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(name + INDEX_SUFFIX + '.tmp', self.transport.get(self._index._filename),)
408
f = self._data._open_file()
410
transport.put(name + DATA_SUFFIX, f)
413
# move the copied index into place
414
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
416
def create_empty(self, name, transport, mode=None):
417
return KnitVersionedFile(name, transport, factory=self.factory, delta=self.delta, create=True)
419
def _fix_parents(self, version, new_parents):
420
"""Fix the parents list for version.
422
This is done by appending a new version to the index
423
with identical data except for the parents list.
424
the parents list must be a superset of the current
427
current_values = self._index._cache[version]
428
assert set(current_values[4]).difference(set(new_parents)) == set()
429
self._index.add_version(version,
435
def get_delta(self, version_id):
436
"""Get a delta for constructing version from some other version."""
437
if not self.has_version(version_id):
438
raise RevisionNotPresent(version_id, self.filename)
440
parents = self.get_parents(version_id)
445
data_pos, data_size = self._index.get_position(version_id)
446
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
447
version_idx = self._index.lookup(version_id)
448
noeol = 'no-eol' in self._index.get_options(version_id)
449
if 'fulltext' == self._index.get_method(version_id):
450
new_content = self.factory.parse_fulltext(data, version_idx)
451
if parent is not None:
452
reference_content = self._get_content(parent)
453
old_texts = reference_content.text()
456
new_texts = new_content.text()
457
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
458
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
460
delta = self.factory.parse_line_delta(data, version_idx)
461
return parent, sha1, noeol, delta
463
def get_graph_with_ghosts(self):
464
"""See VersionedFile.get_graph_with_ghosts()."""
465
graph_items = self._index.get_graph()
466
return dict(graph_items)
468
def get_sha1(self, version_id):
469
"""See VersionedFile.get_sha1()."""
470
record_map = self._get_record_map([version_id])
471
method, content, digest, next = record_map[version_id]
476
"""See VersionedFile.get_suffixes()."""
477
return [DATA_SUFFIX, INDEX_SUFFIX]
479
def has_ghost(self, version_id):
480
"""True if there is a ghost reference in the file to version_id."""
482
if self.has_version(version_id):
484
# optimisable if needed by memoising the _ghosts set.
485
items = self._index.get_graph()
486
for node, parents in items:
487
for parent in parents:
488
if parent not in self._index._cache:
489
if parent == version_id:
494
"""See VersionedFile.versions."""
495
return self._index.get_versions()
497
def has_version(self, version_id):
498
"""See VersionedFile.has_version."""
499
return self._index.has_version(version_id)
501
__contains__ = has_version
503
def _merge_annotations(self, content, parents, parent_texts={},
504
delta=None, annotated=None):
505
"""Merge annotations for content. This is done by comparing
506
the annotations based on changed to the text.
510
for parent_id in parents:
511
merge_content = self._get_content(parent_id, parent_texts)
512
seq = KnitSequenceMatcher(None, merge_content.text(), content.text())
513
if delta_seq is None:
514
# setup a delta seq to reuse.
516
for i, j, n in seq.get_matching_blocks():
519
# this appears to copy (origin, text) pairs across to the new
520
# content for any line that matches the last-checked parent.
521
# FIXME: save the sequence control data for delta compression
522
# against the most relevant parent rather than rediffing.
523
content._lines[j:j+n] = merge_content._lines[i:i+n]
526
reference_content = self._get_content(parents[0], parent_texts)
527
new_texts = content.text()
528
old_texts = reference_content.text()
529
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
530
return self._make_line_delta(delta_seq, content)
532
def _make_line_delta(self, delta_seq, new_content):
533
"""Generate a line delta from delta_seq and new_content."""
535
for op in delta_seq.get_opcodes():
538
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
541
def _get_components_positions(self, version_ids):
542
"""Produce a map of position data for the components of versions.
544
This data is intended to be used for retrieving the knit records.
546
A dict of version_id to (method, data_pos, data_size, next) is
548
method is the way referenced data should be applied.
549
data_pos is the position of the data in the knit.
550
data_size is the size of the data in the knit.
551
next is the build-parent of the version, or None for fulltexts.
554
for version_id in version_ids:
557
while cursor is not None and cursor not in component_data:
558
method = self._index.get_method(cursor)
559
if method == 'fulltext':
562
next = self.get_parents(cursor)[0]
563
data_pos, data_size = self._index.get_position(cursor)
564
component_data[cursor] = (method, data_pos, data_size, next)
566
return component_data
568
def _get_content(self, version_id, parent_texts={}):
569
"""Returns a content object that makes up the specified
571
if not self.has_version(version_id):
572
raise RevisionNotPresent(version_id, self.filename)
574
cached_version = parent_texts.get(version_id, None)
575
if cached_version is not None:
576
return cached_version
578
text_map, contents_map = self._get_content_maps([version_id])
579
return contents_map[version_id]
581
def _check_versions_present(self, version_ids):
582
"""Check that all specified versions are present."""
583
version_ids = set(version_ids)
584
for r in list(version_ids):
585
if self._index.has_version(r):
586
version_ids.remove(r)
588
raise RevisionNotPresent(list(version_ids)[0], self.filename)
590
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
591
"""See VersionedFile.add_lines_with_ghosts()."""
