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
82
from bzrlib.errors import FileExists, NoSuchFile, KnitError, \
83
InvalidRevisionId, KnitCorrupt, KnitHeaderError, \
84
RevisionNotPresent, RevisionAlreadyPresent
85
from bzrlib.tuned_gzip import GzipFile
86
from bzrlib.trace import mutter
87
from bzrlib.osutils import contains_whitespace, contains_linebreaks, \
89
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
90
from bzrlib.tsort import topo_sort
92
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
95
# TODO: Split out code specific to this format into an associated object.
97
# TODO: Can we put in some kind of value to check that the index and data
98
# files belong together?
100
# TODO: accommodate binaries, perhaps by storing a byte count
102
# TODO: function to check whole file
104
# TODO: atomically append data, then measure backwards from the cursor
105
# position after writing to work out where it was located. we may need to
106
# bypass python file buffering.
108
DATA_SUFFIX = '.knit'
109
INDEX_SUFFIX = '.kndx'
112
class KnitContent(object):
113
"""Content of a knit version to which deltas can be applied."""
115
def __init__(self, lines):
118
def annotate_iter(self):
119
"""Yield tuples of (origin, text) for each content line."""
120
for origin, text in self._lines:
124
"""Return a list of (origin, text) tuples."""
125
return list(self.annotate_iter())
127
def line_delta_iter(self, new_lines):
128
"""Generate line-based delta from this content to new_lines."""
129
new_texts = [text for origin, text in new_lines._lines]
130
old_texts = [text for origin, text in self._lines]
131
s = KnitSequenceMatcher(None, old_texts, new_texts)
132
for op in s.get_opcodes():
135
# ofrom oto length data
136
yield (op[1], op[2], op[4]-op[3], new_lines._lines[op[3]:op[4]])
138
def line_delta(self, new_lines):
139
return list(self.line_delta_iter(new_lines))
142
return [text for origin, text in self._lines]
145
return KnitContent(self._lines[:])
148
class _KnitFactory(object):
149
"""Base factory for creating content objects."""
151
def make(self, lines, version):
152
num_lines = len(lines)
153
return KnitContent(zip([version] * num_lines, lines))
156
class KnitAnnotateFactory(_KnitFactory):
157
"""Factory for creating annotated Content objects."""
161
def parse_fulltext(self, content, version):
162
"""Convert fulltext to internal representation
164
fulltext content is of the format
165
revid(utf8) plaintext\n
166
internal representation is of the format:
169
decode_utf8 = cache_utf8.decode
172
origin, text = line.split(' ', 1)
173
lines.append((decode_utf8(origin), text))
174
return KnitContent(lines)
176
def parse_line_delta_iter(self, lines):
177
for result_item in self.parse_line_delta[lines]:
180
def parse_line_delta(self, lines, version):
181
"""Convert a line based delta into internal representation.
183
line delta is in the form of:
184
intstart intend intcount
186
revid(utf8) newline\n
187
internal representation is
188
(start, end, count, [1..count tuples (revid, newline)])
190
decode_utf8 = cache_utf8.decode
194
# walk through the lines parsing.
196
start, end, count = [int(n) for n in header.split(',')]
200
origin, text = next().split(' ', 1)
202
contents.append((decode_utf8(origin), text))
203
result.append((start, end, count, contents))
206
def lower_fulltext(self, content):
207
"""convert a fulltext content record into a serializable form.
209
see parse_fulltext which this inverts.
211
encode_utf8 = cache_utf8.encode
212
return ['%s %s' % (encode_utf8(o), t) for o, t in content._lines]
214
def lower_line_delta(self, delta):
215
"""convert a delta into a serializable form.
217
See parse_line_delta which this inverts.
219
encode_utf8 = cache_utf8.encode
221
for start, end, c, lines in delta:
222
out.append('%d,%d,%d\n' % (start, end, c))
223
out.extend(encode_utf8(origin) + ' ' + text
224
for origin, text in lines)
228
class KnitPlainFactory(_KnitFactory):
229
"""Factory for creating plain Content objects."""
233
def parse_fulltext(self, content, version):
234
"""This parses an unannotated fulltext.
236
Note that this is not a noop - the internal representation
237
has (versionid, line) - its just a constant versionid.
239
return self.make(content, version)
241
def parse_line_delta_iter(self, lines, version):
243
header = lines.pop(0)
244
start, end, c = [int(n) for n in header.split(',')]
245
yield start, end, c, zip([version] * c, lines[:c])
248
def parse_line_delta(self, lines, version):
249
return list(self.parse_line_delta_iter(lines, version))
251
def lower_fulltext(self, content):
252
return content.text()
254
def lower_line_delta(self, delta):
256
for start, end, c, lines in delta:
257
out.append('%d,%d,%d\n' % (start, end, c))
258
out.extend([text for origin, text in lines])
262
def make_empty_knit(transport, relpath):
263
"""Construct a empty knit at the specified location."""
264
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
268
class KnitVersionedFile(VersionedFile):
269
"""Weave-like structure with faster random access.
271
A knit stores a number of texts and a summary of the relationships
272
between them. Texts are identified by a string version-id. Texts
273
are normally stored and retrieved as a series of lines, but can
274
also be passed as single strings.
276
Lines are stored with the trailing newline (if any) included, to
277
avoid special cases for files with no final newline. Lines are
278
composed of 8-bit characters, not unicode. The combination of
279
these approaches should mean any 'binary' file can be safely
280
stored and retrieved.
283
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
284
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
285
create=False, create_parent_dir=False, delay_create=False,
287
"""Construct a knit at location specified by relpath.
289
:param create: If not True, only open an existing knit.
290
:param create_parent_dir: If True, create the parent directory if
291
creating the file fails. (This is used for stores with
292
hash-prefixes that may not exist yet)
293
:param delay_create: The calling code is aware that the knit won't
294
actually be created until the first data is stored.
296
if deprecated_passed(basis_knit):
297
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
298
" deprecated as of bzr 0.9.",
299
DeprecationWarning, stacklevel=2)
300
if access_mode is None:
302
super(KnitVersionedFile, self).__init__(access_mode)
303
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
304
self.transport = transport
305
self.filename = relpath
306
self.factory = factory or KnitAnnotateFactory()
307
self.writable = (access_mode == 'w')
310
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
311
access_mode, create=create, file_mode=file_mode,
312
create_parent_dir=create_parent_dir, delay_create=delay_create,
314
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
315
access_mode, create=create and not len(self), file_mode=file_mode,
316
create_parent_dir=create_parent_dir, delay_create=delay_create,
320
return '%s(%s)' % (self.__class__.__name__,
321
self.transport.abspath(self.filename))
323
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
324
"""See VersionedFile._add_delta()."""
