
270
A knit stores a number of texts and a summary of the relationships
271
between them. Texts are identified by a string version-id. Texts
272
are normally stored and retrieved as a series of lines, but can
273
also be passed as single strings.
275
Lines are stored with the trailing newline (if any) included, to
276
avoid special cases for files with no final newline. Lines are
277
composed of 8-bit characters, not unicode. The combination of
278
these approaches should mean any 'binary' file can be safely
279
stored and retrieved.
282
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
283
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
285
"""Construct a knit at location specified by relpath.
287
:param create: If not True, only open an existing knit.
289
if deprecated_passed(basis_knit):
290
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
291
" deprecated as of bzr 0.9.",
292
DeprecationWarning, stacklevel=2)
293
if access_mode is None:
295
super(KnitVersionedFile, self).__init__(access_mode)
296
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
297
self.transport = transport
298
self.filename = relpath
299
self.factory = factory or KnitAnnotateFactory()
300
self.writable = (access_mode == 'w')
303
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
304
access_mode, create=create, file_mode=file_mode)
305
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
306
access_mode, create=create and not len(self), file_mode=file_mode)
309
return '%s(%s)' % (self.__class__.__name__,
310
self.transport.abspath(self.filename))
312
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
313
"""See VersionedFile._add_delta()."""
314
self._check_add(version_id, []) # should we check the lines ?
315
self._check_versions_present(parents)
319
for parent in parents:
320
if not self.has_version(parent):
321
ghosts.append(parent)
323
present_parents.append(parent)
325
if delta_parent is None:
326
# reconstitute as full text.
327
assert len(delta) == 1 or len(delta) == 0
329
assert delta[0][0] == 0
330
assert delta[0][1] == 0, delta[0][1]
331
return super(KnitVersionedFile, self)._add_delta(version_id,
342
options.append('no-eol')
344
if delta_parent is not None:
345
# determine the current delta chain length.
346
# To speed the extract of texts the delta chain is limited
347
# to a fixed number of deltas. This should minimize both
348
# I/O and the time spend applying deltas.
350
delta_parents = [delta_parent]
352
parent = delta_parents[0]
353
method = self._index.get_method(parent)
354
if method == 'fulltext':
356
delta_parents = self._index.get_parents(parent)
358
if method == 'line-delta':
359
# did not find a fulltext in the delta limit.
360
# just do a normal insertion.
361
return super(KnitVersionedFile, self)._add_delta(version_id,
368
options.append('line-delta')
369
store_lines = self.factory.lower_line_delta(delta)
371
where, size = self._data.add_record(version_id, digest, store_lines)
372
self._index.add_version(version_id, options, where, size, parents)
374
def _add_raw_records(self, records, data):
375
"""Add all the records 'records' with data pre-joined in 'data'.
377
:param records: A list of tuples(version_id, options, parents, size).
378
:param data: The data for the records. When it is written, the records
379
are adjusted to have pos pointing into data by the sum of
380
the preceding records sizes.
383
pos = self._data.add_raw_record(data)
386
for (version_id, options, parents, size) in records:
387
index_entries.append((version_id, options, pos+offset,
389
if self._data._do_cache:
390
self._data._cache[version_id] = data[offset:offset+size]
392
self._index.add_versions(index_entries)
394
def enable_cache(self):
395
"""Start caching data for this knit"""
396
self._data.enable_cache()
398
def clear_cache(self):
399
"""Clear the data cache only."""
400
self._data.clear_cache()
402
def copy_to(self, name, transport):
403
"""See VersionedFile.copy_to()."""
404
# copy the current index to a temp index to avoid racing with local
406
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
407
self.transport.get(self._index._filename))
409
f = self._data._open_file()
411
transport.put_file(name + DATA_SUFFIX, f)
414
# move the copied index into place
415
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
417
def create_empty(self, name, transport, mode=None):
418
return KnitVersionedFile(name, transport, factory=self.factory,
419
delta=self.delta, create=True)
421
def _fix_parents(self, version, new_parents):
422
"""Fix the parents list for version.
424
This is done by appending a new version to the index
425
with identical data except for the parents list.
426
the parents list must be a superset of the current
429
current_values = self._index._cache[version]
430
assert set(current_values[4]).difference(set(new_parents)) == set()
431
self._index.add_version(version,
437
def get_delta(self, version_id):
438
"""Get a delta for constructing version from some other version."""
439
if not self.has_version(version_id):
440
raise RevisionNotPresent(version_id, self.filename)
442
parents = self.get_parents(version_id)
447
data_pos, data_size = self._index.get_position(version_id)
448
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
449
version_idx = self._index.lookup(version_id)
450
noeol = 'no-eol' in self._index.get_options(version_id)
451
if 'fulltext' == self._index.get_method(version_id):
452
new_content = self.factory.parse_fulltext(data, version_idx)
453
if parent is not None:
454
reference_content = self._get_content(parent)
455
old_texts = reference_content.text()
458
new_texts = new_content.text()
459
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
460
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
462
delta = self.factory.parse_line_delta(data, version_idx)
463
return parent, sha1, noeol, delta
465
def get_graph_with_ghosts(self):
466
"""See VersionedFile.get_graph_with_ghosts()."""
467
graph_items = self._index.get_graph()
468
return dict(graph_items)
470
def get_sha1(self, version_id):
471
"""See VersionedFile.get_sha1()."""
472
record_map = self._get_record_map([version_id])
473
method, content, digest, next = record_map[version_id]
478
"""See VersionedFile.get_suffixes()."""
479
return [DATA_SUFFIX, INDEX_SUFFIX]
481
def has_ghost(self, version_id):
482
"""True if there is a ghost reference in the file to version_id."""
