1
# Copyright (C) 2005, 2006 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
83
from bzrlib.errors import FileExists, NoSuchFile, KnitError, \
84
InvalidRevisionId, KnitCorrupt, KnitHeaderError, \
85
RevisionNotPresent, RevisionAlreadyPresent
86
from bzrlib.tuned_gzip import GzipFile
87
from bzrlib.trace import mutter
88
from bzrlib.osutils import contains_whitespace, contains_linebreaks, \
90
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
91
from bzrlib.tsort import topo_sort
93
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
96
# TODO: Split out code specific to this format into an associated object.
98
# TODO: Can we put in some kind of value to check that the index and data
99
# files belong together?
101
# TODO: accommodate binaries, perhaps by storing a byte count
103
# TODO: function to check whole file
105
# TODO: atomically append data, then measure backwards from the cursor
106
# position after writing to work out where it was located. we may need to
107
# bypass python file buffering.
109
DATA_SUFFIX = '.knit'
110
INDEX_SUFFIX = '.kndx'
113
class KnitContent(object):
114
"""Content of a knit version to which deltas can be applied."""
116
def __init__(self, lines):
119
def annotate_iter(self):
120
"""Yield tuples of (origin, text) for each content line."""
121
for origin, text in self._lines:
125
"""Return a list of (origin, text) tuples."""
126
return list(self.annotate_iter())
128
def line_delta_iter(self, new_lines):
129
"""Generate line-based delta from this content to new_lines."""
130
new_texts = [text for origin, text in new_lines._lines]
131
old_texts = [text for origin, text in self._lines]
132
s = KnitSequenceMatcher(None, old_texts, new_texts)
133
for op in s.get_opcodes():
136
# ofrom oto length data
137
yield (op[1], op[2], op[4]-op[3], new_lines._lines[op[3]:op[4]])
139
def line_delta(self, new_lines):
140
return list(self.line_delta_iter(new_lines))
143
return [text for origin, text in self._lines]
146
return KnitContent(self._lines[:])
149
class _KnitFactory(object):
150
"""Base factory for creating content objects."""
152
def make(self, lines, version):
153
num_lines = len(lines)
154
return KnitContent(zip([version] * num_lines, lines))
157
class KnitAnnotateFactory(_KnitFactory):
158
"""Factory for creating annotated Content objects."""
162
def parse_fulltext(self, content, version):
163
"""Convert fulltext to internal representation
165
fulltext content is of the format
166
revid(utf8) plaintext\n
167
internal representation is of the format:
170
decode_utf8 = cache_utf8.decode
173
origin, text = line.split(' ', 1)
174
lines.append((decode_utf8(origin), text))
175
return KnitContent(lines)
177
def parse_line_delta_iter(self, lines):
178
for result_item in self.parse_line_delta[lines]:
181
def parse_line_delta(self, lines, version):
182
"""Convert a line based delta into internal representation.
184
line delta is in the form of:
185
intstart intend intcount
187
revid(utf8) newline\n
188
internal representation is
189
(start, end, count, [1..count tuples (revid, newline)])
191
decode_utf8 = cache_utf8.decode
195
# walk through the lines parsing.
197
start, end, count = [int(n) for n in header.split(',')]
201
origin, text = next().split(' ', 1)
203
contents.append((decode_utf8(origin), text))
204
result.append((start, end, count, contents))
207
def lower_fulltext(self, content):
208
"""convert a fulltext content record into a serializable form.
210
see parse_fulltext which this inverts.
212
encode_utf8 = cache_utf8.encode
213
return ['%s %s' % (encode_utf8(o), t) for o, t in content._lines]
215
def lower_line_delta(self, delta):
216
"""convert a delta into a serializable form.
218
See parse_line_delta which this inverts.
220
encode_utf8 = cache_utf8.encode
222
for start, end, c, lines in delta:
223
out.append('%d,%d,%d\n' % (start, end, c))
224
out.extend(encode_utf8(origin) + ' ' + text
225
for origin, text in lines)
229
class KnitPlainFactory(_KnitFactory):
230
"""Factory for creating plain Content objects."""
234
def parse_fulltext(self, content, version):
235
"""This parses an unannotated fulltext.
237
Note that this is not a noop - the internal representation
238
has (versionid, line) - its just a constant versionid.
240
return self.make(content, version)
242
def parse_line_delta_iter(self, lines, version):
244
header = lines.pop(0)
245
start, end, c = [int(n) for n in header.split(',')]
246
yield start, end, c, zip([version] * c, lines[:c])
249
def parse_line_delta(self, lines, version):
250
return list(self.parse_line_delta_iter(lines, version))
252
def lower_fulltext(self, content):
253
return content.text()
255
def lower_line_delta(self, delta):
257
for start, end, c, lines in delta:
258
out.append('%d,%d,%d\n' % (start, end, c))
259
out.extend([text for origin, text in lines])
263
def make_empty_knit(transport, relpath):
264
"""Construct a empty knit at the specified location."""
265
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
269
class KnitVersionedFile(VersionedFile):
270
"""Weave-like structure with faster random access.
272
A knit stores a number of texts and a summary of the relationships
273
between them. Texts are identified by a string version-id. Texts
274
are normally stored and retrieved as a series of lines, but can
275
also be passed as single strings.
277
Lines are stored with the trailing newline (if any) included, to
278
avoid special cases for files with no final newline. Lines are
279
composed of 8-bit characters, not unicode. The combination of
280
these approaches should mean any 'binary' file can be safely
281
stored and retrieved.
284
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
285
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
286
create=False, create_parent_dir=False, delay_create=False,
288
"""Construct a knit at location specified by relpath.
290
:param create: If not True, only open an existing knit.
291
:param create_parent_dir: If True, create the parent directory if
292
creating the file fails. (This is used for stores with
293
hash-prefixes that may not exist yet)
294
:param delay_create: The calling code is aware that the knit won't
295
actually be created until the first data is stored.
297
if deprecated_passed(basis_knit):
298
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
299
" deprecated as of bzr 0.9.",
300
DeprecationWarning, stacklevel=2)
301
if access_mode is None:
303
super(KnitVersionedFile, self).__init__(access_mode)
304
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
305
self.transport = transport
306
self.filename = relpath
307
self.factory = factory or KnitAnnotateFactory()
308
self.writable = (access_mode == 'w')
311
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
312
access_mode, create=create, file_mode=file_mode,
313
create_parent_dir=create_parent_dir, delay_create=delay_create,
315
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
316
access_mode, create=create and not len(self), file_mode=file_mode,
317
create_parent_dir=create_parent_dir, delay_create=delay_create,
321
return '%s(%s)' % (self.__class__.__name__,
322
self.transport.abspath(self.filename))
324
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
325
"""See VersionedFile._add_delta()."""