325
self._check_add(version_id, []) # should we check the lines ?
326
self._check_versions_present(parents)
330
for parent in parents:
331
if not self.has_version(parent):
332
ghosts.append(parent)
334
present_parents.append(parent)
336
if delta_parent is None:
337
# reconstitute as full text.
338
assert len(delta) == 1 or len(delta) == 0
340
assert delta[0][0] == 0
341
assert delta[0][1] == 0, delta[0][1]
342
return super(KnitVersionedFile, self)._add_delta(version_id,
353
options.append('no-eol')
355
if delta_parent is not None:
356
# determine the current delta chain length.
357
# To speed the extract of texts the delta chain is limited
358
# to a fixed number of deltas. This should minimize both
359
# I/O and the time spend applying deltas.
361
delta_parents = [delta_parent]
363
parent = delta_parents[0]
364
method = self._index.get_method(parent)
365
if method == 'fulltext':
367
delta_parents = self._index.get_parents(parent)
369
if method == 'line-delta':
370
# did not find a fulltext in the delta limit.
371
# just do a normal insertion.
372
return super(KnitVersionedFile, self)._add_delta(version_id,
379
options.append('line-delta')
380
store_lines = self.factory.lower_line_delta(delta)
382
where, size = self._data.add_record(version_id, digest, store_lines)
383
self._index.add_version(version_id, options, where, size, parents)
385
def _add_raw_records(self, records, data):
386
"""Add all the records 'records' with data pre-joined in 'data'.
388
:param records: A list of tuples(version_id, options, parents, size).
389
:param data: The data for the records. When it is written, the records
390
are adjusted to have pos pointing into data by the sum of
391
the preceding records sizes.
394
pos = self._data.add_raw_record(data)
397
for (version_id, options, parents, size) in records:
398
index_entries.append((version_id, options, pos+offset,
400
if self._data._do_cache:
401
self._data._cache[version_id] = data[offset:offset+size]
403
self._index.add_versions(index_entries)
405
def enable_cache(self):
406
"""Start caching data for this knit"""
407
self._data.enable_cache()
409
def clear_cache(self):
410
"""Clear the data cache only."""
411
self._data.clear_cache()
413
def copy_to(self, name, transport):
414
"""See VersionedFile.copy_to()."""
415
# copy the current index to a temp index to avoid racing with local
417
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
418
self.transport.get(self._index._filename))
420
f = self._data._open_file()
422
transport.put_file(name + DATA_SUFFIX, f)
425
# move the copied index into place
426
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
428
def create_empty(self, name, transport, mode=None):
429
return KnitVersionedFile(name, transport, factory=self.factory,
430
delta=self.delta, create=True)
432
def _fix_parents(self, version, new_parents):
433
"""Fix the parents list for version.
435
This is done by appending a new version to the index
436
with identical data except for the parents list.
437
the parents list must be a superset of the current
440
current_values = self._index._cache[version]
441
assert set(current_values[4]).difference(set(new_parents)) == set()
442
self._index.add_version(version,
448
def get_delta(self, version_id):
449
"""Get a delta for constructing version from some other version."""
450
if not self.has_version(version_id):
451
raise RevisionNotPresent(version_id, self.filename)
453
parents = self.get_parents(version_id)
458
data_pos, data_size = self._index.get_position(version_id)
459
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
460
version_idx = self._index.lookup(version_id)
461
noeol = 'no-eol' in self._index.get_options(version_id)
462
if 'fulltext' == self._index.get_method(version_id):
463
new_content = self.factory.parse_fulltext(data, version_idx)
464
if parent is not None:
465
reference_content = self._get_content(parent)
466
old_texts = reference_content.text()
469
new_texts = new_content.text()
470
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
471
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
473
delta = self.factory.parse_line_delta(data, version_idx)
474
return parent, sha1, noeol, delta
476
def get_graph_with_ghosts(self):
477
"""See VersionedFile.get_graph_with_ghosts()."""
478
graph_items = self._index.get_graph()
479
return dict(graph_items)
481
def get_sha1(self, version_id):
482
"""See VersionedFile.get_sha1()."""
483
record_map = self._get_record_map([version_id])
484
method, content, digest, next = record_map[version_id]
489
"""See VersionedFile.get_suffixes()."""
490
return [DATA_SUFFIX, INDEX_SUFFIX]
492
def has_ghost(self, version_id):
493
"""True if there is a ghost reference in the file to version_id."""
495
if self.has_version(version_id):
497
# optimisable if needed by memoising the _ghosts set.
498
items = self._index.get_graph()
499
for node, parents in items:
500
for parent in parents:
501
if parent not in self._index._cache:
502
if parent == version_id:
507
"""See VersionedFile.versions."""
508
return self._index.get_versions()
510
def has_version(self, version_id):
511
"""See VersionedFile.has_version."""
512
return self._index.has_version(version_id)
514
__contains__ = has_version
516
def _merge_annotations(self, content, parents, parent_texts={},
517
delta=None, annotated=None):
518
"""Merge annotations for content. This is done by comparing
519
the annotations based on changed to the text.
523
for parent_id in parents:
524
merge_content = self._get_content(parent_id, parent_texts)
525
seq = KnitSequenceMatcher(None, merge_content.text(), content.text())
526
if delta_seq is None:
527
# setup a delta seq to reuse.
529
for i, j, n in seq.get_matching_blocks():
532
# this appears to copy (origin, text) pairs across to the new
533
# content for any line that matches the last-checked parent.
534
# FIXME: save the sequence control data for delta compression
535
# against the most relevant parent rather than rediffing.
536
content._lines[j:j+n] = merge_content._lines[i:i+n]
539
reference_content = self._get_content(parents[0], parent_texts)
540
new_texts = content.text()
541
old_texts = reference_content.text()
542
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
543
return self._make_line_delta(delta_seq, content)
545
def _make_line_delta(self, delta_seq, new_content):
546
"""Generate a line delta from delta_seq and new_content."""
548
for op in delta_seq.get_opcodes():
551
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
554
def _get_components_positions(self, version_ids):
555
"""Produce a map of position data for the components of versions.