484
if self.has_version(version_id):
486
# optimisable if needed by memoising the _ghosts set.
487
items = self._index.get_graph()
488
for node, parents in items:
489
for parent in parents:
490
if parent not in self._index._cache:
491
if parent == version_id:
496
"""See VersionedFile.versions."""
497
return self._index.get_versions()
499
def has_version(self, version_id):
500
"""See VersionedFile.has_version."""
501
return self._index.has_version(version_id)
503
__contains__ = has_version
505
def _merge_annotations(self, content, parents, parent_texts={},
506
delta=None, annotated=None):
507
"""Merge annotations for content. This is done by comparing
508
the annotations based on changed to the text.
512
for parent_id in parents:
513
merge_content = self._get_content(parent_id, parent_texts)
514
seq = KnitSequenceMatcher(None, merge_content.text(), content.text())
515
if delta_seq is None:
516
# setup a delta seq to reuse.
518
for i, j, n in seq.get_matching_blocks():
521
# this appears to copy (origin, text) pairs across to the new
522
# content for any line that matches the last-checked parent.
523
# FIXME: save the sequence control data for delta compression
524
# against the most relevant parent rather than rediffing.
525
content._lines[j:j+n] = merge_content._lines[i:i+n]
528
reference_content = self._get_content(parents[0], parent_texts)
529
new_texts = content.text()
530
old_texts = reference_content.text()
531
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
532
return self._make_line_delta(delta_seq, content)
534
def _make_line_delta(self, delta_seq, new_content):
535
"""Generate a line delta from delta_seq and new_content."""
537
for op in delta_seq.get_opcodes():
540
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
543
def _get_components_positions(self, version_ids):
544
"""Produce a map of position data for the components of versions.
546
This data is intended to be used for retrieving the knit records.
548
A dict of version_id to (method, data_pos, data_size, next) is
550
method is the way referenced data should be applied.
551
data_pos is the position of the data in the knit.
552
data_size is the size of the data in the knit.
553
next is the build-parent of the version, or None for fulltexts.
556
for version_id in version_ids:
559
while cursor is not None and cursor not in component_data:
560
method = self._index.get_method(cursor)
561
if method == 'fulltext':
564
next = self.get_parents(cursor)[0]
565
data_pos, data_size = self._index.get_position(cursor)
566
component_data[cursor] = (method, data_pos, data_size, next)
568
return component_data
570
def _get_content(self, version_id, parent_texts={}):
571
"""Returns a content object that makes up the specified
573
if not self.has_version(version_id):
574
raise RevisionNotPresent(version_id, self.filename)
576
cached_version = parent_texts.get(version_id, None)
577
if cached_version is not None:
578
return cached_version
580
text_map, contents_map = self._get_content_maps([version_id])
581
return contents_map[version_id]
583
def _check_versions_present(self, version_ids):
584
"""Check that all specified versions are present."""
585
version_ids = set(version_ids)
586
for r in list(version_ids):
587
if self._index.has_version(r):
588
version_ids.remove(r)
590
raise RevisionNotPresent(list(version_ids)[0], self.filename)
592
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
593
"""See VersionedFile.add_lines_with_ghosts()."""
594
self._check_add(version_id, lines)
595
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
597
def _add_lines(self, version_id, parents, lines, parent_texts):
598
"""See VersionedFile.add_lines."""
599
self._check_add(version_id, lines)
600
self._check_versions_present(parents)
601
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
603
def _check_add(self, version_id, lines):
604
"""check that version_id and lines are safe to add."""
605
assert self.writable, "knit is not opened for write"
606
### FIXME escape. RBC 20060228
607
if contains_whitespace(version_id):
608
raise InvalidRevisionId(version_id, self.filename)
609
if self.has_version(version_id):
610
raise RevisionAlreadyPresent(version_id, self.filename)
611
self._check_lines_not_unicode(lines)
612
self._check_lines_are_lines(lines)
614
def _add(self, version_id, lines, parents, delta, parent_texts):
615
"""Add a set of lines on top of version specified by parents.
617
If delta is true, compress the text as a line-delta against
620
Any versions not present will be converted into ghosts.
622
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
623
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
624
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
625
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
626
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
627
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
628
# +1383 0 8.0370 8.0370 +<len>
629
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
630
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
631
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
632
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
636
if parent_texts is None:
638
for parent in parents:
639
if not self.has_version(parent):
640
ghosts.append(parent)
642
present_parents.append(parent)
644
if delta and not len(present_parents):
647
digest = sha_strings(lines)
650
if lines[-1][-1] != '\n':
651
options.append('no-eol')
652
lines[-1] = lines[-1] + '\n'
654
if len(present_parents) and delta:
655
# To speed the extract of texts the delta chain is limited
656
# to a fixed number of deltas. This should minimize both
657
# I/O and the time spend applying deltas.
659
delta_parents = present_parents
661
parent = delta_parents[0]
662
method = self._index.get_method(parent)
663
if method == 'fulltext':
665
delta_parents = self._index.get_parents(parent)
667
if method == 'line-delta':
670
lines = self.factory.make(lines, version_id)
671
if delta or (self.factory.annotated and len(present_parents) > 0):
672
# Merge annotations from parent texts if so is needed.
673
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
674
delta, self.factory.annotated)
677
options.append('line-delta')
678
store_lines = self.factory.lower_line_delta(delta_hunks)
680
options.append('fulltext')
681
store_lines = self.factory.lower_fulltext(lines)
683
where, size = self._data.add_record(version_id, digest, store_lines)
684
self._index.add_version(version_id, options, where, size, parents)
687
def check(self, progress_bar=None):
688
"""See VersionedFile.check()."""