326
self._check_add(version_id, []) # should we check the lines ?
327
self._check_versions_present(parents)
331
for parent in parents:
332
if not self.has_version(parent):
333
ghosts.append(parent)
335
present_parents.append(parent)
337
if delta_parent is None:
338
# reconstitute as full text.
339
assert len(delta) == 1 or len(delta) == 0
341
assert delta[0][0] == 0
342
assert delta[0][1] == 0, delta[0][1]
343
return super(KnitVersionedFile, self)._add_delta(version_id,
354
options.append('no-eol')
356
if delta_parent is not None:
357
# determine the current delta chain length.
358
# To speed the extract of texts the delta chain is limited
359
# to a fixed number of deltas. This should minimize both
360
# I/O and the time spend applying deltas.
362
delta_parents = [delta_parent]
364
parent = delta_parents[0]
365
method = self._index.get_method(parent)
366
if method == 'fulltext':
368
delta_parents = self._index.get_parents(parent)
370
if method == 'line-delta':
371
# did not find a fulltext in the delta limit.
372
# just do a normal insertion.
373
return super(KnitVersionedFile, self)._add_delta(version_id,
380
options.append('line-delta')
381
store_lines = self.factory.lower_line_delta(delta)
383
where, size = self._data.add_record(version_id, digest, store_lines)
384
self._index.add_version(version_id, options, where, size, parents)
386
def _add_raw_records(self, records, data):
387
"""Add all the records 'records' with data pre-joined in 'data'.
389
:param records: A list of tuples(version_id, options, parents, size).
390
:param data: The data for the records. When it is written, the records
391
are adjusted to have pos pointing into data by the sum of
392
the preceding records sizes.
395
pos = self._data.add_raw_record(data)
398
for (version_id, options, parents, size) in records:
399
index_entries.append((version_id, options, pos+offset,
401
if self._data._do_cache:
402
self._data._cache[version_id] = data[offset:offset+size]
404
self._index.add_versions(index_entries)
406
def enable_cache(self):
407
"""Start caching data for this knit"""
408
self._data.enable_cache()
410
def clear_cache(self):
411
"""Clear the data cache only."""
412
self._data.clear_cache()
414
def copy_to(self, name, transport):
415
"""See VersionedFile.copy_to()."""
416
# copy the current index to a temp index to avoid racing with local
418
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
419
self.transport.get(self._index._filename))
421
f = self._data._open_file()
423
transport.put_file(name + DATA_SUFFIX, f)
426
# move the copied index into place
427
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
429
def create_empty(self, name, transport, mode=None):
430
return KnitVersionedFile(name, transport, factory=self.factory,
431
delta=self.delta, create=True)
433
def _fix_parents(self, version, new_parents):
434
"""Fix the parents list for version.
436
This is done by appending a new version to the index
437
with identical data except for the parents list.
438
the parents list must be a superset of the current
441
current_values = self._index._cache[version]
442
assert set(current_values[4]).difference(set(new_parents)) == set()
443
self._index.add_version(version,
449
def get_delta(self, version_id):
450
"""Get a delta for constructing version from some other version."""
451
if not self.has_version(version_id):
452
raise RevisionNotPresent(version_id, self.filename)
454
parents = self.get_parents(version_id)
459
data_pos, data_size = self._index.get_position(version_id)
460
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
461
version_idx = self._index.lookup(version_id)
462
noeol = 'no-eol' in self._index.get_options(version_id)
463
if 'fulltext' == self._index.get_method(version_id):
464
new_content = self.factory.parse_fulltext(data, version_idx)
465
if parent is not None:
466
reference_content = self._get_content(parent)
467
old_texts = reference_content.text()
470
new_texts = new_content.text()
471
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
472
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
474
delta = self.factory.parse_line_delta(data, version_idx)
475
return parent, sha1, noeol, delta
477
def get_graph_with_ghosts(self):
478
"""See VersionedFile.get_graph_with_ghosts()."""
479
graph_items = self._index.get_graph()
480
return dict(graph_items)
482
def get_sha1(self, version_id):
483
"""See VersionedFile.get_sha1()."""
484
record_map = self._get_record_map([version_id])
485
method, content, digest, next = record_map[version_id]
490
"""See VersionedFile.get_suffixes()."""
491
return [DATA_SUFFIX, INDEX_SUFFIX]
493
def has_ghost(self, version_id):
494
"""True if there is a ghost reference in the file to version_id."""
496
if self.has_version(version_id):
498
# optimisable if needed by memoising the _ghosts set.
499
items = self._index.get_graph()
500
for node, parents in items:
501
for parent in parents:
502
if parent not in self._index._cache:
503
if parent == version_id:
508
"""See VersionedFile.versions."""
509
return self._index.get_versions()
511
def has_version(self, version_id):
512
"""See VersionedFile.has_version."""
513
return self._index.has_version(version_id)
515
__contains__ = has_version
517
def _merge_annotations(self, content, parents, parent_texts={},
518
delta=None, annotated=None):
519
"""Merge annotations for content. This is done by comparing
520
the annotations based on changed to the text.
524
for parent_id in parents:
525
merge_content = self._get_content(parent_id, parent_texts)
526
seq = patiencediff.PatienceSequenceMatcher(
527
None, merge_content.text(), content.text())
528
if delta_seq is None:
529
# setup a delta seq to reuse.
531
for i, j, n in seq.get_matching_blocks():
534
# this appears to copy (origin, text) pairs across to the new
535
# content for any line that matches the last-checked parent.
536
# FIXME: save the sequence control data for delta compression
537
# against the most relevant parent rather than rediffing.
538
content._lines[j:j+n] = merge_content._lines[i:i+n]
541
reference_content = self._get_content(parents[0], parent_texts)
542
new_texts = content.text()
543
old_texts = reference_content.text()
544
delta_seq = patiencediff.PatienceSequenceMatcher(
545
None, old_texts, new_texts)
546
return self._make_line_delta(delta_seq, content)
548
def _make_line_delta(self, delta_seq, new_content):
549
"""Generate a line delta from delta_seq and new_content."""