557
This data is intended to be used for retrieving the knit records.
559
A dict of version_id to (method, data_pos, data_size, next) is
561
method is the way referenced data should be applied.
562
data_pos is the position of the data in the knit.
563
data_size is the size of the data in the knit.
564
next is the build-parent of the version, or None for fulltexts.
567
for version_id in version_ids:
570
while cursor is not None and cursor not in component_data:
571
method = self._index.get_method(cursor)
572
if method == 'fulltext':
575
next = self.get_parents(cursor)[0]
576
data_pos, data_size = self._index.get_position(cursor)
577
component_data[cursor] = (method, data_pos, data_size, next)
579
return component_data
581
def _get_content(self, version_id, parent_texts={}):
582
"""Returns a content object that makes up the specified
584
if not self.has_version(version_id):
585
raise RevisionNotPresent(version_id, self.filename)
587
cached_version = parent_texts.get(version_id, None)
588
if cached_version is not None:
589
return cached_version
591
text_map, contents_map = self._get_content_maps([version_id])
592
return contents_map[version_id]
594
def _check_versions_present(self, version_ids):
595
"""Check that all specified versions are present."""
596
version_ids = set(version_ids)
597
for r in list(version_ids):
598
if self._index.has_version(r):
599
version_ids.remove(r)
601
raise RevisionNotPresent(list(version_ids)[0], self.filename)
603
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
604
"""See VersionedFile.add_lines_with_ghosts()."""
605
self._check_add(version_id, lines)
606
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
608
def _add_lines(self, version_id, parents, lines, parent_texts):
609
"""See VersionedFile.add_lines."""
610
self._check_add(version_id, lines)
611
self._check_versions_present(parents)
612
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
614
def _check_add(self, version_id, lines):
615
"""check that version_id and lines are safe to add."""
616
assert self.writable, "knit is not opened for write"
617
### FIXME escape. RBC 20060228
618
if contains_whitespace(version_id):
619
raise InvalidRevisionId(version_id, self.filename)
620
if self.has_version(version_id):
621
raise RevisionAlreadyPresent(version_id, self.filename)
622
self._check_lines_not_unicode(lines)
623
self._check_lines_are_lines(lines)
625
def _add(self, version_id, lines, parents, delta, parent_texts):
626
"""Add a set of lines on top of version specified by parents.
628
If delta is true, compress the text as a line-delta against
631
Any versions not present will be converted into ghosts.
633
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
634
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
635
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
636
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
637
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
638
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
639
# +1383 0 8.0370 8.0370 +<len>
640
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
641
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
642
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
643
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
647
if parent_texts is None:
649
for parent in parents:
650
if not self.has_version(parent):
651
ghosts.append(parent)
653
present_parents.append(parent)
655
if delta and not len(present_parents):
658
digest = sha_strings(lines)
661
if lines[-1][-1] != '\n':
662
options.append('no-eol')
663
lines[-1] = lines[-1] + '\n'
665
if len(present_parents) and delta:
666
# To speed the extract of texts the delta chain is limited
667
# to a fixed number of deltas. This should minimize both
668
# I/O and the time spend applying deltas.
670
delta_parents = present_parents
672
parent = delta_parents[0]
673
method = self._index.get_method(parent)
674
if method == 'fulltext':
676
delta_parents = self._index.get_parents(parent)
678
if method == 'line-delta':
681
lines = self.factory.make(lines, version_id)
682
if delta or (self.factory.annotated and len(present_parents) > 0):
683
# Merge annotations from parent texts if so is needed.
684
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
685
delta, self.factory.annotated)
688
options.append('line-delta')
689
store_lines = self.factory.lower_line_delta(delta_hunks)
691
options.append('fulltext')
692
store_lines = self.factory.lower_fulltext(lines)
694
where, size = self._data.add_record(version_id, digest, store_lines)
695
self._index.add_version(version_id, options, where, size, parents)
698
def check(self, progress_bar=None):
699
"""See VersionedFile.check()."""
701
def _clone_text(self, new_version_id, old_version_id, parents):
702
"""See VersionedFile.clone_text()."""
703
# FIXME RBC 20060228 make fast by only inserting an index with null
705
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
707
def get_lines(self, version_id):
708
"""See VersionedFile.get_lines()."""
709
return self.get_line_list([version_id])[0]
711
def _get_record_map(self, version_ids):
712
"""Produce a dictionary of knit records.
714
The keys are version_ids, the values are tuples of (method, content,
716
method is the way the content should be applied.
717
content is a KnitContent object.
718
digest is the SHA1 digest of this version id after all steps are done
719
next is the build-parent of the version, i.e. the leftmost ancestor.
720
If the method is fulltext, next will be None.
722
position_map = self._get_components_positions(version_ids)
723
# c = component_id, m = method, p = position, s = size, n = next
724
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
726
for component_id, content, digest in \
727
self._data.read_records_iter(records):
728
method, position, size, next = position_map[component_id]
729
record_map[component_id] = method, content, digest, next
733
def get_text(self, version_id):
734
"""See VersionedFile.get_text"""
735
return self.get_texts([version_id])[0]
737
def get_texts(self, version_ids):
738
return [''.join(l) for l in self.get_line_list(version_ids)]
740
def get_line_list(self, version_ids):
741
"""Return the texts of listed versions as a list of strings."""
742
text_map, content_map = self._get_content_maps(version_ids)
743
return [text_map[v] for v in version_ids]
745
def _get_content_maps(self, version_ids):
746
"""Produce maps of text and KnitContents
748
:return: (text_map, content_map) where text_map contains the texts for
749
the requested versions and content_map contains the KnitContents.
750
Both dicts take version_ids as their keys.