592
self._check_add(version_id, lines)
593
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
595
def _add_lines(self, version_id, parents, lines, parent_texts):
596
"""See VersionedFile.add_lines."""
597
self._check_add(version_id, lines)
598
self._check_versions_present(parents)
599
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
601
def _check_add(self, version_id, lines):
602
"""check that version_id and lines are safe to add."""
603
assert self.writable, "knit is not opened for write"
604
### FIXME escape. RBC 20060228
605
if contains_whitespace(version_id):
606
raise InvalidRevisionId(version_id, self.filename)
607
if self.has_version(version_id):
608
raise RevisionAlreadyPresent(version_id, self.filename)
609
self._check_lines_not_unicode(lines)
610
self._check_lines_are_lines(lines)
612
def _add(self, version_id, lines, parents, delta, parent_texts):
613
"""Add a set of lines on top of version specified by parents.
615
If delta is true, compress the text as a line-delta against
618
Any versions not present will be converted into ghosts.
620
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
621
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
622
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
623
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
624
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
625
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
626
# +1383 0 8.0370 8.0370 +<len>
627
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
628
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
629
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
630
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
634
if parent_texts is None:
636
for parent in parents:
637
if not self.has_version(parent):
638
ghosts.append(parent)
640
present_parents.append(parent)
642
if delta and not len(present_parents):
645
digest = sha_strings(lines)
648
if lines[-1][-1] != '\n':
649
options.append('no-eol')
650
lines[-1] = lines[-1] + '\n'
652
if len(present_parents) and delta:
653
# To speed the extract of texts the delta chain is limited
654
# to a fixed number of deltas. This should minimize both
655
# I/O and the time spend applying deltas.
657
delta_parents = present_parents
659
parent = delta_parents[0]
660
method = self._index.get_method(parent)
661
if method == 'fulltext':
663
delta_parents = self._index.get_parents(parent)
665
if method == 'line-delta':
668
lines = self.factory.make(lines, version_id)
669
if delta or (self.factory.annotated and len(present_parents) > 0):
670
# Merge annotations from parent texts if so is needed.
671
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
672
delta, self.factory.annotated)
675
options.append('line-delta')
676
store_lines = self.factory.lower_line_delta(delta_hunks)
678
options.append('fulltext')
679
store_lines = self.factory.lower_fulltext(lines)
681
where, size = self._data.add_record(version_id, digest, store_lines)
682
self._index.add_version(version_id, options, where, size, parents)
685
def check(self, progress_bar=None):
686
"""See VersionedFile.check()."""
688
def _clone_text(self, new_version_id, old_version_id, parents):
689
"""See VersionedFile.clone_text()."""
690
# FIXME RBC 20060228 make fast by only inserting an index with null
692
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
694
def get_lines(self, version_id):
695
"""See VersionedFile.get_lines()."""
696
return self.get_line_list([version_id])[0]
698
def _get_record_map(self, version_ids):
699
"""Produce a dictionary of knit records.
701
The keys are version_ids, the values are tuples of (method, content,
703
method is the way the content should be applied.
704
content is a KnitContent object.
705
digest is the SHA1 digest of this version id after all steps are done
706
next is the build-parent of the version, i.e. the leftmost ancestor.
707
If the method is fulltext, next will be None.
709
position_map = self._get_components_positions(version_ids)
710
# c = component_id, m = method, p = position, s = size, n = next
711
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
713
for component_id, content, digest in \
714
self._data.read_records_iter(records):
715
method, position, size, next = position_map[component_id]
716
record_map[component_id] = method, content, digest, next
720
def get_text(self, version_id):
721
"""See VersionedFile.get_text"""
722
return self.get_texts([version_id])[0]
724
def get_texts(self, version_ids):
725
return [''.join(l) for l in self.get_line_list(version_ids)]
727
def get_line_list(self, version_ids):
728
"""Return the texts of listed versions as a list of strings."""
729
text_map, content_map = self._get_content_maps(version_ids)
730
return [text_map[v] for v in version_ids]
732
def _get_content_maps(self, version_ids):
733
"""Produce maps of text and KnitContents
735
:return: (text_map, content_map) where text_map contains the texts for
736
the requested versions and content_map contains the KnitContents.
737
Both dicts take version_ids as their keys.
739
for version_id in version_ids:
740
if not self.has_version(version_id):
741
raise RevisionNotPresent(version_id, self.filename)
742
record_map = self._get_record_map(version_ids)
747
for version_id in version_ids:
750
while cursor is not None:
751
method, data, digest, next = record_map[cursor]
752
components.append((cursor, method, data, digest))
753
if cursor in content_map:
758
for component_id, method, data, digest in reversed(components):
759
if component_id in content_map:
760
content = content_map[component_id]
762
version_idx = self._index.lookup(component_id)
763
if method == 'fulltext':
764
assert content is None
765
content = self.factory.parse_fulltext(data, version_idx)
766
elif method == 'line-delta':
767
delta = self.factory.parse_line_delta(data[:],
769
content = content.copy()
770
content._lines = self._apply_delta(content._lines,
772
content_map[component_id] = content
774
if 'no-eol' in self._index.get_options(version_id):
775
content = content.copy()
776
line = content._lines[-1][1].rstrip('\n')
777
content._lines[-1] = (content._lines[-1][0], line)
