1
# Copyright (C) 2005, 2006, 2007 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
"""Knit versionedfile implementation.
19
A knit is a versioned file implementation that supports efficient append only
23
lifeless: the data file is made up of "delta records". each delta record has a delta header
24
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
25
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
26
end-marker; simply "end VERSION"
28
delta can be line or full contents.a
29
... the 8's there are the index number of the annotation.
30
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
34
8 e.set('executable', 'yes')
36
8 if elt.get('executable') == 'yes':
37
8 ie.executable = True
38
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
42
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
43
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
44
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
45
09:33 < lifeless> right
46
09:33 < jrydberg> lifeless: the position and size is the range in the data file
49
so the index sequence is the dictionary compressed sequence number used
50
in the deltas to provide line annotation
55
# 10:16 < lifeless> make partial index writes safe
56
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
57
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
59
# move sha1 out of the content so that join is faster at verifying parents
60
# record content length ?
64
from cStringIO import StringIO
66
from itertools import izip, chain
82
from bzrlib.errors import (
90
RevisionAlreadyPresent,
92
from bzrlib.tuned_gzip import GzipFile
93
from bzrlib.trace import mutter
94
from bzrlib.osutils import (
99
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
100
from bzrlib.tsort import topo_sort
103
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
106
# TODO: Split out code specific to this format into an associated object.
108
# TODO: Can we put in some kind of value to check that the index and data
109
# files belong together?
111
# TODO: accommodate binaries, perhaps by storing a byte count
113
# TODO: function to check whole file
115
# TODO: atomically append data, then measure backwards from the cursor
116
# position after writing to work out where it was located. we may need to
117
# bypass python file buffering.
119
DATA_SUFFIX = '.knit'
120
INDEX_SUFFIX = '.kndx'
123
class KnitContent(object):
124
"""Content of a knit version to which deltas can be applied."""
126
def __init__(self, lines):
129
def annotate_iter(self):
130
"""Yield tuples of (origin, text) for each content line."""
131
return iter(self._lines)
134
"""Return a list of (origin, text) tuples."""
135
return list(self.annotate_iter())
137
def line_delta_iter(self, new_lines):
138
"""Generate line-based delta from this content to new_lines."""
139
new_texts = new_lines.text()
140
old_texts = self.text()
141
s = KnitSequenceMatcher(None, old_texts, new_texts)
142
for tag, i1, i2, j1, j2 in s.get_opcodes():
145
# ofrom, oto, length, data
146
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
148
def line_delta(self, new_lines):
149
return list(self.line_delta_iter(new_lines))
152
return [text for origin, text in self._lines]
155
return KnitContent(self._lines[:])
158
class _KnitFactory(object):
159
"""Base factory for creating content objects."""
161
def make(self, lines, version_id):
162
num_lines = len(lines)
163
return KnitContent(zip([version_id] * num_lines, lines))
166
class KnitAnnotateFactory(_KnitFactory):
167
"""Factory for creating annotated Content objects."""
171
def parse_fulltext(self, content, version_id):
172
"""Convert fulltext to internal representation
174
fulltext content is of the format
175
revid(utf8) plaintext\n
176
internal representation is of the format:
179
# TODO: jam 20070209 The tests expect this to be returned as tuples,
180
# but the code itself doesn't really depend on that.
181
# Figure out a way to not require the overhead of turning the
182
# list back into tuples.
183
lines = [tuple(line.split(' ', 1)) for line in content]
184
return KnitContent(lines)
186
def parse_line_delta_iter(self, lines):
187
return iter(self.parse_line_delta(lines))
189
def parse_line_delta(self, lines, version_id):
190
"""Convert a line based delta into internal representation.
192
line delta is in the form of:
193
intstart intend intcount
195
revid(utf8) newline\n
196
internal representation is
197
(start, end, count, [1..count tuples (revid, newline)])
204
def cache_and_return(line):
205
origin, text = line.split(' ', 1)
206
return cache.setdefault(origin, origin), text
208
# walk through the lines parsing.
210
start, end, count = [int(n) for n in header.split(',')]
211
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
212
result.append((start, end, count, contents))
215
def get_fulltext_content(self, lines):
216
"""Extract just the content lines from a fulltext."""
217
return (line.split(' ', 1)[1] for line in lines)
219
def get_linedelta_content(self, lines):
220
"""Extract just the content from a line delta.
222
This doesn't return all of the extra information stored in a delta.
223
Only the actual content lines.
228
header = header.split(',')
229
count = int(header[2])
230
for i in xrange(count):
231
origin, text = next().split(' ', 1)
234
def lower_fulltext(self, content):
235
"""convert a fulltext content record into a serializable form.
237
see parse_fulltext which this inverts.
239
# TODO: jam 20070209 We only do the caching thing to make sure that
240
# the origin is a valid utf-8 line, eventually we could remove it
241
return ['%s %s' % (o, t) for o, t in content._lines]
243
def lower_line_delta(self, delta):
244
"""convert a delta into a serializable form.
246
See parse_line_delta which this inverts.
248
# TODO: jam 20070209 We only do the caching thing to make sure that
249
# the origin is a valid utf-8 line, eventually we could remove it
251
for start, end, c, lines in delta:
252
out.append('%d,%d,%d\n' % (start, end, c))
253
out.extend(origin + ' ' + text
254
for origin, text in lines)
258
class KnitPlainFactory(_KnitFactory):
259
"""Factory for creating plain Content objects."""
263
def parse_fulltext(self, content, version_id):
264
"""This parses an unannotated fulltext.
266
Note that this is not a noop - the internal representation
267
has (versionid, line) - its just a constant versionid.
269
return self.make(content, version_id)
271
def parse_line_delta_iter(self, lines, version_id):
273
num_lines = len(lines)
274
while cur < num_lines:
277
start, end, c = [int(n) for n in header.split(',')]
278
yield start, end, c, zip([version_id] * c, lines[cur:cur+c])
281
def parse_line_delta(self, lines, version_id):
282
return list(self.parse_line_delta_iter(lines, version_id))
284
def get_fulltext_content(self, lines):
285
"""Extract just the content lines from a fulltext."""
288
def get_linedelta_content(self, lines):
289
"""Extract just the content from a line delta.
291
This doesn't return all of the extra information stored in a delta.
292
Only the actual content lines.
297
header = header.split(',')
298
count = int(header[2])
299
for i in xrange(count):
302
def lower_fulltext(self, content):
303
return content.text()
305
def lower_line_delta(self, delta):
307
for start, end, c, lines in delta:
308
out.append('%d,%d,%d\n' % (start, end, c))
309
out.extend([text for origin, text in lines])
313
def make_empty_knit(transport, relpath):
314
"""Construct a empty knit at the specified location."""