752
for version_id in version_ids:
753
if not self.has_version(version_id):
754
raise RevisionNotPresent(version_id, self.filename)
755
record_map = self._get_record_map(version_ids)
760
for version_id in version_ids:
763
while cursor is not None:
764
method, data, digest, next = record_map[cursor]
765
components.append((cursor, method, data, digest))
766
if cursor in content_map:
771
for component_id, method, data, digest in reversed(components):
772
if component_id in content_map:
773
content = content_map[component_id]
775
version_idx = self._index.lookup(component_id)
776
if method == 'fulltext':
777
assert content is None
778
content = self.factory.parse_fulltext(data, version_idx)
779
elif method == 'line-delta':
780
delta = self.factory.parse_line_delta(data[:],
782
content = content.copy()
783
content._lines = self._apply_delta(content._lines,
785
content_map[component_id] = content
787
if 'no-eol' in self._index.get_options(version_id):
788
content = content.copy()
789
line = content._lines[-1][1].rstrip('\n')
790
content._lines[-1] = (content._lines[-1][0], line)
791
final_content[version_id] = content
793
# digest here is the digest from the last applied component.
794
text = content.text()
795
if sha_strings(text) != digest:
796
raise KnitCorrupt(self.filename,
797
'sha-1 does not match %s' % version_id)
799
text_map[version_id] = text
800
return text_map, final_content
802
def iter_lines_added_or_present_in_versions(self, version_ids=None,
804
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
805
if version_ids is None:
806
version_ids = self.versions()
808
pb = progress.DummyProgress()
809
# we don't care about inclusions, the caller cares.
810
# but we need to setup a list of records to visit.
811
# we need version_id, position, length
812
version_id_records = []
813
requested_versions = list(version_ids)
814
# filter for available versions
815
for version_id in requested_versions:
816
if not self.has_version(version_id):
817
raise RevisionNotPresent(version_id, self.filename)
818
# get a in-component-order queue:
820
for version_id in self.versions():
821
if version_id in requested_versions:
822
version_ids.append(version_id)
823
data_pos, length = self._index.get_position(version_id)
824
version_id_records.append((version_id, data_pos, length))
827
total = len(version_id_records)
828
pb.update('Walking content.', count, total)
829
for version_id, data, sha_value in \
830
self._data.read_records_iter(version_id_records):
831
pb.update('Walking content.', count, total)
832
method = self._index.get_method(version_id)
833
version_idx = self._index.lookup(version_id)
834
assert method in ('fulltext', 'line-delta')
835
if method == 'fulltext':
836
content = self.factory.parse_fulltext(data, version_idx)
837
for line in content.text():
840
delta = self.factory.parse_line_delta(data, version_idx)
841
for start, end, count, lines in delta:
842
for origin, line in lines:
845
pb.update('Walking content.', total, total)
847
def num_versions(self):
848
"""See VersionedFile.num_versions()."""
849
return self._index.num_versions()
851
__len__ = num_versions
853
def annotate_iter(self, version_id):
854
"""See VersionedFile.annotate_iter."""
855
content = self._get_content(version_id)
856
for origin, text in content.annotate_iter():
859
def get_parents(self, version_id):
860
"""See VersionedFile.get_parents."""
863
# 52554 calls in 1264 872 internal down from 3674
865
return self._index.get_parents(version_id)
867
raise RevisionNotPresent(version_id, self.filename)
869
def get_parents_with_ghosts(self, version_id):
870
"""See VersionedFile.get_parents."""
872
return self._index.get_parents_with_ghosts(version_id)
874
raise RevisionNotPresent(version_id, self.filename)
876
def get_ancestry(self, versions):
877
"""See VersionedFile.get_ancestry."""
878
if isinstance(versions, basestring):
879
versions = [versions]
882
self._check_versions_present(versions)
883
return self._index.get_ancestry(versions)
885
def get_ancestry_with_ghosts(self, versions):
886
"""See VersionedFile.get_ancestry_with_ghosts."""
887
if isinstance(versions, basestring):
888
versions = [versions]
891
self._check_versions_present(versions)
892
return self._index.get_ancestry_with_ghosts(versions)
894
#@deprecated_method(zero_eight)
895
def walk(self, version_ids):
896
"""See VersionedFile.walk."""
897
# We take the short path here, and extract all relevant texts
898
# and put them in a weave and let that do all the work. Far
899
# from optimal, but is much simpler.
900
# FIXME RB 20060228 this really is inefficient!
901
from bzrlib.weave import Weave
903
w = Weave(self.filename)
904
ancestry = self.get_ancestry(version_ids)
905
sorted_graph = topo_sort(self._index.get_graph())
906
version_list = [vid for vid in sorted_graph if vid in ancestry]
908
for version_id in version_list:
909
lines = self.get_lines(version_id)
910
w.add_lines(version_id, self.get_parents(version_id), lines)
912
for lineno, insert_id, dset, line in w.walk(version_ids):
913
yield lineno, insert_id, dset, line
915
def plan_merge(self, ver_a, ver_b):
916
"""See VersionedFile.plan_merge."""
917
ancestors_b = set(self.get_ancestry(ver_b))
918
def status_a(revision, text):
919
if revision in ancestors_b:
920
return 'killed-b', text
924
ancestors_a = set(self.get_ancestry(ver_a))
925
def status_b(revision, text):
926
if revision in ancestors_a:
927
return 'killed-a', text
931
annotated_a = self.annotate(ver_a)
932
annotated_b = self.annotate(ver_b)
933
plain_a = [t for (a, t) in annotated_a]
934
plain_b = [t for (a, t) in annotated_b]
935
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
938
for ai, bi, l in blocks:
939
# process all mismatched sections
940
# (last mismatched section is handled because blocks always
941
# includes a 0-length last block)
942
for revision, text in annotated_a[a_cur:ai]:
943
yield status_a(revision, text)
944
for revision, text in annotated_b[b_cur:bi]:
945
yield status_b(revision, text)
947
# and now the matched section
950
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
951
assert text_a == text_b
952
yield "unchanged", text_a
955
class _KnitComponentFile(object):
956
"""One of the files used to implement a knit database"""
958
def __init__(self, transport, filename, mode, file_mode=None,
959
create_parent_dir=False, dir_mode=None):
960
self._transport = transport
961
self._filename = filename
963
self._file_mode = file_mode
964
self._dir_mode = dir_mode
965
self._create_parent_dir = create_parent_dir
966
self._need_to_create = False
968
def check_header(self, fp):
970
if line != self.HEADER:
971
raise KnitHeaderError(badline=line)
974
"""Commit is a nop."""
977
return '%s(%s)' % (self.__class__.__name__, self._filename)
980
class _KnitIndex(_KnitComponentFile):
981
"""Manages knit index file.
983
The index is already kept in memory and read on startup, to enable
984
fast lookups of revision information. The cursor of the index
985
file is always pointing to the end, making it easy to append
988
_cache is a cache for fast mapping from version id to a Index
991
_history is a cache for fast mapping from indexes to version ids.
993
The index data format is dictionary compressed when it comes to
994
parent references; a index entry may only have parents that with a
995
lover index number. As a result, the index is topological sorted.