778
final_content[version_id] = content
780
# digest here is the digest from the last applied component.
781
text = content.text()
782
if sha_strings(text) != digest:
783
raise KnitCorrupt(self.filename,
784
'sha-1 does not match %s' % version_id)
786
text_map[version_id] = text
787
return text_map, final_content
789
def iter_lines_added_or_present_in_versions(self, version_ids=None):
790
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
791
if version_ids is None:
792
version_ids = self.versions()
793
# we don't care about inclusions, the caller cares.
794
# but we need to setup a list of records to visit.
795
# we need version_id, position, length
796
version_id_records = []
797
requested_versions = list(version_ids)
798
# filter for available versions
799
for version_id in requested_versions:
800
if not self.has_version(version_id):
801
raise RevisionNotPresent(version_id, self.filename)
802
# get a in-component-order queue:
804
for version_id in self.versions():
805
if version_id in requested_versions:
806
version_ids.append(version_id)
807
data_pos, length = self._index.get_position(version_id)
808
version_id_records.append((version_id, data_pos, length))
810
pb = bzrlib.ui.ui_factory.nested_progress_bar()
812
total = len(version_id_records)
814
pb.update('Walking content.', count, total)
815
for version_id, data, sha_value in \
816
self._data.read_records_iter(version_id_records):
817
pb.update('Walking content.', count, total)
818
method = self._index.get_method(version_id)
819
version_idx = self._index.lookup(version_id)
820
assert method in ('fulltext', 'line-delta')
821
if method == 'fulltext':
822
content = self.factory.parse_fulltext(data, version_idx)
823
for line in content.text():
826
delta = self.factory.parse_line_delta(data, version_idx)
827
for start, end, count, lines in delta:
828
for origin, line in lines:
831
pb.update('Walking content.', total, total)
834
pb.update('Walking content.', total, total)
838
def num_versions(self):
839
"""See VersionedFile.num_versions()."""
840
return self._index.num_versions()
842
__len__ = num_versions
844
def annotate_iter(self, version_id):
845
"""See VersionedFile.annotate_iter."""
846
content = self._get_content(version_id)
847
for origin, text in content.annotate_iter():
850
def get_parents(self, version_id):
851
"""See VersionedFile.get_parents."""
854
# 52554 calls in 1264 872 internal down from 3674
856
return self._index.get_parents(version_id)
858
raise RevisionNotPresent(version_id, self.filename)
860
def get_parents_with_ghosts(self, version_id):
861
"""See VersionedFile.get_parents."""
863
return self._index.get_parents_with_ghosts(version_id)
865
raise RevisionNotPresent(version_id, self.filename)
867
def get_ancestry(self, versions):
868
"""See VersionedFile.get_ancestry."""
869
if isinstance(versions, basestring):
870
versions = [versions]
873
self._check_versions_present(versions)
874
return self._index.get_ancestry(versions)
876
def get_ancestry_with_ghosts(self, versions):
877
"""See VersionedFile.get_ancestry_with_ghosts."""
878
if isinstance(versions, basestring):
879
versions = [versions]
882
self._check_versions_present(versions)
883
return self._index.get_ancestry_with_ghosts(versions)
885
#@deprecated_method(zero_eight)
886
def walk(self, version_ids):
887
"""See VersionedFile.walk."""
888
# We take the short path here, and extract all relevant texts
889
# and put them in a weave and let that do all the work. Far
890
# from optimal, but is much simpler.
891
# FIXME RB 20060228 this really is inefficient!
892
from bzrlib.weave import Weave
894
w = Weave(self.filename)
895
ancestry = self.get_ancestry(version_ids)
896
sorted_graph = topo_sort(self._index.get_graph())
897
version_list = [vid for vid in sorted_graph if vid in ancestry]
899
for version_id in version_list:
900
lines = self.get_lines(version_id)
901
w.add_lines(version_id, self.get_parents(version_id), lines)
903
for lineno, insert_id, dset, line in w.walk(version_ids):
904
yield lineno, insert_id, dset, line
906
def plan_merge(self, ver_a, ver_b):
907
"""See VersionedFile.plan_merge."""
908
ancestors_b = set(self.get_ancestry(ver_b))
909
def status_a(revision, text):
910
if revision in ancestors_b:
911
return 'killed-b', text
915
ancestors_a = set(self.get_ancestry(ver_a))
916
def status_b(revision, text):
917
if revision in ancestors_a:
918
return 'killed-a', text
922
annotated_a = self.annotate(ver_a)
923
annotated_b = self.annotate(ver_b)
924
plain_a = [t for (a, t) in annotated_a]
925
plain_b = [t for (a, t) in annotated_b]
926
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
929
for ai, bi, l in blocks:
930
# process all mismatched sections
931
# (last mismatched section is handled because blocks always
932
# includes a 0-length last block)
933
for revision, text in annotated_a[a_cur:ai]:
934
yield status_a(revision, text)
935
for revision, text in annotated_b[b_cur:bi]:
936
yield status_b(revision, text)
938
# and now the matched section
941
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
942
assert text_a == text_b
943
yield "unchanged", text_a
946
class _KnitComponentFile(object):
947
"""One of the files used to implement a knit database"""
949
def __init__(self, transport, filename, mode, file_mode=None):
950
self._transport = transport
951
self._filename = filename
953
self._file_mode=file_mode
955
def write_header(self):
956
if self._transport.append(self._filename, StringIO(self.HEADER),
957
mode=self._file_mode):
958
raise KnitCorrupt(self._filename, 'misaligned after writing header')
960
def check_header(self, fp):
962
if line != self.HEADER:
963
raise KnitHeaderError(badline=line)
966
"""Commit is a nop."""