690
def _clone_text(self, new_version_id, old_version_id, parents):
691
"""See VersionedFile.clone_text()."""
692
# FIXME RBC 20060228 make fast by only inserting an index with null
694
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
696
def get_lines(self, version_id):
697
"""See VersionedFile.get_lines()."""
698
return self.get_line_list([version_id])[0]
700
def _get_record_map(self, version_ids):
701
"""Produce a dictionary of knit records.
703
The keys are version_ids, the values are tuples of (method, content,
705
method is the way the content should be applied.
706
content is a KnitContent object.
707
digest is the SHA1 digest of this version id after all steps are done
708
next is the build-parent of the version, i.e. the leftmost ancestor.
709
If the method is fulltext, next will be None.
711
position_map = self._get_components_positions(version_ids)
712
# c = component_id, m = method, p = position, s = size, n = next
713
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
715
for component_id, content, digest in \
716
self._data.read_records_iter(records):
717
method, position, size, next = position_map[component_id]
718
record_map[component_id] = method, content, digest, next
722
def get_text(self, version_id):
723
"""See VersionedFile.get_text"""
724
return self.get_texts([version_id])[0]
726
def get_texts(self, version_ids):
727
return [''.join(l) for l in self.get_line_list(version_ids)]
729
def get_line_list(self, version_ids):
730
"""Return the texts of listed versions as a list of strings."""
731
text_map, content_map = self._get_content_maps(version_ids)
732
return [text_map[v] for v in version_ids]
734
def _get_content_maps(self, version_ids):
735
"""Produce maps of text and KnitContents
737
:return: (text_map, content_map) where text_map contains the texts for
738
the requested versions and content_map contains the KnitContents.
739
Both dicts take version_ids as their keys.
741
for version_id in version_ids:
742
if not self.has_version(version_id):
743
raise RevisionNotPresent(version_id, self.filename)
744
record_map = self._get_record_map(version_ids)
749
for version_id in version_ids:
752
while cursor is not None:
753
method, data, digest, next = record_map[cursor]
754
components.append((cursor, method, data, digest))
755
if cursor in content_map:
760
for component_id, method, data, digest in reversed(components):
761
if component_id in content_map:
762
content = content_map[component_id]
764
version_idx = self._index.lookup(component_id)
765
if method == 'fulltext':
766
assert content is None
767
content = self.factory.parse_fulltext(data, version_idx)
768
elif method == 'line-delta':
769
delta = self.factory.parse_line_delta(data[:],
771
content = content.copy()
772
content._lines = self._apply_delta(content._lines,
774
content_map[component_id] = content
776
if 'no-eol' in self._index.get_options(version_id):
777
content = content.copy()
778
line = content._lines[-1][1].rstrip('\n')
779
content._lines[-1] = (content._lines[-1][0], line)
780
final_content[version_id] = content
782
# digest here is the digest from the last applied component.
783
text = content.text()
784
if sha_strings(text) != digest:
785
raise KnitCorrupt(self.filename,
786
'sha-1 does not match %s' % version_id)
788
text_map[version_id] = text
789
return text_map, final_content
791
def iter_lines_added_or_present_in_versions(self, version_ids=None):
792
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
793
if version_ids is None:
794
version_ids = self.versions()
795
# we don't care about inclusions, the caller cares.
796
# but we need to setup a list of records to visit.
797
# we need version_id, position, length
798
version_id_records = []
799
requested_versions = list(version_ids)
800
# filter for available versions
801
for version_id in requested_versions:
802
if not self.has_version(version_id):
803
raise RevisionNotPresent(version_id, self.filename)
804
# get a in-component-order queue:
806
for version_id in self.versions():
807
if version_id in requested_versions:
808
version_ids.append(version_id)
809
data_pos, length = self._index.get_position(version_id)
810
version_id_records.append((version_id, data_pos, length))
812
pb = bzrlib.ui.ui_factory.nested_progress_bar()
814
total = len(version_id_records)
816
pb.update('Walking content.', count, total)
817
for version_id, data, sha_value in \
818
self._data.read_records_iter(version_id_records):
819
pb.update('Walking content.', count, total)
820
method = self._index.get_method(version_id)
821
version_idx = self._index.lookup(version_id)
822
assert method in ('fulltext', 'line-delta')
823
if method == 'fulltext':
824
content = self.factory.parse_fulltext(data, version_idx)
825
for line in content.text():
828
delta = self.factory.parse_line_delta(data, version_idx)
829
for start, end, count, lines in delta:
830
for origin, line in lines:
833
pb.update('Walking content.', total, total)
836
pb.update('Walking content.', total, total)
840
def num_versions(self):
841
"""See VersionedFile.num_versions()."""
842
return self._index.num_versions()
844
__len__ = num_versions
846
def annotate_iter(self, version_id):
847
"""See VersionedFile.annotate_iter."""
848
content = self._get_content(version_id)
849
for origin, text in content.annotate_iter():
852
def get_parents(self, version_id):
853
"""See VersionedFile.get_parents."""
856
# 52554 calls in 1264 872 internal down from 3674
858
return self._index.get_parents(version_id)
860
raise RevisionNotPresent(version_id, self.filename)
862
def get_parents_with_ghosts(self, version_id):
863
"""See VersionedFile.get_parents."""
865
return self._index.get_parents_with_ghosts(version_id)
867
raise RevisionNotPresent(version_id, self.filename)
869
def get_ancestry(self, versions):
870
"""See VersionedFile.get_ancestry."""