315
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
319
class KnitVersionedFile(VersionedFile):
320
"""Weave-like structure with faster random access.
322
A knit stores a number of texts and a summary of the relationships
323
between them. Texts are identified by a string version-id. Texts
324
are normally stored and retrieved as a series of lines, but can
325
also be passed as single strings.
327
Lines are stored with the trailing newline (if any) included, to
328
avoid special cases for files with no final newline. Lines are
329
composed of 8-bit characters, not unicode. The combination of
330
these approaches should mean any 'binary' file can be safely
331
stored and retrieved.
334
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
335
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
336
create=False, create_parent_dir=False, delay_create=False,
338
"""Construct a knit at location specified by relpath.
340
:param create: If not True, only open an existing knit.
341
:param create_parent_dir: If True, create the parent directory if
342
creating the file fails. (This is used for stores with
343
hash-prefixes that may not exist yet)
344
:param delay_create: The calling code is aware that the knit won't
345
actually be created until the first data is stored.
347
if deprecated_passed(basis_knit):
348
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
349
" deprecated as of bzr 0.9.",
350
DeprecationWarning, stacklevel=2)
351
if access_mode is None:
353
super(KnitVersionedFile, self).__init__(access_mode)
354
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
355
self.transport = transport
356
self.filename = relpath
357
self.factory = factory or KnitAnnotateFactory()
358
self.writable = (access_mode == 'w')
361
self._max_delta_chain = 200
363
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
364
access_mode, create=create, file_mode=file_mode,
365
create_parent_dir=create_parent_dir, delay_create=delay_create,
367
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
368
access_mode, create=create and not len(self), file_mode=file_mode,
369
create_parent_dir=create_parent_dir, delay_create=delay_create,
373
return '%s(%s)' % (self.__class__.__name__,
374
self.transport.abspath(self.filename))
376
def _check_should_delta(self, first_parents):
377
"""Iterate back through the parent listing, looking for a fulltext.
379
This is used when we want to decide whether to add a delta or a new
380
fulltext. It searches for _max_delta_chain parents. When it finds a
381
fulltext parent, it sees if the total size of the deltas leading up to
382
it is large enough to indicate that we want a new full text anyway.
384
Return True if we should create a new delta, False if we should use a
389
delta_parents = first_parents
390
for count in xrange(self._max_delta_chain):
391
parent = delta_parents[0]
392
method = self._index.get_method(parent)
393
pos, size = self._index.get_position(parent)
394
if method == 'fulltext':
398
delta_parents = self._index.get_parents(parent)
400
# We couldn't find a fulltext, so we must create a new one
403
return fulltext_size > delta_size
405
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
406
"""See VersionedFile._add_delta()."""
407
self._check_add(version_id, []) # should we check the lines ?
408
self._check_versions_present(parents)
412
for parent in parents:
413
if not self.has_version(parent):
414
ghosts.append(parent)
416
present_parents.append(parent)
418
if delta_parent is None:
419
# reconstitute as full text.
420
assert len(delta) == 1 or len(delta) == 0
422
assert delta[0][0] == 0
423
assert delta[0][1] == 0, delta[0][1]
424
return super(KnitVersionedFile, self)._add_delta(version_id,
435
options.append('no-eol')
437
if delta_parent is not None:
438
# determine the current delta chain length.
439
# To speed the extract of texts the delta chain is limited
440
# to a fixed number of deltas. This should minimize both
441
# I/O and the time spend applying deltas.
442
# The window was changed to a maximum of 200 deltas, but also added
443
# was a check that the total compressed size of the deltas is
444
# smaller than the compressed size of the fulltext.
445
if not self._check_should_delta([delta_parent]):
446
# We don't want a delta here, just do a normal insertion.
447
return super(KnitVersionedFile, self)._add_delta(version_id,
454
options.append('line-delta')
455
store_lines = self.factory.lower_line_delta(delta)
457
where, size = self._data.add_record(version_id, digest, store_lines)
458
self._index.add_version(version_id, options, where, size, parents)
460
def _add_raw_records(self, records, data):
461
"""Add all the records 'records' with data pre-joined in 'data'.
463
:param records: A list of tuples(version_id, options, parents, size).
464
:param data: The data for the records. When it is written, the records
465
are adjusted to have pos pointing into data by the sum of
466
the preceding records sizes.
469
pos = self._data.add_raw_record(data)
472
for (version_id, options, parents, size) in records:
473
index_entries.append((version_id, options, pos+offset,
475
if self._data._do_cache:
476
self._data._cache[version_id] = data[offset:offset+size]
478
self._index.add_versions(index_entries)
480
def enable_cache(self):
481
"""Start caching data for this knit"""
482
self._data.enable_cache()
484
def clear_cache(self):
485
"""Clear the data cache only."""
486
self._data.clear_cache()
488
def copy_to(self, name, transport):
489
"""See VersionedFile.copy_to()."""
490
# copy the current index to a temp index to avoid racing with local
492
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
493
self.transport.get(self._index._filename))
495
f = self._data._open_file()
497
transport.put_file(name + DATA_SUFFIX, f)
500
# move the copied index into place
501
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
503
def create_empty(self, name, transport, mode=None):
504
return KnitVersionedFile(name, transport, factory=self.factory,
505
delta=self.delta, create=True)
507
def _fix_parents(self, version_id, new_parents):
508
"""Fix the parents list for version.
510
This is done by appending a new version to the index
511
with identical data except for the parents list.
512
the parents list must be a superset of the current
515
current_values = self._index._cache[version_id]
516
assert set(current_values[4]).difference(set(new_parents)) == set()
517
self._index.add_version(version_id,
523
def get_delta(self, version_id):
524
"""Get a delta for constructing version from some other version."""
525
version_id = osutils.safe_revision_id(version_id)
526
self.check_not_reserved_id(version_id)
527
if not self.has_version(version_id):
528
raise RevisionNotPresent(version_id, self.filename)
530
parents = self.get_parents(version_id)
535
data_pos, data_size = self._index.get_position(version_id)
536
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
537
noeol = 'no-eol' in self._index.get_options(version_id)
538
if 'fulltext' == self._index.get_method(version_id):
539
new_content = self.factory.parse_fulltext(data, version_id)
540
if parent is not None:
541
reference_content = self._get_content(parent)
542
old_texts = reference_content.text()
545
new_texts = new_content.text()
546
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
547
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
549
delta = self.factory.parse_line_delta(data, version_id)
550
return parent, sha1, noeol, delta
552
def get_graph_with_ghosts(self):
553
"""See VersionedFile.get_graph_with_ghosts()."""