969
return '%s(%s)' % (self.__class__.__name__, self._filename)
972
class _KnitIndex(_KnitComponentFile):
973
"""Manages knit index file.
975
The index is already kept in memory and read on startup, to enable
976
fast lookups of revision information. The cursor of the index
977
file is always pointing to the end, making it easy to append
980
_cache is a cache for fast mapping from version id to a Index
983
_history is a cache for fast mapping from indexes to version ids.
985
The index data format is dictionary compressed when it comes to
986
parent references; a index entry may only have parents that with a
987
lover index number. As a result, the index is topological sorted.
989
Duplicate entries may be written to the index for a single version id
990
if this is done then the latter one completely replaces the former:
991
this allows updates to correct version and parent information.
992
Note that the two entries may share the delta, and that successive
993
annotations and references MUST point to the first entry.
995
The index file on disc contains a header, followed by one line per knit
996
record. The same revision can be present in an index file more than once.
997
The first occurrence gets assigned a sequence number starting from 0.
999
The format of a single line is
1000
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1001
REVISION_ID is a utf8-encoded revision id
1002
FLAGS is a comma separated list of flags about the record. Values include
1003
no-eol, line-delta, fulltext.
1004
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1005
that the the compressed data starts at.
1006
LENGTH is the ascii representation of the length of the data file.
1007
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1009
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1010
revision id already in the knit that is a parent of REVISION_ID.
1011
The ' :' marker is the end of record marker.
1014
when a write is interrupted to the index file, it will result in a line that
1015
does not end in ' :'. If the ' :' is not present at the end of a line, or at
1016
the end of the file, then the record that is missing it will be ignored by
1019
When writing new records to the index file, the data is preceded by '\n'
1020
to ensure that records always start on new lines even if the last write was
1021
interrupted. As a result its normal for the last line in the index to be
1022
missing a trailing newline. One can be added with no harmful effects.
1025
HEADER = "# bzr knit index 8\n"
1027
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1028
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1030
def _cache_version(self, version_id, options, pos, size, parents):
1031
"""Cache a version record in the history array and index cache.
1033
This is inlined into __init__ for performance. KEEP IN SYNC.
1034
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1037
# only want the _history index to reference the 1st index entry
1039
if version_id not in self._cache:
1040
index = len(self._history)
1041
self._history.append(version_id)
1043
index = self._cache[version_id][5]
1044
self._cache[version_id] = (version_id,
1051
def __init__(self, transport, filename, mode, create=False, file_mode=None):
1052
_KnitComponentFile.__init__(self, transport, filename, mode, file_mode)
1054
# position in _history is the 'official' index for a revision
1055
# but the values may have come from a newer entry.
1056
# so - wc -l of a knit index is != the number of unique names
1059
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1064
pb.update('read knit index', count, total)
1065
fp = self._transport.get(self._filename)
1067
self.check_header(fp)
1068
# readlines reads the whole file at once:
1069
# bad for transports like http, good for local disk
1070
# we save 60 ms doing this one change (
1071
# from calling readline each time to calling
1073
# probably what we want for nice behaviour on
1074
# http is a incremental readlines that yields, or
1075
# a check for local vs non local indexes,
1076
for l in fp.readlines():
1078
if len(rec) < 5 or rec[-1] != ':':
1080
# FIXME: in the future we should determine if its a
1081
# short write - and ignore it
1082
# or a different failure, and raise. RBC 20060407
1086
#pb.update('read knit index', count, total)
1087
# See self._parse_parents
1089
for value in rec[4:-1]:
1091
# uncompressed reference
1092
parents.append(value[1:])
1094
# this is 15/4000ms faster than isinstance,
1096
# this function is called thousands of times a
1097
# second so small variations add up.
1098
assert value.__class__ is str
1099
parents.append(self._history[int(value)])
1100
# end self._parse_parents
1101
# self._cache_version(rec[0],
1102
# rec[1].split(','),
1106
# --- self._cache_version
1107
# only want the _history index to reference the 1st
1108
# index entry for version_id
1110
if version_id not in self._cache:
1111
index = len(self._history)
1112
self._history.append(version_id)
1114
index = self._cache[version_id][5]
1115
self._cache[version_id] = (version_id,
1121
# --- self._cache_version
1124
except NoSuchFile, e:
1125
if mode != 'w' or not create:
1129
pb.update('read knit index', total, total)
1132
def _parse_parents(self, compressed_parents):
1133
"""convert a list of string parent values into version ids.