551
for op in delta_seq.get_opcodes():
554
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
557
def _get_components_positions(self, version_ids):
558
"""Produce a map of position data for the components of versions.
560
This data is intended to be used for retrieving the knit records.
562
A dict of version_id to (method, data_pos, data_size, next) is
564
method is the way referenced data should be applied.
565
data_pos is the position of the data in the knit.
566
data_size is the size of the data in the knit.
567
next is the build-parent of the version, or None for fulltexts.
570
for version_id in version_ids:
573
while cursor is not None and cursor not in component_data:
574
method = self._index.get_method(cursor)
575
if method == 'fulltext':
578
next = self.get_parents(cursor)[0]
579
data_pos, data_size = self._index.get_position(cursor)
580
component_data[cursor] = (method, data_pos, data_size, next)
582
return component_data
584
def _get_content(self, version_id, parent_texts={}):
585
"""Returns a content object that makes up the specified
587
if not self.has_version(version_id):
588
raise RevisionNotPresent(version_id, self.filename)
590
cached_version = parent_texts.get(version_id, None)
591
if cached_version is not None:
592
return cached_version
594
text_map, contents_map = self._get_content_maps([version_id])
595
return contents_map[version_id]
597
def _check_versions_present(self, version_ids):
598
"""Check that all specified versions are present."""
599
version_ids = set(version_ids)
600
for r in list(version_ids):
601
if self._index.has_version(r):
602
version_ids.remove(r)
604
raise RevisionNotPresent(list(version_ids)[0], self.filename)
606
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
607
"""See VersionedFile.add_lines_with_ghosts()."""
608
self._check_add(version_id, lines)
609
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
611
def _add_lines(self, version_id, parents, lines, parent_texts):
612
"""See VersionedFile.add_lines."""
613
self._check_add(version_id, lines)
614
self._check_versions_present(parents)
615
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
617
def _check_add(self, version_id, lines):
618
"""check that version_id and lines are safe to add."""
619
assert self.writable, "knit is not opened for write"
620
### FIXME escape. RBC 20060228
621
if contains_whitespace(version_id):
622
raise InvalidRevisionId(version_id, self.filename)
623
if self.has_version(version_id):
624
raise RevisionAlreadyPresent(version_id, self.filename)
625
self._check_lines_not_unicode(lines)
626
self._check_lines_are_lines(lines)
628
def _add(self, version_id, lines, parents, delta, parent_texts):
629
"""Add a set of lines on top of version specified by parents.
631
If delta is true, compress the text as a line-delta against
634
Any versions not present will be converted into ghosts.
636
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
637
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
638
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
639
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
640
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
641
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
642
# +1383 0 8.0370 8.0370 +<len>
643
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
644
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
645
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
646
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
650
if parent_texts is None:
652
for parent in parents:
653
if not self.has_version(parent):
654
ghosts.append(parent)
656
present_parents.append(parent)
658
if delta and not len(present_parents):
661
digest = sha_strings(lines)
664
if lines[-1][-1] != '\n':
665
options.append('no-eol')
666
lines[-1] = lines[-1] + '\n'
668
if len(present_parents) and delta:
669
# To speed the extract of texts the delta chain is limited
670
# to a fixed number of deltas. This should minimize both
671
# I/O and the time spend applying deltas.
673
delta_parents = present_parents
675
parent = delta_parents[0]
676
method = self._index.get_method(parent)
677
if method == 'fulltext':
679
delta_parents = self._index.get_parents(parent)
681
if method == 'line-delta':
684
lines = self.factory.make(lines, version_id)
685
if delta or (self.factory.annotated and len(present_parents) > 0):
686
# Merge annotations from parent texts if so is needed.
687
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
688
delta, self.factory.annotated)
691
options.append('line-delta')
692
store_lines = self.factory.lower_line_delta(delta_hunks)
694
options.append('fulltext')
695
store_lines = self.factory.lower_fulltext(lines)
697
where, size = self._data.add_record(version_id, digest, store_lines)
698
self._index.add_version(version_id, options, where, size, parents)
701
def check(self, progress_bar=None):
702
"""See VersionedFile.check()."""
704
def _clone_text(self, new_version_id, old_version_id, parents):
705
"""See VersionedFile.clone_text()."""
706
# FIXME RBC 20060228 make fast by only inserting an index with null
708
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
710
def get_lines(self, version_id):
711
"""See VersionedFile.get_lines()."""
712
return self.get_line_list([version_id])[0]
714
def _get_record_map(self, version_ids):
715
"""Produce a dictionary of knit records.
717
The keys are version_ids, the values are tuples of (method, content,
719
method is the way the content should be applied.
720
content is a KnitContent object.
721
digest is the SHA1 digest of this version id after all steps are done
722
next is the build-parent of the version, i.e. the leftmost ancestor.
723
If the method is fulltext, next will be None.
725
position_map = self._get_components_positions(version_ids)
726
# c = component_id, m = method, p = position, s = size, n = next
727
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
729
for component_id, content, digest in \
730
self._data.read_records_iter(records):
731
method, position, size, next = position_map[component_id]
732
record_map[component_id] = method, content, digest, next
736
def get_text(self, version_id):
737
"""See VersionedFile.get_text"""
738
return self.get_texts([version_id])[0]
740
def get_texts(self, version_ids):
741
return [''.join(l) for l in self.get_line_list(version_ids)]
743
def get_line_list(self, version_ids):
744
"""Return the texts of listed versions as a list of strings."""
745
text_map, content_map = self._get_content_maps(version_ids)
746
return [text_map[v] for v in version_ids]
748
def _get_content_maps(self, version_ids):
749
"""Produce maps of text and KnitContents
751
:return: (text_map, content_map) where text_map contains the texts for
752
the requested versions and content_map contains the KnitContents.
753
Both dicts take version_ids as their keys.