554
graph_items = self._index.get_graph()
555
return dict(graph_items)
557
def get_sha1(self, version_id):
558
"""See VersionedFile.get_sha1()."""
559
version_id = osutils.safe_revision_id(version_id)
560
record_map = self._get_record_map([version_id])
561
method, content, digest, next = record_map[version_id]
566
"""See VersionedFile.get_suffixes()."""
567
return [DATA_SUFFIX, INDEX_SUFFIX]
569
def has_ghost(self, version_id):
570
"""True if there is a ghost reference in the file to version_id."""
571
version_id = osutils.safe_revision_id(version_id)
573
if self.has_version(version_id):
575
# optimisable if needed by memoising the _ghosts set.
576
items = self._index.get_graph()
577
for node, parents in items:
578
for parent in parents:
579
if parent not in self._index._cache:
580
if parent == version_id:
585
"""See VersionedFile.versions."""
586
return self._index.get_versions()
588
def has_version(self, version_id):
589
"""See VersionedFile.has_version."""
590
version_id = osutils.safe_revision_id(version_id)
591
return self._index.has_version(version_id)
593
__contains__ = has_version
595
def _merge_annotations(self, content, parents, parent_texts={},
596
delta=None, annotated=None):
597
"""Merge annotations for content. This is done by comparing
598
the annotations based on changed to the text.
602
for parent_id in parents:
603
merge_content = self._get_content(parent_id, parent_texts)
604
seq = patiencediff.PatienceSequenceMatcher(
605
None, merge_content.text(), content.text())
606
if delta_seq is None:
607
# setup a delta seq to reuse.
609
for i, j, n in seq.get_matching_blocks():
612
# this appears to copy (origin, text) pairs across to the new
613
# content for any line that matches the last-checked parent.
614
# FIXME: save the sequence control data for delta compression
615
# against the most relevant parent rather than rediffing.
616
content._lines[j:j+n] = merge_content._lines[i:i+n]
619
reference_content = self._get_content(parents[0], parent_texts)
620
new_texts = content.text()
621
old_texts = reference_content.text()
622
delta_seq = patiencediff.PatienceSequenceMatcher(
623
None, old_texts, new_texts)
624
return self._make_line_delta(delta_seq, content)
626
def _make_line_delta(self, delta_seq, new_content):
627
"""Generate a line delta from delta_seq and new_content."""
629
for op in delta_seq.get_opcodes():
632
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
635
def _get_components_positions(self, version_ids):
636
"""Produce a map of position data for the components of versions.
638
This data is intended to be used for retrieving the knit records.
640
A dict of version_id to (method, data_pos, data_size, next) is
642
method is the way referenced data should be applied.
643
data_pos is the position of the data in the knit.
644
data_size is the size of the data in the knit.
645
next is the build-parent of the version, or None for fulltexts.
648
for version_id in version_ids:
651
while cursor is not None and cursor not in component_data:
652
method = self._index.get_method(cursor)
653
if method == 'fulltext':
656
next = self.get_parents(cursor)[0]
657
data_pos, data_size = self._index.get_position(cursor)
658
component_data[cursor] = (method, data_pos, data_size, next)
660
return component_data
662
def _get_content(self, version_id, parent_texts={}):
663
"""Returns a content object that makes up the specified
665
if not self.has_version(version_id):
666
raise RevisionNotPresent(version_id, self.filename)
668
cached_version = parent_texts.get(version_id, None)
669
if cached_version is not None:
670
return cached_version
672
text_map, contents_map = self._get_content_maps([version_id])
673
return contents_map[version_id]
675
def _check_versions_present(self, version_ids):
676
"""Check that all specified versions are present."""
677
self._index.check_versions_present(version_ids)
679
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
680
"""See VersionedFile.add_lines_with_ghosts()."""
681
self._check_add(version_id, lines)
682
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
684
def _add_lines(self, version_id, parents, lines, parent_texts):
685
"""See VersionedFile.add_lines."""
686
self._check_add(version_id, lines)
687
self._check_versions_present(parents)
688
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
690
def _check_add(self, version_id, lines):
691
"""check that version_id and lines are safe to add."""
692
assert self.writable, "knit is not opened for write"
693
### FIXME escape. RBC 20060228
694
if contains_whitespace(version_id):
695
raise InvalidRevisionId(version_id, self.filename)
696
self.check_not_reserved_id(version_id)
697
if self.has_version(version_id):
698
raise RevisionAlreadyPresent(version_id, self.filename)
699
self._check_lines_not_unicode(lines)
700
self._check_lines_are_lines(lines)
702
def _add(self, version_id, lines, parents, delta, parent_texts):
703
"""Add a set of lines on top of version specified by parents.
705
If delta is true, compress the text as a line-delta against
708
Any versions not present will be converted into ghosts.
710
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
711
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
712
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
713
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
714
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
715
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
716
# +1383 0 8.0370 8.0370 +<len>
717
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
718
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
719
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
720
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
724
if parent_texts is None:
726
for parent in parents:
727
if not self.has_version(parent):
728
ghosts.append(parent)
730
present_parents.append(parent)
732
if delta and not len(present_parents):
735
digest = sha_strings(lines)
738
if lines[-1][-1] != '\n':
739
options.append('no-eol')
740
lines[-1] = lines[-1] + '\n'
742
if len(present_parents) and delta:
743
# To speed the extract of texts the delta chain is limited
744
# to a fixed number of deltas. This should minimize both
745
# I/O and the time spend applying deltas.
746
delta = self._check_should_delta(present_parents)
748
assert isinstance(version_id, str)
749
lines = self.factory.make(lines, version_id)
750
if delta or (self.factory.annotated and len(present_parents) > 0):
751
# Merge annotations from parent texts if so is needed.
752
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
753
delta, self.factory.annotated)
756
options.append('line-delta')
757
store_lines = self.factory.lower_line_delta(delta_hunks)
759
options.append('fulltext')
760
store_lines = self.factory.lower_fulltext(lines)
762
where, size = self._data.add_record(version_id, digest, store_lines)
763
self._index.add_version(version_id, options, where, size, parents)
766
def check(self, progress_bar=None):
767
"""See VersionedFile.check()."""
769
def _clone_text(self, new_version_id, old_version_id, parents):
770
"""See VersionedFile.clone_text()."""