1135
ints are looked up in the index.
1136
.FOO values are ghosts and converted in to FOO.
1138
NOTE: the function is retained here for clarity, and for possible
1139
use in partial index reads. However bulk processing now has
1140
it inlined in __init__ for inner-loop optimisation.
1143
for value in compressed_parents:
1144
if value[-1] == '.':
1145
# uncompressed reference
1146
result.append(value[1:])
1148
# this is 15/4000ms faster than isinstance,
1149
# this function is called thousands of times a
1150
# second so small variations add up.
1151
assert value.__class__ is str
1152
result.append(self._history[int(value)])
1155
def get_graph(self):
1157
for version_id, index in self._cache.iteritems():
1158
graph.append((version_id, index[4]))
1161
def get_ancestry(self, versions):
1162
"""See VersionedFile.get_ancestry."""
1163
# get a graph of all the mentioned versions:
1165
pending = set(versions)
1167
version = pending.pop()
1168
parents = self._cache[version][4]
1169
# got the parents ok
1171
parents = [parent for parent in parents if parent in self._cache]
1172
for parent in parents:
1173
# if not completed and not a ghost
1174
if parent not in graph:
1176
graph[version] = parents
1177
return topo_sort(graph.items())
1179
def get_ancestry_with_ghosts(self, versions):
1180
"""See VersionedFile.get_ancestry_with_ghosts."""
1181
# get a graph of all the mentioned versions:
1183
pending = set(versions)
1185
version = pending.pop()
1187
parents = self._cache[version][4]
1193
# got the parents ok
1194
for parent in parents:
1195
if parent not in graph:
1197
graph[version] = parents
1198
return topo_sort(graph.items())
1200
def num_versions(self):
1201
return len(self._history)
1203
__len__ = num_versions
1205
def get_versions(self):
1206
return self._history
1208
def idx_to_name(self, idx):
1209
return self._history[idx]
1211
def lookup(self, version_id):
1212
assert version_id in self._cache
1213
return self._cache[version_id][5]
1215
def _version_list_to_index(self, versions):
1216
encode_utf8 = cache_utf8.encode
1218
for version in versions:
1219
if version in self._cache:
1220
# -- inlined lookup() --
1221
result_list.append(str(self._cache[version][5]))
1222
# -- end lookup () --
1224
result_list.append('.' + encode_utf8(version))
1225
return ' '.join(result_list)
1227
def add_version(self, version_id, options, pos, size, parents):
1228
"""Add a version record to the index."""
1229
self.add_versions(((version_id, options, pos, size, parents),))
1231
def add_versions(self, versions):
1232
"""Add multiple versions to the index.
1234
:param versions: a list of tuples:
1235
(version_id, options, pos, size, parents).
1238
encode_utf8 = cache_utf8.encode
1239
for version_id, options, pos, size, parents in versions:
1240
line = "\n%s %s %s %s %s :" % (encode_utf8(version_id),
1244
self._version_list_to_index(parents))
1245
assert isinstance(line, str), \
1246
'content must be utf-8 encoded: %r' % (line,)
1248
self._transport.append(self._filename, StringIO(''.join(lines)))
1249
# cache after writing, so that a failed write leads to missing cache
1250
# entries not extra ones. XXX TODO: RBC 20060502 in the event of a
1251
# failure, reload the index or flush it or some such, to prevent
1252
# writing records that did complete twice.
1253
for version_id, options, pos, size, parents in versions:
1254
self._cache_version(version_id, options, pos, size, parents)
1256
def has_version(self, version_id):
1257
"""True if the version is in the index."""
1258
return self._cache.has_key(version_id)
1260
def get_position(self, version_id):
1261
"""Return data position and size of specified version."""
1262
return (self._cache[version_id][2], \
1263
self._cache[version_id][3])
1265
def get_method(self, version_id):
1266
"""Return compression method of specified version."""
1267
options = self._cache[version_id][1]
1268
if 'fulltext' in options:
1271
assert 'line-delta' in options
1274
def get_options(self, version_id):
1275
return self._cache[version_id][1]
1277
def get_parents(self, version_id):
1278
"""Return parents of specified version ignoring ghosts."""
1279
return [parent for parent in self._cache[version_id][4]
1280
if parent in self._cache]
1282
def get_parents_with_ghosts(self, version_id):
1283
"""Return parents of specified version with ghosts."""
1284
return self._cache[version_id][4]
1286
def check_versions_present(self, version_ids):
1287
"""Check that all specified versions are present."""
1288
version_ids = set(version_ids)
1289
for version_id in list(version_ids):
1290
if version_id in self._cache:
1291
version_ids.remove(version_id)
1293
raise RevisionNotPresent(list(version_ids)[0], self.filename)
1296
class _KnitData(_KnitComponentFile):
1297
"""Contents of the knit data file"""
1299
HEADER = "# bzr knit data 8\n"
1301
def __init__(self, transport, filename, mode, create=False, file_mode=None):
1302
_KnitComponentFile.__init__(self, transport, filename, mode)
1303
self._checked = False
1304
# TODO: jam 20060713 conceptually, this could spill to disk
1305
# if the cached size gets larger than a certain amount
1306
# but it complicates the model a bit, so for now just use
1307
# a simple dictionary
1309
self._do_cache = False
1311
self._transport.put(self._filename, StringIO(''), mode=file_mode)
1313
def enable_cache(self):
1314
"""Enable caching of reads."""