997
Duplicate entries may be written to the index for a single version id
998
if this is done then the latter one completely replaces the former:
999
this allows updates to correct version and parent information.
1000
Note that the two entries may share the delta, and that successive
1001
annotations and references MUST point to the first entry.
1003
The index file on disc contains a header, followed by one line per knit
1004
record. The same revision can be present in an index file more than once.
1005
The first occurrence gets assigned a sequence number starting from 0.
1007
The format of a single line is
1008
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1009
REVISION_ID is a utf8-encoded revision id
1010
FLAGS is a comma separated list of flags about the record. Values include
1011
no-eol, line-delta, fulltext.
1012
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1013
that the the compressed data starts at.
1014
LENGTH is the ascii representation of the length of the data file.
1015
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1017
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1018
revision id already in the knit that is a parent of REVISION_ID.
1019
The ' :' marker is the end of record marker.
1022
when a write is interrupted to the index file, it will result in a line that
1023
does not end in ' :'. If the ' :' is not present at the end of a line, or at
1024
the end of the file, then the record that is missing it will be ignored by
1027
When writing new records to the index file, the data is preceded by '\n'
1028
to ensure that records always start on new lines even if the last write was
1029
interrupted. As a result its normal for the last line in the index to be
1030
missing a trailing newline. One can be added with no harmful effects.
1033
HEADER = "# bzr knit index 8\n"
1035
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1036
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1038
def _cache_version(self, version_id, options, pos, size, parents):
1039
"""Cache a version record in the history array and index cache.
1041
This is inlined into __init__ for performance. KEEP IN SYNC.
1042
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1045
# only want the _history index to reference the 1st index entry
1047
if version_id not in self._cache:
1048
index = len(self._history)
1049
self._history.append(version_id)
1051
index = self._cache[version_id][5]
1052
self._cache[version_id] = (version_id,
1059
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1060
create_parent_dir=False, delay_create=False, dir_mode=None):
1061
_KnitComponentFile.__init__(self, transport, filename, mode,
1062
file_mode=file_mode,
1063
create_parent_dir=create_parent_dir,
1066
# position in _history is the 'official' index for a revision
1067
# but the values may have come from a newer entry.
1068
# so - wc -l of a knit index is != the number of unique names
1071
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1076
pb.update('read knit index', count, total)
1077
fp = self._transport.get(self._filename)
1079
self.check_header(fp)
1080
# readlines reads the whole file at once:
1081
# bad for transports like http, good for local disk
1082
# we save 60 ms doing this one change (
1083
# from calling readline each time to calling
1085
# probably what we want for nice behaviour on
1086
# http is a incremental readlines that yields, or
1087
# a check for local vs non local indexes,
1088
for l in fp.readlines():
1090
if len(rec) < 5 or rec[-1] != ':':
1092
# FIXME: in the future we should determine if its a
1093
# short write - and ignore it
1094
# or a different failure, and raise. RBC 20060407
1098
#pb.update('read knit index', count, total)
1099
# See self._parse_parents
1101
for value in rec[4:-1]:
1103
# uncompressed reference
1104
parents.append(value[1:])
1106
# this is 15/4000ms faster than isinstance,
1108
# this function is called thousands of times a
1109
# second so small variations add up.
1110
assert value.__class__ is str
1111
parents.append(self._history[int(value)])
1112
# end self._parse_parents
1113
# self._cache_version(rec[0],
1114
# rec[1].split(','),
1118
# --- self._cache_version
1119
# only want the _history index to reference the 1st
1120
# index entry for version_id
1122
if version_id not in self._cache:
1123
index = len(self._history)
1124
self._history.append(version_id)
1126
index = self._cache[version_id][5]
1127
self._cache[version_id] = (version_id,
1133
# --- self._cache_version
1136
except NoSuchFile, e:
1137
if mode != 'w' or not create:
1140
self._need_to_create = True
1142
self._transport.put_bytes_non_atomic(self._filename,
1143
self.HEADER, mode=self._file_mode)
1146
pb.update('read knit index', total, total)
1149
def _parse_parents(self, compressed_parents):
1150
"""convert a list of string parent values into version ids.
1152
ints are looked up in the index.
1153
.FOO values are ghosts and converted in to FOO.
1155
NOTE: the function is retained here for clarity, and for possible
1156
use in partial index reads. However bulk processing now has
1157
it inlined in __init__ for inner-loop optimisation.
1160
for value in compressed_parents:
1161
if value[-1] == '.':
1162
# uncompressed reference
1163
result.append(value[1:])
1165
# this is 15/4000ms faster than isinstance,
1166
# this function is called thousands of times a
1167
# second so small variations add up.
1168
assert value.__class__ is str
1169
result.append(self._history[int(value)])
1172
def get_graph(self):
1174
for version_id, index in self._cache.iteritems():
1175
graph.append((version_id, index[4]))
1178
def get_ancestry(self, versions):
1179
"""See VersionedFile.get_ancestry."""
1180
# get a graph of all the mentioned versions:
1182
pending = set(versions)
1184
version = pending.pop()
1185
parents = self._cache[version][4]
1186
# got the parents ok
1188
parents = [parent for parent in parents if parent in self._cache]
1189
for parent in parents:
1190
# if not completed and not a ghost
1191
if parent not in graph:
1193
graph[version] = parents
1194
return topo_sort(graph.items())
1196
def get_ancestry_with_ghosts(self, versions):
1197
"""See VersionedFile.get_ancestry_with_ghosts."""
1198
# get a graph of all the mentioned versions:
1200
pending = set(versions)
1202
version = pending.pop()
1204
parents = self._cache[version][4]
1210
# got the parents ok
1211
for parent in parents:
1212
if parent not in graph:
1214
graph[version] = parents
1215
return topo_sort(graph.items())
1217
def num_versions(self):
1218
return len(self._history)
1220
__len__ = num_versions
1222
def get_versions(self):
1223
return self._history
1225
def idx_to_name(self, idx):
1226
return self._history[idx]
1228
def lookup(self, version_id):
1229
assert version_id in self._cache
1230
return self._cache[version_id][5]
1232
def _version_list_to_index(self, versions):
1233
encode_utf8 = cache_utf8.encode
1235
for version in versions:
1236
if version in self._cache:
1237
# -- inlined lookup() --
1238
result_list.append(str(self._cache[version][5]))
1239
# -- end lookup () --
1241
result_list.append('.' + encode_utf8(version))
1242
return ' '.join(result_list)
1244
def add_version(self, version_id, options, pos, size, parents):
1245
"""Add a version record to the index."""