871
if isinstance(versions, basestring):
872
versions = [versions]
875
self._check_versions_present(versions)
876
return self._index.get_ancestry(versions)
878
def get_ancestry_with_ghosts(self, versions):
879
"""See VersionedFile.get_ancestry_with_ghosts."""
880
if isinstance(versions, basestring):
881
versions = [versions]
884
self._check_versions_present(versions)
885
return self._index.get_ancestry_with_ghosts(versions)
887
#@deprecated_method(zero_eight)
888
def walk(self, version_ids):
889
"""See VersionedFile.walk."""
890
# We take the short path here, and extract all relevant texts
891
# and put them in a weave and let that do all the work. Far
892
# from optimal, but is much simpler.
893
# FIXME RB 20060228 this really is inefficient!
894
from bzrlib.weave import Weave
896
w = Weave(self.filename)
897
ancestry = self.get_ancestry(version_ids)
898
sorted_graph = topo_sort(self._index.get_graph())
899
version_list = [vid for vid in sorted_graph if vid in ancestry]
901
for version_id in version_list:
902
lines = self.get_lines(version_id)
903
w.add_lines(version_id, self.get_parents(version_id), lines)
905
for lineno, insert_id, dset, line in w.walk(version_ids):
906
yield lineno, insert_id, dset, line
908
def plan_merge(self, ver_a, ver_b):
909
"""See VersionedFile.plan_merge."""
910
ancestors_b = set(self.get_ancestry(ver_b))
911
def status_a(revision, text):
912
if revision in ancestors_b:
913
return 'killed-b', text
917
ancestors_a = set(self.get_ancestry(ver_a))
918
def status_b(revision, text):
919
if revision in ancestors_a:
920
return 'killed-a', text
924
annotated_a = self.annotate(ver_a)
925
annotated_b = self.annotate(ver_b)
926
plain_a = [t for (a, t) in annotated_a]
927
plain_b = [t for (a, t) in annotated_b]
928
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
931
for ai, bi, l in blocks:
932
# process all mismatched sections
933
# (last mismatched section is handled because blocks always
934
# includes a 0-length last block)
935
for revision, text in annotated_a[a_cur:ai]:
936
yield status_a(revision, text)
937
for revision, text in annotated_b[b_cur:bi]:
938
yield status_b(revision, text)
940
# and now the matched section
943
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
944
assert text_a == text_b
945
yield "unchanged", text_a
948
class _KnitComponentFile(object):
949
"""One of the files used to implement a knit database"""
951
def __init__(self, transport, filename, mode, file_mode=None):
952
self._transport = transport
953
self._filename = filename
955
self._file_mode=file_mode
957
def write_header(self):
958
if self._transport.put_bytes_non_atomic(self._filename, self.HEADER,
959
mode=self._file_mode):
960
raise KnitCorrupt(self._filename, 'misaligned after writing header')
962
def check_header(self, fp):
964
if line != self.HEADER:
965
raise KnitHeaderError(badline=line)
968
"""Commit is a nop."""
971
return '%s(%s)' % (self.__class__.__name__, self._filename)
974
class _KnitIndex(_KnitComponentFile):
975
"""Manages knit index file.
977
The index is already kept in memory and read on startup, to enable
978
fast lookups of revision information. The cursor of the index
979
file is always pointing to the end, making it easy to append
982
_cache is a cache for fast mapping from version id to a Index
985
_history is a cache for fast mapping from indexes to version ids.
987
The index data format is dictionary compressed when it comes to
988
parent references; a index entry may only have parents that with a
989
lover index number. As a result, the index is topological sorted.
991
Duplicate entries may be written to the index for a single version id
992
if this is done then the latter one completely replaces the former:
993
this allows updates to correct version and parent information.
994
Note that the two entries may share the delta, and that successive
995
annotations and references MUST point to the first entry.
997
The index file on disc contains a header, followed by one line per knit
998
record. The same revision can be present in an index file more than once.
999
The first occurrence gets assigned a sequence number starting from 0.
1001
The format of a single line is
1002
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1003
REVISION_ID is a utf8-encoded revision id
1004
FLAGS is a comma separated list of flags about the record. Values include
1005
no-eol, line-delta, fulltext.
1006
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1007
that the the compressed data starts at.
1008
LENGTH is the ascii representation of the length of the data file.
1009
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1011
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1012
revision id already in the knit that is a parent of REVISION_ID.
1013
The ' :' marker is the end of record marker.
1016
when a write is interrupted to the index file, it will result in a line that
1017
does not end in ' :'. If the ' :' is not present at the end of a line, or at
1018
the end of the file, then the record that is missing it will be ignored by
1021
When writing new records to the index file, the data is preceded by '\n'
1022
to ensure that records always start on new lines even if the last write was
1023
interrupted. As a result its normal for the last line in the index to be
1024
missing a trailing newline. One can be added with no harmful effects.
1027
HEADER = "# bzr knit index 8\n"
1029
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1030
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1032
def _cache_version(self, version_id, options, pos, size, parents):
1033
"""Cache a version record in the history array and index cache.
1035
This is inlined into __init__ for performance. KEEP IN SYNC.