755
for version_id in version_ids:
756
if not self.has_version(version_id):
757
raise RevisionNotPresent(version_id, self.filename)
758
record_map = self._get_record_map(version_ids)
763
for version_id in version_ids:
766
while cursor is not None:
767
method, data, digest, next = record_map[cursor]
768
components.append((cursor, method, data, digest))
769
if cursor in content_map:
774
for component_id, method, data, digest in reversed(components):
775
if component_id in content_map:
776
content = content_map[component_id]
778
version_idx = self._index.lookup(component_id)
779
if method == 'fulltext':
780
assert content is None
781
content = self.factory.parse_fulltext(data, version_idx)
782
elif method == 'line-delta':
783
delta = self.factory.parse_line_delta(data[:],
785
content = content.copy()
786
content._lines = self._apply_delta(content._lines,
788
content_map[component_id] = content
790
if 'no-eol' in self._index.get_options(version_id):
791
content = content.copy()
792
line = content._lines[-1][1].rstrip('\n')
793
content._lines[-1] = (content._lines[-1][0], line)
794
final_content[version_id] = content
796
# digest here is the digest from the last applied component.
797
text = content.text()
798
if sha_strings(text) != digest:
799
raise KnitCorrupt(self.filename,
800
'sha-1 does not match %s' % version_id)
802
text_map[version_id] = text
803
return text_map, final_content
805
def iter_lines_added_or_present_in_versions(self, version_ids=None,
807
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
808
if version_ids is None:
809
version_ids = self.versions()
811
pb = progress.DummyProgress()
812
# we don't care about inclusions, the caller cares.
813
# but we need to setup a list of records to visit.
814
# we need version_id, position, length
815
version_id_records = []
816
requested_versions = list(version_ids)
817
# filter for available versions
818
for version_id in requested_versions:
819
if not self.has_version(version_id):
820
raise RevisionNotPresent(version_id, self.filename)
821
# get a in-component-order queue:
823
for version_id in self.versions():
824
if version_id in requested_versions:
825
version_ids.append(version_id)
826
data_pos, length = self._index.get_position(version_id)
827
version_id_records.append((version_id, data_pos, length))
830
total = len(version_id_records)
831
pb.update('Walking content.', count, total)
832
for version_id, data, sha_value in \
833
self._data.read_records_iter(version_id_records):
834
pb.update('Walking content.', count, total)
835
method = self._index.get_method(version_id)
836
version_idx = self._index.lookup(version_id)
837
assert method in ('fulltext', 'line-delta')
838
if method == 'fulltext':
839
content = self.factory.parse_fulltext(data, version_idx)
840
for line in content.text():
843
delta = self.factory.parse_line_delta(data, version_idx)
844
for start, end, count, lines in delta:
845
for origin, line in lines:
848
pb.update('Walking content.', total, total)
850
def num_versions(self):
851
"""See VersionedFile.num_versions()."""
852
return self._index.num_versions()
854
__len__ = num_versions
856
def annotate_iter(self, version_id):
857
"""See VersionedFile.annotate_iter."""
858
content = self._get_content(version_id)
859
for origin, text in content.annotate_iter():
862
def get_parents(self, version_id):
863
"""See VersionedFile.get_parents."""
866
# 52554 calls in 1264 872 internal down from 3674
868
return self._index.get_parents(version_id)
870
raise RevisionNotPresent(version_id, self.filename)
872
def get_parents_with_ghosts(self, version_id):
873
"""See VersionedFile.get_parents."""
875
return self._index.get_parents_with_ghosts(version_id)
877
raise RevisionNotPresent(version_id, self.filename)
879
def get_ancestry(self, versions):
880
"""See VersionedFile.get_ancestry."""
881
if isinstance(versions, basestring):
882
versions = [versions]
885
self._check_versions_present(versions)
886
return self._index.get_ancestry(versions)
888
def get_ancestry_with_ghosts(self, versions):
889
"""See VersionedFile.get_ancestry_with_ghosts."""
890
if isinstance(versions, basestring):
891
versions = [versions]
894
self._check_versions_present(versions)
895
return self._index.get_ancestry_with_ghosts(versions)
897
#@deprecated_method(zero_eight)
898
def walk(self, version_ids):
899
"""See VersionedFile.walk."""
900
# We take the short path here, and extract all relevant texts
901
# and put them in a weave and let that do all the work. Far
902
# from optimal, but is much simpler.
903
# FIXME RB 20060228 this really is inefficient!
904
from bzrlib.weave import Weave
906
w = Weave(self.filename)
907
ancestry = self.get_ancestry(version_ids)
908
sorted_graph = topo_sort(self._index.get_graph())
909
version_list = [vid for vid in sorted_graph if vid in ancestry]
911
for version_id in version_list:
912
lines = self.get_lines(version_id)
913
w.add_lines(version_id, self.get_parents(version_id), lines)
915
for lineno, insert_id, dset, line in w.walk(version_ids):
916
yield lineno, insert_id, dset, line
918
def plan_merge(self, ver_a, ver_b):
919
"""See VersionedFile.plan_merge."""
920
ancestors_b = set(self.get_ancestry(ver_b))
921
def status_a(revision, text):
922
if revision in ancestors_b:
923
return 'killed-b', text
927
ancestors_a = set(self.get_ancestry(ver_a))
928
def status_b(revision, text):
929
if revision in ancestors_a:
930
return 'killed-a', text
934
annotated_a = self.annotate(ver_a)
935
annotated_b = self.annotate(ver_b)
936
plain_a = [t for (a, t) in annotated_a]
937
plain_b = [t for (a, t) in annotated_b]
938
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
941
for ai, bi, l in blocks:
942
# process all mismatched sections
943
# (last mismatched section is handled because blocks always
944
# includes a 0-length last block)
945
for revision, text in annotated_a[a_cur:ai]:
946
yield status_a(revision, text)
947
for revision, text in annotated_b[b_cur:bi]:
948
yield status_b(revision, text)
950
# and now the matched section
953
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
954
assert text_a == text_b
955
yield "unchanged", text_a
958
class _KnitComponentFile(object):
959
"""One of the files used to implement a knit database"""
961
def __init__(self, transport, filename, mode, file_mode=None,
962
create_parent_dir=False, dir_mode=None):
963
self._transport = transport
964
self._filename = filename
966
self._file_mode = file_mode
967
self._dir_mode = dir_mode
968
self._create_parent_dir = create_parent_dir
969
self._need_to_create = False
971
def check_header(self, fp):
973
if line != self.HEADER:
974
raise KnitHeaderError(badline=line)
977
"""Commit is a nop."""
980
return '%s(%s)' % (self.__class__.__name__, self._filename)
983
class _KnitIndex(_KnitComponentFile):
984
"""Manages knit index file.
986
The index is already kept in memory and read on startup, to enable
987
fast lookups of revision information. The cursor of the index
988
file is always pointing to the end, making it easy to append
991
_cache is a cache for fast mapping from version id to a Index
994
_history is a cache for fast mapping from indexes to version ids.