1246
self.add_versions(((version_id, options, pos, size, parents),))
1248
def add_versions(self, versions):
1249
"""Add multiple versions to the index.
1251
:param versions: a list of tuples:
1252
(version_id, options, pos, size, parents).
1255
encode_utf8 = cache_utf8.encode
1256
for version_id, options, pos, size, parents in versions:
1257
line = "\n%s %s %s %s %s :" % (encode_utf8(version_id),
1261
self._version_list_to_index(parents))
1262
assert isinstance(line, str), \
1263
'content must be utf-8 encoded: %r' % (line,)
1265
if not self._need_to_create:
1266
self._transport.append_bytes(self._filename, ''.join(lines))
1269
sio.write(self.HEADER)
1270
sio.writelines(lines)
1272
self._transport.put_file_non_atomic(self._filename, sio,
1273
create_parent_dir=self._create_parent_dir,
1274
mode=self._file_mode,
1275
dir_mode=self._dir_mode)
1276
self._need_to_create = False
1278
# cache after writing, so that a failed write leads to missing cache
1279
# entries not extra ones. XXX TODO: RBC 20060502 in the event of a
1280
# failure, reload the index or flush it or some such, to prevent
1281
# writing records that did complete twice.
1282
for version_id, options, pos, size, parents in versions:
1283
self._cache_version(version_id, options, pos, size, parents)
1285
def has_version(self, version_id):
1286
"""True if the version is in the index."""
1287
return (version_id in self._cache)
1289
def get_position(self, version_id):
1290
"""Return data position and size of specified version."""
1291
return (self._cache[version_id][2], \
1292
self._cache[version_id][3])
1294
def get_method(self, version_id):
1295
"""Return compression method of specified version."""
1296
options = self._cache[version_id][1]
1297
if 'fulltext' in options:
1300
assert 'line-delta' in options
1303
def get_options(self, version_id):
1304
return self._cache[version_id][1]
1306
def get_parents(self, version_id):
1307
"""Return parents of specified version ignoring ghosts."""
1308
return [parent for parent in self._cache[version_id][4]
1309
if parent in self._cache]
1311
def get_parents_with_ghosts(self, version_id):
1312
"""Return parents of specified version with ghosts."""
1313
return self._cache[version_id][4]
1315
def check_versions_present(self, version_ids):
1316
"""Check that all specified versions are present."""
1317
version_ids = set(version_ids)
1318
for version_id in list(version_ids):
1319
if version_id in self._cache:
1320
version_ids.remove(version_id)
1322
raise RevisionNotPresent(list(version_ids)[0], self.filename)
1325
class _KnitData(_KnitComponentFile):
1326
"""Contents of the knit data file"""
1328
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1329
create_parent_dir=False, delay_create=False,
1331
_KnitComponentFile.__init__(self, transport, filename, mode,
1332
file_mode=file_mode,
1333
create_parent_dir=create_parent_dir,
1335
self._checked = False
1336
# TODO: jam 20060713 conceptually, this could spill to disk
1337
# if the cached size gets larger than a certain amount
1338
# but it complicates the model a bit, so for now just use
1339
# a simple dictionary
1341
self._do_cache = False
1344
self._need_to_create = create
1346
self._transport.put_bytes_non_atomic(self._filename, '',
1347
mode=self._file_mode)
1349
def enable_cache(self):
1350
"""Enable caching of reads."""
1351
self._do_cache = True
1353
def clear_cache(self):
1354
"""Clear the record cache."""
1355
self._do_cache = False
1358
def _open_file(self):
1360
return self._transport.get(self._filename)
1365
def _record_to_data(self, version_id, digest, lines):
1366
"""Convert version_id, digest, lines into a raw data block.
1368
:return: (len, a StringIO instance with the raw data ready to read.)
1371
data_file = GzipFile(None, mode='wb', fileobj=sio)
1373
version_id_utf8 = cache_utf8.encode(version_id)
1374
data_file.writelines(chain(
1375
["version %s %d %s\n" % (version_id_utf8,
1379
["end %s\n" % version_id_utf8]))
1386
def add_raw_record(self, raw_data):
1387
"""Append a prepared record to the data file.
1389
:return: the offset in the data file raw_data was written.
1391
assert isinstance(raw_data, str), 'data must be plain bytes'
1392
if not self._need_to_create:
1393
return self._transport.append_bytes(self._filename, raw_data)
1395
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1396
create_parent_dir=self._create_parent_dir,
1397
mode=self._file_mode,
1398
dir_mode=self._dir_mode)
1399
self._need_to_create = False
1402
def add_record(self, version_id, digest, lines):
1403
"""Write new text record to disk. Returns the position in the
1404
file where it was written."""
1405
size, sio = self._record_to_data(version_id, digest, lines)
1407
if not self._need_to_create:
1408
start_pos = self._transport.append_file(self._filename, sio)
1410
self._transport.put_file_non_atomic(self._filename, sio,
1411
create_parent_dir=self._create_parent_dir,
1412
mode=self._file_mode,
1413
dir_mode=self._dir_mode)
1414
self._need_to_create = False
1417
self._cache[version_id] = sio.getvalue()
1418
return start_pos, size
1420
def _parse_record_header(self, version_id, raw_data):
1421
"""Parse a record header for consistency.
1423
:return: the header and the decompressor stream.
1424
as (stream, header_record)
1426
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1427
rec = df.readline().split()
1429
raise KnitCorrupt(self._filename, 'unexpected number of elements in record header')
1430
if cache_utf8.decode(rec[1]) != version_id:
1431
raise KnitCorrupt(self._filename,
1432
'unexpected version, wanted %r, got %r' % (
1433
version_id, rec[1]))
1436
def _parse_record(self, version_id, data):
1438
# 4168 calls in 2880 217 internal
1439
# 4168 calls to _parse_record_header in 2121
1440
# 4168 calls to readlines in 330
1441
df, rec = self._parse_record_header(version_id, data)
1442
record_contents = df.readlines()
1443
l = record_contents.pop()
1444
assert len(record_contents) == int(rec[2])
1445
if l != 'end %s\n' % cache_utf8.encode(version_id):
1446
raise KnitCorrupt(self._filename, 'unexpected version end line %r, wanted %r'
1449
return record_contents, rec[3]
1451
def read_records_iter_raw(self, records):
1452
"""Read text records from data file and yield raw data.