771
# FIXME RBC 20060228 make fast by only inserting an index with null
773
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
775
def get_lines(self, version_id):
776
"""See VersionedFile.get_lines()."""
777
return self.get_line_list([version_id])[0]
779
def _get_record_map(self, version_ids):
780
"""Produce a dictionary of knit records.
782
The keys are version_ids, the values are tuples of (method, content,
784
method is the way the content should be applied.
785
content is a KnitContent object.
786
digest is the SHA1 digest of this version id after all steps are done
787
next is the build-parent of the version, i.e. the leftmost ancestor.
788
If the method is fulltext, next will be None.
790
position_map = self._get_components_positions(version_ids)
791
# c = component_id, m = method, p = position, s = size, n = next
792
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
794
for component_id, content, digest in \
795
self._data.read_records_iter(records):
796
method, position, size, next = position_map[component_id]
797
record_map[component_id] = method, content, digest, next
801
def get_text(self, version_id):
802
"""See VersionedFile.get_text"""
803
return self.get_texts([version_id])[0]
805
def get_texts(self, version_ids):
806
return [''.join(l) for l in self.get_line_list(version_ids)]
808
def get_line_list(self, version_ids):
809
"""Return the texts of listed versions as a list of strings."""
810
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
811
for version_id in version_ids:
812
self.check_not_reserved_id(version_id)
813
text_map, content_map = self._get_content_maps(version_ids)
814
return [text_map[v] for v in version_ids]
816
def _get_content_maps(self, version_ids):
817
"""Produce maps of text and KnitContents
819
:return: (text_map, content_map) where text_map contains the texts for
820
the requested versions and content_map contains the KnitContents.
821
Both dicts take version_ids as their keys.
823
for version_id in version_ids:
824
if not self.has_version(version_id):
825
raise RevisionNotPresent(version_id, self.filename)
826
record_map = self._get_record_map(version_ids)
831
for version_id in version_ids:
834
while cursor is not None:
835
method, data, digest, next = record_map[cursor]
836
components.append((cursor, method, data, digest))
837
if cursor in content_map:
842
for component_id, method, data, digest in reversed(components):
843
if component_id in content_map:
844
content = content_map[component_id]
846
if method == 'fulltext':
847
assert content is None
848
content = self.factory.parse_fulltext(data, version_id)
849
elif method == 'line-delta':
850
delta = self.factory.parse_line_delta(data, version_id)
851
content = content.copy()
852
content._lines = self._apply_delta(content._lines,
854
content_map[component_id] = content
856
if 'no-eol' in self._index.get_options(version_id):
857
content = content.copy()
858
line = content._lines[-1][1].rstrip('\n')
859
content._lines[-1] = (content._lines[-1][0], line)
860
final_content[version_id] = content
862
# digest here is the digest from the last applied component.
863
text = content.text()
864
if sha_strings(text) != digest:
865
raise KnitCorrupt(self.filename,
866
'sha-1 does not match %s' % version_id)
868
text_map[version_id] = text
869
return text_map, final_content
871
def iter_lines_added_or_present_in_versions(self, version_ids=None,
873
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
874
if version_ids is None:
875
version_ids = self.versions()
877
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
879
pb = progress.DummyProgress()
880
# we don't care about inclusions, the caller cares.
881
# but we need to setup a list of records to visit.
882
# we need version_id, position, length
883
version_id_records = []
884
requested_versions = set(version_ids)
885
# filter for available versions
886
for version_id in requested_versions:
887
if not self.has_version(version_id):
888
raise RevisionNotPresent(version_id, self.filename)
889
# get a in-component-order queue:
890
for version_id in self.versions():
891
if version_id in requested_versions:
892
data_pos, length = self._index.get_position(version_id)
893
version_id_records.append((version_id, data_pos, length))
895
total = len(version_id_records)
896
for version_idx, (version_id, data, sha_value) in \
897
enumerate(self._data.read_records_iter(version_id_records)):
898
pb.update('Walking content.', version_idx, total)
899
method = self._index.get_method(version_id)
901
assert method in ('fulltext', 'line-delta')
902
if method == 'fulltext':
903
line_iterator = self.factory.get_fulltext_content(data)
905
line_iterator = self.factory.get_linedelta_content(data)
906
for line in line_iterator:
909
pb.update('Walking content.', total, total)
911
def num_versions(self):
912
"""See VersionedFile.num_versions()."""
913
return self._index.num_versions()
915
__len__ = num_versions
917
def annotate_iter(self, version_id):
918
"""See VersionedFile.annotate_iter."""
919
version_id = osutils.safe_revision_id(version_id)
920
content = self._get_content(version_id)
921
for origin, text in content.annotate_iter():
924
def get_parents(self, version_id):
925
"""See VersionedFile.get_parents."""
928
# 52554 calls in 1264 872 internal down from 3674
929
version_id = osutils.safe_revision_id(version_id)
931
return self._index.get_parents(version_id)
933
raise RevisionNotPresent(version_id, self.filename)
935
def get_parents_with_ghosts(self, version_id):
936
"""See VersionedFile.get_parents."""
937
version_id = osutils.safe_revision_id(version_id)
939
return self._index.get_parents_with_ghosts(version_id)
941
raise RevisionNotPresent(version_id, self.filename)
943
def get_ancestry(self, versions, topo_sorted=True):
944
"""See VersionedFile.get_ancestry."""
945
if isinstance(versions, basestring):
946
versions = [versions]
949
versions = [osutils.safe_revision_id(v) for v in versions]
950
return self._index.get_ancestry(versions, topo_sorted)
952
def get_ancestry_with_ghosts(self, versions):
953
"""See VersionedFile.get_ancestry_with_ghosts."""
954
if isinstance(versions, basestring):
955
versions = [versions]
958
versions = [osutils.safe_revision_id(v) for v in versions]
959
return self._index.get_ancestry_with_ghosts(versions)
961
#@deprecated_method(zero_eight)
962
def walk(self, version_ids):
963
"""See VersionedFile.walk."""
964
# We take the short path here, and extract all relevant texts
965
# and put them in a weave and let that do all the work. Far
966
# from optimal, but is much simpler.
967
# FIXME RB 20060228 this really is inefficient!
968
from bzrlib.weave import Weave
970
w = Weave(self.filename)
971
ancestry = set(self.get_ancestry(version_ids, topo_sorted=False))
972
sorted_graph = topo_sort(self._index.get_graph())
973
version_list = [vid for vid in sorted_graph if vid in ancestry]
975
for version_id in version_list:
976
lines = self.get_lines(version_id)
977
w.add_lines(version_id, self.get_parents(version_id), lines)
979
for lineno, insert_id, dset, line in w.walk(version_ids):
980
yield lineno, insert_id, dset, line
982
def plan_merge(self, ver_a, ver_b):
983
"""See VersionedFile.plan_merge."""