996
The index data format is dictionary compressed when it comes to
997
parent references; a index entry may only have parents that with a
998
lover index number. As a result, the index is topological sorted.
1000
Duplicate entries may be written to the index for a single version id
1001
if this is done then the latter one completely replaces the former:
1002
this allows updates to correct version and parent information.
1003
Note that the two entries may share the delta, and that successive
1004
annotations and references MUST point to the first entry.
1006
The index file on disc contains a header, followed by one line per knit
1007
record. The same revision can be present in an index file more than once.
1008
The first occurrence gets assigned a sequence number starting from 0.
1010
The format of a single line is
1011
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1012
REVISION_ID is a utf8-encoded revision id
1013
FLAGS is a comma separated list of flags about the record. Values include
1014
no-eol, line-delta, fulltext.
1015
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1016
that the the compressed data starts at.
1017
LENGTH is the ascii representation of the length of the data file.
1018
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1020
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1021
revision id already in the knit that is a parent of REVISION_ID.
1022
The ' :' marker is the end of record marker.
1025
when a write is interrupted to the index file, it will result in a line that
1026
does not end in ' :'. If the ' :' is not present at the end of a line, or at
1027
the end of the file, then the record that is missing it will be ignored by
1030
When writing new records to the index file, the data is preceded by '\n'
1031
to ensure that records always start on new lines even if the last write was
1032
interrupted. As a result its normal for the last line in the index to be
1033
missing a trailing newline. One can be added with no harmful effects.
1036
HEADER = "# bzr knit index 8\n"
1038
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1039
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1041
def _cache_version(self, version_id, options, pos, size, parents):
1042
"""Cache a version record in the history array and index cache.
1044
This is inlined into __init__ for performance. KEEP IN SYNC.
1045
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1048
# only want the _history index to reference the 1st index entry
1050
if version_id not in self._cache:
1051
index = len(self._history)
1052
self._history.append(version_id)
1054
index = self._cache[version_id][5]
1055
self._cache[version_id] = (version_id,
1062
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1063
create_parent_dir=False, delay_create=False, dir_mode=None):
1064
_KnitComponentFile.__init__(self, transport, filename, mode,
1065
file_mode=file_mode,
1066
create_parent_dir=create_parent_dir,
1069
# position in _history is the 'official' index for a revision
1070
# but the values may have come from a newer entry.
1071
# so - wc -l of a knit index is != the number of unique names
1074
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1079
pb.update('read knit index', count, total)
1080
fp = self._transport.get(self._filename)
1082
self.check_header(fp)
1083
# readlines reads the whole file at once:
1084
# bad for transports like http, good for local disk
1085
# we save 60 ms doing this one change (
1086
# from calling readline each time to calling
1088
# probably what we want for nice behaviour on
1089
# http is a incremental readlines that yields, or
1090
# a check for local vs non local indexes,
1091
for l in fp.readlines():
1093
if len(rec) < 5 or rec[-1] != ':':
1095
# FIXME: in the future we should determine if its a
1096
# short write - and ignore it
1097
# or a different failure, and raise. RBC 20060407
1101
#pb.update('read knit index', count, total)
1102
# See self._parse_parents
1104
for value in rec[4:-1]:
1106
# uncompressed reference
1107
parents.append(value[1:])
1109
# this is 15/4000ms faster than isinstance,
1111
# this function is called thousands of times a
1112
# second so small variations add up.
1113
assert value.__class__ is str
1114
parents.append(self._history[int(value)])
1115
# end self._parse_parents
1116
# self._cache_version(rec[0],
1117
# rec[1].split(','),
1121
# --- self._cache_version
1122
# only want the _history index to reference the 1st
1123
# index entry for version_id
1125
if version_id not in self._cache:
1126
index = len(self._history)
1127
self._history.append(version_id)
1129
index = self._cache[version_id][5]
1130
self._cache[version_id] = (version_id,
1136
# --- self._cache_version
1139
except NoSuchFile, e:
1140
if mode != 'w' or not create:
1143
self._need_to_create = True
1145
self._transport.put_bytes_non_atomic(self._filename,
1146
self.HEADER, mode=self._file_mode)
1149
pb.update('read knit index', total, total)
1152
def _parse_parents(self, compressed_parents):
1153
"""convert a list of string parent values into version ids.
1155
ints are looked up in the index.
1156
.FOO values are ghosts and converted in to FOO.
1158
NOTE: the function is retained here for clarity, and for possible
1159
use in partial index reads. However bulk processing now has
1160
it inlined in __init__ for inner-loop optimisation.
1163
for value in compressed_parents:
1164
if value[-1] == '.':
1165
# uncompressed reference
1166
result.append(value[1:])
1168
# this is 15/4000ms faster than isinstance,
1169
# this function is called thousands of times a
1170
# second so small variations add up.
1171
assert value.__class__ is str
1172
result.append(self._history[int(value)])
1175
def get_graph(self):
1177
for version_id, index in self._cache.iteritems():
1178
graph.append((version_id, index[4]))
1181
def get_ancestry(self, versions):
1182
"""See VersionedFile.get_ancestry."""
1183
# get a graph of all the mentioned versions:
1185
pending = set(versions)
1187
version = pending.pop()
1188
parents = self._cache[version][4]
1189
# got the parents ok
1191
parents = [parent for parent in parents if parent in self._cache]
1192
for parent in parents:
1193
# if not completed and not a ghost
1194
if parent not in graph:
1196
graph[version] = parents
1197
return topo_sort(graph.items())
1199
def get_ancestry_with_ghosts(self, versions):
1200
"""See VersionedFile.get_ancestry_with_ghosts."""
1201
# get a graph of all the mentioned versions:
1203
pending = set(versions)
1205
version = pending.pop()
1207
parents = self._cache[version][4]
1213
# got the parents ok
1214
for parent in parents:
1215
if parent not in graph:
1217
graph[version] = parents
1218
return topo_sort(graph.items())
1220
def num_versions(self):
1221
return len(self._history)
1223
__len__ = num_versions
1225
def get_versions(self):
1226
return self._history
1228
def idx_to_name(self, idx):
1229
return self._history[idx]
1231
def lookup(self, version_id):
1232
assert version_id in self._cache
1233
return self._cache[version_id][5]
1235
def _version_list_to_index(self, versions):
1236
encode_utf8 = cache_utf8.encode
1238
for version in versions:
1239
if version in self._cache:
1240
# -- inlined lookup() --
1241
result_list.append(str(self._cache[version][5]))
1242
# -- end lookup () --
1244
result_list.append('.' + encode_utf8(version))
1245
return ' '.join(result_list)
1247
def add_version(self, version_id, options, pos, size, parents):
1248
"""Add a version record to the index."""