1036
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1039
# only want the _history index to reference the 1st index entry
1041
if version_id not in self._cache:
1042
index = len(self._history)
1043
self._history.append(version_id)
1045
index = self._cache[version_id][5]
1046
self._cache[version_id] = (version_id,
1053
def __init__(self, transport, filename, mode, create=False, file_mode=None):
1054
_KnitComponentFile.__init__(self, transport, filename, mode, file_mode)
1056
# position in _history is the 'official' index for a revision
1057
# but the values may have come from a newer entry.
1058
# so - wc -l of a knit index is != the number of unique names
1061
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1066
pb.update('read knit index', count, total)
1067
fp = self._transport.get(self._filename)
1069
self.check_header(fp)
1070
# readlines reads the whole file at once:
1071
# bad for transports like http, good for local disk
1072
# we save 60 ms doing this one change (
1073
# from calling readline each time to calling
1075
# probably what we want for nice behaviour on
1076
# http is a incremental readlines that yields, or
1077
# a check for local vs non local indexes,
1078
for l in fp.readlines():
1080
if len(rec) < 5 or rec[-1] != ':':
1082
# FIXME: in the future we should determine if its a
1083
# short write - and ignore it
1084
# or a different failure, and raise. RBC 20060407
1088
#pb.update('read knit index', count, total)
1089
# See self._parse_parents
1091
for value in rec[4:-1]:
1093
# uncompressed reference
1094
parents.append(value[1:])
1096
# this is 15/4000ms faster than isinstance,
1098
# this function is called thousands of times a
1099
# second so small variations add up.
1100
assert value.__class__ is str
1101
parents.append(self._history[int(value)])
1102
# end self._parse_parents
1103
# self._cache_version(rec[0],
1104
# rec[1].split(','),
1108
# --- self._cache_version
1109
# only want the _history index to reference the 1st
1110
# index entry for version_id
1112
if version_id not in self._cache:
1113
index = len(self._history)
1114
self._history.append(version_id)
1116
index = self._cache[version_id][5]
1117
self._cache[version_id] = (version_id,
1123
# --- self._cache_version
1126
except NoSuchFile, e:
1127
if mode != 'w' or not create:
1131
pb.update('read knit index', total, total)
1134
def _parse_parents(self, compressed_parents):
1135
"""convert a list of string parent values into version ids.
1137
ints are looked up in the index.
1138
.FOO values are ghosts and converted in to FOO.
1140
NOTE: the function is retained here for clarity, and for possible
1141
use in partial index reads. However bulk processing now has
1142
it inlined in __init__ for inner-loop optimisation.
1145
for value in compressed_parents:
1146
if value[-1] == '.':
1147
# uncompressed reference
1148
result.append(value[1:])
1150
# this is 15/4000ms faster than isinstance,
1151
# this function is called thousands of times a
1152
# second so small variations add up.
1153
assert value.__class__ is str
1154
result.append(self._history[int(value)])
1157
def get_graph(self):
1159
for version_id, index in self._cache.iteritems():
1160
graph.append((version_id, index[4]))
1163
def get_ancestry(self, versions):
1164
"""See VersionedFile.get_ancestry."""
1165
# get a graph of all the mentioned versions:
1167
pending = set(versions)
1169
version = pending.pop()
1170
parents = self._cache[version][4]
1171
# got the parents ok
1173
parents = [parent for parent in parents if parent in self._cache]
1174
for parent in parents:
1175
# if not completed and not a ghost
1176
if parent not in graph:
1178
graph[version] = parents
1179
return topo_sort(graph.items())
1181
def get_ancestry_with_ghosts(self, versions):
1182
"""See VersionedFile.get_ancestry_with_ghosts."""
1183
# get a graph of all the mentioned versions:
1185
pending = set(versions)
1187
version = pending.pop()
1189
parents = self._cache[version][4]
1195
# got the parents ok
1196
for parent in parents:
1197
if parent not in graph:
1199
graph[version] = parents
1200
return topo_sort(graph.items())
1202
def num_versions(self):
1203
return len(self._history)
1205
__len__ = num_versions
1207
def get_versions(self):
1208
return self._history
1210
def idx_to_name(self, idx):
1211
return self._history[idx]
1213
def lookup(self, version_id):
1214
assert version_id in self._cache
1215
return self._cache[version_id][5]
1217
def _version_list_to_index(self, versions):
1218
encode_utf8 = cache_utf8.encode
1220
for version in versions:
1221
if version in self._cache:
1222
# -- inlined lookup() --
1223
result_list.append(str(self._cache[version][5]))
1224
# -- end lookup () --
1226
result_list.append('.' + encode_utf8(version))
1227
return ' '.join(result_list)
1229
def add_version(self, version_id, options, pos, size, parents):
1230
"""Add a version record to the index."""
1231
self.add_versions(((version_id, options, pos, size, parents),))
1233
def add_versions(self, versions):
1234
"""Add multiple versions to the index.
1236
:param versions: a list of tuples:
1237
(version_id, options, pos, size, parents).
1240
encode_utf8 = cache_utf8.encode
1241
for version_id, options, pos, size, parents in versions:
1242
line = "\n%s %s %s %s %s :" % (encode_utf8(version_id),
1246
self._version_list_to_index(parents))
1247
assert isinstance(line, str), \
1248
'content must be utf-8 encoded: %r' % (line,)
1250
self._transport.append_bytes(self._filename, ''.join(lines))
1251
# cache after writing, so that a failed write leads to missing cache
1252
# entries not extra ones. XXX TODO: RBC 20060502 in the event of a
1253
# failure, reload the index or flush it or some such, to prevent
1254
# writing records that did complete twice.
1255
for version_id, options, pos, size, parents in versions:
1256
self._cache_version(version_id, options, pos, size, parents)
1258
def has_version(self, version_id):
1259
"""True if the version is in the index."""
1260
return (version_id in self._cache)
1262
def get_position(self, version_id):
1263
"""Return data position and size of specified version."""
1264
return (self._cache[version_id][2], \
1265
self._cache[version_id][3])
1267
def get_method(self, version_id):
1268
"""Return compression method of specified version."""