984
ver_a = osutils.safe_revision_id(ver_a)
985
ver_b = osutils.safe_revision_id(ver_b)
986
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
988
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
989
annotated_a = self.annotate(ver_a)
990
annotated_b = self.annotate(ver_b)
991
return merge._plan_annotate_merge(annotated_a, annotated_b,
992
ancestors_a, ancestors_b)
995
class _KnitComponentFile(object):
996
"""One of the files used to implement a knit database"""
998
def __init__(self, transport, filename, mode, file_mode=None,
999
create_parent_dir=False, dir_mode=None):
1000
self._transport = transport
1001
self._filename = filename
1003
self._file_mode = file_mode
1004
self._dir_mode = dir_mode
1005
self._create_parent_dir = create_parent_dir
1006
self._need_to_create = False
1008
def _full_path(self):
1009
"""Return the full path to this file."""
1010
return self._transport.base + self._filename
1012
def check_header(self, fp):
1013
line = fp.readline()
1015
# An empty file can actually be treated as though the file doesn't
1017
raise errors.NoSuchFile(self._full_path())
1018
if line != self.HEADER:
1019
raise KnitHeaderError(badline=line,
1020
filename=self._transport.abspath(self._filename))
1023
"""Commit is a nop."""
1026
return '%s(%s)' % (self.__class__.__name__, self._filename)
1029
class _KnitIndex(_KnitComponentFile):
1030
"""Manages knit index file.
1032
The index is already kept in memory and read on startup, to enable
1033
fast lookups of revision information. The cursor of the index
1034
file is always pointing to the end, making it easy to append
1037
_cache is a cache for fast mapping from version id to a Index
1040
_history is a cache for fast mapping from indexes to version ids.
1042
The index data format is dictionary compressed when it comes to
1043
parent references; a index entry may only have parents that with a
1044
lover index number. As a result, the index is topological sorted.
1046
Duplicate entries may be written to the index for a single version id
1047
if this is done then the latter one completely replaces the former:
1048
this allows updates to correct version and parent information.
1049
Note that the two entries may share the delta, and that successive
1050
annotations and references MUST point to the first entry.
1052
The index file on disc contains a header, followed by one line per knit
1053
record. The same revision can be present in an index file more than once.
1054
The first occurrence gets assigned a sequence number starting from 0.
1056
The format of a single line is
1057
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1058
REVISION_ID is a utf8-encoded revision id
1059
FLAGS is a comma separated list of flags about the record. Values include
1060
no-eol, line-delta, fulltext.
1061
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1062
that the the compressed data starts at.
1063
LENGTH is the ascii representation of the length of the data file.
1064
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1066
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1067
revision id already in the knit that is a parent of REVISION_ID.
1068
The ' :' marker is the end of record marker.
1071
when a write is interrupted to the index file, it will result in a line
1072
that does not end in ' :'. If the ' :' is not present at the end of a line,
1073
or at the end of the file, then the record that is missing it will be
1074
ignored by the parser.
1076
When writing new records to the index file, the data is preceded by '\n'
1077
to ensure that records always start on new lines even if the last write was
1078
interrupted. As a result its normal for the last line in the index to be
1079
missing a trailing newline. One can be added with no harmful effects.
1082
HEADER = "# bzr knit index 8\n"
1084
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1085
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1087
def _cache_version(self, version_id, options, pos, size, parents):
1088
"""Cache a version record in the history array and index cache.
1090
This is inlined into _load_data for performance. KEEP IN SYNC.
1091
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1094
# only want the _history index to reference the 1st index entry
1096
if version_id not in self._cache:
1097
index = len(self._history)
1098
self._history.append(version_id)
1100
index = self._cache[version_id][5]
1101
self._cache[version_id] = (version_id,
1108
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1109
create_parent_dir=False, delay_create=False, dir_mode=None):
1110
_KnitComponentFile.__init__(self, transport, filename, mode,
1111
file_mode=file_mode,
1112
create_parent_dir=create_parent_dir,
1115
# position in _history is the 'official' index for a revision
1116
# but the values may have come from a newer entry.
1117
# so - wc -l of a knit index is != the number of unique names
1121
fp = self._transport.get(self._filename)
1123
# _load_data may raise NoSuchFile if the target knit is
1125
_load_data(self, fp)
1129
if mode != 'w' or not create:
1132
self._need_to_create = True
1134
self._transport.put_bytes_non_atomic(
1135
self._filename, self.HEADER, mode=self._file_mode)
1137
def get_graph(self):
1138
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1140
def get_ancestry(self, versions, topo_sorted=True):
1141
"""See VersionedFile.get_ancestry."""
1142
# get a graph of all the mentioned versions:
1144
pending = set(versions)
1147
version = pending.pop()
1150
parents = [p for p in cache[version][4] if p in cache]
1152
raise RevisionNotPresent(version, self._filename)
1153
# if not completed and not a ghost
1154
pending.update([p for p in parents if p not in graph])
1155
graph[version] = parents
1158
return topo_sort(graph.items())
1160
def get_ancestry_with_ghosts(self, versions):
1161
"""See VersionedFile.get_ancestry_with_ghosts."""
1162
# get a graph of all the mentioned versions:
1163
self.check_versions_present(versions)
1166
pending = set(versions)
1168
version = pending.pop()
1170
parents = cache[version][4]
1176
pending.update([p for p in parents if p not in graph])
1177
graph[version] = parents
1178
return topo_sort(graph.items())
1180
def num_versions(self):
1181
return len(self._history)
1183
__len__ = num_versions
1185
def get_versions(self):
1186
return self._history
1188
def idx_to_name(self, idx):
1189
return self._history[idx]
1191
def lookup(self, version_id):
1192
assert version_id in self._cache
1193
return self._cache[version_id][5]
1195
def _version_list_to_index(self, versions):
1198
for version in versions:
1199
if version in cache:
1200
# -- inlined lookup() --
1201
result_list.append(str(cache[version][5]))
1202
# -- end lookup () --
1204
result_list.append('.' + version)
1205
return ' '.join(result_list)
1207
def add_version(self, version_id, options, pos, size, parents):
1208
"""Add a version record to the index."""