1249
self.add_versions(((version_id, options, pos, size, parents),))
1251
def add_versions(self, versions):
1252
"""Add multiple versions to the index.
1254
:param versions: a list of tuples:
1255
(version_id, options, pos, size, parents).
1258
encode_utf8 = cache_utf8.encode
1259
orig_history = self._history[:]
1260
orig_cache = self._cache.copy()
1263
for version_id, options, pos, size, parents in versions:
1264
line = "\n%s %s %s %s %s :" % (encode_utf8(version_id),
1268
self._version_list_to_index(parents))
1269
assert isinstance(line, str), \
1270
'content must be utf-8 encoded: %r' % (line,)
1272
self._cache_version(version_id, options, pos, size, parents)
1273
if not self._need_to_create:
1274
self._transport.append_bytes(self._filename, ''.join(lines))
1277
sio.write(self.HEADER)
1278
sio.writelines(lines)
1280
self._transport.put_file_non_atomic(self._filename, sio,
1281
create_parent_dir=self._create_parent_dir,
1282
mode=self._file_mode,
1283
dir_mode=self._dir_mode)
1284
self._need_to_create = False
1286
# If any problems happen, restore the original values and re-raise
1287
self._history = orig_history
1288
self._cache = orig_cache
1291
def has_version(self, version_id):
1292
"""True if the version is in the index."""
1293
return (version_id in self._cache)
1295
def get_position(self, version_id):
1296
"""Return data position and size of specified version."""
1297
return (self._cache[version_id][2], \
1298
self._cache[version_id][3])
1300
def get_method(self, version_id):
1301
"""Return compression method of specified version."""
1302
options = self._cache[version_id][1]
1303
if 'fulltext' in options:
1306
assert 'line-delta' in options
1309
def get_options(self, version_id):
1310
return self._cache[version_id][1]
1312
def get_parents(self, version_id):
1313
"""Return parents of specified version ignoring ghosts."""
1314
return [parent for parent in self._cache[version_id][4]
1315
if parent in self._cache]
1317
def get_parents_with_ghosts(self, version_id):
1318
"""Return parents of specified version with ghosts."""
1319
return self._cache[version_id][4]
1321
def check_versions_present(self, version_ids):
1322
"""Check that all specified versions are present."""
1323
version_ids = set(version_ids)
1324
for version_id in list(version_ids):
1325
if version_id in self._cache:
1326
version_ids.remove(version_id)
1328
raise RevisionNotPresent(list(version_ids)[0], self.filename)
1331
class _KnitData(_KnitComponentFile):
1332
"""Contents of the knit data file"""
1334
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1335
create_parent_dir=False, delay_create=False,
1337
_KnitComponentFile.__init__(self, transport, filename, mode,
1338
file_mode=file_mode,
1339
create_parent_dir=create_parent_dir,
1341
self._checked = False
1342
# TODO: jam 20060713 conceptually, this could spill to disk
1343
# if the cached size gets larger than a certain amount
1344
# but it complicates the model a bit, so for now just use
1345
# a simple dictionary
1347
self._do_cache = False
1350
self._need_to_create = create
1352
self._transport.put_bytes_non_atomic(self._filename, '',
1353
mode=self._file_mode)
1355
def enable_cache(self):
1356
"""Enable caching of reads."""
1357
self._do_cache = True
1359
def clear_cache(self):
1360
"""Clear the record cache."""
1361
self._do_cache = False
1364
def _open_file(self):
1366
return self._transport.get(self._filename)
1371
def _record_to_data(self, version_id, digest, lines):
1372
"""Convert version_id, digest, lines into a raw data block.
1374
:return: (len, a StringIO instance with the raw data ready to read.)
1377
data_file = GzipFile(None, mode='wb', fileobj=sio)
1379
version_id_utf8 = cache_utf8.encode(version_id)
1380
data_file.writelines(chain(
1381
["version %s %d %s\n" % (version_id_utf8,
1385
["end %s\n" % version_id_utf8]))
1392
def add_raw_record(self, raw_data):
1393
"""Append a prepared record to the data file.
1395
:return: the offset in the data file raw_data was written.
1397
assert isinstance(raw_data, str), 'data must be plain bytes'
1398
if not self._need_to_create:
1399
return self._transport.append_bytes(self._filename, raw_data)
1401
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1402
create_parent_dir=self._create_parent_dir,
1403
mode=self._file_mode,
1404
dir_mode=self._dir_mode)
1405
self._need_to_create = False
1408
def add_record(self, version_id, digest, lines):
1409
"""Write new text record to disk. Returns the position in the
1410
file where it was written."""
1411
size, sio = self._record_to_data(version_id, digest, lines)
1413
if not self._need_to_create:
1414
start_pos = self._transport.append_file(self._filename, sio)
1416
self._transport.put_file_non_atomic(self._filename, sio,
1417
create_parent_dir=self._create_parent_dir,
1418
mode=self._file_mode,
1419
dir_mode=self._dir_mode)
1420
self._need_to_create = False
1423
self._cache[version_id] = sio.getvalue()
1424
return start_pos, size
1426
def _parse_record_header(self, version_id, raw_data):
1427
"""Parse a record header for consistency.
1429
:return: the header and the decompressor stream.
1430
as (stream, header_record)
1432
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1433
rec = df.readline().split()
1435
raise KnitCorrupt(self._filename, 'unexpected number of elements in record header')
1436
if cache_utf8.decode(rec[1]) != version_id:
1437
raise KnitCorrupt(self._filename,
1438
'unexpected version, wanted %r, got %r' % (
1439
version_id, rec[1]))
1442
def _parse_record(self, version_id, data):
1444
# 4168 calls in 2880 217 internal
1445
# 4168 calls to _parse_record_header in 2121
1446
# 4168 calls to readlines in 330
1447
df, rec = self._parse_record_header(version_id, data)
1448
record_contents = df.readlines()
1449
l = record_contents.pop()
1450
assert len(record_contents) == int(rec[2])
1451
if l != 'end %s\n' % cache_utf8.encode(version_id):
1452
raise KnitCorrupt(self._filename, 'unexpected version end line %r, wanted %r'
1455
return record_contents, rec[3]
1457
def read_records_iter_raw(self, records):
1458
"""Read text records from data file and yield raw data.