1454
This unpacks enough of the text record to validate the id is
1455
as expected but thats all.
1457
# setup an iterator of the external records:
1458
# uses readv so nice and fast we hope.
1460
# grab the disk data needed.
1462
# Don't check _cache if it is empty
1463
needed_offsets = [(pos, size) for version_id, pos, size
1465
if version_id not in self._cache]
1467
needed_offsets = [(pos, size) for version_id, pos, size
1470
raw_records = self._transport.readv(self._filename, needed_offsets)
1473
for version_id, pos, size in records:
1474
if version_id in self._cache:
1475
# This data has already been validated
1476
data = self._cache[version_id]
1478
pos, data = raw_records.next()
1480
self._cache[version_id] = data
1482
# validate the header
1483
df, rec = self._parse_record_header(version_id, data)
1485
yield version_id, data
1487
def read_records_iter(self, records):
1488
"""Read text records from data file and yield result.
1490
The result will be returned in whatever is the fastest to read.
1491
Not by the order requested. Also, multiple requests for the same
1492
record will only yield 1 response.
1493
:param records: A list of (version_id, pos, len) entries
1494
:return: Yields (version_id, contents, digest) in the order
1495
read, not the order requested
1501
# Skip records we have alread seen
1502
yielded_records = set()
1503
needed_records = set()
1504
for record in records:
1505
if record[0] in self._cache:
1506
if record[0] in yielded_records:
1508
yielded_records.add(record[0])
1509
data = self._cache[record[0]]
1510
content, digest = self._parse_record(record[0], data)
1511
yield (record[0], content, digest)
1513
needed_records.add(record)
1514
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1516
needed_records = sorted(set(records), key=operator.itemgetter(1))
1518
if not needed_records:
1521
# The transport optimizes the fetching as well
1522
# (ie, reads continuous ranges.)
1523
readv_response = self._transport.readv(self._filename,
1524
[(pos, size) for version_id, pos, size in needed_records])
1526
for (version_id, pos, size), (pos, data) in \
1527
izip(iter(needed_records), readv_response):
1528
content, digest = self._parse_record(version_id, data)
1530
self._cache[version_id] = data
1531
yield version_id, content, digest
1533
def read_records(self, records):
1534
"""Read records into a dictionary."""
1536
for record_id, content, digest in \
1537
self.read_records_iter(records):
1538
components[record_id] = (content, digest)
1542
class InterKnit(InterVersionedFile):
1543
"""Optimised code paths for knit to knit operations."""
1545
_matching_file_from_factory = KnitVersionedFile
1546
_matching_file_to_factory = KnitVersionedFile
1549
def is_compatible(source, target):
1550
"""Be compatible with knits. """
1552
return (isinstance(source, KnitVersionedFile) and
1553
isinstance(target, KnitVersionedFile))
1554
except AttributeError:
1557
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1558
"""See InterVersionedFile.join."""
1559
assert isinstance(self.source, KnitVersionedFile)
1560
assert isinstance(self.target, KnitVersionedFile)
1562
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1567
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1569
version_ids = list(version_ids)
1570
if None in version_ids:
1571
version_ids.remove(None)
1573
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1574
this_versions = set(self.target._index.get_versions())
1575
needed_versions = self.source_ancestry - this_versions
1576
cross_check_versions = self.source_ancestry.intersection(this_versions)
1577
mismatched_versions = set()
1578
for version in cross_check_versions:
1579
# scan to include needed parents.
1580
n1 = set(self.target.get_parents_with_ghosts(version))
1581
n2 = set(self.source.get_parents_with_ghosts(version))
1583
# FIXME TEST this check for cycles being introduced works
1584
# the logic is we have a cycle if in our graph we are an
1585
# ancestor of any of the n2 revisions.
1591
parent_ancestors = self.source.get_ancestry(parent)
1592
if version in parent_ancestors:
1593
raise errors.GraphCycleError([parent, version])
1594
# ensure this parent will be available later.
1595
new_parents = n2.difference(n1)
1596
needed_versions.update(new_parents.difference(this_versions))
1597
mismatched_versions.add(version)
1599
if not needed_versions and not mismatched_versions:
1601
full_list = topo_sort(self.source.get_graph())
1603
version_list = [i for i in full_list if (not self.target.has_version(i)
1604
and i in needed_versions)]
1608
copy_queue_records = []
1610
for version_id in version_list:
1611
options = self.source._index.get_options(version_id)
1612
parents = self.source._index.get_parents_with_ghosts(version_id)
1613
# check that its will be a consistent copy:
1614
for parent in parents:
1615
# if source has the parent, we must :
1616
# * already have it or
1617
# * have it scheduled already
1618
# otherwise we don't care
1619
assert (self.target.has_version(parent) or
1620
parent in copy_set or
1621
not self.source.has_version(parent))
1622
data_pos, data_size = self.source._index.get_position(version_id)
1623
copy_queue_records.append((version_id, data_pos, data_size))
1624
copy_queue.append((version_id, options, parents))
1625
copy_set.add(version_id)
1627
# data suck the join:
1629
total = len(version_list)
1632
for (version_id, raw_data), \
1633
(version_id2, options, parents) in \
1634
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1636
assert version_id == version_id2, 'logic error, inconsistent results'
1638
pb.update("Joining knit", count, total)
1639
raw_records.append((version_id, options, parents, len(raw_data)))
1640
raw_datum.append(raw_data)
1641
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1643
for version in mismatched_versions:
1644
# FIXME RBC 20060309 is this needed?
1645
n1 = set(self.target.get_parents_with_ghosts(version))
1646
n2 = set(self.source.get_parents_with_ghosts(version))
1647
# write a combined record to our history preserving the current
1648
# parents as first in the list
1649
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1650
self.target.fix_parents(version, new_parents)
1656
InterVersionedFile.register_optimiser(InterKnit)
1659
class WeaveToKnit(InterVersionedFile):
1660
"""Optimised code paths for weave to knit operations."""
1662
_matching_file_from_factory = bzrlib.weave.WeaveFile
1663
_matching_file_to_factory = KnitVersionedFile
1666
def is_compatible(source, target):
1667
"""Be compatible with weaves to knits."""