1209
self.add_versions(((version_id, options, pos, size, parents),))
1211
def add_versions(self, versions):
1212
"""Add multiple versions to the index.
1214
:param versions: a list of tuples:
1215
(version_id, options, pos, size, parents).
1218
orig_history = self._history[:]
1219
orig_cache = self._cache.copy()
1222
for version_id, options, pos, size, parents in versions:
1223
line = "\n%s %s %s %s %s :" % (version_id,
1227
self._version_list_to_index(parents))
1228
assert isinstance(line, str), \
1229
'content must be utf-8 encoded: %r' % (line,)
1231
self._cache_version(version_id, options, pos, size, parents)
1232
if not self._need_to_create:
1233
self._transport.append_bytes(self._filename, ''.join(lines))
1236
sio.write(self.HEADER)
1237
sio.writelines(lines)
1239
self._transport.put_file_non_atomic(self._filename, sio,
1240
create_parent_dir=self._create_parent_dir,
1241
mode=self._file_mode,
1242
dir_mode=self._dir_mode)
1243
self._need_to_create = False
1245
# If any problems happen, restore the original values and re-raise
1246
self._history = orig_history
1247
self._cache = orig_cache
1250
def has_version(self, version_id):
1251
"""True if the version is in the index."""
1252
return version_id in self._cache
1254
def get_position(self, version_id):
1255
"""Return data position and size of specified version."""
1256
entry = self._cache[version_id]
1257
return entry[2], entry[3]
1259
def get_method(self, version_id):
1260
"""Return compression method of specified version."""
1261
options = self._cache[version_id][1]
1262
if 'fulltext' in options:
1265
if 'line-delta' not in options:
1266
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1269
def get_options(self, version_id):
1270
return self._cache[version_id][1]
1272
def get_parents(self, version_id):
1273
"""Return parents of specified version ignoring ghosts."""
1274
return [parent for parent in self._cache[version_id][4]
1275
if parent in self._cache]
1277
def get_parents_with_ghosts(self, version_id):
1278
"""Return parents of specified version with ghosts."""
1279
return self._cache[version_id][4]
1281
def check_versions_present(self, version_ids):
1282
"""Check that all specified versions are present."""
1284
for version_id in version_ids:
1285
if version_id not in cache:
1286
raise RevisionNotPresent(version_id, self._filename)
1289
class _KnitData(_KnitComponentFile):
1290
"""Contents of the knit data file"""
1292
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1293
create_parent_dir=False, delay_create=False,
1295
_KnitComponentFile.__init__(self, transport, filename, mode,
1296
file_mode=file_mode,
1297
create_parent_dir=create_parent_dir,
1299
self._checked = False
1300
# TODO: jam 20060713 conceptually, this could spill to disk
1301
# if the cached size gets larger than a certain amount
1302
# but it complicates the model a bit, so for now just use
1303
# a simple dictionary
1305
self._do_cache = False
1308
self._need_to_create = create
1310
self._transport.put_bytes_non_atomic(self._filename, '',
1311
mode=self._file_mode)
1313
def enable_cache(self):
1314
"""Enable caching of reads."""
1315
self._do_cache = True
1317
def clear_cache(self):
1318
"""Clear the record cache."""
1319
self._do_cache = False
1322
def _open_file(self):
1324
return self._transport.get(self._filename)
1329
def _record_to_data(self, version_id, digest, lines):
1330
"""Convert version_id, digest, lines into a raw data block.
1332
:return: (len, a StringIO instance with the raw data ready to read.)
1335
data_file = GzipFile(None, mode='wb', fileobj=sio)
1337
assert isinstance(version_id, str)
1338
data_file.writelines(chain(
1339
["version %s %d %s\n" % (version_id,
1343
["end %s\n" % version_id]))
1350
def add_raw_record(self, raw_data):
1351
"""Append a prepared record to the data file.
1353
:return: the offset in the data file raw_data was written.
1355
assert isinstance(raw_data, str), 'data must be plain bytes'
1356
if not self._need_to_create:
1357
return self._transport.append_bytes(self._filename, raw_data)
1359
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1360
create_parent_dir=self._create_parent_dir,
1361
mode=self._file_mode,
1362
dir_mode=self._dir_mode)
1363
self._need_to_create = False
1366
def add_record(self, version_id, digest, lines):
1367
"""Write new text record to disk. Returns the position in the
1368
file where it was written."""
1369
size, sio = self._record_to_data(version_id, digest, lines)
1371
if not self._need_to_create:
1372
start_pos = self._transport.append_file(self._filename, sio)
1374
self._transport.put_file_non_atomic(self._filename, sio,
1375
create_parent_dir=self._create_parent_dir,
1376
mode=self._file_mode,
1377
dir_mode=self._dir_mode)
1378
self._need_to_create = False
1381
self._cache[version_id] = sio.getvalue()
1382
return start_pos, size
1384
def _parse_record_header(self, version_id, raw_data):
1385
"""Parse a record header for consistency.
1387
:return: the header and the decompressor stream.
1388
as (stream, header_record)
1390
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1392
rec = self._check_header(version_id, df.readline())
1393
except Exception, e:
1394
raise KnitCorrupt(self._filename,
1395
"While reading {%s} got %s(%s)"
1396
% (version_id, e.__class__.__name__, str(e)))
1399
def _check_header(self, version_id, line):
1402
raise KnitCorrupt(self._filename,
1403
'unexpected number of elements in record header')
1404
if rec[1] != version_id:
1405
raise KnitCorrupt(self._filename,
1406
'unexpected version, wanted %r, got %r'
1407
% (version_id, rec[1]))
1410
def _parse_record(self, version_id, data):
1412
# 4168 calls in 2880 217 internal
1413
# 4168 calls to _parse_record_header in 2121
1414
# 4168 calls to readlines in 330
1415
df = GzipFile(mode='rb', fileobj=StringIO(data))
1418
record_contents = df.readlines()
1419
except Exception, e:
1420
raise KnitCorrupt(self._filename,
1421
"While reading {%s} got %s(%s)"
1422
% (version_id, e.__class__.__name__, str(e)))
1423
header = record_contents.pop(0)
1424
rec = self._check_header(version_id, header)
1426
last_line = record_contents.pop()
1427
if len(record_contents) != int(rec[2]):
1428
raise KnitCorrupt(self._filename,
1429
'incorrect number of lines %s != %s'
1431
% (len(record_contents), int(rec[2]),
1433
if last_line != 'end %s\n' % rec[1]:
1434
raise KnitCorrupt(self._filename,
1435
'unexpected version end line %r, wanted %r'
1436
% (last_line, version_id))
1438
return record_contents, rec[3]
1440
def read_records_iter_raw(self, records):
1441
"""Read text records from data file and yield raw data.