1269
options = self._cache[version_id][1]
1270
if 'fulltext' in options:
1273
assert 'line-delta' in options
1276
def get_options(self, version_id):
1277
return self._cache[version_id][1]
1279
def get_parents(self, version_id):
1280
"""Return parents of specified version ignoring ghosts."""
1281
return [parent for parent in self._cache[version_id][4]
1282
if parent in self._cache]
1284
def get_parents_with_ghosts(self, version_id):
1285
"""Return parents of specified version with ghosts."""
1286
return self._cache[version_id][4]
1288
def check_versions_present(self, version_ids):
1289
"""Check that all specified versions are present."""
1290
version_ids = set(version_ids)
1291
for version_id in list(version_ids):
1292
if version_id in self._cache:
1293
version_ids.remove(version_id)
1295
raise RevisionNotPresent(list(version_ids)[0], self.filename)
1298
class _KnitData(_KnitComponentFile):
1299
"""Contents of the knit data file"""
1301
HEADER = "# bzr knit data 8\n"
1303
def __init__(self, transport, filename, mode, create=False, file_mode=None):
1304
_KnitComponentFile.__init__(self, transport, filename, mode)
1305
self._checked = False
1306
# TODO: jam 20060713 conceptually, this could spill to disk
1307
# if the cached size gets larger than a certain amount
1308
# but it complicates the model a bit, so for now just use
1309
# a simple dictionary
1311
self._do_cache = False
1313
self._transport.put_bytes_non_atomic(self._filename, '',
1316
def enable_cache(self):
1317
"""Enable caching of reads."""
1318
self._do_cache = True
1320
def clear_cache(self):
1321
"""Clear the record cache."""
1322
self._do_cache = False
1325
def _open_file(self):
1327
return self._transport.get(self._filename)
1332
def _record_to_data(self, version_id, digest, lines):
1333
"""Convert version_id, digest, lines into a raw data block.
1335
:return: (len, a StringIO instance with the raw data ready to read.)
1338
data_file = GzipFile(None, mode='wb', fileobj=sio)
1340
version_id_utf8 = cache_utf8.encode(version_id)
1341
data_file.writelines(chain(
1342
["version %s %d %s\n" % (version_id_utf8,
1346
["end %s\n" % version_id_utf8]))
1353
def add_raw_record(self, raw_data):
1354
"""Append a prepared record to the data file.
1356
:return: the offset in the data file raw_data was written.
1358
assert isinstance(raw_data, str), 'data must be plain bytes'
1359
return self._transport.append_bytes(self._filename, raw_data)
1361
def add_record(self, version_id, digest, lines):
1362
"""Write new text record to disk. Returns the position in the
1363
file where it was written."""
1364
size, sio = self._record_to_data(version_id, digest, lines)
1366
start_pos = self._transport.append_file(self._filename, sio)
1368
self._cache[version_id] = sio.getvalue()
1369
return start_pos, size
1371
def _parse_record_header(self, version_id, raw_data):
1372
"""Parse a record header for consistency.
1374
:return: the header and the decompressor stream.
1375
as (stream, header_record)
1377
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1378
rec = df.readline().split()
1380
raise KnitCorrupt(self._filename, 'unexpected number of elements in record header')
1381
if cache_utf8.decode(rec[1]) != version_id:
1382
raise KnitCorrupt(self._filename,
1383
'unexpected version, wanted %r, got %r' % (
1384
version_id, rec[1]))
1387
def _parse_record(self, version_id, data):
1389
# 4168 calls in 2880 217 internal
1390
# 4168 calls to _parse_record_header in 2121
1391
# 4168 calls to readlines in 330
1392
df, rec = self._parse_record_header(version_id, data)
1393
record_contents = df.readlines()
1394
l = record_contents.pop()
1395
assert len(record_contents) == int(rec[2])
1396
if l != 'end %s\n' % cache_utf8.encode(version_id):
1397
raise KnitCorrupt(self._filename, 'unexpected version end line %r, wanted %r'
1400
return record_contents, rec[3]
1402
def read_records_iter_raw(self, records):
1403
"""Read text records from data file and yield raw data.
1405
This unpacks enough of the text record to validate the id is
1406
as expected but thats all.
1408
# setup an iterator of the external records:
1409
# uses readv so nice and fast we hope.
1411
# grab the disk data needed.
1413
# Don't check _cache if it is empty
1414
needed_offsets = [(pos, size) for version_id, pos, size
1416
if version_id not in self._cache]
1418
needed_offsets = [(pos, size) for version_id, pos, size
1421
raw_records = self._transport.readv(self._filename, needed_offsets)
1424
for version_id, pos, size in records:
1425
if version_id in self._cache:
1426
# This data has already been validated
1427
data = self._cache[version_id]
1429
pos, data = raw_records.next()
1431
self._cache[version_id] = data
1433
# validate the header
1434
df, rec = self._parse_record_header(version_id, data)
1436
yield version_id, data
1438
def read_records_iter(self, records):
1439
"""Read text records from data file and yield result.
1441
The result will be returned in whatever is the fastest to read.
1442
Not by the order requested. Also, multiple requests for the same
1443
record will only yield 1 response.