1460
This unpacks enough of the text record to validate the id is
1461
as expected but thats all.
1463
# setup an iterator of the external records:
1464
# uses readv so nice and fast we hope.
1466
# grab the disk data needed.
1468
# Don't check _cache if it is empty
1469
needed_offsets = [(pos, size) for version_id, pos, size
1471
if version_id not in self._cache]
1473
needed_offsets = [(pos, size) for version_id, pos, size
1476
raw_records = self._transport.readv(self._filename, needed_offsets)
1479
for version_id, pos, size in records:
1480
if version_id in self._cache:
1481
# This data has already been validated
1482
data = self._cache[version_id]
1484
pos, data = raw_records.next()
1486
self._cache[version_id] = data
1488
# validate the header
1489
df, rec = self._parse_record_header(version_id, data)
1491
yield version_id, data
1493
def read_records_iter(self, records):
1494
"""Read text records from data file and yield result.
1496
The result will be returned in whatever is the fastest to read.
1497
Not by the order requested. Also, multiple requests for the same
1498
record will only yield 1 response.
1499
:param records: A list of (version_id, pos, len) entries
1500
:return: Yields (version_id, contents, digest) in the order
1501
read, not the order requested
1507
# Skip records we have alread seen
1508
yielded_records = set()
1509
needed_records = set()
1510
for record in records:
1511
if record[0] in self._cache:
1512
if record[0] in yielded_records:
1514
yielded_records.add(record[0])
1515
data = self._cache[record[0]]
1516
content, digest = self._parse_record(record[0], data)
1517
yield (record[0], content, digest)
1519
needed_records.add(record)
1520
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1522
needed_records = sorted(set(records), key=operator.itemgetter(1))
1524
if not needed_records:
1527
# The transport optimizes the fetching as well
1528
# (ie, reads continuous ranges.)
1529
readv_response = self._transport.readv(self._filename,
1530
[(pos, size) for version_id, pos, size in needed_records])
1532
for (version_id, pos, size), (pos, data) in \
1533
izip(iter(needed_records), readv_response):
1534
content, digest = self._parse_record(version_id, data)
1536
self._cache[version_id] = data
1537
yield version_id, content, digest
1539
def read_records(self, records):
1540
"""Read records into a dictionary."""
1542
for record_id, content, digest in \
1543
self.read_records_iter(records):
1544
components[record_id] = (content, digest)
1548
class InterKnit(InterVersionedFile):
1549
"""Optimised code paths for knit to knit operations."""
1551
_matching_file_from_factory = KnitVersionedFile
1552
_matching_file_to_factory = KnitVersionedFile
1555
def is_compatible(source, target):
1556
"""Be compatible with knits. """
1558
return (isinstance(source, KnitVersionedFile) and
1559
isinstance(target, KnitVersionedFile))
1560
except AttributeError:
1563
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1564
"""See InterVersionedFile.join."""
1565
assert isinstance(self.source, KnitVersionedFile)
1566
assert isinstance(self.target, KnitVersionedFile)
1568
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1573
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1575
version_ids = list(version_ids)
1576
if None in version_ids:
1577
version_ids.remove(None)
1579
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1580
this_versions = set(self.target._index.get_versions())
1581
needed_versions = self.source_ancestry - this_versions
1582
cross_check_versions = self.source_ancestry.intersection(this_versions)
1583
mismatched_versions = set()
1584
for version in cross_check_versions:
1585
# scan to include needed parents.
1586
n1 = set(self.target.get_parents_with_ghosts(version))
1587
n2 = set(self.source.get_parents_with_ghosts(version))
1589
# FIXME TEST this check for cycles being introduced works
1590
# the logic is we have a cycle if in our graph we are an
1591
# ancestor of any of the n2 revisions.
1597
parent_ancestors = self.source.get_ancestry(parent)
1598
if version in parent_ancestors:
1599
raise errors.GraphCycleError([parent, version])
1600
# ensure this parent will be available later.
1601
new_parents = n2.difference(n1)
1602
needed_versions.update(new_parents.difference(this_versions))
1603
mismatched_versions.add(version)
1605
if not needed_versions and not mismatched_versions:
1607
full_list = topo_sort(self.source.get_graph())
1609
version_list = [i for i in full_list if (not self.target.has_version(i)
1610
and i in needed_versions)]
1614
copy_queue_records = []
1616
for version_id in version_list:
1617
options = self.source._index.get_options(version_id)
1618
parents = self.source._index.get_parents_with_ghosts(version_id)
1619
# check that its will be a consistent copy:
1620
for parent in parents:
1621
# if source has the parent, we must :
1622
# * already have it or
1623
# * have it scheduled already
1624
# otherwise we don't care
1625
assert (self.target.has_version(parent) or
1626
parent in copy_set or
1627
not self.source.has_version(parent))
1628
data_pos, data_size = self.source._index.get_position(version_id)
1629
copy_queue_records.append((version_id, data_pos, data_size))
1630
copy_queue.append((version_id, options, parents))
1631
copy_set.add(version_id)
1633
# data suck the join:
1635
total = len(version_list)
1638
for (version_id, raw_data), \
1639
(version_id2, options, parents) in \
1640
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1642
assert version_id == version_id2, 'logic error, inconsistent results'
1644
pb.update("Joining knit", count, total)
1645
raw_records.append((version_id, options, parents, len(raw_data)))
1646
raw_datum.append(raw_data)
1647
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1649
for version in mismatched_versions:
1650
# FIXME RBC 20060309 is this needed?
1651
n1 = set(self.target.get_parents_with_ghosts(version))
1652
n2 = set(self.source.get_parents_with_ghosts(version))
1653
# write a combined record to our history preserving the current
1654
# parents as first in the list
1655
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1656
self.target.fix_parents(version, new_parents)
1662
InterVersionedFile.register_optimiser(InterKnit)
1665
class WeaveToKnit(InterVersionedFile):
1666
"""Optimised code paths for weave to knit operations."""
1668
_matching_file_from_factory = bzrlib.weave.WeaveFile
1669
_matching_file_to_factory = KnitVersionedFile
1672
def is_compatible(source, target):
1673
"""Be compatible with weaves to knits."""