1443
This unpacks enough of the text record to validate the id is
1444
as expected but thats all.
1446
# setup an iterator of the external records:
1447
# uses readv so nice and fast we hope.
1449
# grab the disk data needed.
1451
# Don't check _cache if it is empty
1452
needed_offsets = [(pos, size) for version_id, pos, size
1454
if version_id not in self._cache]
1456
needed_offsets = [(pos, size) for version_id, pos, size
1459
raw_records = self._transport.readv(self._filename, needed_offsets)
1461
for version_id, pos, size in records:
1462
if version_id in self._cache:
1463
# This data has already been validated
1464
data = self._cache[version_id]
1466
pos, data = raw_records.next()
1468
self._cache[version_id] = data
1470
# validate the header
1471
df, rec = self._parse_record_header(version_id, data)
1473
yield version_id, data
1475
def read_records_iter(self, records):
1476
"""Read text records from data file and yield result.
1478
The result will be returned in whatever is the fastest to read.
1479
Not by the order requested. Also, multiple requests for the same
1480
record will only yield 1 response.
1481
:param records: A list of (version_id, pos, len) entries
1482
:return: Yields (version_id, contents, digest) in the order
1483
read, not the order requested
1489
# Skip records we have alread seen
1490
yielded_records = set()
1491
needed_records = set()
1492
for record in records:
1493
if record[0] in self._cache:
1494
if record[0] in yielded_records:
1496
yielded_records.add(record[0])
1497
data = self._cache[record[0]]
1498
content, digest = self._parse_record(record[0], data)
1499
yield (record[0], content, digest)
1501
needed_records.add(record)
1502
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1504
needed_records = sorted(set(records), key=operator.itemgetter(1))
1506
if not needed_records:
1509
# The transport optimizes the fetching as well
1510
# (ie, reads continuous ranges.)
1511
readv_response = self._transport.readv(self._filename,
1512
[(pos, size) for version_id, pos, size in needed_records])
1514
for (version_id, pos, size), (pos, data) in \
1515
izip(iter(needed_records), readv_response):
1516
content, digest = self._parse_record(version_id, data)
1518
self._cache[version_id] = data
1519
yield version_id, content, digest
1521
def read_records(self, records):
1522
"""Read records into a dictionary."""
1524
for record_id, content, digest in \
1525
self.read_records_iter(records):
1526
components[record_id] = (content, digest)
1530
class InterKnit(InterVersionedFile):
1531
"""Optimised code paths for knit to knit operations."""
1533
_matching_file_from_factory = KnitVersionedFile
1534
_matching_file_to_factory = KnitVersionedFile
1537
def is_compatible(source, target):
1538
"""Be compatible with knits. """
1540
return (isinstance(source, KnitVersionedFile) and
1541
isinstance(target, KnitVersionedFile))
1542
except AttributeError:
1545
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1546
"""See InterVersionedFile.join."""
1547
assert isinstance(self.source, KnitVersionedFile)
1548
assert isinstance(self.target, KnitVersionedFile)
1550
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1555
pb = ui.ui_factory.nested_progress_bar()
1557
version_ids = list(version_ids)
1558
if None in version_ids:
1559
version_ids.remove(None)
1561
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1562
this_versions = set(self.target._index.get_versions())
1563
needed_versions = self.source_ancestry - this_versions
1564
cross_check_versions = self.source_ancestry.intersection(this_versions)
1565
mismatched_versions = set()
1566
for version in cross_check_versions:
1567
# scan to include needed parents.
1568
n1 = set(self.target.get_parents_with_ghosts(version))
1569
n2 = set(self.source.get_parents_with_ghosts(version))
1571
# FIXME TEST this check for cycles being introduced works
1572
# the logic is we have a cycle if in our graph we are an
1573
# ancestor of any of the n2 revisions.
1579
parent_ancestors = self.source.get_ancestry(parent)
1580
if version in parent_ancestors:
1581
raise errors.GraphCycleError([parent, version])
1582
# ensure this parent will be available later.
1583
new_parents = n2.difference(n1)
1584
needed_versions.update(new_parents.difference(this_versions))
1585
mismatched_versions.add(version)
1587
if not needed_versions and not mismatched_versions:
1589
full_list = topo_sort(self.source.get_graph())
1591
version_list = [i for i in full_list if (not self.target.has_version(i)
1592
and i in needed_versions)]
1596
copy_queue_records = []
1598
for version_id in version_list:
1599
options = self.source._index.get_options(version_id)
1600
parents = self.source._index.get_parents_with_ghosts(version_id)
1601
# check that its will be a consistent copy:
1602
for parent in parents:
1603
# if source has the parent, we must :
1604
# * already have it or
1605
# * have it scheduled already
1606
# otherwise we don't care
1607
assert (self.target.has_version(parent) or
1608
parent in copy_set or
1609
not self.source.has_version(parent))
1610
data_pos, data_size = self.source._index.get_position(version_id)
1611
copy_queue_records.append((version_id, data_pos, data_size))
1612
copy_queue.append((version_id, options, parents))
1613
copy_set.add(version_id)
1615
# data suck the join:
1617
total = len(version_list)
1620
for (version_id, raw_data), \
1621
(version_id2, options, parents) in \
1622
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1624
assert version_id == version_id2, 'logic error, inconsistent results'
1626
pb.update("Joining knit", count, total)
1627
raw_records.append((version_id, options, parents, len(raw_data)))
1628
raw_datum.append(raw_data)
1629
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1631
for version in mismatched_versions:
1632
# FIXME RBC 20060309 is this needed?
1633
n1 = set(self.target.get_parents_with_ghosts(version))
1634
n2 = set(self.source.get_parents_with_ghosts(version))
1635
# write a combined record to our history preserving the current
1636
# parents as first in the list
1637
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1638
self.target.fix_parents(version, new_parents)
1644
InterVersionedFile.register_optimiser(InterKnit)
1647
class WeaveToKnit(InterVersionedFile):
1648
"""Optimised code paths for weave to knit operations."""
1650
_matching_file_from_factory = bzrlib.weave.WeaveFile
1651
_matching_file_to_factory = KnitVersionedFile
1654
def is_compatible(source, target):
1655
"""Be compatible with weaves to knits."""