1315
self._do_cache = True
1317
def clear_cache(self):
1318
"""Clear the record cache."""
1319
self._do_cache = False
1322
def _open_file(self):
1324
return self._transport.get(self._filename)
1329
def _record_to_data(self, version_id, digest, lines):
1330
"""Convert version_id, digest, lines into a raw data block.
1332
:return: (len, a StringIO instance with the raw data ready to read.)
1335
data_file = GzipFile(None, mode='wb', fileobj=sio)
1337
version_id_utf8 = cache_utf8.encode(version_id)
1338
data_file.writelines(chain(
1339
["version %s %d %s\n" % (version_id_utf8,
1343
["end %s\n" % version_id_utf8]))
1350
def add_raw_record(self, raw_data):
1351
"""Append a prepared record to the data file.
1353
:return: the offset in the data file raw_data was written.
1355
assert isinstance(raw_data, str), 'data must be plain bytes'
1356
return self._transport.append(self._filename, StringIO(raw_data))
1358
def add_record(self, version_id, digest, lines):
1359
"""Write new text record to disk. Returns the position in the
1360
file where it was written."""
1361
size, sio = self._record_to_data(version_id, digest, lines)
1363
start_pos = self._transport.append(self._filename, sio)
1365
self._cache[version_id] = sio.getvalue()
1366
return start_pos, size
1368
def _parse_record_header(self, version_id, raw_data):
1369
"""Parse a record header for consistency.
1371
:return: the header and the decompressor stream.
1372
as (stream, header_record)
1374
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1375
rec = df.readline().split()
1377
raise KnitCorrupt(self._filename, 'unexpected number of elements in record header')
1378
if cache_utf8.decode(rec[1]) != version_id:
1379
raise KnitCorrupt(self._filename,
1380
'unexpected version, wanted %r, got %r' % (
1381
version_id, rec[1]))
1384
def _parse_record(self, version_id, data):
1386
# 4168 calls in 2880 217 internal
1387
# 4168 calls to _parse_record_header in 2121
1388
# 4168 calls to readlines in 330
1389
df, rec = self._parse_record_header(version_id, data)
1390
record_contents = df.readlines()
1391
l = record_contents.pop()
1392
assert len(record_contents) == int(rec[2])
1393
if l != 'end %s\n' % cache_utf8.encode(version_id):
1394
raise KnitCorrupt(self._filename, 'unexpected version end line %r, wanted %r'
1397
return record_contents, rec[3]
1399
def read_records_iter_raw(self, records):
1400
"""Read text records from data file and yield raw data.
1402
This unpacks enough of the text record to validate the id is
1403
as expected but thats all.
1405
# setup an iterator of the external records:
1406
# uses readv so nice and fast we hope.
1408
# grab the disk data needed.
1410
# Don't check _cache if it is empty
1411
needed_offsets = [(pos, size) for version_id, pos, size
1413
if version_id not in self._cache]
1415
needed_offsets = [(pos, size) for version_id, pos, size
1418
raw_records = self._transport.readv(self._filename, needed_offsets)
1421
for version_id, pos, size in records:
1422
if version_id in self._cache:
1423
# This data has already been validated
1424
data = self._cache[version_id]
1426
pos, data = raw_records.next()
1428
self._cache[version_id] = data
1430
# validate the header
1431
df, rec = self._parse_record_header(version_id, data)
1433
yield version_id, data
1435
def read_records_iter(self, records):
1436
"""Read text records from data file and yield result.
1438
The result will be returned in whatever is the fastest to read.
1439
Not by the order requested. Also, multiple requests for the same
1440
record will only yield 1 response.
1441
:param records: A list of (version_id, pos, len) entries
1442
:return: Yields (version_id, contents, digest) in the order
1443
read, not the order requested
1449
# Skip records we have alread seen
1450
yielded_records = set()
1451
needed_records = set()
1452
for record in records:
1453
if record[0] in self._cache:
1454
if record[0] in yielded_records:
1456
yielded_records.add(record[0])
1457
data = self._cache[record[0]]
1458
content, digest = self._parse_record(record[0], data)
1459
yield (record[0], content, digest)
1461
needed_records.add(record)
1462
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1464
needed_records = sorted(set(records), key=operator.itemgetter(1))
1466
if not needed_records:
1469
# The transport optimizes the fetching as well
1470
# (ie, reads continuous ranges.)
1471
readv_response = self._transport.readv(self._filename,
1472
[(pos, size) for version_id, pos, size in needed_records])
1474
for (version_id, pos, size), (pos, data) in \
1475
izip(iter(needed_records), readv_response):
1476
content, digest = self._parse_record(version_id, data)
1478
self._cache[version_id] = data
1479
yield version_id, content, digest
1481
def read_records(self, records):
1482
"""Read records into a dictionary."""
1484
for record_id, content, digest in \
1485
self.read_records_iter(records):
1486
components[record_id] = (content, digest)
1490
class InterKnit(InterVersionedFile):
1491
"""Optimised code paths for knit to knit operations."""