1669
return (isinstance(source, bzrlib.weave.Weave) and
1670
isinstance(target, KnitVersionedFile))
1671
except AttributeError:
1674
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1675
"""See InterVersionedFile.join."""
1676
assert isinstance(self.source, bzrlib.weave.Weave)
1677
assert isinstance(self.target, KnitVersionedFile)
1679
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1684
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1686
version_ids = list(version_ids)
1688
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1689
this_versions = set(self.target._index.get_versions())
1690
needed_versions = self.source_ancestry - this_versions
1691
cross_check_versions = self.source_ancestry.intersection(this_versions)
1692
mismatched_versions = set()
1693
for version in cross_check_versions:
1694
# scan to include needed parents.
1695
n1 = set(self.target.get_parents_with_ghosts(version))
1696
n2 = set(self.source.get_parents(version))
1697
# if all of n2's parents are in n1, then its fine.
1698
if n2.difference(n1):
1699
# FIXME TEST this check for cycles being introduced works
1700
# the logic is we have a cycle if in our graph we are an
1701
# ancestor of any of the n2 revisions.
1707
parent_ancestors = self.source.get_ancestry(parent)
1708
if version in parent_ancestors:
1709
raise errors.GraphCycleError([parent, version])
1710
# ensure this parent will be available later.
1711
new_parents = n2.difference(n1)
1712
needed_versions.update(new_parents.difference(this_versions))
1713
mismatched_versions.add(version)
1715
if not needed_versions and not mismatched_versions:
1717
full_list = topo_sort(self.source.get_graph())
1719
version_list = [i for i in full_list if (not self.target.has_version(i)
1720
and i in needed_versions)]
1724
total = len(version_list)
1725
for version_id in version_list:
1726
pb.update("Converting to knit", count, total)
1727
parents = self.source.get_parents(version_id)
1728
# check that its will be a consistent copy:
1729
for parent in parents:
1730
# if source has the parent, we must already have it
1731
assert (self.target.has_version(parent))
1732
self.target.add_lines(
1733
version_id, parents, self.source.get_lines(version_id))
1736
for version in mismatched_versions:
1737
# FIXME RBC 20060309 is this needed?
1738
n1 = set(self.target.get_parents_with_ghosts(version))
1739
n2 = set(self.source.get_parents(version))
1740
# write a combined record to our history preserving the current
1741
# parents as first in the list
1742
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1743
self.target.fix_parents(version, new_parents)
1749
InterVersionedFile.register_optimiser(WeaveToKnit)
1752
class KnitSequenceMatcher(difflib.SequenceMatcher):
1753
"""Knit tuned sequence matcher.
1755
This is based on profiling of difflib which indicated some improvements
1756
for our usage pattern.
1759
def find_longest_match(self, alo, ahi, blo, bhi):
1760
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1762
If isjunk is not defined:
1764
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1765
alo <= i <= i+k <= ahi
1766
blo <= j <= j+k <= bhi
1767
and for all (i',j',k') meeting those conditions,
1770
and if i == i', j <= j'
1772
In other words, of all maximal matching blocks, return one that
1773
starts earliest in a, and of all those maximal matching blocks that
1774
start earliest in a, return the one that starts earliest in b.
1776
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1777
>>> s.find_longest_match(0, 5, 0, 9)
1780
If isjunk is defined, first the longest matching block is
1781
determined as above, but with the additional restriction that no
1782
junk element appears in the block. Then that block is extended as
1783
far as possible by matching (only) junk elements on both sides. So
1784
the resulting block never matches on junk except as identical junk
1785
happens to be adjacent to an "interesting" match.
1787
Here's the same example as before, but considering blanks to be
1788
junk. That prevents " abcd" from matching the " abcd" at the tail
1789
end of the second sequence directly. Instead only the "abcd" can
1790
match, and matches the leftmost "abcd" in the second sequence:
1792
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1793
>>> s.find_longest_match(0, 5, 0, 9)
1796
If no blocks match, return (alo, blo, 0).
1798
>>> s = SequenceMatcher(None, "ab", "c")
1799
>>> s.find_longest_match(0, 2, 0, 1)
1803
# CAUTION: stripping common prefix or suffix would be incorrect.
1807
# Longest matching block is "ab", but if common prefix is
1808
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1809
# strip, so ends up claiming that ab is changed to acab by
1810
# inserting "ca" in the middle. That's minimal but unintuitive:
1811
# "it's obvious" that someone inserted "ac" at the front.
1812
# Windiff ends up at the same place as diff, but by pairing up
1813
# the unique 'b's and then matching the first two 'a's.
1815
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1816
besti, bestj, bestsize = alo, blo, 0
1817
# find longest junk-free match
1818
# during an iteration of the loop, j2len[j] = length of longest
1819
# junk-free match ending with a[i-1] and b[j]
1823
for i in xrange(alo, ahi):
1824
# look at all instances of a[i] in b; note that because
1825
# b2j has no junk keys, the loop is skipped if a[i] is junk
1826
j2lenget = j2len.get
1829
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1830
# following improvement
1831
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1832
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1833
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1835
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1836
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1837
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
1849
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1851
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1854
# Extend the best by non-junk elements on each end. In particular,
1855
# "popular" non-junk elements aren't in b2j, which greatly speeds
1856
# the inner loop above, but also means "the best" match so far
1857
# doesn't contain any junk *or* popular non-junk elements.
1858
while besti > alo and bestj > blo and \
1859
not isbjunk(b[bestj-1]) and \
1860
a[besti-1] == b[bestj-1]:
1861
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1862
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1863
not isbjunk(b[bestj+bestsize]) and \
1864
a[besti+bestsize] == b[bestj+bestsize]:
1867
# Now that we have a wholly interesting match (albeit possibly
1868
# empty!), we may as well suck up the matching junk on each
1869
# side of it too. Can't think of a good reason not to, and it
1870
# saves post-processing the (possibly considerable) expense of
1871
# figuring out what to do with it. In the case of an empty
1872
# interesting match, this is clearly the right thing to do,
1873
# because no other kind of match is possible in the regions.
1874
while besti > alo and bestj > blo and \
1875
isbjunk(b[bestj-1]) and \
1876
a[besti-1] == b[bestj-1]:
1877
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1878
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1879
isbjunk(b[bestj+bestsize]) and \
1880
a[besti+bestsize] == b[bestj+bestsize]:
1881
bestsize = bestsize + 1
1883
return besti, bestj, bestsize
added
removed
bzrlib/knit.py