1444
:param records: A list of (version_id, pos, len) entries
1445
:return: Yields (version_id, contents, digest) in the order
1446
read, not the order requested
1452
# Skip records we have alread seen
1453
yielded_records = set()
1454
needed_records = set()
1455
for record in records:
1456
if record[0] in self._cache:
1457
if record[0] in yielded_records:
1459
yielded_records.add(record[0])
1460
data = self._cache[record[0]]
1461
content, digest = self._parse_record(record[0], data)
1462
yield (record[0], content, digest)
1464
needed_records.add(record)
1465
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1467
needed_records = sorted(set(records), key=operator.itemgetter(1))
1469
if not needed_records:
1472
# The transport optimizes the fetching as well
1473
# (ie, reads continuous ranges.)
1474
readv_response = self._transport.readv(self._filename,
1475
[(pos, size) for version_id, pos, size in needed_records])
1477
for (version_id, pos, size), (pos, data) in \
1478
izip(iter(needed_records), readv_response):
1479
content, digest = self._parse_record(version_id, data)
1481
self._cache[version_id] = data
1482
yield version_id, content, digest
1484
def read_records(self, records):
1485
"""Read records into a dictionary."""
1487
for record_id, content, digest in \
1488
self.read_records_iter(records):
1489
components[record_id] = (content, digest)
1493
class InterKnit(InterVersionedFile):
1494
"""Optimised code paths for knit to knit operations."""
1496
_matching_file_from_factory = KnitVersionedFile
1497
_matching_file_to_factory = KnitVersionedFile
1500
def is_compatible(source, target):
1501
"""Be compatible with knits. """
1503
return (isinstance(source, KnitVersionedFile) and
1504
isinstance(target, KnitVersionedFile))
1505
except AttributeError:
1508
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1509
"""See InterVersionedFile.join."""
1510
assert isinstance(self.source, KnitVersionedFile)
1511
assert isinstance(self.target, KnitVersionedFile)
1513
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1518
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1520
version_ids = list(version_ids)
1521
if None in version_ids:
1522
version_ids.remove(None)
1524
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1525
this_versions = set(self.target._index.get_versions())
1526
needed_versions = self.source_ancestry - this_versions
1527
cross_check_versions = self.source_ancestry.intersection(this_versions)
1528
mismatched_versions = set()
1529
for version in cross_check_versions:
1530
# scan to include needed parents.
1531
n1 = set(self.target.get_parents_with_ghosts(version))
1532
n2 = set(self.source.get_parents_with_ghosts(version))
1534
# FIXME TEST this check for cycles being introduced works
1535
# the logic is we have a cycle if in our graph we are an
1536
# ancestor of any of the n2 revisions.
1542
parent_ancestors = self.source.get_ancestry(parent)
1543
if version in parent_ancestors:
1544
raise errors.GraphCycleError([parent, version])
1545
# ensure this parent will be available later.
1546
new_parents = n2.difference(n1)
1547
needed_versions.update(new_parents.difference(this_versions))
1548
mismatched_versions.add(version)
1550
if not needed_versions and not mismatched_versions:
1552
full_list = topo_sort(self.source.get_graph())
1554
version_list = [i for i in full_list if (not self.target.has_version(i)
1555
and i in needed_versions)]
1559
copy_queue_records = []
1561
for version_id in version_list:
1562
options = self.source._index.get_options(version_id)
1563
parents = self.source._index.get_parents_with_ghosts(version_id)
1564
# check that its will be a consistent copy:
1565
for parent in parents:
1566
# if source has the parent, we must :
1567
# * already have it or
1568
# * have it scheduled already
1569
# otherwise we don't care
1570
assert (self.target.has_version(parent) or
1571
parent in copy_set or
1572
not self.source.has_version(parent))
1573
data_pos, data_size = self.source._index.get_position(version_id)
1574
copy_queue_records.append((version_id, data_pos, data_size))
1575
copy_queue.append((version_id, options, parents))
1576
copy_set.add(version_id)
1578
# data suck the join:
1580
total = len(version_list)
1583
for (version_id, raw_data), \
1584
(version_id2, options, parents) in \
1585
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1587
assert version_id == version_id2, 'logic error, inconsistent results'
1589
pb.update("Joining knit", count, total)
1590
raw_records.append((version_id, options, parents, len(raw_data)))
1591
raw_datum.append(raw_data)
1592
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1594
for version in mismatched_versions:
1595
# FIXME RBC 20060309 is this needed?
1596
n1 = set(self.target.get_parents_with_ghosts(version))
1597
n2 = set(self.source.get_parents_with_ghosts(version))
1598
# write a combined record to our history preserving the current
1599
# parents as first in the list
1600
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1601
self.target.fix_parents(version, new_parents)
1607
InterVersionedFile.register_optimiser(InterKnit)
1610
class WeaveToKnit(InterVersionedFile):
1611
"""Optimised code paths for weave to knit operations."""
1613
_matching_file_from_factory = bzrlib.weave.WeaveFile
1614
_matching_file_to_factory = KnitVersionedFile
1617
def is_compatible(source, target):
1618
"""Be compatible with weaves to knits."""
1620
return (isinstance(source, bzrlib.weave.Weave) and
1621
isinstance(target, KnitVersionedFile))
1622
except AttributeError:
1625
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1626
"""See InterVersionedFile.join."""