1493
_matching_file_from_factory = KnitVersionedFile
1494
_matching_file_to_factory = KnitVersionedFile
1497
def is_compatible(source, target):
1498
"""Be compatible with knits. """
1500
return (isinstance(source, KnitVersionedFile) and
1501
isinstance(target, KnitVersionedFile))
1502
except AttributeError:
1505
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1506
"""See InterVersionedFile.join."""
1507
assert isinstance(self.source, KnitVersionedFile)
1508
assert isinstance(self.target, KnitVersionedFile)
1510
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1515
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1517
version_ids = list(version_ids)
1518
if None in version_ids:
1519
version_ids.remove(None)
1521
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1522
this_versions = set(self.target._index.get_versions())
1523
needed_versions = self.source_ancestry - this_versions
1524
cross_check_versions = self.source_ancestry.intersection(this_versions)
1525
mismatched_versions = set()
1526
for version in cross_check_versions:
1527
# scan to include needed parents.
1528
n1 = set(self.target.get_parents_with_ghosts(version))
1529
n2 = set(self.source.get_parents_with_ghosts(version))
1531
# FIXME TEST this check for cycles being introduced works
1532
# the logic is we have a cycle if in our graph we are an
1533
# ancestor of any of the n2 revisions.
1539
parent_ancestors = self.source.get_ancestry(parent)
1540
if version in parent_ancestors:
1541
raise errors.GraphCycleError([parent, version])
1542
# ensure this parent will be available later.
1543
new_parents = n2.difference(n1)
1544
needed_versions.update(new_parents.difference(this_versions))
1545
mismatched_versions.add(version)
1547
if not needed_versions and not mismatched_versions:
1549
full_list = topo_sort(self.source.get_graph())
1551
version_list = [i for i in full_list if (not self.target.has_version(i)
1552
and i in needed_versions)]
1556
copy_queue_records = []
1558
for version_id in version_list:
1559
options = self.source._index.get_options(version_id)
1560
parents = self.source._index.get_parents_with_ghosts(version_id)
1561
# check that its will be a consistent copy:
1562
for parent in parents:
1563
# if source has the parent, we must :
1564
# * already have it or
1565
# * have it scheduled already
1566
# otherwise we don't care
1567
assert (self.target.has_version(parent) or
1568
parent in copy_set or
1569
not self.source.has_version(parent))
1570
data_pos, data_size = self.source._index.get_position(version_id)
1571
copy_queue_records.append((version_id, data_pos, data_size))
1572
copy_queue.append((version_id, options, parents))
1573
copy_set.add(version_id)
1575
# data suck the join:
1577
total = len(version_list)
1580
for (version_id, raw_data), \
1581
(version_id2, options, parents) in \
1582
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1584
assert version_id == version_id2, 'logic error, inconsistent results'
1586
pb.update("Joining knit", count, total)
1587
raw_records.append((version_id, options, parents, len(raw_data)))
1588
raw_datum.append(raw_data)
1589
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1591
for version in mismatched_versions:
1592
# FIXME RBC 20060309 is this needed?
1593
n1 = set(self.target.get_parents_with_ghosts(version))
1594
n2 = set(self.source.get_parents_with_ghosts(version))
1595
# write a combined record to our history preserving the current
1596
# parents as first in the list
1597
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1598
self.target.fix_parents(version, new_parents)
1604
InterVersionedFile.register_optimiser(InterKnit)
1607
class WeaveToKnit(InterVersionedFile):
1608
"""Optimised code paths for weave to knit operations."""
1610
_matching_file_from_factory = bzrlib.weave.WeaveFile
1611
_matching_file_to_factory = KnitVersionedFile
1614
def is_compatible(source, target):
1615
"""Be compatible with weaves to knits."""
1617
return (isinstance(source, bzrlib.weave.Weave) and
1618
isinstance(target, KnitVersionedFile))
1619
except AttributeError:
1622
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1623
"""See InterVersionedFile.join."""
1624
assert isinstance(self.source, bzrlib.weave.Weave)
1625
assert isinstance(self.target, KnitVersionedFile)
1627
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1632
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1634
version_ids = list(version_ids)
1636
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1637
this_versions = set(self.target._index.get_versions())
1638
needed_versions = self.source_ancestry - this_versions
1639
cross_check_versions = self.source_ancestry.intersection(this_versions)
1640
mismatched_versions = set()
1641
for version in cross_check_versions:
1642
# scan to include needed parents.
1643
n1 = set(self.target.get_parents_with_ghosts(version))
1644
n2 = set(self.source.get_parents(version))
1645
# if all of n2's parents are in n1, then its fine.
1646
if n2.difference(n1):
1647
# FIXME TEST this check for cycles being introduced works
1648
# the logic is we have a cycle if in our graph we are an
1649
# ancestor of any of the n2 revisions.
1655
parent_ancestors = self.source.get_ancestry(parent)
1656
if version in parent_ancestors:
1657
raise errors.GraphCycleError([parent, version])