1657
return (isinstance(source, bzrlib.weave.Weave) and
1658
isinstance(target, KnitVersionedFile))
1659
except AttributeError:
1662
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1663
"""See InterVersionedFile.join."""
1664
assert isinstance(self.source, bzrlib.weave.Weave)
1665
assert isinstance(self.target, KnitVersionedFile)
1667
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1672
pb = ui.ui_factory.nested_progress_bar()
1674
version_ids = list(version_ids)
1676
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1677
this_versions = set(self.target._index.get_versions())
1678
needed_versions = self.source_ancestry - this_versions
1679
cross_check_versions = self.source_ancestry.intersection(this_versions)
1680
mismatched_versions = set()
1681
for version in cross_check_versions:
1682
# scan to include needed parents.
1683
n1 = set(self.target.get_parents_with_ghosts(version))
1684
n2 = set(self.source.get_parents(version))
1685
# if all of n2's parents are in n1, then its fine.
1686
if n2.difference(n1):
1687
# FIXME TEST this check for cycles being introduced works
1688
# the logic is we have a cycle if in our graph we are an
1689
# ancestor of any of the n2 revisions.
1695
parent_ancestors = self.source.get_ancestry(parent)
1696
if version in parent_ancestors:
1697
raise errors.GraphCycleError([parent, version])
1698
# ensure this parent will be available later.
1699
new_parents = n2.difference(n1)
1700
needed_versions.update(new_parents.difference(this_versions))
1701
mismatched_versions.add(version)
1703
if not needed_versions and not mismatched_versions:
1705
full_list = topo_sort(self.source.get_graph())
1707
version_list = [i for i in full_list if (not self.target.has_version(i)
1708
and i in needed_versions)]
1712
total = len(version_list)
1713
for version_id in version_list:
1714
pb.update("Converting to knit", count, total)
1715
parents = self.source.get_parents(version_id)
1716
# check that its will be a consistent copy:
1717
for parent in parents:
1718
# if source has the parent, we must already have it
1719
assert (self.target.has_version(parent))
1720
self.target.add_lines(
1721
version_id, parents, self.source.get_lines(version_id))
1724
for version in mismatched_versions:
1725
# FIXME RBC 20060309 is this needed?
1726
n1 = set(self.target.get_parents_with_ghosts(version))
1727
n2 = set(self.source.get_parents(version))
1728
# write a combined record to our history preserving the current
1729
# parents as first in the list
1730
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1731
self.target.fix_parents(version, new_parents)
1737
InterVersionedFile.register_optimiser(WeaveToKnit)
1740
class KnitSequenceMatcher(difflib.SequenceMatcher):
1741
"""Knit tuned sequence matcher.
1743
This is based on profiling of difflib which indicated some improvements
1744
for our usage pattern.
1747
def find_longest_match(self, alo, ahi, blo, bhi):
1748
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1750
If isjunk is not defined:
1752
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1753
alo <= i <= i+k <= ahi
1754
blo <= j <= j+k <= bhi
1755
and for all (i',j',k') meeting those conditions,
1758
and if i == i', j <= j'
1760
In other words, of all maximal matching blocks, return one that
1761
starts earliest in a, and of all those maximal matching blocks that
1762
start earliest in a, return the one that starts earliest in b.
1764
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1765
>>> s.find_longest_match(0, 5, 0, 9)
1768
If isjunk is defined, first the longest matching block is
1769
determined as above, but with the additional restriction that no
1770
junk element appears in the block. Then that block is extended as
1771
far as possible by matching (only) junk elements on both sides. So
1772
the resulting block never matches on junk except as identical junk
1773
happens to be adjacent to an "interesting" match.
1775
Here's the same example as before, but considering blanks to be
1776
junk. That prevents " abcd" from matching the " abcd" at the tail
1777
end of the second sequence directly. Instead only the "abcd" can
1778
match, and matches the leftmost "abcd" in the second sequence:
1780
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1781
>>> s.find_longest_match(0, 5, 0, 9)
1784
If no blocks match, return (alo, blo, 0).
1786
>>> s = SequenceMatcher(None, "ab", "c")
1787
>>> s.find_longest_match(0, 2, 0, 1)
1791
# CAUTION: stripping common prefix or suffix would be incorrect.
1795
# Longest matching block is "ab", but if common prefix is
1796
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1797
# strip, so ends up claiming that ab is changed to acab by
1798
# inserting "ca" in the middle. That's minimal but unintuitive:
1799
# "it's obvious" that someone inserted "ac" at the front.
1800
# Windiff ends up at the same place as diff, but by pairing up
1801
# the unique 'b's and then matching the first two 'a's.
1803
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1804
besti, bestj, bestsize = alo, blo, 0
1805
# find longest junk-free match
1806
# during an iteration of the loop, j2len[j] = length of longest
1807
# junk-free match ending with a[i-1] and b[j]
1811
for i in xrange(alo, ahi):
1812
# look at all instances of a[i] in b; note that because
1813
# b2j has no junk keys, the loop is skipped if a[i] is junk
1814
j2lenget = j2len.get
1817
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1818
# following improvement
1819
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1820
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1821
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1823
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1824
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1825
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
1837
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1839
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1842
# Extend the best by non-junk elements on each end. In particular,
1843
# "popular" non-junk elements aren't in b2j, which greatly speeds
1844
# the inner loop above, but also means "the best" match so far
1845
# doesn't contain any junk *or* popular non-junk elements.
1846
while besti > alo and bestj > blo and \
1847
not isbjunk(b[bestj-1]) and \
1848
a[besti-1] == b[bestj-1]:
1849
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1850
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1851
not isbjunk(b[bestj+bestsize]) and \
1852
a[besti+bestsize] == b[bestj+bestsize]:
1855
# Now that we have a wholly interesting match (albeit possibly
1856
# empty!), we may as well suck up the matching junk on each
1857
# side of it too. Can't think of a good reason not to, and it
1858
# saves post-processing the (possibly considerable) expense of
1859
# figuring out what to do with it. In the case of an empty
1860
# interesting match, this is clearly the right thing to do,
1861
# because no other kind of match is possible in the regions.
1862
while besti > alo and bestj > blo and \
1863
isbjunk(b[bestj-1]) and \
1864
a[besti-1] == b[bestj-1]:
1865
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1866
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1867
isbjunk(b[bestj+bestsize]) and \
1868
a[besti+bestsize] == b[bestj+bestsize]:
1869
bestsize = bestsize + 1
1871
return besti, bestj, bestsize
1875
from bzrlib._knit_load_data_c import _load_data_c as _load_data
1877
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