1675
return (isinstance(source, bzrlib.weave.Weave) and
1676
isinstance(target, KnitVersionedFile))
1677
except AttributeError:
1680
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1681
"""See InterVersionedFile.join."""
1682
assert isinstance(self.source, bzrlib.weave.Weave)
1683
assert isinstance(self.target, KnitVersionedFile)
1685
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1690
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1692
version_ids = list(version_ids)
1694
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1695
this_versions = set(self.target._index.get_versions())
1696
needed_versions = self.source_ancestry - this_versions
1697
cross_check_versions = self.source_ancestry.intersection(this_versions)
1698
mismatched_versions = set()
1699
for version in cross_check_versions:
1700
# scan to include needed parents.
1701
n1 = set(self.target.get_parents_with_ghosts(version))
1702
n2 = set(self.source.get_parents(version))
1703
# if all of n2's parents are in n1, then its fine.
1704
if n2.difference(n1):
1705
# FIXME TEST this check for cycles being introduced works
1706
# the logic is we have a cycle if in our graph we are an
1707
# ancestor of any of the n2 revisions.
1713
parent_ancestors = self.source.get_ancestry(parent)
1714
if version in parent_ancestors:
1715
raise errors.GraphCycleError([parent, version])
1716
# ensure this parent will be available later.
1717
new_parents = n2.difference(n1)
1718
needed_versions.update(new_parents.difference(this_versions))
1719
mismatched_versions.add(version)
1721
if not needed_versions and not mismatched_versions:
1723
full_list = topo_sort(self.source.get_graph())
1725
version_list = [i for i in full_list if (not self.target.has_version(i)
1726
and i in needed_versions)]
1730
total = len(version_list)
1731
for version_id in version_list:
1732
pb.update("Converting to knit", count, total)
1733
parents = self.source.get_parents(version_id)
1734
# check that its will be a consistent copy:
1735
for parent in parents:
1736
# if source has the parent, we must already have it
1737
assert (self.target.has_version(parent))
1738
self.target.add_lines(
1739
version_id, parents, self.source.get_lines(version_id))
1742
for version in mismatched_versions:
1743
# FIXME RBC 20060309 is this needed?
1744
n1 = set(self.target.get_parents_with_ghosts(version))
1745
n2 = set(self.source.get_parents(version))
1746
# write a combined record to our history preserving the current
1747
# parents as first in the list
1748
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1749
self.target.fix_parents(version, new_parents)
1755
InterVersionedFile.register_optimiser(WeaveToKnit)
1758
class KnitSequenceMatcher(difflib.SequenceMatcher):
1759
"""Knit tuned sequence matcher.
1761
This is based on profiling of difflib which indicated some improvements
1762
for our usage pattern.
1765
def find_longest_match(self, alo, ahi, blo, bhi):
1766
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1768
If isjunk is not defined:
1770
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1771
alo <= i <= i+k <= ahi
1772
blo <= j <= j+k <= bhi
1773
and for all (i',j',k') meeting those conditions,
1776
and if i == i', j <= j'
1778
In other words, of all maximal matching blocks, return one that
1779
starts earliest in a, and of all those maximal matching blocks that
1780
start earliest in a, return the one that starts earliest in b.
1782
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1783
>>> s.find_longest_match(0, 5, 0, 9)
1786
If isjunk is defined, first the longest matching block is
1787
determined as above, but with the additional restriction that no
1788
junk element appears in the block. Then that block is extended as
1789
far as possible by matching (only) junk elements on both sides. So
1790
the resulting block never matches on junk except as identical junk
1791
happens to be adjacent to an "interesting" match.
1793
Here's the same example as before, but considering blanks to be
1794
junk. That prevents " abcd" from matching the " abcd" at the tail
1795
end of the second sequence directly. Instead only the "abcd" can
1796
match, and matches the leftmost "abcd" in the second sequence:
1798
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1799
>>> s.find_longest_match(0, 5, 0, 9)
1802
If no blocks match, return (alo, blo, 0).
1804
>>> s = SequenceMatcher(None, "ab", "c")
1805
>>> s.find_longest_match(0, 2, 0, 1)
1809
# CAUTION: stripping common prefix or suffix would be incorrect.
1813
# Longest matching block is "ab", but if common prefix is
1814
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1815
# strip, so ends up claiming that ab is changed to acab by
1816
# inserting "ca" in the middle. That's minimal but unintuitive:
1817
# "it's obvious" that someone inserted "ac" at the front.
1818
# Windiff ends up at the same place as diff, but by pairing up
1819
# the unique 'b's and then matching the first two 'a's.
1821
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1822
besti, bestj, bestsize = alo, blo, 0
1823
# find longest junk-free match
1824
# during an iteration of the loop, j2len[j] = length of longest
1825
# junk-free match ending with a[i-1] and b[j]
1829
for i in xrange(alo, ahi):
1830
# look at all instances of a[i] in b; note that because
1831
# b2j has no junk keys, the loop is skipped if a[i] is junk
1832
j2lenget = j2len.get
1835
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1836
# following improvement
1837
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1838
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1839
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1841
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1842
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1843
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
1855
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1857
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1860
# Extend the best by non-junk elements on each end. In particular,
1861
# "popular" non-junk elements aren't in b2j, which greatly speeds
1862
# the inner loop above, but also means "the best" match so far
1863
# doesn't contain any junk *or* popular non-junk elements.
1864
while besti > alo and bestj > blo and \
1865
not isbjunk(b[bestj-1]) and \
1866
a[besti-1] == b[bestj-1]:
1867
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1868
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1869
not isbjunk(b[bestj+bestsize]) and \
1870
a[besti+bestsize] == b[bestj+bestsize]:
1873
# Now that we have a wholly interesting match (albeit possibly
1874
# empty!), we may as well suck up the matching junk on each
1875
# side of it too. Can't think of a good reason not to, and it
1876
# saves post-processing the (possibly considerable) expense of
1877
# figuring out what to do with it. In the case of an empty
1878
# interesting match, this is clearly the right thing to do,
1879
# because no other kind of match is possible in the regions.
1880
while besti > alo and bestj > blo and \
1881
isbjunk(b[bestj-1]) and \
1882
a[besti-1] == b[bestj-1]:
1883
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1884
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1885
isbjunk(b[bestj+bestsize]) and \
1886
a[besti+bestsize] == b[bestj+bestsize]:
1887
bestsize = bestsize + 1
1889
return besti, bestj, bestsize
added
removed
bzrlib/knit.py