1658
# ensure this parent will be available later.
1659
new_parents = n2.difference(n1)
1660
needed_versions.update(new_parents.difference(this_versions))
1661
mismatched_versions.add(version)
1663
if not needed_versions and not mismatched_versions:
1665
full_list = topo_sort(self.source.get_graph())
1667
version_list = [i for i in full_list if (not self.target.has_version(i)
1668
and i in needed_versions)]
1672
total = len(version_list)
1673
for version_id in version_list:
1674
pb.update("Converting to knit", count, total)
1675
parents = self.source.get_parents(version_id)
1676
# check that its will be a consistent copy:
1677
for parent in parents:
1678
# if source has the parent, we must already have it
1679
assert (self.target.has_version(parent))
1680
self.target.add_lines(
1681
version_id, parents, self.source.get_lines(version_id))
1684
for version in mismatched_versions:
1685
# FIXME RBC 20060309 is this needed?
1686
n1 = set(self.target.get_parents_with_ghosts(version))
1687
n2 = set(self.source.get_parents(version))
1688
# write a combined record to our history preserving the current
1689
# parents as first in the list
1690
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1691
self.target.fix_parents(version, new_parents)
1697
InterVersionedFile.register_optimiser(WeaveToKnit)
1700
class KnitSequenceMatcher(difflib.SequenceMatcher):
1701
"""Knit tuned sequence matcher.
1703
This is based on profiling of difflib which indicated some improvements
1704
for our usage pattern.
1707
def find_longest_match(self, alo, ahi, blo, bhi):
1708
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1710
If isjunk is not defined:
1712
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1713
alo <= i <= i+k <= ahi
1714
blo <= j <= j+k <= bhi
1715
and for all (i',j',k') meeting those conditions,
1718
and if i == i', j <= j'
1720
In other words, of all maximal matching blocks, return one that
1721
starts earliest in a, and of all those maximal matching blocks that
1722
start earliest in a, return the one that starts earliest in b.
1724
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1725
>>> s.find_longest_match(0, 5, 0, 9)
1728
If isjunk is defined, first the longest matching block is
1729
determined as above, but with the additional restriction that no
1730
junk element appears in the block. Then that block is extended as
1731
far as possible by matching (only) junk elements on both sides. So
1732
the resulting block never matches on junk except as identical junk
1733
happens to be adjacent to an "interesting" match.
1735
Here's the same example as before, but considering blanks to be
1736
junk. That prevents " abcd" from matching the " abcd" at the tail
1737
end of the second sequence directly. Instead only the "abcd" can
1738
match, and matches the leftmost "abcd" in the second sequence:
1740
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1741
>>> s.find_longest_match(0, 5, 0, 9)
1744
If no blocks match, return (alo, blo, 0).
1746
>>> s = SequenceMatcher(None, "ab", "c")
1747
>>> s.find_longest_match(0, 2, 0, 1)
1751
# CAUTION: stripping common prefix or suffix would be incorrect.
1755
# Longest matching block is "ab", but if common prefix is
1756
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1757
# strip, so ends up claiming that ab is changed to acab by
1758
# inserting "ca" in the middle. That's minimal but unintuitive:
1759
# "it's obvious" that someone inserted "ac" at the front.
1760
# Windiff ends up at the same place as diff, but by pairing up
1761
# the unique 'b's and then matching the first two 'a's.
1763
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1764
besti, bestj, bestsize = alo, blo, 0
1765
# find longest junk-free match
1766
# during an iteration of the loop, j2len[j] = length of longest
1767
# junk-free match ending with a[i-1] and b[j]
1771
for i in xrange(alo, ahi):
1772
# look at all instances of a[i] in b; note that because
1773
# b2j has no junk keys, the loop is skipped if a[i] is junk
1774
j2lenget = j2len.get
1777
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1778
# following improvement
1779
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1780
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1781
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1783
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1784
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1785
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
1797
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1799
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1802
# Extend the best by non-junk elements on each end. In particular,
1803
# "popular" non-junk elements aren't in b2j, which greatly speeds
1804
# the inner loop above, but also means "the best" match so far
1805
# doesn't contain any junk *or* popular non-junk elements.
1806
while besti > alo and bestj > blo and \
1807
not isbjunk(b[bestj-1]) and \
1808
a[besti-1] == b[bestj-1]:
1809
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1810
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1811
not isbjunk(b[bestj+bestsize]) and \
1812
a[besti+bestsize] == b[bestj+bestsize]:
1815
# Now that we have a wholly interesting match (albeit possibly
1816
# empty!), we may as well suck up the matching junk on each
1817
# side of it too. Can't think of a good reason not to, and it
1818
# saves post-processing the (possibly considerable) expense of
1819
# figuring out what to do with it. In the case of an empty
1820
# interesting match, this is clearly the right thing to do,
1821
# because no other kind of match is possible in the regions.
1822
while besti > alo and bestj > blo and \
1823
isbjunk(b[bestj-1]) and \
1824
a[besti-1] == b[bestj-1]:
1825
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1826
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1827
isbjunk(b[bestj+bestsize]) and \
1828
a[besti+bestsize] == b[bestj+bestsize]:
1829
bestsize = bestsize + 1
1831
return besti, bestj, bestsize
added
removed
bzrlib/knit.py