1627
assert isinstance(self.source, bzrlib.weave.Weave)
1628
assert isinstance(self.target, KnitVersionedFile)
1630
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1635
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1637
version_ids = list(version_ids)
1639
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1640
this_versions = set(self.target._index.get_versions())
1641
needed_versions = self.source_ancestry - this_versions
1642
cross_check_versions = self.source_ancestry.intersection(this_versions)
1643
mismatched_versions = set()
1644
for version in cross_check_versions:
1645
# scan to include needed parents.
1646
n1 = set(self.target.get_parents_with_ghosts(version))
1647
n2 = set(self.source.get_parents(version))
1648
# if all of n2's parents are in n1, then its fine.
1649
if n2.difference(n1):
1650
# FIXME TEST this check for cycles being introduced works
1651
# the logic is we have a cycle if in our graph we are an
1652
# ancestor of any of the n2 revisions.
1658
parent_ancestors = self.source.get_ancestry(parent)
1659
if version in parent_ancestors:
1660
raise errors.GraphCycleError([parent, version])
1661
# ensure this parent will be available later.
1662
new_parents = n2.difference(n1)
1663
needed_versions.update(new_parents.difference(this_versions))
1664
mismatched_versions.add(version)
1666
if not needed_versions and not mismatched_versions:
1668
full_list = topo_sort(self.source.get_graph())
1670
version_list = [i for i in full_list if (not self.target.has_version(i)
1671
and i in needed_versions)]
1675
total = len(version_list)
1676
for version_id in version_list:
1677
pb.update("Converting to knit", count, total)
1678
parents = self.source.get_parents(version_id)
1679
# check that its will be a consistent copy:
1680
for parent in parents:
1681
# if source has the parent, we must already have it
1682
assert (self.target.has_version(parent))
1683
self.target.add_lines(
1684
version_id, parents, self.source.get_lines(version_id))
1687
for version in mismatched_versions:
1688
# FIXME RBC 20060309 is this needed?
1689
n1 = set(self.target.get_parents_with_ghosts(version))
1690
n2 = set(self.source.get_parents(version))
1691
# write a combined record to our history preserving the current
1692
# parents as first in the list
1693
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1694
self.target.fix_parents(version, new_parents)
1700
InterVersionedFile.register_optimiser(WeaveToKnit)
1703
class KnitSequenceMatcher(difflib.SequenceMatcher):
1704
"""Knit tuned sequence matcher.
1706
This is based on profiling of difflib which indicated some improvements
1707
for our usage pattern.
1710
def find_longest_match(self, alo, ahi, blo, bhi):
1711
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1713
If isjunk is not defined:
1715
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1716
alo <= i <= i+k <= ahi
1717
blo <= j <= j+k <= bhi
1718
and for all (i',j',k') meeting those conditions,
1721
and if i == i', j <= j'
1723
In other words, of all maximal matching blocks, return one that
1724
starts earliest in a, and of all those maximal matching blocks that
1725
start earliest in a, return the one that starts earliest in b.
1727
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1728
>>> s.find_longest_match(0, 5, 0, 9)
1731
If isjunk is defined, first the longest matching block is
1732
determined as above, but with the additional restriction that no
1733
junk element appears in the block. Then that block is extended as
1734
far as possible by matching (only) junk elements on both sides. So
1735
the resulting block never matches on junk except as identical junk
1736
happens to be adjacent to an "interesting" match.
1738
Here's the same example as before, but considering blanks to be
1739
junk. That prevents " abcd" from matching the " abcd" at the tail
1740
end of the second sequence directly. Instead only the "abcd" can
1741
match, and matches the leftmost "abcd" in the second sequence:
1743
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1744
>>> s.find_longest_match(0, 5, 0, 9)
1747
If no blocks match, return (alo, blo, 0).
1749
>>> s = SequenceMatcher(None, "ab", "c")
1750
>>> s.find_longest_match(0, 2, 0, 1)
1754
# CAUTION: stripping common prefix or suffix would be incorrect.
1758
# Longest matching block is "ab", but if common prefix is
1759
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1760
# strip, so ends up claiming that ab is changed to acab by
1761
# inserting "ca" in the middle. That's minimal but unintuitive:
1762
# "it's obvious" that someone inserted "ac" at the front.
1763
# Windiff ends up at the same place as diff, but by pairing up
1764
# the unique 'b's and then matching the first two 'a's.
1766
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1767
besti, bestj, bestsize = alo, blo, 0
1768
# find longest junk-free match
1769
# during an iteration of the loop, j2len[j] = length of longest
1770
# junk-free match ending with a[i-1] and b[j]
1774
for i in xrange(alo, ahi):
1775
# look at all instances of a[i] in b; note that because
1776
# b2j has no junk keys, the loop is skipped if a[i] is junk
1777
j2lenget = j2len.get
1780
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1781
# following improvement
1782
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1783
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1784
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1786
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1787
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1788
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
1800
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1802
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1805
# Extend the best by non-junk elements on each end. In particular,
1806
# "popular" non-junk elements aren't in b2j, which greatly speeds
1807
# the inner loop above, but also means "the best" match so far
1808
# doesn't contain any junk *or* popular non-junk elements.
1809
while besti > alo and bestj > blo and \
1810
not isbjunk(b[bestj-1]) and \
1811
a[besti-1] == b[bestj-1]:
1812
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1813
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1814
not isbjunk(b[bestj+bestsize]) and \
1815
a[besti+bestsize] == b[bestj+bestsize]:
1818
# Now that we have a wholly interesting match (albeit possibly
1819
# empty!), we may as well suck up the matching junk on each
1820
# side of it too. Can't think of a good reason not to, and it
1821
# saves post-processing the (possibly considerable) expense of
1822
# figuring out what to do with it. In the case of an empty
1823
# interesting match, this is clearly the right thing to do,
1824
# because no other kind of match is possible in the regions.
1825
while besti > alo and bestj > blo and \
1826
isbjunk(b[bestj-1]) and \
1827
a[besti-1] == b[bestj-1]:
1828
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1829
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1830
isbjunk(b[bestj+bestsize]) and \
1831
a[besti+bestsize] == b[bestj+bestsize]:
1832
bestsize = bestsize + 1
1834
return besti, bestj, bestsize
added
removed
bzrlib/knit.py
